betaFeedback

WorksPhysics and Astronomy

Twister After Dark: Diagnosing a Damaging Nocturnal Tornado in Iloilo, Philippines (08 August 2026)

Abstract

At ~1430 UTC (2230 LST) on 08 August 2026, a significant tornado struck the Mandurriao and Jaro districts of Iloilo City, Philippines. Its parent supercell developed within the southwest monsoon enhanced by Typhoon Dolphin, under weak synoptic forcing, as a nocturnal low-level jet of ~40 kt near 1.5-3 km advected warm, moist marine air onshore. ERA5 depicted MUCAPE of ~2000 J kg−1^{-1}, negligible inhibition, an LCL below 300 m, and nearly purely streamwise low-level vorticity, but deep-layer shear (DLS, 0-6 km bulk wind difference) below 3 m s−1^{-1}. Hodographs from the newly operational Iloilo S-band dual-polarization radar (S-ILO), ~5-7 km from the storm, revealed low-level shear (LLS, 0-1 km bulk wind difference) of ~16 m s−1^{-1} and 0-1 km storm-relative helicity of 124-211 m2^2 s−2^{-2}. The storm exhibited a hook echo with a weak echo hole, a ZDRZ_{DR} arc and ring, an elevated 55-60 dBZ rear-flank core consistent with a descending reflectivity core, and a ~50 dBZ ring interpreted as a debris ball. A survey of 31 damage indicators documented a ~2.9 km long and up to ~340 m wide path rated IF2 on the International Fujita Scale. Processor smoothing of the radial velocity and a ρHVρ_{HV} threshold for rainfall estimation masked any tornadic vortex or debris signature, although a radial velocity simulation shows that a narrow vortex (core radius ~50-100 m) with a rotational velocity of ~35-45 m s−1^{-1} could have been concealed. This case shows that significant tornadoes can develop at night during the Philippine monsoon season and that operational radar quality control should be revised to retain severe-storm signatures.

Highlights

  • Documents a nocturnal IF2 tornado that struck urban Iloilo City within ∼7\sim7 km of an S-band dual-polarization radar.

  • The parent supercell formed in a typhoon-enhanced southwest monsoon with weak deep-layer shear (DLS) but strong, highly streamwise low-level shear (LLS).

  • A nocturnal low-level jet surge with warm, moist advection strengthened the LLS after sunset.

  • An elevated rear-flank reflectivity core, consistent with a descending reflectivity core, capped the weak echo hole and debris ball.

Introduction

Annually, severe convective storms such as tornadic supercell thunderstorms occur in the Philippine Archipelago, particularly along the Greater Metro Manila Region (GMMR), Western Visayas, and Cotabato provinces. These hazards are by-products of the combination of ample vertical wind shear and atmospheric instability, including a lifting mechanism (i.e., forcing) either from plain-to-mountain flow or lowering of the low- and mid-level geopotential heights (e.g., mountains, tropical cyclones [TCs], easterly waves; [60, 24] (Capuli et al. 2026), and references therein). Although previous climatological analyses using documented severe weather reports have emphasized the spatial and temporal occurrences of these severe weather events (hereafter SWE; [24]), the propensity of these severe convective storms remains poorly understood in the region. Still, the documentation of tornadoes remains serendipitous, occurring typically when tornadoes strike cities amidst the mostly sparsely populated areas ([3, 88] (Oliviera et al. 2022)). There is evidence, though, that tornadic events in rural lands and forests occur regularly, but are rarely reported, especially when the damage is slight or restricted to vegetation with few damage indicators or low population density ([113, 121]).1

Many studies have examined how environmental conditions relate to severe thunderstorms that produce tornadoes or tornado outbreaks (e.g., Thompson et al. 2012; Lee et al. 2013; Coniglio and Parker [34]; Bagaglini et al. [8]; Pilguj et al. [104]). This interest has lasted for decades because tornadic storms are among the most destructive natural hazards. Their footprint is local, but their social and economic impacts can be severe (e.g., Antonescu et al. [6]; Anderson-Frey and Brooks [4]; Púčik et al. [79]). In the Philippines, however, tornadoes have been sparsely documented, so little is known about the environments and storm characteristics behind them. Tornado environments have been described much more thoroughly elsewhere, notably in the United States (Evans and Doswell [40]; Bunkers et al. [19]; Boustead et al. [14]) and Europe (Groenemeijer and Kühne [45]; Antonescu et al. [6, 5]). More recently, Mexico and Indonesia have also drawn attention (León-Cruz et al. [67]; León-Cruz [66]; Firdaus et al. [41]). Their tropical settings and the way their SWEs vary over time make them useful comparisons for the Philippines.

Generally, severe convective storms (SCS) develop where four ingredients occur together, namely (1) sufficient atmospheric instability, (2) strong vertical wind shear, (3) ample moisture in the lower troposphere and boundary layer, and (4) a lifting mechanism that starts deep moist convection. The lift can come from orography, a frontal boundary, a dryline, or a low-level convergence zone (Rasmussen [80]; Thompson et al. [111]; Mercer et al. [91]; Sherburn and Parker [87]). Both Markowski and Richardson [72] and Davies-Jones [36] note that tornadic supercells tend to form where there is considerable overlap between substantial convective available potential energy (CAPE) and deep-layer shear (DLS). CAPE sets the buoyancy available to the updraft while DLS, i.e., the magnitude of the vector wind difference between the surface and 6 km AGL, sustains the mid-level rotation that keeps a supercell organized and long-lived (Rasmussen and Blanchard [81]; Brooks et al. [15]; Coffer et al. [32]). Interestingly, tornadoes are not confined to isolated supercells as both observations (Nielsen and Schumacher [93]) and numerical simulations (Nielsen and Schumacher [92]) show that mesoscale convective systems (MCSs) can also produce them.

Case studies that used similar approaches are a useful starting point for understanding severe weather events (SWEs). Studies from several regions show how synoptic-scale and mesoscale environments shape tornadic storms. Thompson and Edwards [109] analyzed the 3 May 1999 outbreak over the U.S. Great Plains, which produced the Bridge Creek–Moore, Oklahoma, tornado. Lopes et al. [69] documented the 11–12 June 2018 severe weather outbreak along the Argentina–Brazil border. In Asia, Choo et al. [28] examined the significant tornado of 6 May 2012 in Japan, while Meng et al. [77] documented the deadliest tornado on record in China, which struck Funing County, Yancheng, Jiangsu Province. These studies show that tornadic storms can develop under both strong and weak synoptic forcing, for example from frontal systems, upper-level vortices, or jet streaks. They also need a favorable combination of instability and vertical wind shear, with the addition of complex terrain and mesoscale boundaries that can reinforce vertical motion and help tornadogenesis. Examples include orographic lifting, local outflow boundaries, and storm interactions such as “nudgers” and mergers (Kirshbaum et al. [60]; Nixon et al. [96]). Recent work has also examined the radar signatures of tornadic supercells (Clark [29]; Wakimoto et al. [114]; Lopes et al. [69]). Using dual-polarization radar fields such as differential reflectivity (ZDRZ_{\mathrm{DR}}) and the copolar correlation coefficient (ρHV\rho_{\mathrm{HV}}), these studies identified tornadic debris signatures (TDSs), i.e., an area of low ρHV\rho_{\mathrm{HV}} (< 0.80) that coincides with high reflectivity, and these studies discussed TDSs alongside other radar-derived storm structures such as ZDRZ_{\mathrm{DR}} columns in the parent storm. In related work, Yuan et al. [124] found reduced polarimetric signatures within the low-level mesocyclone of the 14 May 2021 Jiangsu tornadic supercell.

Recent efforts to systematically document severe storm phenomena and develop severe storm report databases have begun in the Philippines through Project Severe Weather Archive of the Philippines (SWAP; Capuli [24]). Despite this effort, tornado damage surveys in the country remain scarce. Recent examples of tornado-focused research in the Philippines include Capuli et al. (2026) and Capuli (2026). Among these studies, Capuli et al. (2026) described the synoptic and mesoscale conditions (including radar features such as BWER and hook echo feature) present during the development of a semi-discrete supercell storm that spawned an EF2 tornado in the rural area of Candating in Arayat, Pampanga on 27 May 2024, located in a favorable area for tornadic storms in the GMMR. The authors showed that the tornado developed in a modest synoptic forcing characterized by lowering of 500-hPa mid-level heights as induced by the migratory TC. This synoptic pattern where mid-level height falls are in-place tend to be associated with major tornado outbreaks found in the U. S. Great Plains’ Tornado Alley due to the intense and rapid low-level atmospheric response (Mercer et al. [91]; Jiang et al. [55]). The lower troposphere exhibited a moist boundary layer, leading to low lifting condensation levels (LCLs), and sufficient southwesterly windflow (or Southwest Monsoon) that generated ample near-surface wind shear (LLS) and DLS in both speed and direction manifested as a curvature in the hodograph profile. It has been long established that “sickle-shaped” hodographs (or hook shaped; where a layer of strong LLS sits beneath an abrupt clockwise turning of the shear vector with height) have become accepted as favorable for significant (≥ EF2) tornadoes along cyclonic, right-moving supercells (Weisman and Klemp [117]; Thompson and Edwards [109]; Nixon and Allen [95]), especially those that feature strong shear near the surface (Coffer et al. [33]). More recently, Capuli [26] documented a localized outbreak of near-simultaneous tornadic supercells in Camarines Norte that developed within an easterly-wave regime, demonstrating that Philippine tornadic environments are not restricted to the southwest monsoon.

On the night of 08 August 2026, a severe thunderstorm developed over the southern coast of Panay Island and produced a significant tornado that tracked through the Mandurriao and Jaro districts of Iloilo City at approximately 2230 LST (1430 UTC). The tornado uprooted and snapped trees, toppled utility poles, and damaged residential and commercial structures, leaving 148 families (or 587 individuals) affected after their homes and properties sustained damage, as confirmed by the City Disaster Risk Reduction and Management Office of Iloilo (CDRRMO-Iloilo) and local media.

The 8 August 2026 Iloilo tornado is a rare documented case of a tornadic storm in the Philippines. Western Visayas is one of the country’s recognized tornado hotspots, yet previous Philippine case studies have focused mostly on Luzon, and few tornadoes from this region have been examined in detail. That makes it a valuable opportunity to study the meteorological processes that govern tornado initiation and development in a tropical environment, in a region that has received little research attention. Although the atmospheric ingredients required for tornadic supercell formation are the same in any part of the world, typical conditions that lead to SCSs in this area of the Philippines may exhibit characteristics that differ from their well-documented U. S. Great Plains or European severe storm environments (Taszarek et al. [105]). Thus, an improved documentation of how these ingredients are brought together, in the moist tropical environment of the Philippines, can add to the overall knowledge of the distinct atmospheric regimes that are capable of producing tornadic supercells.

Early reports and media of the event were scattered across social media platforms. Project SWAP collected, verified, and archived these materials systematically, and they are included in its fourth data release (DR4). The tornado struck at night, so extremely few photographs or videos of it exist. Damage reports, however, were plentiful and have been compiled into a comprehensive dataset (§ 5) that provides essential context for assessing the tornado’s impact. This study examines the meteorological conditions that led to tornadogenesis, covering the synoptic, mesoscale, and kinematic aspects of the parent storm’s environment. Similar to Capuli (2026), this tornado occurred within about 7 km of an operational dual-polarization radar, thus dual-polarization variables can be used to examine the storm’s structure, characteristics, and potential debris signatures in greater detail. Using extensive observational data and sounding-derived parameters, this study provides a foundation for future research on tropical tornadoes.

The paper is organized as follows. § 3 outlines the materials and methods, including the reanalysis dataset, atmospheric profile parameters, and the satellite, radar, surface, and high-resolution optical imagery used in the study. § 5 presents the environmental context (from the synoptic scale to the mesoscale), the storm evolution and radar-derived kinematics, the surface observations, and the damage analysis. Finally, § 6 provides the summary of findings and an extended discussion of the results.

Methodology

Environmental Reanalysis

To characterize the environmental conditions associated with the 08 August 2026 tornado event, this study utilized reanalysis data to assess the potential for SCS development. Reanalysis datasets are widely employed to diagnose atmospheric conditions conducive to severe weather such as tornadoes, damaging winds, and hail-producing storms. They are particularly useful for calculating instability indices and other parameters essential to SCS forecasting (Taszarek et al. [107]).

For this case, environmental reanalysis data were obtained from the fifth-generation European Centre for Medium-Range Weather Forecast (ECMWF) reanalysis (ERA5) accessed through the Climate Data Store (CDS; Hersbach et al. [48]). ERA5 has demonstrated good performance in depicting vertical profiles of convective environments, especially across the United States, Europe, and even other countries along Southeast Asia (Coffer et al. [33]; Taszarek et al. [108]; Pilguj et al. [103]; Firdaus et al. [41]). However, some known biases persist, particularly in the boundary layer. These include discrepancies in low-level parcel characteristics and vertical shear parameters, especially near surface boundaries (King and Kennedy [59]). Such biases are influenced by geographic location and surface elevation. Despite these limitations, ERA5 remains among the most reliable and accessible datasets for investigating severe convective environments, globally (Coffer et al. [33]; Taszarek et al. [106]). For large-scale synoptic analysis, the study domain extended from 4∘4^\circ–22∘22^\circN latitude and 110∘110^\circ–135∘135^\circE longitude. Meanwhile, the mesoscale sector covered 8∘8^\circ–14∘14^\circN and 119∘119^\circ–126∘126^\circE. Both domains used a spatial resolution of 0.25∘×0.25∘0.25^\circ\times0.25^\circ. An hourly temporal resolution was used to allow detailed analysis of the atmospheric evolution surrounding the event, which occurred around 14 UTC. The time steps selected for in-depth analysis were 13, 14, and 15 UTC, enabling a focused examination of pre-, during, and post-storm conditions. Throughout this paper, times are given in UTC, with Philippine local standard time (LST = UTC + 8 h) in parentheses.

Although ERA5 provides several convective parameters, including CAPE, it has been shown to substantially underestimate the severity of convective environments. Capuli (2025) demonstrated this for a Philippine hailstorm case in which native ERA5 CAPE of ∼2000\sim2000–2500 J kg−12500\ \mathrm{J\ kg^{-1}} rose to nearly 5000 J kg−15000\ \mathrm{J\ kg^{-1}} when recomputed from the model-level profile with a surface-based (SB) parcel. The difference between a marginally severe environment and one capable of the large hail observed. The bias likely stems from the ERA5 formulation, which lifts candidate parcels from a series of model levels within the lowest ∼350 hPa\sim350\ \mathrm{hPa} rather than from the near surface, so the selected parcel is frequently more dilute than the true boundary-layer parcel. To address this, both the ERA5 single-level and 137 hybrid sigma-pressure model-level fields were combined and re-gridded onto a common grid, and CAPE was recalculated using most-unstable (MU) parcel which is the parcel yielding the largest CAPE among candidates lifted from levels within the lowest 300 hPa, with the lowest candidate taken from the 2 m AGL. Likewise, other kinematic parameters such as Storm-relative Helicity (SRH), which is not available via the standard data field in ERA5, were computed for the 0–1 and 0–3 km layers using the xcape codebase [65]. Kinematic parameters, such as SRH3^{3}, the Bunkers et al. [20] internal-dynamics storm motion (B2K), and the resulting storm-relative wind (VSRV_{\mathrm{SR}}), are computed via the right-moving vector (B2KRM), consistent with standard practice for the Northern Hemisphere. This assumes that the storm moves to the right of the non-pressure-weighted mean wind in the 0–6 km layer. Finally, all xcape-derived thermodynamic quantities (e.g., MUCAPE, SBCIN), including those underlying the ERA5 mesoscale spatial maps, assume liquid-only pseudoadiabatic ascent, with condensate removed on formation and ice processes neglected as chosen for computational efficiency and consistency across the domain.

The interpretation of all reanalysis-derived parameters (including derived through the Skew-T Hodograph in § 3) is nonetheless constrained by the resolution of ERA5 relative to the scale of the setting. Spatial averaging across a land–sea boundary and across smoothed terrain tends to dilute local thermodynamic extremes and to remove the small-scale flow deformation that enhances low-level rotation, so that ERA5-derived (and the recalculated from previous section) CAPE and SRH may either overestimate or underestimate the near-storm environment. Thus, the parameters reported here are interpreted as characterizing the mesoscale environment within which the storms developed.

Proximity Sounding Profile

Because no radiosonde was launched near the time and location of the event, a model-derived proximity profile was extracted from ERA5 at the S-ILO site (10.772∘N10.772^{\circ}\mathrm{N}, 122.579∘E122.579^{\circ}\mathrm{E}) for 1400 UTC (2200 LST) using a combination of single-level and 137 hybrid sigma-pressure model-level fields. This profile was analyzed together with the observed soundings from the PAGASA Iloilo station (WMO 98637) at 0000 UTC and the Mactan station (WMO 98646), ∼161\sim161 km east-southeast of S-ILO, at 1200 UTC to assess the thermodynamic and kinematic conditions preceding and at the time of the tornadic event.

While the ERA5 recalculation described in § 3 utilizes a liquid-only pseudoadiabatic ascent, the parcel profiles shown on the Skew-T diagrams were computed assuming a non-entraining, irreversible mixed-phase adiabatic process following [102]. Under pseudoadiabatic ascent, all condensate is assumed to fall out of the parcel immediately [38]. Meanwhile, the present formulation for the Skew-Ts instead retains condensate and represents the mixed-phase layer between the 0 °C level and the level of homogeneous freezing (∼−40\sim-40 °C), where liquid and ice coexist. Because it enforces conservation of energy rather than of moist entropy, it avoids the biases that arise when entropy-based formulations are applied across the mixed-phase region. Support for retaining condensate in a tropical context comes from Xu and Emanuel (1989), who found that mean tropical soundings are nearly neutral with respect to reversible moist adiabatic ascent and are therefore only weakly conditionally unstable, indicating that the tropical environmental temperature profile is more closely matched by an adiabatic than by a pseudoadiabatic parcel. We note that their analysis employed a reversible liquid-only parcel including water loading rather than the mixed-phase formulation adopted here. Thus, it is cited as physical motivation for retaining condensate, not as direct validation of the present parcel model.

In addition, an irreversible adiabatic ascent with entrainment was also employed to compute the Entraining CAPE (ECAPE; [101], readers are referred to the paper for the full derivation). ECAPE accounts for dilution of updraft plumes due to environmental entrainment, particularly under conditions of limited storm-relative inflow and/or mid- to upper-tropospheric dryness. It therefore provides a more realistic estimate of updraft intensity than the conventional, undiluted CAPE, which can overestimate buoyancy in environments where entrainment is significant. All undiluted Skew-T profiles are calculated with virtual temperature correction. Depending on meteorological context, either MU or SB parcels were selected for key thermodynamic parameters. CAPE and the level of free convection (LFC) were computed using MU parcel profiles, while convective inhibition (CIN) and LCL were determined using SB profiles. Although mixed-layer parcels are often preferred for their robustness, SB and MU parcels were preferred in this study due to their better representation of near-surface instability (or stability) and lowest cloud base heights. These factors are critical in distinguishing SB from elevated convection, particularly in tornadic environments.

To ascertain the kinematic conditions within this tornado event, the hodograph was used. In a conventional, ground-relative hodograph, no transformation is made to the modeled uu and vv wind components [71, 97]. However, this frame does not show the wind profile as experienced by the storm, which governs storm inflow and the ingestion of environmental vorticity. To place the kinematic analysis in the frame of the storm rather than the ground, the wind profile is expressed in storm-relative coordinates. The transformation is a rigid translation of the hodograph by the storm motion vector c=(cu,cv)\mathbf{c} = (c_u,c_v),

VSR(z)=V(z)−c(1)\mathbf{V}_{\mathrm{SR}}(z) = \mathbf{V}(z) - \mathbf{c} \tag*{(1)}

where V(z)=[u(z),v(z)]\mathbf{V}(z) = [u(z),v(z)] is the ground-relative wind. Because c\mathbf{c} is independent of height, the vertical shear vector and the horizontal vorticity are unchanged by the transformation as only the origin of the hodograph moves. Here, c\mathbf{c} is the B2KRM_{\mathrm{RM}} computed with the non-pressure-weighted mean wind and shear of the 0–3 km layer rather than the conventional 0–6 km layer, for consistency with the radar-derived hodographs, whose reliable portion extends only to ∼3.5\sim3.5 km AGL (§ 3§ 3(iii)).

The kinematic parameters were calculated from the interpolated uu- and vv-wind components from each sounding profile. The bulk wind difference (BWD) was calculated across several vertical layers to assess shear magnitude. These included the layers from 0–500 m (BWD500_{500}), 0–1 km (BWD01_{01}; LLS), 0–3 km (BWD03_{03}; MLS), and 0–6 km (BWD06_{06}; DLS). SRH is of the greatest dynamical importance in this regard [35]. The SRH was computed for several layers, namely 0–500 m (SRH500_{500}), 0–1 km (SRH01_{01}), 0–3 km (SRH03_{03}), and 1–3 km (SRH13_{13}). For the near-ground layer (0–1 km), the mean VSR\mathbf{V}_{\mathrm{SR}} was calculated relative to the B2KRM_{\mathrm{RM}}. This analysis also included the mean streamwise vorticity integrated over 0–500 m and 0–1 km layer represented by ωs500\omega_{s500}, and ωs01\omega_{s01}, respectively. These quantities were calculated as

ωs=(∇×(v−c))⋅(v−c)∥v−c∥(2)\omega_s = \frac{(\nabla\times(\mathbf{v}-\mathbf{c})) \cdot(\mathbf{v}-\mathbf{c})}{\lVert\mathbf{v}-\mathbf{c}\rVert} \tag*{(2)}

where v\mathbf{v} is the three-dimensional environmental wind and c\mathbf{c} is the storm motion vector under the assumption of the B2K method. Furthermore, the vertical vorticity along the low-level mesocyclone was calculated. Based on the first-order approximation of tilting efficiency proposed by Coffer et al. [31], the vertical vorticity generated from the tilting of streamwise vorticity in a purely streamwise environment can be expressed as

ζLLM=ωswVSR(3)\zeta_{\mathrm{LLM}} = \omega_s \frac{w}{V_{\mathrm{SR}}} \tag*{(3)}

where ωs\omega_s is the streamwise vorticity4, ww is the vertical velocity of the updraft5, and VSRV_{\mathrm{SR}} is the storm-relative wind in the lowest 1 km. This formulation highlights how the conversion of environmental streamwise vorticity into vertical vorticity depends on the interaction and balance between vertical and horizontal motions [35]. It provides a physically intuitive approximation for assessing the potential for mesocyclone development and subsequent tornadogenesis.

The “streamwiseness” (ω~s\widetilde{\omega}_s) of horizontal vorticity was also calculated by dividing the integrated streamwise component of horizontal vorticity by the integrated total magnitude of the horizontal vorticity in a given layer,

ω~s=ωsω(4)\widetilde{\omega}_s = \frac{\omega_s}{\omega} \tag*{(4)}

The use of ω~s\widetilde{\omega}_s is justified for two reasons. First, near-zero shear conditions are rare in tornado environments [33]. Second, the parameter offers intuitive interpretation e.g., a value of 0.95 means that 95% of the horizontal vorticity is streamwise in nature. More fundamentally, the degree to which vorticity is streamwise remains a fundamental aspect of supercell dynamics and tornadogenesis [30].

Finally, the Deviant Tornado Motion (DTM) was calculated based on the method developed by Nixon and Allen [94]. This parameter evaluates the potential deviation of tornado motion from both the low-level advective flow and the parent storm motion, and offers insight into tornado track behavior relative to ambient wind conditions. All proximity soundings associated with this event were analyzed using SounderPy (v3.1.0) by Gillett [42]. A list of the evaluated kinematic and thermodynamic parameters, along with their definitions and computation methods, is presented in Table 1. Storm-relative quantities in Table 1 (SRH, VSRV_{\mathrm{SR}}, ωs\omega_s, streamwiseness, DTM, the supercell composite parameter [SCP], and the significant tornado parameter [STP]; Thompson et al. [111, 112]) were computed against the B2KRM_{\mathrm{RM}}, and the SCP and STP are the effective-inflow-layer formulations, which use MU parcel quantities rather than the mixed-layer quantities of the fixed-layer STP.

ParameterDefinitionNotes00 UTC ILO OBS12 UTC MAC OBS14 UTC ILO ERA5
CAPEConvective Available Potential EnergyMU parcel1260 J kg−1^{-1}2532 J kg−1^{-1}2014 J kg−1^{-1}
CAPE03_{03}0–3 km CAPEMU parcel100 J kg−1^{-1}128 J kg−1^{-1}130 J kg−1^{-1}
ECAPEEntraining CAPEMU parcel*1736 J kg−1^{-1}1632 J kg−1^{-1}
CINConvective InhibitionSB parcel0 J kg−1^{-1}−21 J kg−1^{-1}0 J kg−1^{-1}
LCLLifted Condensation LevelSB parcel910 m622 m295 m
LFCLevel of Free ConvectionMU parcel958 m1081 m295 m
LR03_{03}0–3 km Lapse RateLow-level Lapse Rate5.17 °C km−1^{-1}6.58 °C km−1^{-1}5.81 °C km−1^{-1}
LR36_{36}3–6 km Lapse RateMid-level Lapse Rate5.57 °C km−1^{-1}6.07 °C km−1^{-1}5.94 °C km−1^{-1}
RH13_{13}Mean 1–3 km Relative Humidity97.20%82.96%84.61%
RH16_{16}Mean 1–6 km RH96.97%81.98%81.91%
BWD01_{01}0–1 km Bulk Wind DifferenceLow-level Shear (LLS)5.16 m s−1^{-1} / 10.04 kt10.03 m s−1^{-1} / 19.51 kt10.50 m s−1^{-1} / 20.42 kt
BWD03_{03}0–3 km BWDMid-level Shear (MLS)10.02 m s−1^{-1} / 19.49 kt16.08 m s−1^{-1} / 31.26 kt17.64 m s−1^{-1} / 34.29 kt
BWD06_{06}0–6 km BWDDeep-layer Shear (DLS)5.83 m s−1^{-1} / 11.34 kt8.55 m s−1^{-1} / 16.62 kt2.46 m s−1^{-1} / 4.79 kt
B2KRMBunkers RM Storm MotionRM: Right Mover8.89 m s−1^{-1} / 17.28 kt9.09 m s−1^{-1} / 17.66 kt13.60 m s−1^{-1} / 26.43 kt
VSRV_{\mathrm{SR}}Mean 0–1 km Storm-relative WindUsing B2KRM8.33 m s−1^{-1} / 16.20 kt10.27 m s−1^{-1} / 19.96 kt11.86 m s−1^{-1} / 23.06 kt
SRH500_{500}0–500 m Storm-relative HelicityUsing B2KRM20 m2^{2} s−2^{-2}57 m2^{2} s−2^{-2}79 m2^{2} s−2^{-2}
SRH01_{01}0–1 km SRHUsing B2KRM40 m2^{2} s−2^{-2}101 m2^{2} s−2^{-2}127 m2^{2} s−2^{-2}
SRH03_{03}0–3 km SRHUsing B2KRM82 m2^{2} s−2^{-2}247 m2^{2} s−2^{-2}259 m2^{2} s−2^{-2}
SRH13_{13}1–3 km SRHUsing B2KRM42 m2^{2} s−2^{-2}146 m2^{2} s−2^{-2}131 m2^{2} s−2^{-2}
ωs500\omega_{s500}0–500 m Streamwise VorticityUsing B2KRM0.005 s−1^{-1}0.012 s−1^{-1}0.014 s−1^{-1}
ωes500\omega_{es500}0–500 m StreamwisenessUsing B2KRM91.61%99.15%98.62%
ωs01\omega_{s01}0–1 km Streamwise VorticityUsing B2KRM0.005 s−1^{-1}0.010 s−1^{-1}0.011 s−1^{-1}
ωes01\omega_{es01}0–1 km StreamwisenessUsing B2KRM95.35%97.61%98.54%
ζLLM\zeta_{\mathrm{LLM}}0–1 km Low-level Vertical VorticityMU parcel and B2KRM0.008 s−1^{-1}0.016 s−1^{-1}0.015 s−1^{-1}
DTMDeviant Tornado MotionUsing B2KRM4.65 m s−1^{-1} / 9.04 kt7.06 m s−1^{-1} / 13.73 kt10.05 m s−1^{-1} / 19.53 kt
WMAXSHEARUndiluted Updraft Velocity × BWD06_{06}MU parcel293.02 m2^{2} s−2^{-2}608.46 m2^{2} s−2^{-2}156.25 m2^{2} s−2^{-2}
SCPSupercell Composite ParameterEffective inflow1.133.327.44
STPSignificant Tornado ParameterEffective inflow0.250.742.28

Table 1. Environmental sounding parameters evaluated in this case. An asterisk indicates that the computation failed.

Remote Sensing and In-Situ Data

HIMAWARI-9

Convective-system evolution was monitored using the Advanced Himawari Imager (AHI) on board Himawari-9 [11]. AHI observes in 16 spectral bands, namely three visible (central wavelengths 0.47–0.64 μ\mum), three near-infrared (0.86–2.3 μ\mum), and ten infrared (3.9–13.3 μ\mum). The tornado occurred about two hours after local sunset, so the visible and near-infrared bands carried no useful signal. The analysis therefore relies only on the 10.4-μ\mum longwave infrared window channel (Band 13, B13), which works the same by day and night. B13 is well suited to delineating cold cloud tops of deep convection [brightness temperature (BT) < 233 K (−40 °C)] because water vapor absorbs little at this wavelength Hirose et al. [49]. It is also commonly used for quickly identifying convective features linked to intense, short-lived rainfall Wang et al. [116]. The nominal resolution of B13 is 2 km at the sub-satellite point. Full-disk B13 imagery on the GEOS141 fixed grid was retrieved every 10 min from 1400 to 1500 UTC (2200–2300 LST) on 08 August 2026 through the National Computational Infrastructure (NCI) Australia THREDDS data server, although the 1440 UTC scan was unavailable in the archive. This window brackets the S-ILO radar volumes analyzed (2200–2230 LST) and the tornado’s lifetime.

Polarimetric Radar

Iloilo Radar and Specifics

DOST-PAGASA operates a nationwide network of 17 weather radars, consisting of seven dual-polarization (dual-pol) C-band radars, three dual-pol S-band radars, and seven single-polarization (single-pol) S-band radars. Together, these systems support the agency’s capacity for early detection and monitoring of weather systems across the archipelago. The 08

August 2026 tornadic supercell tracked through the domain of the Iloilo S-band radar (hereafter S-ILO), one of the three dual-pol S-band radars, located in Jaro, Iloilo (10.773056°N, 122.579444°E; 34 m above mean sea level) on Panay Island (Figure 1). S-ILO was inaugurated on 06 March 2026 and began operations on 22 May 2026, less than three months before the event. It provides extensive coverage over the Western Visayas, a region frequently affected by TCs and the southwest monsoon. S-ILO employs a modified Volume Coverage Pattern 11 (VCP11), which optimizes sampling across both low and high elevation angles and is particularly well suited to deep tropical convection. Each volume consists of 14 tilts (0.5–19.5°) completed every 10 min, with 360 radials at 1° azimuthal spacing and 250 m range gates out to a Doppler scanning radius of 240 km. The technical specifications of S-ILO are listed in Table 2.

Elevation map encompassing S-ILO and its radar coverage

Figure 1. Elevation map encompassing S-ILO (red circle) and its 240 km radar coverage with increments of 60 km each. The red dashed line serves as a reference for the dynamical vertical cross section of various atmospheric variables along 11.0°N (Figure 3).

Frequency BandS-band, 2.868 GHz (λ=10.453\lambda= 10.453 cm)
PolarizationDual-polarization (Dual-pol)
Position10.773056°N, 122.579444°E
Altitude34 m above mean sea level
Pulsing Scheme4:3 dual-PRF on the lowest 6 tilts (0.5–5.3°); single PRF on the upper 8 tilts (6.2–19.5°)
Low / High PRF600 / 800 Hz (1.667 / 1.250 ms) dual-PRF; 300 Hz (3.333 ms) single PRF
Nyquist Velocity±15.68 m s−1^{-1} (600 Hz), ±20.91 m s−1^{-1} (800 Hz); extended ±62.72 m s−1^{-1} on dual-PRF tilts; ±7.84 m s−1^{-1} on single-PRF tilts
Maximum Range / Doppler Range240 km / 240 km
Beam Width / Range Resolution (Gate Spacing)1.33° (hh and vv) / 250 m
Azimuthal Sampling1.0° (360 rays per sweep)
Elevation Angles (14)0.5°, 1.5°, 2.4°, 3.4°, 4.3°, 5.3°, 6.2°, 7.8°, 8.7°, 10°, 12°, 14°, 16.7°, 19.5°
Volume Cycle Interval10 min
Start of Operation22 May 2026

Table 2. Technical specifications of S-ILO.

The volumes were archived in Universal Format (UF), and several of their header entries were found to be inconsistent with the data. These entries were therefore rewritten before analysis, although the data values themselves were not modified. The stored wavelength was unusable and was replaced with λ=0.10453\lambda= 0.10453 m, and the pulse repetition time (PRT) slot, which carried the pulse repetition frequency (PRF) in Hz, was converted to PRT = 1/PRF. The Nyquist velocity of each sweep was then recomputed from its PRF as vN=λPRF/4v_N = \lambda\mathrm{PRF}/4. With λ=0.10453\lambda= 0.10453 m, the PRFs repro duce both scanning modes of S-ILO exactly. The lowest six tilts (0.5–5.3°) use 4:3 staggered dual-PRF sampling at 600/800 Hz, which yields base Nyquist velocities of ±15.68 and ±20.91 m s−1^{-1} and an extended Nyquist velocity of ±62.72 m s−1^{-1}. The remaining eight tilts (6.2–19.5°) operate at a single PRF of 300 Hz, with a Nyquist velocity of only ±7.84 m s−1^{-1}. The header Nyquist velocity of 30.0 m s−1^{-1} on the 1.5–5.3° tilts was rejected, since VRV_\mathrm{R} reaches ±47–61 m s−1^{-1} on these tilts and never exceeds the extended limit of ±62.72 m s−1^{-1}.

Beam height above radar level (hh), and ground range (ss) were computed with the 4/3-Earth model (Doviak and Zrnić [37]),

h=r2+(kea)2+2rkeasin⁡θ−kea,s=keaarcsin⁡(rcos⁡θkea+h),(5)\begin{aligned} h &= \sqrt{r^{2} + (k_{\mathrm{e}}a)^{2} + 2r k_{\mathrm{e}}a \sin\theta} - k_{\mathrm{e}}a,\\ s &= k_{\mathrm{e}}a \arcsin\left(\frac{r\cos\theta}{k_{\mathrm{e}}a+h}\right), \tag*{(5)} \end{aligned}

where rr is the slant range, θ\theta is the elevation angle, a=6371a = 6371 km, and ke=4/3k_{\mathrm{e}} = 4/3.

(ii) Radar Quality Control S-ILO’s radar data underwent quality control based on a fuzzy logic approach designed to detect and eliminate interference patterns. This method draws on prior algorithm assessments (e.g., [58, 99 and references therein]) and was adapted for both single- and dual-pol radar products following the framework described in Lin et al. [68]. In addition, these radar volumes had also been subjected to a dual-polarimetric quality control, applied as part of an ongoing project to quality-control radar volumes for operational quantitative precipitation estimation (QPE). Among its criteria, gates with ρHV<0.80\rho_{\mathrm{HV}} < 0.80 were omitted, i.e., no gate in the ρHV\rho_{\mathrm{HV}} field falls below this threshold, and masked across all radar fields such as in ZHZ_{\mathrm{H}}, ZDRZ_{\mathrm{DR}}, and KDPK_{\mathrm{DP}} alike. For QPE, this threshold is a sound choice, since it efficiently removes non-meteorological echoes such as ground clutter, biological targets and clear-air returns that would otherwise contaminate rainfall estimates. For tornadic storms, however, it is detrimental. Lofted debris is characterized precisely by low ρHV\rho_{\mathrm{HV}}, typically < 0.80 ([84, 13]), so the same threshold that cleans the volume for rainfall estimation also removes the gates where a TDS would be expected. A TDS can therefore neither be confirmed nor ruled out from the supplied data.

No further correction was applied to the polarimetric fields. Attenuation is negligible at S-band for the short ranges involved, as the storm remained within ~22 km of the radar throughout the analysis period. In addition, S-ILO had only recently been made operational, so no specific attenuation correction, systematic-bias calibration, or wet-radome correction had yet been established for it. Unlike the polarimetric fields, the VRV_\mathrm{R} required substantial correction. Two distinct velocity artifacts were present in the S-ILO volumes and were corrected sequentially. The dual-PRF tilts (0.5–5.3°) are subject to processor errors, in which the unfolding of a gate between the two PRF estimates fails and the retrieved velocity is displaced by an integer multiple of twice one of the base Nyquist velocities. The single-PRF tilts (6.2–19.5°) are instead subject to conventional aliasing, which is severe at S-ILO, since a ±7.84 m s−1^{-1} Nyquist interval is easily exceeded by the 15–25 m s−1^{-1} environmental flow. Before either correction, three filters were applied: (1) velocity gates without valid ZHZ_\mathrm{H} were removed, (2) gates with ∣VR∣<1|V_\mathrm{R}| < 1 m s−1^{-1} on tilts at or below 1.5° were removed as residual ground or sea clutter, and (3) isolated velocity patches smaller than 10 contiguous gates were removed with Py-ART’s despeckling routine.

The dual-PRF processor errors were corrected first, using the method of Alford et al. [2], which builds on the median-reference approach of Joe and May [56] and Holleman and Beekhuis (2003). In this method, a gate whose departure from the local reference matches one of the expected error multiples of 2vL2v_L or 2vH2v_H is shifted back by that multiple. Only the 0.50.5–5.3∘5.3^\circ tilts were treated, using vL=15.68 m s−1v_L = 15.68\ \mathrm{m\ s^{-1}} and vH=20.91 m s−1v_H = 20.91\ \mathrm{m\ s^{-1}}. Once corrected, the dual-PRF tilts cannot fold at the observed speeds, and they therefore served as the unambiguous anchor for dealiasing the single-PRF tilts. Continuity-only dealiasing algorithm such as UNRAVEL (Louf et al. [70]) and Py-ART’s region-based method lose their reference between the isolated cells of this case, so each folded tilt was instead dealiased against the already-corrected tilts directly below it. This is similar to the vertical-continuity step of four-dimensional dealiasing (James and Houze [54]), but anchored to the radar’s own dual-PRF tilts rather than to a sounding. Adjacent gates with ∣ΔV∣<0.5vN|\Delta V| < 0.5v_N were first joined into regions. Since a fold produces a jump of about 2vN2v_N, a region never straddles a fold and receives a single fold count nkn_k, so that

Vi=Viraw+2nkvN,nk∈{−4,…,4},(6)V_i = V_i^{\mathrm{raw}} + 2n_k v_N,\qquad n_k \in\{-4,\ldots,4\}, \tag*{(6)}

For every gate, a local reference (RiR_i) was taken from the nearest corrected tilt bb below, as a 3×53 \times5 ray-by-gate median at the same azimuth and ground range. This reference was projected to the elevation of tilt ss and corrected for the environmental shear between the two beam heights,

Ri=[V~bcos⁡θb+Venv(hs,ϕ)−Venv(hb,ϕ)]cos⁡θs,(7)R_i = \left[\frac{\widetilde{V}_b}{\cos\theta_b} + V_{\mathrm{env}}(h_s,\phi) - V_{\mathrm{env}}(h_b,\phi)\right]\cos\theta_s, \tag*{(7)}

where Venv(h,ϕ)=uenv(h)sin⁡ϕ+venv(h)cos⁡ϕV_{\mathrm{env}}(h,\phi) = u_{\mathrm{env}}(h)\sin\phi+ v_{\mathrm{env}}(h)\cos\phi is the horizontal radial component of a quadratic fit to the dual-PRF-only wind profile (§ 3§ 3(iii)). The fold counts were then chosen to minimize

E(nk)=∑i∈kmin⁡(∣Vi−Ri∣,vN)+∑(i,j)min⁡(∣Vi−Vj∣,vN)+λ∑i∈kmin⁡(∣Vi−Venv,i∣,vN),(8)E(n_k) = \sum_{i\in k}\min(|V_i-R_i|,v_N) + \sum_{(i,j)}\min(|V_i-V_j|,v_N) + \lambda\sum_{i\in k}\min(|V_i-V_{\mathrm{env},i}|,v_N), \tag*{(8)}

where the first term anchors the region to the tilt below, the second enforces continuity across region boundaries, and the third is a weak environmental prior (λ=0.2\lambda= 0.2). Each term is capped at vNv_N, so that a few poor references or a genuine sharp shear zone, such as a mesocyclone, cannot dominate the solution. Equation (8) was minimized by iterated conditional modes (Besag [10]), sweeping the tilts from the bottom up, and a second pass added the reference from the newly corrected tilts above. For the near-storm analysis, a stricter variant was applied as regions were joined only where ∣ΔV∣<0.25vN|\Delta V| < 0.25v_N and the reference came from one tilt below only. This was followed by a suspect-region mask that removes regions in which more than half of the gates depart by more than vNv_N from the 7×217 \times21-gate median of their own sweep or by more than 1.5vN1.5v_N from both adjacent tilts. The scheme was verified by folding the dual-PRF tilts artificially at 7.84 m s−17.84\ \mathrm{m\ s^{-1}} and recovering them; only 1.61.6–4.8%4.8\% of the gates received a wrong fold, compared with 88–77%77\% for Py-ART’s region-based method and UNRAVEL on the same test.

Despite these corrections, the VRV_R data of S-ILO remain problematic for tornado-scale analysis, for four reasons identified upon checking.

  • Velocity sign. The velocity is stored with the opposite sign to the CF convention (positive toward the radar). With the sign as stored, the retrieved low-level winds blow from the north-northeast and the B2KRM_\mathrm{RM} moves toward the southwest. This contradicts both the east-northeastward storm motion and the southwesterly winds recorded by the automatic weather station (AWS). Therefore, all velocities were multiplied by −1-1.

  • Aliasing on the upper tilts. The upper eight tilts are so heavily aliased that 3434–81%81\% of the raw gates jump by more than vNv_N relative to the tilt below. Even after dealiasing, 88–29%29\% of the gates on these tilts still depart from the wind profile by more than vNv_N. Thus, the single-PRF tilts were used only for the wind profile, not for vortex-scale diagnostics.

  • Processor smoothing. The velocity field has been strongly smoothed by the signal processor. Fitting a separable Gaussian to the autocorrelation of the small-scale velocity residual yields an effective filter of σz=1.66\sigma_z = 1.66 radials and σr=1.50\sigma_r = 1.50 gates. This suppresses gate-to-gate shear and can almost entirely hide a vortex narrower than about 150 m150\ \mathrm{m}.

  • Missing data at the debris. As previously mentioned, the VRV_R gates without reflectivity were discarded, and the ρHV<0.80\rho_{\mathrm{HV}} < 0.80 gates are missing in ZHZ_H and other dual-pol fields. Hence, the VRV_R field is also missing at the gates where debris would be expected.

Consequently, no tornadic vortex signature (TVS; Brown et al. [16]) could be identified directly in VRV_R, which motivates the velocity simulation described in subsequent section.

Radar Techniques

Because the echoes in this case were patchy, most single range rings never reached the azimuthal coverage required by a classic velocity–azimuth display (VAD; Browning and Wexler [18]). The wind profile was therefore retrieved with a volume velocity processing (VVP)-type layer fit (Waldteufel and Corbin [115]). Layer by layer, all usable gates of all tilts below 45∘45^\circ within 55–120 km120\ \mathrm{km} slant range were fitted to a horizontally uniform wind,

VR=usin⁡ϕcos⁡θ+vcos⁡ϕcos⁡θ+wesin⁡θ(9)V_R = u \sin\phi\cos\theta+ v \cos\phi\cos\theta+ w_e \sin\theta \tag*{(9)}

where ϕ\phi is the azimuth and wew_e combines vertical air motion and hydrometeor fall speed. Layers were centered every 250 m from 125 m to 10 km above ground level (AGL), with a half-thickness of 250 m, and gate heights were computed from Eq. (5). The fit was solved by least squares with iterative rejection of residuals larger than max⁡(2.5σ,2 m s−1)\max(2.5\sigma, 2\ \mathrm{m\ s^{-1}}). A layer was accepted only if it met all of the following criteria, namely (1) at least 100 gates, (2) at least 5 of 12 occupied 30° azimuth sectors, (3) a condition number of the (u,v)(u,v) geometry ≤30\le30, (4) a residual RMS ≤5 m s−1\le5\ \mathrm{m\ s^{-1}}, and (5) at least 60% of the gates retained.

The retrieval was performed in two passes. The first pass used only the dual-PRF gates, which cannot alias. The layer-by-layer median of this fit over three consecutive volumes (±10 min) defined a reference profile extending to about 3.4–3.6 km AGL. The second pass added the dealiased single-PRF gates, and a layer was retained only if its wind stayed within 5 m s−1^{-1} of the reference. Otherwise, the dual-PRF-only fit was used, and where that layer was also missing, the time-filled reference was adopted (e.g., the 125 m layer at 1410 UTC was taken from 1420 UTC). The 10 m AWS wind nearest to each scan was then added as the lowest level of the profile, and the profile was linearly interpolated to a 10 m grid without vertical smoothing.

Storm motion was computed with the internal-dynamics method of Bunkers et al. [20] using the non-pressure-weighted mean wind of the 0–3 km layer, because the reliable portion of the radar-derived profile did not extend to 6 km, so the shear vector was taken between the 2.5–3.0 km and 0–0.5 km mean winds, with a deviation of 7.5 m s−1^{-1}. The DTM was computed as the average of the right-moving vector and the 0–500 m mean wind (Nixon and Allen [94]). As an independent estimate, the displacement of the column-maximum reflectivity composite (2 km grid, ZH>10Z_H > 10 dBZ) between the 1410 and 1420 UTC volumes was determined by two-dimensional cross-correlation, yielding an observed storm motion of 13.7 m s−1^{-1} toward 076°. The BWD, SRH, critical angle (Esterheld and Giuliano [39]), mean VSRV_{\mathrm{SR}}, and the streamwise vorticity and streamwiseness defined in § 3 were then computed for each volume relative to the observed storm motion (Table 3), and the hodographs were plotted in the same frame.

Kinematic parameter1410 UTC1420 UTC1430 UTC
BWD01_{01}16.063 m s−1^{-1} / 31.224 kt16.250 m s−1^{-1} / 31.588 kt14.343 m s−1^{-1} / 27.881 kt
BWD03_{03}10.641 m s−1^{-1} / 20.684 kt11.966 m s−1^{-1} / 23.261 kt16.064 m s−1^{-1} / 31.225 kt
BWD06_{06}———
B2KRM_{\mathrm{RM}}13.535 m s−1^{-1} / 26.310 kt13.827 m s−1^{-1} / 26.878 kt12.506 m s−1^{-1} / 24.310 kt
Storm-relative (observed storm motion of 13.700 m s−1^{-1} / 26.631 kt)
VSRV_{\mathrm{SR}}14.671 m s−1^{-1} / 28.518 kt14.514 m s−1^{-1} / 28.212 kt17.707 m s−1^{-1} / 34.419 kt
SRH500_{500}114.901 m2^{2} s−2^{-2}114.995 m2^{2} s−2^{-2}100.717 m2^{2} s−2^{-2}
SRH01_{01}211.451 m2^{2} s−2^{-2}124.379 m2^{2} s−2^{-2}72.898 m2^{2} s−2^{-2}
SRH03_{03}291.815 m2^{2} s−2^{-2}265.186 m2^{2} s−2^{-2}281.987 m2^{2} s−2^{-2}
ωs500\omega_{s500}0.0193 s−1^{-1}0.0195 s−1^{-1}0.0173 s−1^{-1}
ω~s500\widetilde{\omega}_{s500}0.7383 / 73.83%0.7589 / 75.89%0.5456 / 54.56%
ωs01\omega_{s01}0.0150 s−1^{-1}0.0103 s−1^{-1}0.0074 s−1^{-1}
ω~s01\widetilde{\omega}_{s01}0.7224 / 72.24%0.5607 / 56.07%0.3467 / 34.67%

Table 3. Radar-derived kinematics from the volume velocity processing (VVP) wind profiles and hodographs. Storm-relative quantities are computed relative to the observed storm motion (13.7 m s−1^{-1} toward 076°).

Since a ρHV\rho_{\mathrm{HV}}-based TDS could not be used, the tornado was tracked through its reflectivity signature instead. On each tilt of the 1420 and 1430 UTC volumes, the weak echo hole (WEH) at the tip of the hook echo was identified in the ZHZ_H field and its center was marked manually. Briefly, the WEH is a local ZHZ_H minimum, partly as missing data, enclosed by the high-reflectivity ring of the debris ball. Each center was converted to azimuth and range from the radar, to beam height with Eq. (5), and to latitude and longitude. It was time-stamped from the ray times of its tilt, since successive tilts are scanned about 30 s apart. At 1420 UTC, the WEH could be followed from 2.4° to 10.0° (0.48–1.42 km above radar level), and at 1430 UTC from 2.4° to 5.3° (0.24–0.43 km). The WEH was enclosed by circles of 0.6 km radius at 1420 UTC and 0.46–0.58 km radius at 1430 UTC.

The vertical structure of the WEH was further examined at 1420 UTC with two vertical cross-sections of ZHZ_H, ZDRZ_{\mathrm{DR}}, and ρHV\rho_{\mathrm{HV}} through its center, from the surface to 2 km, one north–south (10.628–10.809°N along 122.499°E) and one east–west (122.442–122.616°E along 10.717°N). The tracked WEH positions were finally compared with the centerline of the surveyed damage path. The centerline was obtained by projecting the damage points onto their principal axis and smoothing their cross-track offsets with a locally linear Gaussian kernel (0.30 km bandwidth).

To examine the three-dimensional structure of the storm, the reflectivity (ZHZ_H) of the 1420 UTC volume was objectively analyzed onto a Cartesian grid with Py-ART ([47]), using a Barnes weighting function, a radius of influence that increases with the beam width (minimum of 300 m), and a grid spacing of 0.25 km in the horizontal and 0.2 km in the vertical over a 20 km × 20 km domain extending to 4 km ARL. Before gridding, every gate was shifted back to the start of the 2.4° tilt by the storm motion (13.7 m s−1^{-1} toward 076°), because the tilts of one volume are scanned ∼31 s apart and a feature moving with the storm otherwise appears displaced downstream on each successive tilt. The gridded field was lightly smoothed (Gaussian filter with σ=0.5\sigma= 0.5 grid points), and closed isosurfaces of ZHZ_H at 30, 35, 40, 45, and 50 dBZ were extracted with the marching-cubes algorithm and drawn above the 2.4° PPI with a vertical exaggeration of 2. The same procedure, applied to ZHZ_H of the 1410, 1420, and 1430 UTC volumes, was used to quantify the height and intensity of the reflectivity cores near the hook, and the same advection correction was applied to the WEH centers when assessing the tilt of the WEH (§ 5).

(iv) Radial Velocity Simulation Because the observed VRV_R is unreliable at the tornado scale, a simulation was performed for two purposes, namely (1) to show what a tornado consistent with the damage survey would have looked like had it been sampled by S-ILO, and (2) to constrain how strong a vortex could have been hidden by the processing. The simulated vortex follows the WEH tracked in the previous subsection. On each tilt of the 1420 UTC volume (particularly 2.4–5.3°), it was centered on the marked WEH position, so that the simulation tilts with height together with the observed reflectivity signature.

Two tangential wind profiles were used. The first is a Rankine vortex,

Vt(ρ)={VrotρRc,ρ≤Rc,VrotRcρ,ρ>Rc.(10)V_t(\rho) = \begin{cases} V_{\mathrm{rot}}\dfrac{\rho}{R_c}, & \rho\le R_c,\\ V_{\mathrm{rot}}\dfrac{R_c}{\rho}, & \rho> R_c. \end{cases} \tag*{(10)}

The second is a Burgers–Rott vortex, which has a smoother core,

Vt(ρ)=Vrot(1+12α)Rcρ[1−exp⁡(−αρ2Rc2)],α=1.2564,(11)V_t(\rho) = V_{\mathrm{rot}}\left(1+\frac{1}{2\alpha}\right)\frac{R_c}{\rho}\left[1-\exp\left(-\alpha\frac{\rho^2}{R_c^2}\right)\right], \qquad\alpha= 1.2564, \tag*{(11)}

where ρ\rho is the distance from the vortex center, RcR_c the core radius (radius of maximum wind) and VrotV_{\mathrm{rot}} is the peak tangential (rotational) speed. A core radius of Rc=100R_c = 100 m and Vrot=40V_{\mathrm{rot}} = 40 m s−1^{-1} were adopted (see below). The cyclonic tangential wind was projected onto the radar beam, VR=Vt(ρ)(k^×ρ^)⋅r^cos⁡θV_R = V_t(\rho) (\hat{\mathbf{k}} \times\hat{\boldsymbol{\rho}}) \cdot\hat{\mathbf{r}} \cos\theta. It was sampled through the effective radar resolution volume following Wood and Brown [119] and Brown et al. [17], using a two-way Gaussian beam with a 1.33° effective beamwidth, smeared by the 1° antenna rotation per radial, and a 250 m rectangular range gate, evaluated on 31 × 9 sub-beam samples per gate. The simulated VRV_R was added to the observed, quality-controlled velocity field without the processor filter, so the simulation represents what the radar would have displayed in unsmoothed data.

The choice of VrotV_{\mathrm{rot}} was constrained statistically with an injection–recovery test. Simulated vortices spanning Rc=50R_c = 50–400 m and Vrot=0V_{\mathrm{rot}} = 0–80 m s−1^{-1} were passed through the beam model and the fitted processor filter, and through a filter 1.5 times wider. They were added to the observed field at 40 random positions within 0.6 km of the WEH. Each injected field was scanned for the maximum cyclonic azimuthal velocity difference (ΔV\Delta V) within 1 km, using a gate-to-gate detector and a peak detector (radials ≤ 2 km apart).

Detection thresholds were set at the 95th percentile of the same detectors at 400 random storm-core locations (ZH≥35Z_H \ge35 dBZ, > 3 km from the WEH). The detection probability Pdet(Vrot,Rc)P_{\mathrm{det}}(V_{\mathrm{rot}},R_c) obtained from these trials was then combined with the damage rating in a Bayesian framework,

P(Vrot∣EF2, no detection, Rc)∝P(EF2∣Vrot)∏tilts[1−Pdet(Vrot,Rc)]P(Vrot),(12)P(V_{\mathrm{rot}} \mid\mathrm{EF2},\ \text{no detection},\ R_c) \propto P(\mathrm{EF2} \mid V_{\mathrm{rot}}) \prod_{\mathrm{tilts}} [1-P_{\mathrm{det}}(V_{\mathrm{rot}},R_c)]P(V_{\mathrm{rot}}), \tag*{(12)}

where P(EF2∣Vrot)P(\mathrm{EF2} \mid V_{\mathrm{rot}}) is the probability that the peak ground-relative wind, Vrot+cV_{\mathrm{rot}}+c, falls within the EF2 range (49.6–60.4 m s−1^{-1}), with the storm translation cc drawn from N(13.7,1.5)\mathcal{N}(13.7,1.5) m s−1^{-1}, the non-detection term treats the 2.4°, 3.4° and 4.3° tilts as independent looks, with equal weights on the fitted filter and the 1.5-times filter, and the prior P(Vrot)P(V_{\mathrm{rot}}) is uniform over 0–80 m s−1^{-1}. For Rc=100R_c = 100 m, the posterior favors Vrot≈35V_{\mathrm{rot}} \approx35–45 m s−1^{-1}, and Vrot=40V_{\mathrm{rot}} = 40 m s−1^{-1} was adopted for the simulation. The EF2 interval is used here because it is defined for a ground-relative gust, which is the quantity that Vrot+cV_{\mathrm{rot}}+c represents; it spans the central speeds of the IF1.5 and IF2 classes (50 and 60 m s−1^{-1}) assigned along the damage path (§ 5).

PlanetScope Observations

Remote sensing provides innovative technologies for assessing tornado debris (Hoque et al. [51]). These technologies have shown to be effective in damage assessment following disasters, as demonstrated by numerous research (Xie et al. [122]). Aerial images (Schaefer et al. [85]), a form of airborne remote sensing technology, are one important source of data for storm damage assessment. PlanetScope (PS) is a CubeSat 3U form factor (10 cm × 10 cm × 30 cm) satellite constellation operated by Planet Labs, Inc. and was utilized in this study. While in sun-synchronous orbit, these cubesats provide daily 4-band (RGB and NIR) imagery of the Earth at 3 m spatial resolution [78]. Currently in orbit are the Dove Classics, the improved Dove-Rs with more spectral and radiometric precision and sharpness, and the hyperspectral SuperDoves. Most images used in the study were daily Dove Classic imagery, or the newer generation Dove-R images (https://www.planet.com/). These have a Ground Sampling Distance (GSD) of 3–4 m at nadir and a positional accuracy of < 10 m RMSE. We used the level 3B Surface Reflectance (SR; ρ\rho) product, which is orthorectified and atmospherically corrected by Planet Labs with their in-house processing. These provide more consistency across time and location localized atmospheric conditions while minimizing uncertainty in the spectral response.

The high-resolution satellite images that were chosen for this analysis are nearly clear images (cloudless, hazeless, and smokeless) within the areas of interest for this analysis of before (05 August 2026) and after the event (11 August 2026), which is ±3 days only from the event (08 August 2026). Furthermore, we apply spatial masking using a sensor-agnostic deep learning model that segments clouds and cloud shadow. This is applied through OmniCloudMask (OCM; Wright et al. [120]), a python library that consists of two ensembled U-Net (Ronneberger et al. [83]) models. These constituent models leverage the use of Fastai Dynamic U-Net (Howard and Gugger [53]), which allows any CNN to be used as a backbone model. The models chosen for this purpose were the “RegnetY 004” and the “ConvNextV2 nano”, supplied with pre-trained weights sourced from the timm library (Wightman [118]). With the masked data, the PS satellite imagery can be used to detect the damage from a tornado event using Normalized Difference Vegetation Index (NDVI).

After the clouds were removed from the pre- and post-tornado PS imagery, NDVI was calculated for the pre-tornado (NDVIBEFORE_{\mathrm{BEFORE}}) and post-tornado (NDVIAFTER_{\mathrm{AFTER}}) scenes. These were then used to compute the percent reduction in NDVI between the two dates. Because it is normalized by the pre-event NDVI, the percent reduction accounts for the initial condition of the vegetation, so a given loss is weighted by how vegetated a pixel was beforehand. Earlier work instead used a simple NDVI difference i.e., Burow et al. [22] subtracted pre-event NDVI from post-event NDVI to produce an NDVI difference image for three tornadoes from a single outbreak, where vegetation conditions were comparable between images taken one week apart.

NDVI and its percent reduction were calculated as

NDVI=ρNIR−ρRedρNIR+ρRed,(13)\mathrm{NDVI} = \frac{\rho_{\mathrm{NIR}}-\rho_{\mathrm{Red}}}{\rho_{\mathrm{NIR}}+\rho_{\mathrm{Red}}}, \tag*{(13)}
ΔNDVI%=NDVIpre−NDVIpostmax⁡(NDVIpre,0.10)×100.(14)\Delta\mathrm{NDVI}_{\%} = \frac{\mathrm{NDVI}_{\mathrm{pre}}-\mathrm{NDVI}_{\mathrm{post}}}{\max\left(\mathrm{NDVI}_{\mathrm{pre}},0.10\right)} \times100. \tag*{(14)}

Automatic Weather Station

Surface observations were obtained from the AWS collocated with S-ILO (~4.8 m above mean sea level), which records temperature, relative humidity, station pressure, wind, and rainfall at 10-min intervals. The local time stamps were converted to UTC. The dewpoint, water vapor mixing ratio, virtual potential temperature (θv\theta_v), equivalent potential temperature (θe\theta_e), and SB LCL were derived with MetPy, with the LCL height obtained from the hypsometric equation over the dry-adiabatic subcloud layer. Perturbations of pressure (Δp′\Delta p') and θv\theta_v (Δθv\Delta\theta_v) were computed as departures from a centered 3-h running mean, which removes the diurnal cycle and the semidiurnal pressure tide while retaining mesoscale signals such as storm outflow.

The buoyancy of the near-surface air was computed as B=gΔθv/θv,refB = g\Delta\theta_v/\theta_{v,\mathrm{ref}}, where θv,ref=θv−Δθv\theta_{v,\mathrm{ref}} = \theta_v-\Delta\theta_v represents the environmental value. A cold pool was diagnosed where Δθv≤−0.5\Delta\theta_v \leq-0.5 K and Δp′>0\Delta p' > 0, and its density-current speed was estimated as C=(2Δp′/ρ)1/2C = (2\Delta p'/\rho)^{1/2} Benjamin [9], where ρ\rho is the air density.

Results

Overview of Synoptic Environment

The synoptic-scale environment from 1300 to 1500 UTC (2100–2300 LST) is shown in Figure 2. As of 0700 UTC (1500 LST), Typhoon Dolphin was located ~825 km north-northeast of extreme northern Luzon and outside the Philippine Area of Responsibility, so that only the southern periphery of its circulation appears along the northern edge of the analysis domain. Although the typhoon did not directly affect the country, it enhanced the southwest monsoon (locally known as Habagat), which had brought widespread rainfall over Luzon and the Visayas since early August. At 500 hPa (Figure 2aFigure 1–aFigure 3), a westerly wind maximum of 25–40 kt extended along 13°–16°N, whereas the flow over the Visayas was comparatively weak (10–20 kt). Geopotential heights over Panay Island remained near 588–589 dam and changed little during the analysis period. Thus, in contrast to the Candating, Arayat, tornado, which occurred under 500-hPa height falls induced by TC Ewiniar (Capuli et al. 2026), synoptic-scale mid-level forcing for ascent was weak over the area of interest, similar to the easterly-wave regime of the Camarines Norte outbreak (Capuli 2026).

Synoptic environment of the Philippine archipelago before, during, and after the event

Figure 2. Synoptic environment of the Philippine archipelago before (1; 13 UTC/21 LST), during (2; 14 UTC/22 LST), and after (3; 15 UTC/23 LST) the event. (a) 500-hPa geopotential heights (blue; dam) and winds (kt), with wind speeds > 10 kt shaded. (b) 700-hPa geopotential heights (blue; dam), precipitable water (shaded; kg m−2^{-2}), and winds (kt). (c) 850-hPa geopotential heights (blue; dam) and winds (kt), with wind speeds > 10 kt shaded.

ERA5 vertical cross sections showing convergence/divergence, streamwise vorticity, vertical velocity, and potential vorticity

Figure 3. ERA5 vertical cross section along 11.0°N (red dashed line in Figure 1) at 14 UTC/22 LST of (a) convergence/divergence (×10−4\times10^{-4} s−1^{-1}), with hatching where convergence is stronger than −0.50×10−4-0.50 \times10^{-4} s−1^{-1}; (b) streamwise vorticity (×10−2\times10^{-2} s−1^{-1}); (c) vertical velocity (Pa s−1^{-1}), with hatching where ascent is stronger than −0.50-0.50 Pa s−1^{-1}; and (d) potential vorticity (PV; shaded, PVU) with equivalent potential temperature (θe\theta_e; contours every 4 K). All panels show the 40 kt nocturnal wind maximum (green contour), the longitude of the tornado (black dashed line), the terrain (brown), and the tangential and normal wind components (barbs; kt). In (d), the bold 354-K θe\theta_e contour marks the low-level moisture tongue.

The monsoon flow was most evident in the lower troposphere. At 700 hPa (Figure 2bFigure 1–bFigure 3), 30–40 kt westerlies spanned the central Philippines, and precipitable water (PWAT) exceeded 65–75 kg m−2^{-2} along a moisture axis near 14°–17°N, while values of 55–62 kg m−2^{-2} covered the Visayas. At 850 hPa (Figure 2cFigure 1–cFigure 3), a broad band of 30–45 kt westerly to west-southwesterly flow extended from the South China Sea across Luzon and the Visayas toward the Philippine Sea, with 25–35 kt winds over Panay Island. This band acted as a large-scale low-level jet that sustained the advection of warm, moist marine air from the Sulu Sea and the Panay Gulf onto the southern coast of Panay, similar to the role of the southwest monsoon in the Candating, Arayat, case (Capuli et al. 2026). Because the westerlies at 850 and 700 hPa were considerably stronger than the flow at 500 hPa, the vertical wind shear over the Visayas was concentrated in the lowest ∼3 km, a configuration that shaped the hodographs discussed in § 5.

Figure 3 presents a zonal vertical cross section along 11.0°N (red dashed line in Figure 1), ∼30 km north of the damage path, at 1400 UTC (2200 LST). The most prominent feature is a nocturnal low-level jet (LLJ), i.e., a 40 kt westerly wind maximum between ∼1.5 and 3.3 km MSL over the sea west of Panay that descended eastward to ∼1.5–2.8 km over and to the lee of the central Panay mountain range (green contour). The jet coincided with the evening surge of the low-level winds retrieved by S-ILO (§ 5) and, together with the 25–35 kt flow at 850 hPa over Panay (Figure 2cFigure 2), formed the low-level branch of the typhoon-enhanced monsoon. Beneath the jet, the near-surface layer contained positive streamwise vorticity exceeding 1.5 × 10−2^{-2} s−1^{-1} in the lowest ∼0.5 km, with values above 0.5 × 10−2^{-2} s−1^{-1} extending to ∼2 km (Figure 3b), indicating that the horizontal vorticity generated by the strong shear beneath the jet was oriented largely along the storm-relative inflow. Low-level convergence and ascent of 1.0–2.0 Pa s−1^{-1} were concentrated on the windward (western) slopes of the mountain range between 121.8° and 122.2°E (Figure 3a and Figure 3c), consistent with the orographic lifting of the impinging monsoon flow. East of the ridge, near the longitude of the tornado, ERA5 instead depicted descent of 1.0–1.5 Pa s−1^{-1} between ∼1.5 and 4 km, overlain by a sloping layer of convergence at 3.5–5.5 km. This lee-side descent is likely a terrain-induced response of the coarsely resolved flow; because ERA5 cannot resolve individual convective storms, it should not be interpreted as an absence of ascent in the near-storm environment. The θe\theta_e field (Figure 3d) shows that the LLJ was underlain by a tongue of warm, moist air with θe≥354\theta_e \ge354 K below ∼1\sim1 km (bold contour), reaching 358 K near the surface west of Panay, and was overlain by a mid-tropospheric θe\theta_e minimum below 342 K between ∼2\sim2 and 7 km. Because the low-level flow was westerly to southwesterly, the jet and the flow beneath it carried this high-θe\theta_e marine air from the Sulu Sea and the Panay Gulf onshore toward Iloilo City, i.e., the LLJ acted as a conveyor of warm, moist air into the inflow region of the storm. The resulting decrease of θe\theta_e by ∼12\sim12–16 K with height indicates a potentially (convectively) unstable column that modest lifting could release. In addition, PV of 0.5–1.2 PVU was concentrated in the lowest ∼1\sim1 km immediately east of the central Panay range, near and upstream of the event location, whereas PV was generally <0.3< 0.3 PVU elsewhere in the lower troposphere. Low-level PV anomalies in the lee of mountains are commonly generated by frictional and diabatic processes as the flow passes over terrain (Aebischer and Schär [1]), and they imply locally enhanced low-level vorticity and/or static stability. At the resolution of ERA5, however, this signal is viewed as a qualitative indication that the terrain of Panay modified the low-level flow upstream of Iloilo City rather than as a quantitative estimate of the vorticity ingested by the storm.

These features indicate that the Iloilo tornado should be regarded as a nocturnal event not only in its timing, about 4 h after local sunset, but also in its environment. In the midlatitudes, nocturnal LLJs develop when the frictional coupling between the boundary layer and the flow above weakens after sunset, allowing the residual-layer wind to accelerate through an inertial oscillation (Blackadar [12]), aided by the diurnal cycle of heating over sloping terrain (Holton [50]; Shapiro et al. [86]). At ∼10.7∘\sim10.7^\circN, the inertial period is ∼64\sim64 h, so the inertial acceleration alone is slow. The evening surge over Panay is more plausibly explained by the frictional decoupling of an already strong, typhoon-enhanced monsoon flow, combined with the nocturnal reversal of the land–sea and slope circulations over the island. Whatever its cause, the jet increased the LLS and SRH after sunset while advecting high-θe\theta_e air beneath it, which is the combination that sustains supercells and favors tornadoes during the nocturnal transition in the midlatitudes (Mead and Thompson [76]; Gropp and Davenport [46]).

Mesoscale Environment

The mesoscale environment from 1300 to 1500 UTC (2100–2300 LST) is shown in Figure 4. CAPE of 2500–3500 J kg−1^{-1} covered the Sulu Sea, the Panay Gulf, and the Guimaras Strait (Figure 4aFigure 1–aFigure 3), whereas values over the interior of Panay decreased to 1000–2000 J kg−1^{-1} as the land surface cooled after sunset. Iloilo City, on the southeastern coast of Panay, lay along the gradient between these two regimes, with CAPE of ∼2000\sim2000–2500 J kg−1^{-1}. The persistence of large CAPE over the adjacent waters and along the coast reflects the continuous onshore advection of warm, moist marine air by the monsoon flow, which offset the nocturnal stabilization of the boundary layer over land. The ωs01\omega_{s01} reached 0.010–0.016 s−1^{-1} over Panay and the adjacent straits, overlapping the zone of large CAPE along the southern coast. The mean 1–6 km relative humidity (RH16_{16}) was 75%–85% over the Visayas (Figure 4bFigure 1–bFigure 3), with RH13_{13} exceeding 80%, and the 10 m winds were westerly to southwesterly at 10–15 kt. Such a moist free troposphere limits the dilution of updrafts by the entrainment of dry environmental air (Peters et al. [101]), which favors the maintenance of strong updrafts in an environment with only moderate CAPE.

Convective and kinematic environment of the Visayas before, during, and after the event

Figure 4. Convective and kinematic environment of the Visayas before (1; 13 UTC/21 LST), during (2; 14 UTC/22 LST), and after (3; 15 UTC/23 LST) the event. (a) CAPE (shaded; J kg−1^{-1}) and 0–1 km streamwise vorticity (blue dashed contours; s−1^{-1}). (b) Mean 1–6 km relative humidity (shaded; %), mean 1–3 km relative humidity (black dashed contours; %), and 10 m winds (kt). (c) SRH01_{01} (shaded; m2^{2} s−2^{-2}) and SRH03_{03} (black dashed contours; m2^{2} s−2^{-2}).

The SRH01_{01} (Figure 4cFigure 1–cFigure 3) exhibited a maximum exceeding 200 m2^2 s−2^{-2} over northern and central Panay and the Guimaras Strait, with SRH03_{03} of 240–270 m2^2 s−2^{-2}. Iloilo City lay along the southern edge of this maximum, where the SRH01_{01} was ∼120\sim120–170 m2^2 s−2^{-2}. The SRH maximum remained nearly stationary from 1300 to 1500 UTC, indicating that the kinematic environment was sustained by the persistent monsoon flow rather than by a transient mesoscale feature. Collectively, the overlap of moderate CAPE, a moist free troposphere, and enhanced low-level SRH and streamwise vorticity along the southern coast of Panay constituted an environment supportive of supercells capable of producing tornadoes.

The surface observations at the S-ILO AWS confirm that the coastal boundary layer did not stabilize appreciably after sunset (Figure 5). Between 1200 and 1420 UTC (2000–2220 LST), the temperature remained at 28.1–28.8 °C, the dewpoint at 22.7–24.5 °C, and θe\theta_e at ∼353\sim353–359 K under light southwesterly winds. In other words, the near-surface air retained its daytime warmth and moisture rather than cooling radiatively. This persistence reflects the continuous warm, moist advection from the adjacent seas by the onshore monsoon flow beneath the LLJ. Consequently, the storm could continue to ingest SB parcels with negligible CIN (Table 1), while the strengthening jet aloft lengthened the low-level hodograph. Supercells that persist into the evening over the Great Plains are similarly characterized by increasing SRH and only small increases in inhibition during the nocturnal transition (Gropp and Davenport [46]).

Observations from the automatic weather station

Figure 5. Observations from the automatic weather station (AWS) collocated with S-ILO from 12 to 16 UTC at 10-min intervals. (a) Temperature (°C; red line) and Dewpoint Temperature (°C; green dashed line), (b) Pressure Perturbation (hPa; black line), (c) Relative Humidity (%; blue line) with surface wind barbs, (d) Δθv\Delta\theta_v (K; red line), (e) Mixing Ratio (g kg−1^{-1}; green line) and LCL (m; brown dashed line), (f) θe\theta_e (K; purple) and Buoyancy (×10−2\times10^{-2} m s−2^{-2}). Red shading marks the reported tornado period (1420–1435 UTC), and blue shading marks the cold-pool period (Δθv≤−0.5\Delta\theta_v \le-0.5 K and Δp′>0\Delta p' > 0). Perturbations are departures from a centered 3-h running mean.

The thermodynamic and kinematic evolution at the point scale is shown in Figure 6 and summarized in Table 1. The 0000 UTC (0800 LST) Iloilo sounding (Figure 6aFigure 1) was nearly saturated from the surface to ∼300\sim300 hPa (RH13_{13} and RH16_{16} of ∼97%\sim97\%), with weak lapse rates (LR03_{03} of 5.17 °C km−1^{-1} and LR36_{36} of 5.57 °C km−1^{-1}), MUCAPE of 1260 J kg−1^{-1}, and an LCL of 910 m. This moist-adiabatic profile was typical of the rainy morning conditions within the monsoon, and the accompanying hodograph (Figure 6bFigure 1) was small, with an LLS of 5.2 m s−1^{-1} and an SRH01_{01} of only 40 m2^2 s−2^{-2}. By 1200 UTC (2000 LST), the Mactan sounding (Figure 6aFigure 2) displayed a markedly more unstable profile, with MUCAPE of 2532 J kg−1^{-1}, ECAPE of 1736 J kg−1^{-1}, SBCIN of −21-21 J kg−1^{-1}, an LCL of 622 m, and a steeper low-level lapse rate (LR03_{03} of 6.58 °C km−1^{-1}). A dry layer between ∼\sim450 and 350 hPa reduced RH16_{16} to ∼\sim82%$. Its hodograph (Figure 6bFigure 2) had LLS and MLS of 10.0 and 16.1 m s−1^{-1}, respectively, an SRH03_{03} of 247 m2^2 s−2^{-2}, and near-surface horizontal vorticity that was almost purely streamwise (ωes500\omega_{es500} of 99%). Although Mactan is $\sim$161 km east-southeast of S-ILO, it sampled the same monsoon airmass a few hours before the event.

Thermodynamic profiles and hodographs

Figure 6. Thermodynamic profiles and hodographs representing (1) the pre-convective environment from the PAGASA Iloilo radiosonde (WMO 98637) at 00 UTC/08 LST, (2) the environment ∼2\sim2 h before the event from the Mactan radiosonde (WMO 98646), ∼161\sim161 km east-southeast of S-ILO, at 12 UTC/20 LST, and (3) the ERA5 proximity profile at S-ILO at 14 UTC/22 LST. (a) Skew-T log-p diagrams of temperature (red; °C), dewpoint temperature (green; °C), and the non-entraining, irreversible mixed-phase MU parcel profile (blue dashed). (b) Storm-relative hodographs (m s−1^{-1}) relative to the 0–3 km Bunkers right-moving storm motion (RM), with height labels in km AGL, the left-moving (LM) and mean-wind (MW) motions, the deviant tornado motion (DTM), and the SRH03 (blue shading).

The ERA5 proximity profile at 1400 UTC (2200 LST; Figure 6aFigure 3), which is closest in time and space to the event, depicted MUCAPE of 2014 J kg−1^{-1}, ECAPE of 1632 J kg−1^{-1}, no CIN, and an LCL and LFC of only 295 m. The ratio of ECAPE to MUCAPE (∼\sim0.81) indicates that entrainment would have diluted the updraft only modestly, owing to the moist free troposphere (RH13_{13} of 85% and RH16_{16} of 82%) and the strong storm-relative inflow (mean VSRV_{\mathrm{SR}} of 11.9 m s−1^{-1}). The hodograph (Figure 6bFigure 3) exhibited pronounced clockwise curvature in the lowest 3 km, with LLS and MLS of 10.5 and 17.6 m s−1^{-1}, respectively. The SRH500_{500}, SRH01_{01}, and SRH03_{03} were 79, 127, and 259 m2^2 s−2^{-2}, respectively, and the streamwiseness was ∼\sim98%–99% in both the 0–500 m and 0–1 km layers. Above ∼3\sim3 km, however, the winds weakened and backed with height, so that the hodograph folded back toward its low-level segment. The DLS was therefore only 2.5 m s−1^{-1}, and WMAXSHEAR was only 156 m2^{2} s−2^{-2}. This DLS is far below the ∼20\sim20 m s−1^{-1} commonly associated with supercells (e.g., Thompson et al. [110]). Nevertheless, the effective-layer formulations of the SCP (7.44) and STP (2.28), which measure the shear over the lower half of the MU parcel’s depth rather than the fixed 0–6 km layer, remained large. The first-order estimate of low-level vertical vorticity from the tilting of streamwise vorticity (Eq. (3)) yielded ζLLM\zeta_{\mathrm{LLM}} of 0.015 s−1^{-1}, comparable to that at Mactan (0.016 s−1^{-1}) and nearly twice that at Iloilo at 0000 UTC (0.008 s−1^{-1}).

Overall, the three profiles describe a transition from a saturated, weakly unstable morning environment to an evening environment with moderate instability, very low LCLs, and strong, highly streamwise LLS. The combination of low LCLs (< 300 m in ERA5), negligible CIN, and large low-level SRH and streamwise vorticity is characteristic of environments that support significant tornadoes (Thompson et al. [110]; Coffer et al. [32, 33]), whereas the weak DLS is atypical of such environments in the midlatitudes. As shown in § 5, the radar-derived wind profile further indicates that ERA5 underestimated the LLS in the near-storm environment.

Storm Evolution and Life Cycle

Figure 7 shows the evolution of the tornado-producing storm in 10-min Himawari-9 B13 imagery from 1400 to 1500 UTC (2200–2300 LST). Imagery before 1400 UTC is not shown, but the earlier stages of the storm’s life cycle are described below for context.

Himawari-9 AHI 10.4-$\mu$m brightness temperature images at six times

Figure 7. Himawari-9 AHI 10.4-μ\mum (B13) brightness temperature depicting the evolution of the 08 August 2026 tornadic storm at (a) 1400, (b) 1410, (c) 1420, (d) 1430, (e) 1450, and (f) 1500 UTC (2200–2300 LST). Brightness temperatures colder than −40 ∘C-40\,^\circ\mathrm{C} are color-shaded. The 1440 UTC scan was unavailable.

At 1200 UTC (2000 LST; not shown in Figure 7), the convective environment over Western Visayas was characterized by a warm, moist, and conditionally unstable airmass within the typhoon-enhanced southwest monsoon, as sampled by the Mactan sounding (§ 5). This is consistent with Heavy Rainfall Warning No. 15, issued at 1200 UTC by the DOST-PAGASA Visayas PAGASA Regional Services Division (PAGASA-VisPRSD), which reported light to moderate rains with occasional heavy rains over Western Visayas, including Iloilo. As convection persisted over the region in the Himawari-9 imagery, PAGASA-VisPRSD issued a thunderstorm watch at 1400 UTC (2200 LST) for possible thunderstorm activity over parts of Western Visayas, including Negros and Iloilo.

By 1410 UTC (2210 LST), the parent cell was already evident in S-ILO reflectivity west-southwest of the radar, having approached from ∼21.5 km at 1400 UTC to ∼14.2 km from S-ILO at 1410 UTC, yet B13 depicted only a small, relatively warm cloud top along the southern coast of Panay (Figure 7b). The first cloud top colder than −40 °C appeared at 1420 UTC (Figure 7c), with BTs of ∼−55 °C over the southern coast of Panay near 10.7°N, 122.4°E, indicating that the updraft had by then penetrated the upper troposphere. At 1425 UTC (2225 LST), PAGASA-VisPRSD issued a thunderstorm advisory reporting that moderate to heavy rain showers with lightning and strong winds were being experienced over Pavia, Iloilo City, and Oton. The advisory added that these conditions could also affect Dumangas, Zarraga, Leganes, and Santa Barbara in Iloilo, and Buenavista and Jordan in Guimaras. This advisory was issued only minutes before the tornado damage began in Mandurriao (§ 5).

At 1430 UTC (Figure 7d), when the tornado was producing damage in Mandurriao and Jaro, the tornadic storm occupied the eastern tip of a cloud area colder than −40-40 °C that extended ∼25\sim25–30 km west-southwestward along the southern coast of Panay, with a BT of only ∼−45\sim-45 to −55-55 °C over Iloilo City. By 1450 and 1500 UTC (Figure 7e and Figure 7f), the tornadic storm appeared as a small, separate cloud top of ∼−45\sim-45 to −50-50 °C over Iloilo City and the northern Guimaras Strait, east of a much larger and colder cloud shield that had developed over southern Panay and expanded toward the south and southwest over the Panay Gulf, reaching a minimum BT of ∼−68\sim-68 °C ∼25\sim25 km west of Iloilo City at 1450 UTC. In the Mactan sounding, BTs of −45-45 to −55-55 °C correspond to cloud tops at ∼12\sim12–13 km, whereas −68-68 °C corresponds to ∼15\sim15 km and the cold-point tropopause lay

at ∼16.7\sim16.7 km (−82.5-82.5 °C). The tornadic storm was therefore low-topped relative to the surrounding deep convection, with cloud tops ∼2\sim2–3 km lower than those of the adjacent cloud shield and ∼4\sim4 km below the tropopause, and no overshooting top was evident. The southwestward expansion of the larger cloud shield is consistent with the northeasterly to easterly winds of ∼6\sim6–15 m s−1^{-1} between 200 and 150 hPa in the Mactan sounding. Low-topped tornadic supercells are common in tropical environments with moderate CAPE, weak DLS, and strong LLS, such as those of landfalling TCs ([74]), and the shallow cloud top of the Iloilo storm agrees with the moderate CAPE and the low-level concentration of the shear in its environment. Because the updraft occupied only a few ∼2\sim2 km pixels, its BT may also have been warmed by sub-pixel effects, so these cloud-top heights are approximate. The tornado was thus produced by a relatively shallow storm whose cloud top did not stand out in the satellite imagery against the colder convection nearby, which underscores the importance of radar for detecting such storms at night. The radar evolution of the storm is shown in Figure 8 and Figure 9, which compile the upper tilts of the 1410 UTC volume and the 1420 and 1430 UTC volumes into a time sequence with successive tilts ∼30\sim30 s apart. Between 1400 and 1430 UTC, the centroid of the ≥45\ge45 dBZ core moved from ∼21.5\sim21.5 km west-southwest of S-ILO to within ∼4\sim4 km of the radar, consistent with the observed storm motion of 13.7 m s−1^{-1} toward 076∘076^\circ. The upper tilts of the 1410 UTC volume (Figure 8a–Figure 8f), which sampled the tornadic storm between ∼2.1\sim2.1 and 4.7 km above radar level (ARL), show a core exceeding 50–55 dBZ near 10.69∘N10.69^\circ\mathrm{N}, 122.49∘E122.49^\circ\mathrm{E} at every tilt, indicating that large hydrometeors were sustained well above the cloud base by a strong updraft. At 14.0∘14.0^\circ and 16.7∘16.7^\circ (Figure 8d and Figure 8e), this core was curled into a comma shape with a notch of weaker echo along its eastern flank, a configuration consistent with a rotating updraft at midlevels.

Time sequence of Reflectivity (dBZ) PPIs within ∼15 km of the S-ILO radar

Figure 8. Time sequence of Reflectivity (dBZ) PPIs within ∼15 km of the S-ILO radar, 14:15:15–14:34:05 UTC (22:15–22:34 LST), compiled from the 14:10, 14:20 and 14:30 UTC volume scans. Panels are in chronological order (A–T), with successive tilts ∼30 s apart: (A–F) 8.7–19.5° from the 14:10 UTC volume, (G–O) 1.5–10.0° from the 14:20 UTC volume and (P–T) 2.4–6.2° from the 14:30 UTC volume. Each panel is labeled with its sweep time (UTC) and elevation angle. The red dot marks the location of S-ILO.

Figure 9. Similar to Figure 8, but for Differential Reflectivity (dB) PPIs within ∼15 km of the S-ILO radar, 14:15:15–14:33:35 UTC (22:15–22:33 LST), compiled from the 14:10, 14:20 and 14:30 UTC volume scans. Panels are in chronological order (A–T), with successive tilts ∼30 s apart: (A–F) 8.7–19.5° from the 14:10 UTC volume, (G–O) 1.5–10.0° from the 14:20 UTC volume and (P–T) 1.5–5.3° from the 14:30 UTC volume. Each panel is labeled with its sweep time (UTC) and elevation angle.

By the 1420 UTC volume (Figure 8g–Figure 8o), the storm had acquired a classic low-level supercell structure. At the lowest usable tilts (1.5∘1.5^\circ–4.3∘4.3^\circ, ∼0.3\sim0.3–0.80.8 km ARL), the core exceeded 55 dBZ, and a hook-shaped appendage developed along its southwestern flank, with a WEH at the tip of the hook (Figure 10a, Figure 10e, Figure 10i, and Figure 10m). The WEH could be followed from 2.4∘2.4^\circ to 10.0∘10.0^\circ (0.480.48–1.421.42 km ARL) and was enclosed by a ring of 50–55 dBZ, the reflectivity structure commonly described as a debris ball. Its center shifted ∼3.5\sim3.5 km toward the east-northeast between the 2.4∘2.4^\circ and 10.0∘10.0^\circ scans, which were collected ∼3.6\sim3.6 min apart. When each tilt is shifted back to the 2.4∘2.4^\circ scan time using the storm motion, however, the WEH centers from 2.4∘2.4^\circ to 10.0∘10.0^\circ lie within ∼0.5\sim0.5 km of one another (x≈−9.4x \approx-9.4 to −8.8-8.8 km and y≈−6.4y \approx-6.4 to −6.1-6.1 km relative to S-ILO), indicating that the WEH column was nearly vertical in the lowest ∼1.4\sim1.4 km and that the apparent tilt was produced by the translation of the storm during the volume scan. At 6.2°–10.0° (∼0.8–1.4 km ARL; Figure 8l–Figure 8o), a weak echo vault bounded on its southern side by a curved band of high reflectivity was evident near 10.71°N, 122.52°E, indicating a bounded weak echo region (BWER) above the hook where the inflow ascended into the updraft. In the 1430 UTC volume (Figure 8p–Figure 8t), the storm core had reached the near range of the radar, the hook and WEH were located over Jaro at 0.24–0.43 km ARL, and a band of moderate reflectivity trailed to the south-southwest along the rear flank.

Four-by-four panel plot of S-ILO observations and simulated vortex signatures

Figure 10. 1420 UTC S-ILO observations (ZHZ_{\mathrm{H}} and VRV_{\mathrm{R}}) and simulated vortex signatures at the hook echo and weak echo hole of the Jaro tornadic supercell at elevation angles of 2.4∘2.4^\circ, 3.4∘3.4^\circ, 4.3∘4.3^\circ and 5.3∘5.3^\circ. Columns: (A, E, I, M) ZHZ_{\mathrm{H}} (dBZ), (B, F, J, N) observed VRV_{\mathrm{R}} (m s−1^{-1}), (C, G, K, O) observed VRV_{\mathrm{R}} plus a simulated cyclonic vortex with a Rankine profile, and (D, H, L, P) a Burgers–Rott profile. They are sampled with the radar’s beam (1.33∘1.33^\circ, 1∘1^\circ azimuthal sampling) and 250 m gate spacing. The simulated vortices have Rc=100R_{\mathrm{c}} = 100 m and Vrot=40V_{\mathrm{rot}} = 40 m s−1^{-1} and are centered on the WEH (dashed circles); xx and yy are the distances (km) east and north of S-ILO.

The ZDRZ_{\mathrm{DR}} fields (Figure 9) reveal several polarimetric features of the supercell. In the 1410 UTC volume (Figure 9a–Figure 9f), ZDRZ_{\mathrm{DR}} of 1.5–3 dB extended through the ∼2–4.7 km layer within and adjacent to the reflectivity core, indicating large, oblate raindrops lofted by the updraft to just below the environmental 0 °C level (∼5.1 km MSL in the Mactan sounding). Whether this enhancement continued above the melting level as a ZDRZ_{\mathrm{DR}} column could not be determined because the highest tilt (19.5°) reached only ∼4.7 km at that range. In the 1420 UTC volume, an arc of ZDRZ_{\mathrm{DR}} of 2.5–3.5 dB lined the western and southwestern reflectivity gradient adjacent to the hook echo at 2.4°–5.3° (Figure 9h–Figure 9k). This ZDRZ_{\mathrm{DR}} arc is attributed to the size sorting of hydrometeors by strong storm-relative winds and low-level helicity, which concentrates large drops along the inflow side of the forward-flank precipitation [62, 63]. At 6.2°–8.7° (Figure 9l–Figure 9n), enhanced ZDRZ_{\mathrm{DR}} of 2–3 dB also surrounded the weak echo vault identified in the reflectivity field, forming a ZDRZ_{\mathrm{DR}} ring around the BWER that marks the periphery of the updraft [62]. In the 1430 UTC volume (Figure 9p–Figure 9t), ZDRZ_{\mathrm{DR}} near the storm core was mostly < 1.5 dB, and the near-range sampling did not reveal a coherent arc.

The vertical structure of the WEH at 1420 UTC is shown in the cross sections of Figure 11. In the south–north section along 122.499°E (Figure 11a–Figure 11c), a ≥ 45 dBZ column extended from the surface to ∼1 km ARL near 10.72°–10.74°N and tilted slightly northward with height, while missing and low-reflectivity gates occupied the area immediately to its south (10.70°–10.715°N) below ∼0.6 km, i.e., at the location of the WEH. ZDRZ_{\mathrm{DR}} exceeded 3 dB at the base of the high-reflectivity column and was 1.5–2 dB through the lowest ∼0.75 km. In the west–east section along 10.717°N (Figure 11d–Figure 11f), the ≥ 45 dBZ core sloped eastward with height from ∼122.50°E near the surface to ∼122.53°E at 1.25 km ARL, and a ZDRZ_{\mathrm{DR}} maximum exceeding 3 dB occupied the western (inflow) edge of the core between 0.3 and 0.65 km ARL. In both sections, ρHV\rho_{\mathrm{HV}} decreased to ∼0.85–0.93 in the lowest ∼0.6 km around the WEH, compared with >0.97> 0.97 in the surrounding rain. Such reductions may arise from a mixture of large drops and hail, from non-uniform beam filling across the strong reflectivity gradients of the hook, or from small amounts of debris mixed with rain. Because all gates with ρHV<0.80\rho_{\mathrm{HV}} < 0.80 had been removed by the quality control applied

Vertical cross-sections of the Jaro tornadic supercell

Figure 11. Vertical cross-sections of the Jaro tornadic supercell at 2220 LST (1420 UTC), 08 August 2026, along the two transects described in § 3§ 3(iii): (a–c) south–north from 10.63°N, 122.50°E to 10.81°N, 122.50°E, and (d–f) west–east from 10.71°N, 122.44°E to 10.71°N, 122.61°E. (a, d) Reflectivity (dBZ), (b, e) Differential Reflectivity (dB), and (c, f) Cross-Correlation Coefficient (dimensionless). For Panels A and D, solid contours mark reflectivity at 30, 45 and 50 dBZ, while differential reflectivity (b and e) at 1.5 and 3 dB; dashed contours in panels C and F demarcate the reflectivity at 30, 40 and 50 dBZ.

for rainfall estimation (§ 3§ 3(ii)), the absence of lower values cannot be used to exclude a TDS. Likewise, the missing gates at the center of the WEH are consistent both with genuinely weak echo in a hydrometeor-depleted vortex core and with debris gates removed by the threshold.

The three-dimensional structure of the storm in the 1420 UTC volume is shown in Figure 12, in which the tilts were shifted to the 2.4∘^\circ scan time and objectively analyzed onto a Cartesian grid (§ 3§ 3(iii)). The 45 dBZ isosurface encloses the main core of the storm from near the surface to ∼\sim2.5–3 km ARL, above which the isosurfaces are truncated because the highest tilt (19.5∘^\circ) reached only ∼\sim3.9 km at this range. Additional cells to the northwest and east of the tornadic storm are also evident. The most conspicuous feature is a compact core exceeding 55 dBZ (maximum of ∼\sim60 dBZ) between ∼\sim1.6 and 2.6 km ARL on the western (rear) flank of the storm, centered ∼\sim0.4 km north of the WEH column at x≈−9.0x \approx-9.0 km and y≈−5.8y \approx-5.8 km, which is enclosed by the large 50 dBZ (red) isosurface on the western flank of the storm in Figure 12. Directly beneath this core, the reflectivity decreased to ∼\sim15–35 dBZ within the WEH below ∼\sim1.2 km, while a ring of 50–52 dBZ enclosed the WEH on its western and eastern sides between ∼\sim0.2 and 1.0 km ARL, as also seen in the vertical cross sections through the WEH (Figure 11). Thus, the WEH and the surrounding debris-ball ring were capped by an intense, elevated core. Because the same reflectivity structure can be produced by hydrometeors, the debris interpretation of the ring requires polarimetric confirmation ([84, 13]), which the quality control applied to the data precluded.

Three-dimensional reflectivity structure of the Jaro tornadic supercell

Figure 12. Three-dimensional reflectivity structure of the Jaro tornadic supercell in the 1420 UTC (2220 LST) S-ILO volume, viewed from the southwest. Isosurfaces of ZHZ_\mathrm{H} at 30 (blue), 35 (green), 40 (yellow), 45 (orange), and 50 dBZ (red) are drawn above the 2.4∘^\circ PPI (floor; dBZ). Each tilt was shifted to the 2.4∘^\circ scan time using the storm motion (13.7 m s−1^{-1} toward 076∘^\circ), objectively analyzed onto a 0.25 km ×\times 0.25 km ×\times 0.2 km grid, and drawn with a vertical exaggeration of 2; xx and yy are the distances (km) east and north of S-ILO.

The location of the elevated core, on the rear flank of the storm and immediately above the hook and WEH, resembles the descending reflectivity core (DRC) documented by Rasmussen et al. [82], a reflectivity protuberance pendant from the rear-flank echo overhang at 3–6 km that descends into the hook appendage, often before tornadogenesis. The preceding volume supports such a descent. At 1410 UTC, in the storm-relative position corresponding to the hook at 1420 UTC, ≥\geq 55 dBZ echo occupied ∼\sim2.4–5.4 km ARL, while the reflectivity below ∼\sim1.4 km did not exceed 44 dBZ and no hook was present. By 1420 UTC, the core had lowered to ∼\sim1.6–2.6 km, and the hook, WEH, and debris-ball ring had formed beneath it. By 1430 UTC, when the tornado was producing damage in Jaro, the hook region below ∼\sim2.2 km contained 46–51 dBZ without a distinct elevated maximum, consistent with the core having descended and spread through the hook, although the proximity of the storm to the radar limited the sampled depth to ∼\sim2 km. This sequence, a lowering of the rear-flank core between 1410 and 1420 UTC followed by the onset of damage ∼\sim10 min later, is consistent with the timing reported by Kennedy et al. [57], who found that 59% of 64 supercells produced DRCs and that 30% of these DRCs occurred from 10 min before to 5 min after tornadogenesis. The limitations of this inference are discussed in § 6.

Radar-Derived Near-Storm Kinematics and Vortex Simulation

The VVP-derived wind profile from 2100 to 2300 LST (1300–1500 UTC; Figure 13a) documents the evolution of the low-level flow upstream of the storm. Before ∼\sim2130 LST, the winds in the lowest 3 km were southwesterly at ∼\sim10–15 m s−1^{-1}. From ∼\sim2130 to 2230 LST, the winds between ∼\sim1 and 3 km AGL strengthened to 20–25 m s−1^{-1}, whereas the 10 m wind measured at the S-ILO AWS remained only ∼\sim2–5 m s−1^{-1}, so that the LLS increased as the storm approached Iloilo City. This intensification is consistent with the nocturnal wind maximum depicted by ERA5 near 1.5–3 km (Figure 3) and resembles the evening strengthening of the LLJ that enhances LLS and SRH in nocturnal tornado environments elsewhere ([61, 76]).

VVP-derived wind profile and radar-derived hodographs

Figure 13. (a) VVP-derived wind profile (barbs colored by horizontal wind speed; m s−1^{-1}) from 2100 to 2300 LST (1300–1500 UTC) derived from S-ILO. (b–d) Radar-derived hodographs at 1410, 1420, and 1430 UTC (2210, 2220, and 2230 LST; black dashed lines in panel a), plotted relative to the observed storm motion (OBS; 13.7 m s−1^{-1} toward 076°), with height labels in km AGL. The B2KRM, Bunkers left-moving (LM), and mean-wind (MW) motions, the deviant tornado motion (DTM), and the SRH03 (blue shading) are also shown.

The radar-derived hodographs at 1410 and 1420 UTC (Figure 13b and Figure 13c; Table 3), plotted relative to the observed storm motion, exhibited a long, clockwise-curved 0–1 km segment, with an LLS of 16.1–16.3 m s−1^{-1} (∼\sim31 kt) and a mean VSRV_{\mathrm{SR}} of 14.5–14.7 m s−1^{-1}. Relative to the observed motion, the SRH500_{500}, SRH01_{01}, and SRH03_{03} were 115, 211, and 292 m2^2 s−2^{-2} at 1410 UTC and 115, 124, and 265 m2^2 s−2^{-2} at 1420 UTC, respectively, with an ω500\omega_{500} of ∼\sim0.019 s−1^{-1} and a streamwiseness of 74%–76%. Critical angles of 48°–60°, computed against the B2KRM_{\mathrm{RM}}, indicate a substantial streamwise component of the near-ground vorticity, although less than the ∼\sim90° that corresponds to purely streamwise vorticity ([39]). These values exceed their ERA5 counterparts at 1400 UTC (LLS of 10.5 m s−1^{-1} and SRH01_{01} of 127 m2^2 s−2^{-2}), confirming that ERA5 underestimated the LLS near the storm, as is common for reanalyses within the boundary layer and near surface boundaries ([59]). Near-ground SRH and streamwise vorticity of this magnitude are among the best discriminators between tornadic and nontornadic supercell environments in the United States ([32, 33]).

The B2KRM_{\mathrm{RM}} computed from the 0–3 km layer (13.5–13.8 m s−1^{-1} from 230°–234°) closely matched the observed speed but was ∼\sim20° to the left of the observed motion (13.7 m s−1^{-1} toward 076°), i.e., the storm moved to the right of the predicted supercell motion by 5–6 m s−1^{-1}. A rightward deviation of this kind increases the storm-relative inflow and SRH available to the storm and has been associated with tornadic supercells ([21]). In this case, the SRH01_{01} at 1410 UTC increased from 158 m2^2 s−2^{-2} relative to the B2KRM_{\mathrm{RM}} to 211 m2^2 s−2^{-2} relative to the observed motion. By 1430 UTC (Figure 13d), the SRH01_{01} decreased to 73 m2^2 s−2^{-2} and the 0–500 m streamwiseness to 55%. This hodograph, however, does not represent the inflow environment, because the storm core was then within ∼4\sim4 km of the radar and the AWS wind used as its lowest level had been modified by the storm’s outflow (§ 5). The retrieved SRH values carry an uncertainty of approximately ±30\pm30–50 m2^2 s−2^{-2}.

Despite this favorable kinematic environment, the observed radial velocity at the WEH did not resolve a tornado-scale couplet. At 1420 UTC, the sign-corrected radial velocity around the WEH was smooth and predominantly inbound at ∼\sim10–15 m s−1^{-1} from 2.4° to 5.3° (Figure 10b, Figure 10f, Figure 10j, and Figure 10n). The storm-relative velocity showed the correct cyclonic sense, with outbound flow on the north-western side and inbound flow on the southeastern side of a broad shear zone, but the largest cyclonic azimuthal velocity difference within 1 km of the WEH was only 4.0, 4.5, and 4.8 m s−1^{-1} at 2.4°, 3.4°, and 4.3°, respectively, and the gate-to-gate difference was only 1.0–1.5 m s−1^{-1}.

These values are well below the detection thresholds derived from the storm core (7.9–12.2 and 4.7–6.3 m s−1^{-1}, respectively) and lie between the 10th and 45th percentiles of the storm-wide distribution. At 1430 UTC, the cyclonic shear strengthened aloft (ΔV\Delta V of ∼\sim17 m s−1^{-1} at 5.3°), but the 2.4° circle enclosing the WEH contained no valid velocity gates.

Figure 10c, Figure 10g, Figure 10k, and Figure 10o show the same observations with a cyclonic Rankine vortex (Rc=100R_c = 100 m, Vrot=40V_{\mathrm{rot}} = 40 m s−1^{-1}) superimposed at the WEH, and Figure 10d, Figure 10h, Figure 10l, and Figure 10p show the equivalent Burgers–Rott vortex. Sampled through the S-ILO beam without the processor filter, either vortex would have produced a compact couplet of ∼\sim30–40 m s−1^{-1} across adjacent gates at every tilt from 2.4° to 5.3°, centered within the WEH and readily identifiable as a TVS (Brown et al. [16]). The Burgers–Rott profile, with its smoother core and broader outer flow, yields a couplet of similar magnitude that extends over a few more gates. Hence, a tornado consistent with the surveyed damage would have been conspicuous in unfiltered data, and its absence from the archived field is attributable mainly to the processing rather than to the absence of a vortex.

The injection–recovery test quantifies this attribution. With the fitted processor filter, vortices with Rc≥150R_\mathrm{c} \ge150 m and Vrot≥20V_\mathrm{rot} \ge20 m s−1^{-1} would have been detected at ≥90%\ge90\% of the injected positions and are therefore excluded, whereas narrower vortices could have remained hidden up to VrotV_\mathrm{rot} of ∼30\sim30 m s−1^{-1} for Rc=100R_\mathrm{c} = 100 m, ∼40\sim40 m s−1^{-1} for Rc=75R_\mathrm{c} = 75 m, and ∼60\sim60 m s−1^{-1} for Rc=50R_\mathrm{c} = 50 m. In unfiltered data, the corresponding detection limits would have been only 12, 16, and 24 m s−1^{-1}, respectively. Combined with the damage rating (Eq. (12)), the posterior for Rc=100R_\mathrm{c} = 100 m favors Vrot≈35V_\mathrm{rot} \approx35–45 m s−1^{-1}, i.e., the lower end of the range permitted by the damage, because larger values become increasingly difficult to conceal. The width of the damage path (§ 5) independently constrains RcR_\mathrm{c} to ∼30\sim30–115 m, so that the radar non-detection and the damage survey are mutually consistent with a narrow tornado of moderate rotational velocity translating at ∼14\sim14 m s−1^{-1}.

Surface Observations of the Storm Outflow

The S-ILO AWS, located ∼5\sim5–7 km north to north-northeast of the damage path, recorded the pre-storm and outflow conditions (Figure 5). From 1200 to 1420 UTC (2000–2220 LST), the surface air was warm and humid, with temperatures of 28.1–28.8 °C, dewpoints of 22.7–24.5 °C, relative humidity of 70%–79%, mixing ratios of 17.5–19.6 g kg−1^{-1}, and θe\theta_\mathrm{e} of ∼353\sim353–359 K. The corresponding SB LCL ranged from ∼520\sim520 to ∼760\sim760 m, higher than the ∼300\sim300 m depicted by ERA5 but still low. Light southwesterly winds of 2–3 m s−1^{-1} prevailed, consistent with the onshore monsoon flow.

The outflow arrived at 1430 UTC (2230 LST), coincident with the passage of the storm core ∼4\sim4 km south of the radar. Between 1420 and 1440 UTC, the temperature fell by 2.4 K (from 28.3 to 26.0 °C), the dewpoint decreased from 23.8 to 22.0 °C, and θe\theta_e decreased by ∼7.8 K (from 356.3 to 348.4 K). The pressure perturbation peaked at +0.64 hPa at 1430 UTC, the virtual potential temperature perturbation reached −2.1 K at 1440 UTC, and the corresponding buoyancy was −0.068 m s−2^{-2}. The 10-min mean wind increased to 4.7 m s−1^{-1} from 204°, and no rainfall was recorded by the gauge during the event. The cold pool persisted from 1430 to 1450 UTC, after which the temperature and θe\theta_e recovered and the LCL lowered to ∼380–470 m as the surface air moistened. The density-current speed estimated from the pressure and density perturbations was ∼8–11 m s−1^{-1}, slower than the storm motion of 13.7 m s−1^{-1}.

The modest θv\theta_v deficit (∼1–2 K) and the small negative buoyancy indicate that the outflow sampled by the AWS was relatively warm and only weakly negatively buoyant. Markowski et al. [73] found that the rear-flank outflow of tornadic supercells, particularly of those producing significant tornadoes, tended to have smaller θv\theta_v and θe\theta_e deficits than that of nontornadic supercells, because weakly negatively buoyant air can be more readily lifted and stretched by the low-level updraft. Because the AWS sampled the outflow ∼4 km from the storm core and ∼5–7 km from the tornado path, it is not possible to determine whether this air originated from the rear-flank or the forward-flank downdraft. Nevertheless, the observations indicate that the near-surface outflow of the Iloilo storm was not strongly negatively buoyant, consistent with a moist, low-LCL environment in which the evaporative cooling of rain is limited.

Damage Analysis

A ground damage survey was conducted along the affected areas of Mandurriao and Jaro and was supplemented by photographs, videos, and CCTV footage archived by Project SWAP. A total of 31 damage indicators (DIs) were documented, of which 29 were geolocated (Figure 14 and Figure 15). The DIs were rated using the International Fujita (IF) Scale (Groenemeijer et al. [44]), which rates intensity using the instantaneous three-dimensional wind at the height of the damage, admits half-step ratings, and contains DIs that are not calibrated exclusively to construction practices in the United States. The documented DIs included trees of the strong, average, and weak strength classes (TR-S, TR-A, and TR-W), tree stands (TS), building roofs and structures of various sturdiness classes (BR and BS), non-structural building elements (BN), poles (PT), fences (FC), signs (SN), and carports (CP). Of the 31 DIs, 4 were rated IF2, 12 were rated IF1.5, 12 were rated IF1, and 3 were rated IF0.5. Several tree ratings in the IF Scale are open lower bounds (e.g., the uprooting of a strong tree, TR-S DoD 3, is rated ≥ IF1.5), so that the tabulated value represents the minimum wind consistent with the observed failure.

Representative damage from the Iloilo tornado

Figure 14. Representative damage from the Iloilo tornado, ordered left to right and top to bottom: (a) a large tree toppled across a road; (b) a snapped hardwood trunk; (c) an uprooted tree with its root plate exposed; (d) collapsed bamboo-framed stalls and scattered roofing; (e) a snapped concrete utility pole with downed lines and roofing sheets; and (f) snapped and uprooted large trees adjacent to a building.

Map of the tornado damage path and survey points

Figure 15. (a) Map of the tornado damage path and survey points (white circles, numbered as in the damage inventory). Shading denotes the IF rating of the modeled damage swath, and the dotted line is the kernel-smoothed damage centerline. Purple diamonds and dashed circles mark the centers and enclosing circles of the weak echo hole (ZHZ_{\mathrm{H}} debris ball) tracked at 2.4°–10.0° in the 2220 LST volume and at 2.4°–5.3° in the 2230 LST volume, and the red circle marks S-ILO. (b, c) Enlarged sections of the path near (b) the Iloilo Business Park, Mandurriao, and (c) Jaro Plaza to Cubay, Jaro. Barangay boundaries are shown in gray.

The four IF2 indicators occurred in two clusters. Near the Iloilo Business Park in Mandurriao, a utility pole (PT-S, DoD 2; IF2, 60 m s−1^{-1}), a strong tree with a snapped trunk (TR-S, DoD 5; ≥ IF2), and a building of sturdiness class AB (BS-AB, DoD 2; ≥ IF2) reached IF2. Near Jaro Plaza, a sign at the gate of the Jaro Metropolitan Cathedral (SN-M, DoD 2; ≥ IF2) also reached IF2. Neighboring indicators in both areas, including uprooted strong trees and damaged roofs, were rated IF1.5, while the eastern portion of the path near Benedicto and Cubay was dominated by IF1 damage to trees, tree stands, fences, and building structures, with an uprooted strong tree (IF1.5) near Benedicto. The tornado is therefore rated IF2 on the IF Scale, which corresponds approximately to EF2 on the Enhanced Fujita scale. This comparison is indicative only, as the two scales differ in reference height, wind-averaging convention, and indicator inventory. Representative damage is shown in Figure 14.

The surveyed damage path (Figure 15) began near the Iloilo Business Park in Mandurriao (10.7183°N, 122.5449°E) and ended near Barangay Cubay in Jaro (10.7269°N, 122.5693°E). A kernel-smoothed centerline fitted to the survey points yields a path length of ∼2.9 km oriented toward 065°, with a median width of ∼170 m and a maximum width of ∼340 m, the latter near Benedicto, where IF1 damage spread over ∼220 m. At the storm translation speed of 13.7 m s−1^{-1}, this path would have been traversed in ∼3.5 min. A translating Rankine vortex fitted to the ratings along the path (RcR_c of 50–100 m and a translation of 13.7 m s−1^{-1}) required a median VrotV_{\mathrm{rot}} of ∼40 m s−1^{-1} (10th–90th percentiles of 29–50 m s−1^{-1}) and placed the vortex center 30–120 m north of the damage centerline, consistent with the stronger winds on the right side of a translating vortex. The resulting IF swath (Figure 15) contains two IF2 patches, at the Iloilo Business Park and near the Jaro Metropolitan Cathedral, connected by a continuous IF1 corridor. Notably, the path orientation (065°) was ∼10° to the left of the parent-storm motion (076°), consistent in sense with the DTM estimated from the radar hodographs, although 3 smaller in magnitude ([94]).

The WEH positions tracked by S-ILO provide an independent constraint on the location and timing of the tornado. In the 2230 LST (1430 UTC) volume, the WEH centers at 2.4°–5.3° (0.24–0.43 km ARL) were located 0.39–0.87 km north-northwest of the damage centerline, with the edges of their enclosing circles within 0–0.3 km of the centerline. The 2.4° center, scanned at 2232:25 LST, lay only 0.27–0.31 km from the DIs at Jaro Plaza. The increasing offset with height indicates a slight north-northwestward tilt of the WEH with height, while the residual offset at the lowest tilt is comparable to the combined uncertainty of the manual WEH marking, the ∼0.13 km beam width at this range, and the 0.25 km gate spacing. Interpolating between the 2.4° WEH positions of the 2220 and 2230 LST volumes, which were 7.0 km and 599 s apart (11.7 m s−1^{-1} toward 079°), places the WEH over the start of the damage path near 2230–2231 LST and over Jaro Plaza near 2232 LST, so that the tornado likely affected the surveyed area between ∼2230 and ∼2234 LST (1430–1434 UTC), within the reported tornado period of 1420–1435 UTC (Figure 5). This timing is consistent with the arrival of the outflow at the S-ILO AWS at 1430 UTC. In the 2220 LST volume, however, the WEH was already present ∼5 km west-southwest of the start of the surveyed path, where no damage was documented. This suggests either that the low-level circulation preceded tornadogenesis by several minutes or that weak damage along this segment went unreported, and the available data cannot distinguish between these possibilities.

The PlanetScope imagery provides supporting evidence that helps delineate the track of the tornado (Figure 16 and Figure 17). Because the pre- and post-event scenes were acquired only 3 days before and after the event (05 and 11 August), changes in land cover unrelated to the tornado are expected to be minor, so that the changes along the surveyed path can be attributed largely to the tornado. Six sites along the track were examined in paired true-color subsets (Figure 16). The most pronounced change occurred at the Iloilo Business Park (site 1), where the damaged shops and commercial establishments were located. There, the dark and textured canopy of the vegetated lot adjacent to the shops became lighter and more uniform on 11 August, and the small structures along its southern edge appeared altered, consistent with the defoliated and uprooted trees and the roof and structural damage documented by the survey near the IF2 indicators. The NDVI reduction (Figure 17c) likewise showed a coherent area of ∼20%–50% reduction around site 1, which contrasts with the near-zero reduction over the undisturbed vegetation to the north and west. Near Jaro Plaza (site 5), the canopy of the trees surrounding the plaza and the cathedral appeared thinned, consistent with the uprooted and snapped trees documented there, and a cluster of 80%–100% reductions occurred near site 3. At the remaining sites, which are dominated by roofs and pavement, the changes were subtler at the ∼3 m resolution of the imagery, as expected for a narrow (≤ 340 m) path through a densely built-up area where the dominant DIs (roofs, poles, and signs) produce little change in vegetation indices. Taken together, the sites of change are aligned along the damage centerline from the Iloilo Business Park to Jaro Plaza, so the PlanetScope imagery corroborates the location and orientation of the track derived from the ground survey and the WEH positions, as in previous applications of satellite imagery to tornado tracks [22] (Capuli 2026).

PlanetScope true-color composites and enlarged before-and-after pairs

Figure 16. PlanetScope true-color composites from (a) 05 August 2026 and (b) 11 August 2026, before and after the Iloilo tornado, with cloud-masked areas in gray. Yellow boxes (1–6) mark the sites shown in the enlarged before-and-after pairs below, and blue circles mark sites of interest along the damage path (§ 5).

NDVI before, NDVI after, and percent reduction maps

Figure 17. (a) NDVIBEFORE\mathrm{NDVI}_{\mathrm{BEFORE}} on 05 August 2026, (b) NDVIAFTER\mathrm{NDVI}_{\mathrm{AFTER}} on 11 August 2026, and (c) percent reduction in NDVI between the two dates (Eq. (14)). Gray areas are cloud-masked. In (c), the numbered blue circles mark the six sites shown in Figure 16.

To place the Iloilo tornado in the context of impact-based forecasting, its environment and damage intensity were compared with the conditional distributions of Smith et al. [89], who related the peak DI wind speeds of damage-surveyed tornadoes in the contiguous United States to the 0.5° VrotV_{\mathrm{rot}} of the nearest WSR-88D, the maximum STP within 80 km of the tornado (STP80km), and the population density, and who grouped these combinations into

five calibrated levels of potential damage intensity. Their box-and-whisker distributions are recreated in Figure 18. The ERA5 proximity profile at 1400 UTC yielded an effective-layer STP of 2.28 (Table 1). For VrotV_{\mathrm{rot}}, the simulated velocity couplets of a vortex consistent with the damage (Rc=100R_{\mathrm{c}} = 100 m and a peak tangential speed of 40 m s−1^{-1}; Figure 10), sampled with the beam and gate of S-ILO, were used. Defined, as in Smith et al. [89], as half the peak cyclonic ΔV\Delta V across the couplet and sampled over all positions of the vortex relative to the radials and range gates, the beam-sampled VrotV_{\mathrm{rot}} at 2.4°–5.3° was ∼12–18 m s−1^{-1} (median ∼13–14 m s−1^{-1}) for the Rankine profile and ∼16–24 m s−1^{-1} (median ∼18–20 m s−1^{-1}) for the Burgers–Rott profile, i.e., ∼23–46 kt with medians of ∼26–38 kt (Table S6 of the supplementary material). The Iloilo tornado therefore falls in the level-2 combination of STP80kmSTP_{80\mathrm{km}} of 1.0–2.9 and VrotV_{\mathrm{rot}} of 25–69 kt (green star in Figure 18). This assignment holds for the median of every profile and tilt, whereas the least favorable sampling of the Rankine vortex (∼12 m s−1\sim12\ \mathrm{m\ s}^{-1}, or ∼23\sim23 kt) falls at the lower edge of this combination. Because these VrotV_{\mathrm{rot}} values come from simulated rather than observed velocity couplets, this placement is illustrative. It indicates the signature that S-ILO would likely have displayed for a vortex consistent with the damage, not an independent radar estimate of the tornado intensity. For this combination, which is the most populated in the sample of Smith et al. [89], the median peak DI wind speed was 93–95 mph (∼42 m s−1\sim42\ \mathrm{m\ s}^{-1}), the interquartile range spanned 79–107 mph

Peak damage indicator wind speeds and damage-intensity levels for nine parameter combinations

Figure 18. Peak damage indicator (DI) wind speed (mph, left; m s−1^{-1}, right) for nine combinations of the significant tornado parameter within 80 km (STP80km), the 0.5∘^\circ rotational velocity (VrotV_{\mathrm{rot}}; kt), and the population density (people km−2^{-2}), recreated from Fig. 13 of Smith et al. [89]. Gray boxes denote the 2009–17 sample and black hollow boxes the independent 2019–20 sample. Boxes span the 25th–75th percentiles, whiskers the 10th–90th percentiles, and the horizontal bars mark the medians, with the percentile values annotated (2019–20 in bold on the left and 2009–17 in gray on the right). Red circles are the lower and upper bounds of the wind speed damage estimate (WSDE), and the blue dotted lines are the EF-scale thresholds. Sample sizes are given in parentheses (top, 2009–17; bottom, 2019–20), and the colored cells denote the damage-intensity levels L1–L5. The green star marks the Iloilo tornado (IF2, ∼60\sim60 m s−1^{-1}) in the STP80km 1.0–2.9 and VrotV_{\mathrm{rot}} 25–69 kt combination (shaded column).

(35–48 m s−1^{-1}), and the 90th percentile was 115–118 mph (51–53 m s−1^{-1}), whereas the wind speed damage estimate (WSDE) ranged from 75 to 135 mph (34–60 m s−1^{-1}). The peak wind speed of the IF2 rating, ∼60 m s−1^{-1} (∼134 mph), therefore lies above the 90th percentile and at the upper bound of the WSDE of this combination, which places the Iloilo tornado in the upper tail of the damage intensities expected for its environment and rotation strength.

Two factors may explain this upper-tail outcome. First, the beam-sampled VrotV_{\mathrm{rot}} underestimates the true rotation of a vortex that is narrower than the radar beam [119]. At ∼10–11 km from S-ILO, the effective beam is ∼0.25 km wide, so the simulated couplets retained only ∼29%–59% of the 40 m s−1^{-1} peak tangential speed. If the same circulation had been sampled at closer range or with finer resolution, yielding a VrotV_{\mathrm{rot}} of ≥ 70 kt (≥ 36 m s−1^{-1}), the tornado would instead fall in the level-3 combination of STP80km of 1.0–5.9 and VrotV_{\mathrm{rot}} ≥ 70 kt. In that combination, ∼134 mph lies close to the median (122–127 mph) and within the interquartile range (111–140 mph) of the peak DI wind speed (Figure 18). Second, the tornado tracked through densely built-up districts, where the population density exceeds the threshold of 20 people km−2^{-2} used by Smith et al. [89] by orders of magnitude. Smith et al. [89] found that, for comparable VrotV_{\mathrm{rot}} and STP80km, tornadoes that affected areas above this threshold produced higher peak DI wind speeds, because more DIs were available to record the strongest winds. The abundance of DIs in Mandurriao and Jaro, from the shops of the Iloilo Business Park to the trees, poles, and signs of Jaro Plaza, likely increased the chance that the strongest winds were captured by the survey. These comparisons are indicative rather than quantitative, because the distributions of Smith et al. [89] were derived from WSR-88D observations at 0.5°, from STP80km rather than from a single proximity profile, and from EF-scale DIs in the United States, whose construction practices differ from those in the Philippines.

The same environment also allows the forecastability of the event to be assessed retrospectively. In the convective outlooks of the U.S. Storm Prediction Center (SPC), the categorical risk is derived from the probability of a tornado, damaging wind, or large hail within 25 miles (40 km) of a point, and a 5% tornado probability on the Day 1 outlook corresponds to a Slight Risk, i.e., level 2 of 5 (Grams and Bunting [43]). The environment of southern Panay on the evening of 08 August combined MUCAPE of ∼2000 J kg−1^{-1}, negligible CIN, LCLs below 300 m, an STP of 2.28, and an SCP of 7.44 with an LLS of ∼16 m s−1^{-1} and SRH01_{01} of up to 211 m2^{2} s−2^{-2} in the radar-derived profiles, while the weak DLS and the weak synoptic-scale forcing favored only isolated supercells within the monsoon. Had such an outlook been issued, these ingredients would most likely have supported a 5% tornado probability, i.e., a Slight Risk (2/5) with a tornado threat, over the southern coast of Panay. A higher category, such as an Enhanced Risk driven by a 10% tornado probability with significant (EF2+) tornado potential, would have been difficult to justify given the expected low coverage of supercells, even though the tornado ultimately reached IF2 intensity. Because the LLJ and the LLS strengthened after sunset, such an outlook would also have needed to emphasize that the threat would peak during the night, when the vulnerability to tornadoes is greatest ([7]).

Summary and Discussion

This study documented the tornado that struck the Mandurriao and Jaro districts of Iloilo City at approximately 1430 UTC (2230 LST) on 08 August 2026, using ERA5 reanalysis, observed and model proximity soundings, Himawari-9 infrared imagery, dual-polarization observations from S-ILO, surface observations, a ground damage survey, and PlanetScope imagery. The tornado was produced by a supercell that developed near the southern coast of Panay and moved east-northeastward at 13.7 m s−1^{-1} across Iloilo City. The damage survey documented a ∼2.9 km long and up to ∼340 m wide path, and the tornado was rated IF2 on the basis of a utility pole, a snapped strong tree, structural damage to a building, and a sign at the Jaro Metropolitan Cathedral. To our knowledge, this is among the first nocturnal tornadoes in the Philippines documented with both a ground damage survey and dual-polarization radar observations, and it adds a well-documented case from Western Visayas, one of the country’s recognized tornado hotspots.

The tornadic storm developed within the southwest monsoon as it was enhanced by Typhoon Dolphin, which was located far to the north-northeast of Luzon. Synoptic-scale forcing for ascent was weak, as 500-hPa heights over the Visayas remained nearly steady, similar to the weakly forced easterly-wave regime of the Camarines Norte outbreak [26] but unlike the 500-hPa height falls that accompanied the Canding, Arayat, tornado (Capuli et al. 2026). Instead, the monsoon acted as a broad LLJ that supplied deep moisture (PWAT of 55–62 kg m−2^{-2}) and concentrated the vertical wind shear in the lowest ∼3 km. The resulting environment combined moderate instability (MUCAPE of ∼2000 J kg−1^{-1} and ECAPE of ∼1600 J kg−1^{-1}), negligible CIN, very low LCLs, and strong, highly streamwise LLS, while the DLS was weak because the hodograph folded back above ∼3 km. Such a hodograph, dominated by LLS rather than DLS, differs from the environments of most tornadic supercells in the U.S. Great Plains. It resembles, however, the tropical-cyclone and high-shear, low-CAPE environments in which tornadic supercells are sustained by strong LLS and storm-relative inflow despite modest DLS [74, 87].

The environment of the Iloilo tornado was distinctly nocturnal. Several factors explain how a tornadic supercell was sustained ∼4\sim4 h after sunset over land, when the stabilization of the boundary layer typically reduces SB instability. First, a nocturnal LLJ of ∼20\sim20 m s−1^{-1} (40 kt) near 1.5–3 km, evident in both ERA5 and the radar-derived wind profiles, surged from ∼2130\sim2130 to 2230 LST and increased the LLS and SRH as the storm approached Iloilo City. A similar evening enhancement of LLS has been documented in nocturnal tornado environments of the United States [61, 76]. Second, the jet and the onshore flow beneath it advected warm, moist marine air (θe\theta_e of 353–359 K at the AWS) from the Panay Gulf, so that the coastal boundary layer barely cooled after sunset and CAPE along the coast remained ∼2000\sim2000–2500 J kg−1^{-1} while it decreased inland. The tornado occurred within ∼3\sim3–4 km of the coast, where this effect was greatest. Third, the moist boundary layer kept the LCL low and limited the negative buoyancy of the outflow, as indicated by the modest Δθv\Delta\theta_v of ∼2\sim2 K at the AWS, which favors the lifting and stretching of near-surface air by the low-level updraft [73]. Fourth, the storm moved ∼20∘\sim20^\circ to the right of the B2KRM, which increased its storm-relative inflow and effective SRH [21].

Terrain may have further modified the low-level flow, as ERA5 depicted enhanced low-level PV and convergence in the vicinity of the central Panay range. However, the resolution of ERA5 is insufficient to determine whether terrain-induced vorticity contributed to tornadogenesis, as suggested for the Candating, Arayat, and Camarines Norte cases (Capuli et al. 2026; Capuli 2026).

The S-ILO observations revealed several hallmark features of a tornadic supercell, including a hook echo with a WEH enclosed by a ring of high reflectivity, a ZDR_{\mathrm{DR}} arc along the inflow side of the forward flank, a ZDR_{\mathrm{DR}} ring around a low-level BWER, and enhanced ZDR_{\mathrm{DR}} extending to just below the 0 °C level. The three-dimensional analysis further showed an elevated rear-flank core of 55–60 dBZ directly above the WEH, which had lowered by ∼1\sim1–2 km since the previous volume and is consistent with a DRC [82, 57]. The WEH tracked across Iloilo City in two consecutive volumes, was nearly vertical in the lowest ∼1.4\sim1.4 km once the scan-time displacement was removed, and passed within a few hundred meters of the surveyed damage path at the time of the damage, indicating that it marked the tornado or its parent low-level circulation. At the same time, this case illustrates how routine data processing can conceal tornado-scale signatures. The archived velocity field had been smoothed by the signal processor, the ρHV<0.80\rho_{\mathrm{HV}} < 0.80 threshold applied for rainfall estimation removed the gates where a TDS would be expected, and the velocity sign stored in the UF files was reversed relative to the CF convention. Consequently, neither a TVS nor a TDS could be identified directly, even though the tornado passed only ∼5\sim5–77 km from the radar. The radial velocity simulation and the injection–recovery analysis showed that a tornado consistent with the damage would have produced a conspicuous couplet in unfiltered data, and that the non-detection is consistent with a narrow vortex (Rc≈50R_{\mathrm{c}} \approx50–100100 m) with VrotV_{\mathrm{rot}} of ∼35\sim35–4545 m s−1^{-1}, in agreement with the width of the damage path and the Rankine swath model. Because radar-based intensity estimates are biased low for vortices comparable to or smaller than the beam [119], these values should be regarded as consistent with, rather than independent of, the damage rating.

It is instructive to compare the Iloilo tornado with the two Philippine tornadic supercells documented previously. The Candating, Arayat, tornado (EF2; Capuli et al. [27]) occurred around midday within the southwest monsoon, which acted as an LLJ, under subtle synoptic-scale ascent influenced by the approach of TC Ewiniar, in an environment of large undiluted instability from steep low-level lapse rates, weak LLS, and highly streamwise near-surface vorticity. The Camarines Norte outbreak (IF2.5, comparable to the EF3 range; Capuli [26]) occurred near 0600 LST within an easterly-wave regime, with MUCAPE of ∼2200\sim2200 J kg−1^{-1}, LCLs below 200 m, and modestly curved but strongly streamwise hodographs. The Iloilo tornado (IF2) occurred at night within the typhoon-enhanced southwest monsoon, with MUCAPE of ∼2000\sim2000 J kg−1^{-1}, LCLs below 300 m, and strong, streamwise LLS enhanced by a nocturnal LLJ. Despite their different synoptic regimes and times of day, the three cases share low LCLs, moderate rather than extreme instability, and near-ground horizontal vorticity that was almost purely streamwise, whereas the DLS varied widely and was weak in the Iloilo case. These shared characteristics suggest that, in the moist tropical environment of the Philippines, the near-ground kinematic profile and the cloud-base height may discriminate tornadic supercells more effectively than the DLS, a hypothesis that should be tested with a larger sample of events from the SWAP database.

Several limitations should be noted. First, no radiosonde was launched near the time and location of the event, so the near-storm thermodynamic profile relies on ERA5, whose resolution smooths the land–sea and terrain contrasts that likely mattered in this case and which underestimated the LLS relative to the radar-derived profile. Second, the radar-derived hodographs extend only to ∼3.5\sim3.5 km AGL, depend on the AWS wind for their lowest level, and were contaminated by outflow at 1430 UTC. Third, the AWS is located ∼5\sim5–77 km from the damage path, so the surface observations describe the storm’s outflow rather than the immediate environment of the tornado. Fourth, the 10-min volume interval of S-ILO is too coarse to resolve the descent of the rear-flank core and the genesis of the low-level vortex, and the three-dimensional analysis relies on ZHZ_{\mathrm{H}} alone. The lowering of the core between 1410 and 1420 UTC is therefore inferred from two volumes rather than observed, whereas mobile radars have documented the descent of such cores at much finer temporal resolution [23], and the storm-relative position assigned to the 1410 UTC core depends on the assumed storm motion. In addition, DRCs also occur in nontornadic supercells and have limited value as a stand-alone tornado precursor [57], so the DRC in this case is interpreted as consistent with, rather than diagnostic of, tornadogenesis. In the long term, rapid-scan X-band radars offer the most promising solution to these sampling limitations. Mobile, rapid-scanning, polarimetric X-band radars can complete a 10-elevation volume in ∼20\sim20 s [100] and have documented tornadogenesis and tornado dissipation in supercells at temporal resolutions of seconds to tens of seconds [52]. Dense networks of small X-band radars spaced tens of kilometers apart sample the lowest kilometers of the atmosphere that long-range radars miss because of the curvature of the Earth, including damaging winds and heavy rainfall [75], while multiparameter phased-array weather radars resolve the three-dimensional structure of storms within ∼30\sim30 s over a ∼60\sim60 km radius for the early detection of torrential rainfall and tornadoes (Takahashi et al. 2019). Such systems, deployed to complement S-band radars such as S-ILO over densely populated areas, would resolve the fine structure not only of tornadic storms but also of other thunderstorms capable of producing hazards, such as downbursts, large hail, and flash floods. Fifth, because the tornado occurred at night, no photographs or videos of the tornado itself were available, and the timing of the damage was inferred from the radar rather than from eyewitness reports. Finally, the absence of high-resolution numerical modeling limits the assessment of terrain effects, of the dynamics of the nocturnal LLJ over Panay, and of the storm-scale processes that led to tornadogenesis. Future work would benefit from convection-permitting or large-eddy simulations (e.g., WRF or CM1) of this event.

Based on this case, we offer several recommendations. First, the quality control applied to S-ILO should be relaxed and made storm-aware so that the structures that identify SCSs and their hazards are not removed or heavily masked. The fixed ρHV<0.80\rho_{\mathrm{HV}} < 0.80 threshold is appropriate for QPE, but in this case it removed exactly the gates in which lofted debris would be expected, together with the corresponding ZHZ_{\mathrm{H}}, ZDRZ_{\mathrm{DR}}, and VRV_{\mathrm{R}}. Rather than deleting such gates, the quality control could flag them while archiving the base moments unaltered, so that the QPE product is computed from the flagged field and the original data remain available for severe-weather analysis. Constraints can then be added to retain low-ρHV\rho_{\mathrm{HV}} gates where they are physically consistent with a TDS, i.e., where they coincide with ZH≥35Z_{\mathrm{H}} \ge35–4040 dBZ, ZDRZ_{\mathrm{DR}} near 0 dB, and a hook echo, mesocyclone, or strong azimuthal shear [84, 13]. Alternatively, a hydrometeor classification algorithm with an explicit debris class can separate debris from clutter and biological scatterers objectively [90]. Likewise, the near-zero-velocity clutter filter could be restricted to clutter-prone gates identified from a static clutter map, and the spatial smoothing of VRV_{\mathrm{R}} by the signal processor should be reduced or, at minimum, the unfiltered velocity and spectrum width should be archived, since the processing in this case concealed a tornado-scale couplet that would otherwise have been conspicuous at a range of only ∼6\sim6–1212 km. Second, the header metadata of the archived UF volumes (wavelength, PRF, and Nyquist velocity) and the sign convention of the velocity field should be verified for each newly commissioned radar before the data are distributed. Third, a severe-weather scanning mode with more frequent low-level updates than the current 10-min volume would allow rapidly evolving features, such as DRCs and low-level mesocyclones, to be resolved, and a network of rapid-scan X-band radars over the major urban centers would provide this capability in the long term. Fourth, as recommended by Capuli (2026), a rapid-response damage survey capability trained in the IF Scale should be activated whenever a tornado is reported, since damage in urban areas is cleared within hours, as occurred in Iloilo City, where clearing operations began on the night of the event. Fifth, the combination of the near-storm environment (e.g., STP) and radar-derived rotation could support impact-based tornado warnings and probabilistic convective outlooks in the Philippines, following the framework of Smith et al. [89] and the SPC outlooks, provided that these relations are recalibrated with local cases from Project SWAP. Moreover, because the tornado occurred at night and only minutes after a thunderstorm advisory was issued, this case underscores the elevated threat posed by nocturnal tornadoes [7] and the value of radar-based nowcasting of low-level mesocyclones near densely populated coastal cities.

In summary, the Iloilo tornado demonstrates that significant tornadoes in the Philippines can develop at night within the typhoon-enhanced southwest monsoon, in environments characterized by a nocturnal LLJ surge with warm, moist advection, moderate instability, very low LCLs, and strong, streamwise LLS but weak DLS. Together with the Candating, Arayat, and Camarines Norte cases, these findings add to the growing evidence that Philippine tornadic environments span multiple regimes, including the southwest monsoon, TCs, and easterly waves, and they emphasize the need for the continued systematic documentation of SCSs through Project SWAP and the national radar network.

Acknowledgments. The author is sincerely grateful to the residents of Iloilo City who reported and documented the tornado in their communities, and whose photographs and accounts made much of this analysis possible. Thanks are also due to Dr. Funing Li (Massachusetts Institute of Technology) and Dr. John T. Allen (Central Michigan University) for valuable guidance on the use of xcape, and to the anonymous reviewers and the editor, whose careful and constructive comments substantially improved the manuscript. This work received no external funding and was carried out on the author’s own time, dedicated to severe convective storm research in the Philippines. Finally, the author thanks family and loved ones for their unwavering support throughout this research. CGHA is supported by Planet Labs PBC (Planet) through the Education and Research Program.

Author Contribution. Generich H. Capuli: Writing – original draft, Writing – review & editing, Visualization, Methodology, Investigation, Formal Analysis, Data curation, Conceptualization, Supervision.

Data availability statement. The ERA5 reanalysis data on single and pressure levels ([48]) are openly available from the Copernicus Climate Data Store (https://cds.climate.copernicus.eu). The individual full-disk Himawari-9 AHI bands were accessed through the THREDDS data server of the National Computational Infrastructure (NCI) Australia. PlanetScope imagery ([78]) was obtained from Planet Labs PBC under its Education and Research Program and cannot be redistributed under the terms of that license. The 2023 administrative boundaries of the Philippines (levels 0–4) of the Philippine Statistics Authority (PSA) and the National Mapping and Resource Information Authority (NAMRIA) are available from the Humanitarian Data Exchange (https://data.humdata.org/dataset/cod-ab-phl), and the Copernicus GLO-30 Digital Elevation Model is distributed by OpenTopography (https://doi.org/10.5069/G9028PQB). Basemap drawn from OpenStreetMap (OSM) data licensed under the ODbL. The one-dimensional ERA5 model sounding profiles used in this study are archived in the fourth data release (DR4) of Project SWAP. The S-ILO radar volumes and the AWS observations are operational holdings of DOST-PAGASA and are not publicly redistributable. They may be requested from DOST-PAGASA through METTS, and the corresponding author will facilitate such requests upon reasonable inquiry. Users of these datasets should follow the attribution and licensing requirements of each provider.

This study made use of several open-source Python packages. The radar volumes were read, processed, and gridded with Py-ART [47]. The sounding and ERA5 convective parameters were computed with MetPy, SounderPy v3.1.0 [42], ecape-parcel, and xcape [65]. The cloud and cloud-shadow masks of the PlanetScope imagery were generated with OmniCloud-Mask [120], and the raster data were processed with Rasterio and rioxarray. The maps were produced with Cartopy, GeoPandas, Shapely, and pyproj, the reflectivity isosurfaces were extracted with scikit-image, and all analyses and figures used NumPy, SciPy, pandas, xarray, netCDF4, Matplotlib, and cmweather. The dual-PRF velocity errors were corrected with the code of Alford et al. [2], and the folded single-PRF tilts were dealiased with a reference-anchored, region-based procedure developed for this study, against which UNRAVEL [70] and the region-based dealiasing of Py-ART were evaluated.

Competing Interest. The author declares no competing interests.

Supplementary Analysis

Supplementary material is provided as a separate document. It contains (1) the complete damage survey inventory, listing the location, damage indicator, degree of damage, and IF rating of each of the 31 surveyed items; (2) the details of the radial velocity simulation, including the radar beam and gate model and the simulated Rankine and Burgers–Rott vortices; and (3) the supporting statistical analyses, namely the injection–recovery test and the Bayesian estimate of the vortex rotational velocity.

References

  1. [1]Aebischer, U., and C. Schär, 1998: Low-level potential vorticity and cyclogenesis to the lee of the Alps. Journal of the Atmospheric Sciences, 55 (2), 186–207.
  2. [2]Alford, A. A., M. I. Biggerstaff, C. L. Ziegler, D. P. Jorgensen, and G. D. Carrie, 2022: A method for correcting staggered pulse repetition time (PRT) and dual pulse repetition frequency (PRF) processor errors in research radar datasets. J. Atmos. Oceanic Technol., 39 (11), 1763–1780, https://doi.org/10.1175/JTECH-D-21-0176.1.
  3. [3]Anderson, C. J., C. K. Wikle, Q. Zhou, and J. A. Royle, 2007: Population Influences Tornado Reports in the United States. Weather and Forecasting, 22 (3), 571–579, https://doi.org/10.1175/WAF997.1.
  4. [4]Anderson-Frey, A. K., and H. Brooks, 2021: Compared to What? Establishing Environmental Baselines for Tornado Warning Skill. Bulletin of the American Meteorological Society, 102 (4), E738–E747, https://doi.org/10.1175/BAMS-D-19-0301.1.
  5. [5]Antonescu, B., H. M. A. M. Ricketts, and D. M. Schultz, 2019: 100 Years Later: Reflecting on Alfred Wegener’s Contributions to Tornado Research in Europe. Bulletin of the American Meteorological Society, 100 (4), 567–578, https://doi.org/10.1175/BAMS-D-17-0316.1.
  6. [6]Antonescu, B., D. M. Schultz, A. Holzer, and P. Groenemeijer, 2017: Tornadoes in Europe: An Underestimated Threat. Bulletin of the American Meteorological Society, 98 (4), 713–728, https://doi.org/10.1175/BAMS-D-16-0171.1.
  7. [7]Ashley, W. S., A. J. Krmenec, and R. Schwantes, 2008: Vulnerability due to nocturnal tornadoes. Weather and Forecasting, 23 (5), 795–807.DOI
  8. [8]Bagaglini, L., R. Ingrosso, and M. M. Miglietta, 2021: Synoptic patterns and mesoscale precursors of Italian tornadoes. Atmospheric Research, 253, 105503, https://doi.org/10.1016/j.atmosres.2021.105503.
  9. [9]Benjamin, T. B., 1968: Gravity currents and related phenomena. Journal of Fluid Mechanics, 31 (2), 209–248.DOI
  10. [10]Besag, J., 1986: On the statistical analysis of dirty pictures. Journal of the Royal Statistical Society: Series B, 48 (3), 259–302.DOI
  11. [11]Bessho, K., and Coauthors, 2016: An Introduction to Himawari-8/9; Japan’s New-Generation Geostationary Meteorological Satellites. Journal of the Meteorological Society of Japan. Ser. II, 94 (2), 151–183, https://doi.org/10.2151/jmsj.2016-009.
  12. [12]Blackadar, A. K., 1957: Boundary layer wind maxima and their significance for the growth of nocturnal inversions. Bulletin of the American Meteorological Society, 38 (5), 283–290.DOI
  13. [13]Bodine, D. J., M. R. Kumjian, R. D. Palmer, P. L. Heinselman, and A. V. Ryzhkov, 2013: Tornado damage estimation using polarimetric radar. Weather and Forecasting, 28 (1), 139–158, https://doi.org/10.1175/WAF-D-11-00158.1.
  14. [14]Boustead, J. M., B. E. Mayes, W. Gargan, J. L. Leighton, G. Phillips, and P. N. Schumacher, 2013: Discriminating Environmental Conditions for Significant Warm Sector and Boundary Tornadoes in Parts of the Great Plains. Weather and Forecasting, 28 (6), 1498–1523, https://doi.org/10.1175/WAF-D-12-00102.1.
  15. [15]Brooks, H. E., C. A. Doswell, and M. P. Kay, 2003: Climatological estimates of local daily tornado probability for the united states. Weather and Forecasting, 18 (4), 626–640, https://doi.org/10.1175/1520-0434(2003)018<0626:CEOLDT>2.0.CO;2.
  16. [16]Brown, R. A., L. R. Lemon, and D. W. Burgess, 1978: Tornado detection by pulsed Doppler radar. Monthly Weather Review, 106 (1), 29–38.doi.org/10.1175/1520-0493(1978)106<0029:tdbpdr>2.0.co;2
  17. [17]Brown, R. A., V. T. Wood, and D. Sirmans, 2002: Improved tornado detection using simulated and actual WSR-88D data with enhanced resolution. Journal of Atmospheric and Oceanic Technology, 19 (11), 1759–1771.doi.org/10.1175/1520-0426(2002)019<1759:itdusa>2.0.co;2
  18. [18]Browning, K. A., and R. Wexler, 1968: The determination of kinematic properties of a wind field using Doppler radar. J. Appl. Meteor., 7 (1), 105–113, https://doi.org/10.1175/1520-0450(1968)007<0105:TDOKPO>2.0.CO;2.
  19. [19]Bunkers, M. J., J. S. Johnson, L. J. Czepyha, J. M. Grzywacz, B. A. Klimowski, and M. R. Hjelmfelt, 2006: An Observational Examination of Long-Lived Supercells. Part II: Environmental Conditions and Forecasting. Weather and Forecasting, 21 (5), 689–714, https://doi.org/10.1175/WAF952.1.
  20. [20]Bunkers, M. J., B. A. Klimowski, J. W. Zeitler, R. L. Thompson, and M. L. Weisman, 2000: Predicting Supercell Motion Using a New Hodograph Technique. Weather and Forecasting, 15 (1), 61 – 79, https://doi.org/10.1175/1520-0434(2000)015⟨0061:PSMUAN⟩2.0.CO;2. Observations. Monthly Weather Review, 148 (12), 4893 – 4915, https://doi.org/10.1175/MWR-D-20-0105.1.
  21. [21]Bunkers, M. J., M. B. Wilson, M. S. Van Den Broeke, and D. J. Healey, 2022: Scan-by-scan storm-motion deviations for concurrent tornadic and nontornadic supercells. Weather and Forecasting, 37 (5), 749–770.
  22. [22]Burow, D., H. V. Herrero, and K. N. Ellis, 2020: Damage Analysis of Three Long-Track Tornadoes Using High-Resolution Satellite Imagery. Atmosphere, 11 (6), https://doi.org/10.3390/atmos11060613.
  23. [23]Byko, Z., P. Markowski, Y. Richardson, J. Wurman, and E. Adler-man, 2009: Descending Reflectivity Cores in Supercell Thunderstorms Observed by Mobile Radars and in a High-Resolution Numerical Simulation. Weather and Forecasting, 24 (1), 155 – 186, https://doi.org/10.1175/2008WAF2222116.1.
  24. [24]Capuli, G. H., 2024: Project Severe Weather Archive of the Philippines (SWAP). Part 1: Establishing a Baseline Climatology for Severe Weather across the Philippine Archipelago. Annals of Geophysics, 67 (5), GC554, https://doi.org/10.4401/ag-9515.
  25. [26]Capuli, G. H., 2026: Localized tornado outbreak at the upstream of a tropical easterly wave in Camarines Norte, Philippines (13 September 2025). Earth Systems and Environment.arxiv.org/abs/2604.20222
  26. [27]Capuli, G. H., M. A. O. Noveno, and M. P. A. Ibañez, 2026: A Case Study of the Tornadic Supercell in the Province of Pampanga, Philippines (27 May 2024). Natural Hazards, 122, 89, https://doi.org/10.1007/s11069-025-07915-8.
  27. [28]Choo, S., K.-H. Min, K.-E. Kim, and G. Lee, 2018: A Case Study of Tsukuba Tornado in Japan on 6 May 2012. J. Korean Earth Sci. Soc., 39 (5), 403–418, https://doi.org/10.5467/JKESS.2018.39.5.403.
  28. [29]Clark, M. R., 2024: Polarimetric radar observations of the Jersey tornadic supercell on 1–2 November 2023. Weather, 79 (11), 362–372, https://doi.org/10.1002/wea.7619.
  29. [30]Coffer, B. E., and M. D. Parker, 2017: Simulated Supercells in Nontornadic and Tornadic VORTEX2 Environments. Monthly Weather Review, 145 (1), 149 – 180, https://doi.org/10.1175/MWR-D-16-0226.1.
  30. [31]Coffer, B. E., M. D. Parker, J. M. Peters, and A. R. Wade, 2023: Supercell Low-Level Mesocyclones: Origins of Inflow and Vorticity. Monthly Weather Review, 151 (9), 2205 – 2232, https://doi.org/10.1175/MWR-D-22-0269.1.
  31. [32]Coffer, B. E., M. D. Parker, R. L. Thompson, B. T. Smith, and R. E. Jew-ell, 2019: Using Near-Ground Storm Relative Helicity in Supercell Tornado Forecasting. Weather and Forecasting, 34 (5), 1417 – 1435, https://doi.org/10.1175/WAF-D-19-0115.1.
  32. [33]Coffer, B. E., M. T. Tazarek, and M. D. Parker, 2020: Near-Ground Wind Profiles of Tornadic and Nontornadic Environments in the United States and Europe from ERA5 Reanalyses. Weather and Forecasting, 35 (6), 2621 – 2638, https://doi.org/10.1175/WAF-D-20-0153.1.
  33. [34]Coniglio, M. C., and M. D. Parker, 2020: Insights into Supercells and Their Environments from Three Decades of Targeted RadiosondeDOI
  34. [35]Davies-Jones, R., 1984: Streamwise Vorticity: The Origin of Up-draft Rotation in Supercell Storms. Journal of Atmospheric Sciences, 41 (20), 2991 – 3006, https://doi.org/10.1175/1520-0469(1984)041⟨2991:SVTOOU⟩2.0.CO;2.
  35. [36]Davies-Jones, R., 2015: A review of supercell and tornado dynamics. Atmospheric Research, 158-159, 274–291, https://doi.org/10.1016/j.atmosres.2014.04.007.
  36. [37]Doviak, R. J., and D. S. Zrnić, 1993: Doppler Radar and Weather Observations. 2nd ed., Academic Press, https://doi.org/10.1016/B978-0-12-221420-2.X5001-7.
  37. [38]Emanuel, K., 1994: Atmospheric Convection. Oxford University Press.
  38. [39]Esterheld, J. M., and D. J. Giuliano, 2008: Discriminating between Tornadic and Non-Tornadic Supercells: A New Hodograph Technique. Electron. J. Sev. Storms Meteorol., 3 (2), 1–50, https://doi.org/10.55599/ejssm.v3i2.15.
  39. [40]Evans, J. S., and C. A. Doswell, 2001: Examination of Derecho Environments Using Proximity Soundings. Weather and Forecasting, 16 (3), 329 – 342, https://doi.org/10.1175/1520-0434(2001)016⟨0329:EODEUP⟩2.0.CO;2.
  40. [41]Firdaus, I. M., T. Yamazaki, M. R. Abdillah, and E. Riawan, 2025: Indonesia tornado database: Tornado climatology of Indonesia. Natural Hazards and Earth System Sciences, 25 (11), 4331–4341, https://doi.org/10.5194/nhess-25-4331-2025.
  41. [42]Gillett, K. J., 2025: Sounderpy: An atmospheric sounding visualization and analysis tool for python. Journal of Open Source Software, 10 (112), 8087, https://doi.org/10.21105/joss.08087.
  42. [43]Grams, J., and B. Bunting, 2026: SPC convective outlooks. Accessed 15 November 2026, https://www.spc.noaa.gov/misc/SPC_probotlk_info.html, NOAA/NWS Storm Prediction Center.
  43. [44]Groenemeijer, P., A. M. Holzer, T. Kühne, and T. Púčik, 2023: The international fujita (if) scale for tornado and wind damage assessments. Tech. rep., European Severe Storms Laboratory (ESSL), 77 pp. URL https://www.essl.org/cms/wp-content/uploads/IF-Scale_v1.0e.pdf.
  44. [45]Groenemeijer, P., and T. Kühne, 2014: A Climatology of Tornadoes in Europe: Results from the European Severe Weather Database. Monthly Weather Review, 142 (12), 4775 – 4790, https://doi.org/10.1175/MWR-D-14-00107.1.
  45. [46]Gropp, M. E., and C. E. Davenport, 2018: The impact of the nocturnal transition on the lifetime and evolution of supercell thunderstorms in the Great Plains. Weather and Forecasting, 33 (4), 1045–1061, https://doi.org/10.1175/WAF-D-17-0150.1.
  46. [47]Helmus, J. J., and S. M. Collis, 2016: The Python ARM Radar Toolkit (Py-ART), a Library for Working with Weather Radar Data in the Python Programming Language. Journal of Open Research Software, 4 (1), https://doi.org/10.5334/jors.119.
  47. [48]Hersbach, H., and Coauthors, 2020: The ERA5 global reanalysis. Quarterly Journal of the Royal Meteorological Society, 146 (730), 1999–2049, https://doi.org/10.1002/qj.3803.
  48. [49]Hirose, H., S. Shige, M. K. Yamamoto, and A. Higuchi, 2019: High Temporal Rainfall Estimations from Himawari-8 Multiband Observations Using the Random-Forest Machine-Learning Method. Journal of the Meteorological Society of Japan. Ser. II, 97 (3), 689–710, https://doi.org/10.2151/jmsj.2019-040. Holleman, I., and H. Beekhuis, 2003: Analysis and correction of dual PRF velocity data. Journal of Atmospheric and Oceanic Technology, 20 (4), 443–453.
  49. [50]Holton, J. R., 1967: The diurnal boundary layer wind oscillation above sloping terrain. Tellus, 19 (2), 200–205.
  50. [51]Hoque, M. A.-A., S. Phinn, C. Roelfsema, and I. Childs, 2017: Tropical cyclone disaster management using remote sensing and spatial analysis: A review. International Journal of Disaster Risk Reduction, 22, 345–354, https://doi.org/10.1016/j.ijdrr.2017.02.008.
  51. [52]Houser, J. L., H. B. Bluestein, and J. C. Snyder, 2015: Rapid-scan, polarimetric, Doppler radar observations of tornadogenesis and tornado dissipation in a tornadic supercell: The “El Reno, Oklahoma” storm of 24 May 2011. Monthly Weather Review, 143 (7), 2685–2710.DOI
  52. [53]Howard, J., and S. Gugger, 2020: Fastai: A Layered API for Deep Learning. Information, 11 (2), https://doi.org/10.3390/info11020108.
  53. [54]James, C. N., and R. A. Houze, Jr., 2001: A real-time four-dimensional Doppler dealiasing scheme. Journal of Atmospheric and Oceanic Technology, 18 (10), 1674–1683.doi.org/10.1175/1520-0426(2001)018<1674:artfdd>2.0.co;2
  54. [55]Jiang, Q., D. T. Dawson II, F. Li, and D. R. Chavas, 2025: Classifying synoptic patterns driving tornadic storms and associated spatial trends in the united states. npj Climate and Atmospheric Science, 8, 7, https://doi.org/10.1038/s41612-025-00897-1.
  55. [56]Joe, P., and P. T. May, 2003: Correction of dual PRF velocity errors for operational Doppler weather radars. Journal of Atmospheric and Oceanic Technology, 20 (4), 429–442.doi.org/10.1175/1520-0426(2003)20<429:codpve>2.0.co;2
  56. [57]Kennedy, A., J. M. Straka, and E. N. Rasmussen, 2007: A statistical study of the association of DRCs with supercells and tornadoes. Weather and Forecasting, 22 (6), 1191–1199, https://doi.org/10.1175/2007WAF2006095.1.
  57. [58]Kilambi, A., F. Fabry, and V. Meunier, 2018: A Simple and Effective Method for Separating Meteorological from Nonmeteorological Targets Using Dual-Polarization Data. Journal of Atmospheric and Oceanic Technology, 35 (7), 1415 – 1424, https://doi.org/10.1175/JTECH-D-17-0175.1.
  58. [59]King, A. T., and A. D. Kennedy, 2019: North American Supercell Environments in Atmospheric Reanalyses and RUC-2. Journal of Applied Meteorology and Climatology, 58 (1), 71 – 92, https://doi.org/10.1175/JAMC-D-18-0015.1.
  59. [60]Kirshbaum, D. J., B. Adler, N. Kalthoff, C. Barthlott, and S. Serafin, 2018: Moist Orographic Convection: Physical Mechanisms and Links to Surface-Exchange Processes. Atmosphere, 9 (3), https://doi.org/10.3390/atmos9030080.
  60. [61]Kis, A. K., and J. M. Straka, 2010: Nocturnal tornado climatology. Weather and Forecasting, 25 (2), 545–561.DOI
  61. [62]Kumjian, M. R., and A. V. Ryzhkov, 2008: Polarimetric Signatures in Supercell Thunderstorms. Journal of Applied Meteorology and Climatology, 47 (7), 1940 – 1961, https://doi.org/10.1175/2007JAMC1874.1.
  62. [63]Kumjian, M. R., and A. V. Ryzhkov, 2009: Storm-Relative Helicity Revealed from Polarimetric Radar Measurements. Journal of the Atmospheric Sciences, 66 (3), 667 – 685, https://doi.org/10.1175/2008JAS2815.1.
  63. [65]Lepore, C., J. Allen, and R. Abernethay, 2021: xcape (vol.1.4). Zenodo, URL https://doi.org/10.5281/zenodo.5270332, https://doi.org/10.5281/zenodo.5270332.
  64. [66]León-Cruz, J. F., 2025: Tornadic environments in mexico. Atmospheric Research, 315, 107916, https://doi.org/10.1016/j.atmosres.2025.107916.
  65. [67]León-Cruz, J. F., L. F. Pineda-Martínez, and N. Carbajal, 2022: Tornado climatology and potentially severe convective environments in Mexico. Climate Research, 87, 147–165, https://doi.org/10.3354/cr01692.
  66. [68]Lin, P.-F., L. K. M. Lagrimas, J.-B. Wang, P.-L. Chang, W.-T. Fang, J. G. C. Almario, B. J.-D. Jou, and E. O. Cayanan, 2021: Removing interference echoes in Philippine radars using a fuzzy logic approach. Terr. Atmos. Ocean. Sci., 32 (5.1), 755–776, https://doi.org/10.3319/TAO.2021.12.20.01.
  67. [69]Lopes, M. M., M. I. de Oliveira, V. Goede, V. Ferreira, and E. de Lima Nascimento, 2025: Damage Survey and Multisensor Analysis of a Cold Season Tornado Outbreak in Southern Brazil. Journal of Applied Meteorology and Climatology, 64 (7), 831 – 855, https://doi.org/10.1175/JAMC-D-24-0146.1.
  68. [70]Louf, V., A. Protat, R. C. Jackson, S. M. Collis, and J. Helmus, 2020: UNRAVEL: A robust modular velocity dealiasing technique for Doppler radar. J. Atmos. Oceanic Technol., 37 (5), 741–758, https://doi.org/10.1175/JTECH-D-19-0020.1.
  69. [71]Markowski, P., C. Hannon, J. Frame, E. Lancaster, A. Pietrycha, R. Edwards, and R. L. Thompson, 2003: Characteristics of Vertical Wind Profiles near Supercells Obtained from the Rapid Update Cycle. Weather and Forecasting, 18 (6), 1262 – 1272, https://doi.org/10.1175/1520-0434(2003)018<1262:COVWPN>2.0.CO;2.
  70. [72]Markowski, P., and Y. Richardson, 2010: Mesoscale Meteorology in Midlatitudes. John Wiley & Sons, https://doi.org/10.1002/9780470682104.
  71. [73]Markowski, P. M., J. M. Straka, and E. N. Rasmussen, 2002: Direct surface thermodynamic observations within the rear-flank downdrafts of nontornadic and tornadic supercells. Monthly Weather Review, 130 (7), 1692 – 1721, https://doi.org/10.1175/1520-0493(2002)130<1692:DSTOWT>2.0.CO;2.
  72. [74]McCaul, E. W., Jr., 1991: Buoyancy and shear characteristics of hurricane-tornado environments. Monthly Weather Review, 119 (8), 1954–1978.doi.org/10.1175/1520-0493(1991)119<1954:bascoh>2.0.co;2
  73. [75]McLaughlin, D., and Coauthors, 2009: Short-wavelength technology and the potential for distributed networks of small radar systems. Bulletin of the American Meteorological Society, 90 (12), 1797–1817, https://doi.org/10.1175/2009BAMS2507.1.
  74. [76]Mead, C. M., and R. L. Thompson, 2011: Environmental characteristics associated with nocturnal significant-tornado events in the Great Plains. Electronic Journal of Severe Storms Meteorology, 6 (6).
  75. [77]Meng, Z., and Coauthors, 2018: The deadliest tornado (EF4) in the past 40 years in China. Weather and Forecasting, 33 (3), 693–713, https://doi.org/10.1175/WAF-D-17-0085.1. Letters, 49 (11), e2022GL098242, https://doi.org/10.1029/2022GL098242, https://agupubs.onlinelibrary.wiley.com/doi/pdf/10.1029/2022GL098242.
  76. [78]Planet Team, 2024: Technical specification: Analysis-ready planetscope (arps). Version 1.0.0, October 2024; accessed April 2026, https://docs.planet.com/data/imagery/arps/techspec/.
  77. [79]Púčik, T., and Coauthors, 2024: The Violent Tornado on 24 June 2021 in Czechia: Damage Survey, Societal Impacts, and Lessons Learned. Weather, Climate, and Society, 16 (3), 411 – 429, https://doi.org/10.1175/WCAS-D-23-0080.1.
  78. [80]Rasmussen, E. N., 2003: Refined supercell and tornado forecast parameters. Weather and Forecasting, 18 (3), 530 – 535, https://doi.org/10.1175/1520-0434(2003)18<530:RSATFP>2.0.CO;2.
  79. [81]Rasmussen, E. N., and D. O. Blanchard, 1998: A Baseline Climatology of Sounding-Derived Supercell and Tornado Forecast Parameters. Weather and Forecasting, 13 (4), 1148 – 1164, https://doi.org/10.1175/1520-0434(1998)013<1148:ABOCSD>2.0.CO;2.
  80. [82]Rasmussen, E. N., J. M. Straka, M. S. Gilmore, and R. Davies-Jones, 2006: A Preliminary Survey of Rear-Flank Descending Reflectivity Cores in Supercell Storms. Weather and Forecasting, 21 (6), 923 – 938, https://doi.org/10.1175/WAF962.1.
  81. [83]Ronneberger, O., P. Fischer, and T. Brox, 2015: U-Net: Convolutional Networks for Biomedical Image Segmentation. Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015, N. Navab, J. Hornegger, W. M. Wells, and A. F. Frangi, Eds., Springer International Publishing, Cham, 234–241, https://doi.org/10.1007/978-3-319-24574-4_28.
  82. [84]Ryzhkov, A. V., T. J. Schuur, D. W. Burgess, and D. S. Zrnic, 2005: Polarimetric Tornado Detection. Journal of Applied Meteorology, 44 (5), 557 – 570, https://doi.org/10.1175/JAM2235.1.
  83. [85]Schaefer, M., R. Teeuw, S. Day, D. Zekkos, P. Weber, T. Meredith, and C. J. van Westen, 2020: Low-cost UAV surveys of hurricane damage in Dominica: automated processing with co-registration of pre-hurricane imagery for change analysis. Natural Hazards, 101, 755–784, https://doi.org/10.1007/s11069-020-03893-1.
  84. [86]Shapiro, A., E. Fedorovich, and S. Rahimi, 2016: A unified theory for the Great Plains nocturnal low-level jet. Journal of the Atmospheric Sciences, 73 (8), 3037–3057.
  85. [87]Sherburn, K. D., and M. D. Parker, 2014: Climatology and Ingredients of Significant Severe Convection in High-Shear, Low-CAPE Environments. Weather and Forecasting, 29 (4), 854 – 877, https://doi.org/10.1175/WAF-D-13-00041.1.
  86. [88]Silva Dias, M. A. F., 2011: An Increase in the Number of Tornado Reports in Brazil. Weather, Climate, and Society, 3 (3), 209 – 217, https://doi.org/10.1175/2011WCAS1095.1.
  87. [89]Smith, B. T., R. L. Thompson, D. A. Speheger, A. R. Dean, C. D. Karstens, and A. K. Anderson-Frey, 2020: WSR-88D tornado intensity estimates. Part II: Real-time applications to tornado warning time scales. Weather and Forecasting, 35 (6), 2493–2506, https://doi.org/10.1175/WAF-D-20-0011.1.
  88. [90]Snyder, J. C., and A. V. Ryzhkov, 2015: Automated detection of polarimetric tornadic debris signatures using a hydrometeor classification algorithm. Journal of Applied Meteorology and Climatology, 54 (9), 1861–1870, https://doi.org/10.1175/JAMC-D-15-0138.1.
  89. [91]Mercer, A. E., C. M. Shafer, C. A. Doswell, L. M. Leslie, and M. B. Richman, 2012: Synoptic Composites of Tornadic and Nontornadic Outbreaks. Monthly Weather Review, 140 (8), 2590 – 2608, https://doi.org/10.1175/MWR-D-12-00029.1.
  90. [92]Nielsen, E. R., and R. S. Schumacher, 2018: Dynamical insights into extreme short-term precipitation associated with supercells and mesovortices. Journal of the Atmospheric Sciences, 75 (9), 2983 – 3009, https://doi.org/10.1175/JAS-D-17-0385.1.
  91. [93]Nielsen, E. R., and R. S. Schumacher, 2020: Dynamical mechanisms supporting extreme rainfall accumulations in the houston “tax day” 2016 flood. Monthly Weather Review, 148 (1), 83 – 109, https://doi.org/10.1175/MWR-D-19-0206.1.
  92. [94]Nixon, C. J., and J. T. Allen, 2021: Anticipating Deviant Tornado Motion Using a Simple Hodograph Technique. Weather and Forecasting, 36 (1), 219 – 235, https://doi.org/10.1175/WAF-D-20-0056.1.
  93. [95]Nixon, C. J., and J. T. Allen, 2022: Distinguishing between Hodographs of Severe Hail and Tornadoes. Weather and Forecasting, 37 (10), 1761 – 1782, https://doi.org/10.1175/WAF-D-21-0136.1.
  94. [96]Nixon, C. J., J. T. Allen, M. B. Wilson, M. J. Bunkers, and M. Taszarek, 2024: Cell Mergers, Boundary Interactions, and Convective Systems in Cases of Significant Tornadoes and Hail. Weather and Forecasting, 39 (10), 1435 – 1458, https://doi.org/10.1175/WAF-D-23-0117.1.
  95. [97]Nowotarski, C. J., and E. A. Jones, 2018: Multivariate Self-Organizing Map Approach to Classifying Supercell Tornado Environments Using Near-Storm, Low-Level Wind and Thermodynamic Profiles. Weather and Forecasting, 33 (3), 661 – 670, https://doi.org/10.1175/WAF-D-17-0189.1.
  96. [99]Overeem, A., R. Uijlenhoet, and H. Leijnse, 2020: Full-Year Evaluation of Nonmeteorological Echo Removal with Dual-Polarization Fuzzy Logic for Two C-Band Radars in a Temperate Climate. Journal of Atmospheric and Oceanic Technology, 37 (9), 1643 – 1660, https://doi.org/10.1175/JTECH-D-19-0149.1.
  97. [100]Pazmany, A. L., J. B. Mead, H. B. Bluestein, J. C. Snyder, and J. B. Houser, 2013: A mobile rapid-scanning X-band polarimetric (RaXPol) Doppler radar system. Journal of Atmospheric and Oceanic Technology, 30 (7), 1398–1413, https://doi.org/10.1175/JTECH-D-12-00166.1.
  98. [101]Peters, J. M., D. R. Chavas, C.-Y. Su, H. Morrison, and B. E. Coffer, 2023: An Analytic Formula for Entraining CAPE in Midlatitude Storm Environments. Journal of the Atmospheric Sciences, 80 (9), 2165 – 2186, https://doi.org/10.1175/JAS-D-23-0003.1.
  99. [102]Peters, J. M., J. P. Mulholland, and D. R. Chavas, 2022: Generalized Lapse Rate Formulas for Use in Entraining CAPE Calculations. Journal of the Atmospheric Sciences, 79 (3), 815 – 836, https://doi.org/10.1175/JAS-D-21-0118.1.
  100. [103]Pilguj, N., M. Taszarek, J. T. Allen, and K. A. Hoogewind, 2022a: Are Trends in Convective Parameters over the United States and Europe Consistent between Reanalyses and Observations? Journal of Climate, 35 (12), 3605 – 3626, https://doi.org/10.1175/JCLI-D-21-0135.1.
  101. [104]Pilguj, N., M. Taszarek, M. Kryza, and H. E. Brooks, 2022b: Reconstruction of Violent Tornado Environments in Europe: High-Resolution Dynamical Downscaling of ERA5. Geophysical Research Takahashi, N., and Coauthors, 2019: Development of multi-parameter phased array weather radar (MP-PAWR) and early detection of torrential rainfall and tornado risk. Journal of Disaster Research, 14 (2), 235–247, https://doi.org/10.20965/jdr.2019.p0235.
  102. [105]Taszarek, M., J. T. Allen, P. Groenemeijer, R. Edwards, H. E. Brooks, V. Chmielewski, and S.-E. Enno, 2020a: Severe Convective Storms across Europe and the United States. Part I: Climatology of Lightning, Large Hail, Severe Wind, and Tornadoes. Journal of Climate, 33 (23), 10 239 – 10 261, https://doi.org/10.1175/JCLI-D-20-0345.1.
  103. [106]Taszarek, M., J. T. Allen, M. Marchio, and H. E. Brooks, 2021a: Global climatology and trends in convective environments from ERA5 and rawinsonde data. npj Climate and Atmospheric Science, 4 (1), 35, https://doi.org/10.1038/s41612-021-00190-x.
  104. [107]Taszarek, M., J. T. Allen, T. Púčik, K. A. Hoogewind, and H. E. Brooks, 2020b: Severe Convective Storms across Europe and the United States. Part II: ERA5 Environments Associated with Lightning, Large Hail, Severe Wind, and Tornadoes. Journal of Climate, 33 (23), 10 263 – 10 286, https://doi.org/10.1175/JCLI-D-20-0346.1.
  105. [108]Taszarek, M., N. Pilguj, J. T. Allen, V. Gensini, H. E. Brooks, and P. Szuster, 2021b: Comparison of Convective Parameters Derived from ERA5 and MERRA-2 with Rawinsonde Data over Europe and North America. Journal of Climate, 34 (8), 3211 – 3237, https://doi.org/10.1175/JCLI-D-20-0484.1.
  106. [109]Thompson, R. L., and R. Edwards, 2000: An Overview of Environmental Conditions and Forecast Implications of the 3 May 1999 Tornado Outbreak. Weather and Forecasting, 15 (6), 682–699, https://doi.org/10.1175/1520-0434(2000)015<0682:AOOECA>2.0.CO;2.
  107. [110]Thompson, R. L., R. Edwards, J. A. Hart, K. L. Elmore, and P. Markowski, 2003: Close Proximity Soundings within Supercell Environments Obtained from the Rapid Update Cycle. Weather and Forecasting, 18 (6), 1243 – 1261, https://doi.org/10.1175/1520-0434(2003)018<1243:CPSWSE>2.0.CO;2.
  108. [111]Thompson, R. L., R. Edwards, and C. M. Mead, 2004: An update to the supercell composite and significant tornado parameters. Preprints, 22nd Conf. on Severe Local Storms, Amer. Meteor. Soc., Hyannis, MA, P8.1.
  109. [112]Thompson, R. L., B. T. Smith, J. S. Grams, A. R. Dean, and C. Broyles, 2012: Convective Modes for Significant Severe Thunderstorms in the Contiguous United States. Part II: Supercell and QLCS Tornado Environments. Weather and Forecasting, 27 (5), 1136 – 1154, https://doi.org/10.1175/WAF-D-11-00116.1.
  110. [113]Wagner, M., R. K. Doe, A. Johnson, Z. Chen, J. Das, and R. S. Cerveny, 2019: Unpiloted Aerial Systems (UASs) Application for Tornado Damage Surveys: Benefits and Procedures. Bulletin of the American Meteorological Society, 100 (12), 2405 – 2409, https://doi.org/10.1175/BAMS-D-19-0124.1.
  111. [114]Wakimoto, R. M., D. J. Bodine, H. B. Bluestein, T. Greenwood, and S. Emmerson, 2025: Kinematic Structure of a Tornado Based on an Analysis of Lofted Debris Speeds, Debris Orientation, and Mobile Radar Data. Monthly Weather Review, 153 (8), 1365 – 1383, https://doi.org/10.1175/MWR-D-24-0176.1.
  112. [115]Waldteufel, P., and H. Corbin, 1979: On the analysis of single-Doppler radar data. Journal of Applied Meteorology, 18 (4), 532–542.doi.org/10.1175/1520-0450(1979)018<0532:otaosd>2.0.co;2
  113. [116]Wang, X., H. Iwabuchi, and T. Yamashita, 2022: Cloud identification and property retrieval from Himawari-8 infrared measurements via a deep neural network. Remote Sensing of Environment, 275, 113 026, https://doi.org/10.1016/j.rse.2022.113026.
  114. [117]Weisman, M. L., and J. B. Klemp, 1986: Characteristics of Isolated Convective Storms, 331–358. American Meteorological Society, Boston, MA, https://doi.org/10.1007/978-1-935704-20-1_15.
  115. [118]Wightman, R., 2019: PyTorch Image Models. GitHub, https://github.com/rwightman/pytorch-image-models, https://doi.org/10.5281/zenodo.4414861.
  116. [119]Wood, V. T., and R. A. Brown, 1997: Effects of radar sampling on single-Doppler velocity signatures of mesocyclones and tornadoes. Wea. Forecasting, 12 (4), 928–938, https://doi.org/10.1175/1520-0434(1997)012<0928:EORSOS>2.0.CO;2.
  117. [120]Wright, N., J. M. A. Duncan, J. N. Callow, S. E. Thompson, and R. J. George, 2025: Training sensor-agnostic deep learning models for remote sensing: Achieving state-of-the-art cloud and cloud shadow identification with OmniCloudMask. Remote Sensing of Environment, 322, 114 694, https://doi.org/10.1016/j.rse.2025.114694.
  118. [121]Wurman, J., K. Kosiba, T. White, and P. Robinson, 2021: Supercell tornadoes are much stronger and wider than damage-based ratings indicate. Proceedings of the National Academy of Sciences, 118 (14), e2021535118, https://doi.org/10.1073/pnas.2021535118.
  119. [122]Xie, M., N. Jean, M. Burke, D. Lobell, and S. Ermon, 2016: Transfer Learning from Deep Features for Remote Sensing and Poverty Mapping. Proceedings of the AAAI Conference on Artificial Intelligence, 30 (1), https://doi.org/10.1609/aaai.v30i1.9906.
  120. [124]Yuan, C., Y. Bai, P. Sun, H. Bai, and L. Xia, 2024: Microphysical Insights into a Tornadic Supercell from Dual-Polarization Radar Observations in Jiangsu, China on 14 May 2021. Journal of Meteorological Research, 38 (2), 303–320, https://doi.org/10.1007/s13351-024-3102-5.

Paper details

vlnc:W_rhz8xtdf
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents