KTLX rainfall and KDP case · Single-case demonstration

Early Detection of Heavy Convective Rainfall from a KTLX K_DP Column: 29 April 2021 Case Demonstration

Archived KTLX S-band dual-polarization observations examined for an early radar signature preceding heavy convective rainfall near Norman, Oklahoma. Ten radar volumes, 01:33:08–02:27:56 UTC on 29 April 2021. All times UTC.

What was processed

Ten KTLX Level-II volumes spanning 01:33:08–02:27:56 UTC on 29 April 2021 were processed for specific differential phase (K_DP), reconstructed into storm-following vertical cross sections, and paired with MRMS surface rainfall at a fixed point near Norman, Oklahoma. This page reports what came out; the figures and the tables carry the detail.

Under the strict quality-control mask, the first K_DP column reaching 1 ° km⁻¹ within 5 km of the tracked core occurs at 01:50:19 UTC. MRMS rainfall at the fixed point begins at 02:02 UTC, and the peak 15-minute accumulation ends at 02:14 UTC. The three volumes before 01:50 extend radar coverage back to 01:33:08 — 28.9 minutes before rainfall onset — and contain no gate meeting the strict column threshold. Under the permissive mask, which admits lower ρ_HV gates, elevated K_DP appears earlier.

The case

Convection moved across the Norman, Oklahoma area late on 28 April into early 29 April 2021, producing a flood report at 20:05 CST (02:05 UTC). The event was warned operationally: NWS Norman issued Flash Flood Warning event 10 at 01:53 UTC for McClain and northwestern Cleveland Counties, valid to 05:00 UTC, naming Norman, Moore, Newcastle, Goldsby, and southern Oklahoma City. The archived product records it as radar-indicated and states that flash flooding was ongoing or expected shortly, with another 1–2 inches possible. The analysis window here, 01:48–02:28 UTC, came from an automated screening pass and was accepted after manual review of corrected radar quick looks.

Sequence of events

Rainfall totals at the fixed target

• Fixed-point rainfall onset: 02:02 UTC.
• Peak 15-minute accumulation: 13.92 mm ending 02:14 UTC, 93.3 % temporal coverage.
• Peak 30-minute accumulation: 15.20 mm ending 02:28 UTC, 100 % coverage.
• 30-minute neighbourhood mean 13.78 mm; neighbourhood maximum 22.35 mm.
• Rolling radar-only one-hour QPE at the target: 15.20 mm at 02:26 UTC.
• The instantaneous target rain rate reaches 150 mm h⁻¹ at 02:06 UTC. This is an instantaneous rate, not an accumulation or an event total.

These are MRMS radar-derived totals at a single fixed pixel plus a 5-by-5-pixel neighbourhood. No return-period statistics were computed and no gauge comparison was made, so the totals stand on their own without a rarity or verification claim attached.

Evolution of the column

Vertical structure was reconstructed as KTLX pseudo-RHIs — vertical cross sections assembled from PPI volume scans, not true operational RHI scans. The panels below are radar-centered: zero range is KTLX and the dashed line marks the moving tracked core. This view keeps the changing range and azimuth visible: over the sequence the tracked core moves from 43.64 km to 6.39 km and back to 14.18 km, so the set of elevations sampled above it changes from volume to volume.

Column metrics

Metrics are computed within 5 km horizontally of the tracked core and below 12 km MSL. A missing threshold-top value means no valid selected gate met that threshold; it does not mean a zero-kilometre column.

Method

Data

Radar is KTLX S-band WSR-88D Level II from the public unidata-nexrad-level2 S3 bucket. The full inventory covers 40 volumes from 00:00–04:00 UTC; the demonstration sequence uses the 10 volumes from 01:33:08–02:27:56 UTC. Native fields used are reflectivity, differential reflectivity, differential phase, and cross-correlation ratio. K_DP is calculated from native KTLX differential phase; MRMS is never used as a K_DP source.

Rainfall context comes from MRMS PrecipRate, RadarOnly_QPE_01H, and RadarQualityIndex (public noaa-mrms-pds bucket) at 2-minute resolution, over 34.8–35.6 °N and 98.0–96.8 °W, with a fixed verification point at 35.22 °N, 97.44 °W. Missing intervals are not filled with zero. Rain rate is integrated by the trapezoidal rule only when both segment endpoints are finite and the gap is no longer than 5 minutes. The accumulation map is masked where temporal coverage falls below 90 %.

Dynamic sweep selection

KTLX changed scan strategy within the window, with 16–23 sweeps per volume, so sweep indices are never copied between volumes. For each volume, every sweep is inspected for coverage of the four required fields; repeated elevations are grouped when fixed angles differ by no more than 0.1°; groups missing required dual-polarization fields are excluded; and within each repeated-elevation group the sweep maximizing the minimum valid fraction among required fields is selected.

K_DP retrieval and gate quality control

Specific differential phase is one half of the range derivative of differential phase, K_DP = ½ dΦ_DP/dr, reported in ° km⁻¹. It responds to liquid-water content and drop-size distribution while being less affected by attenuation and calibration than reflectivity.

Retrieval uses Py-ART 2.3.0 pyart.retrieve.kdp_maesaka in native polar coordinates, one selected sweep at a time, with Clpf=1.0, first_guess=0.01, and finite_order="low". System phase is estimated independently for each sweep with det_sys_phase_gf using the permissive gate filter and first_gate=30. Returned K_DP is explicitly remasked after retrieval, because passing a Py-ART GateFilter does not guarantee the returned Maesaka array retains every QC exclusion. Missing and excluded gates stay masked; they are never converted to zero K_DP.

ρ_HV describes the correlation between horizontally and vertically polarized returns. Values near one generally indicate a more uniform scatterer population. Lower values can occur with hail, melting or mixed-phase particles, biological targets, low signal-to-noise ratio, sidelobes, or nonuniform beam filling. It is a quality indicator, not a unique hail detector.

Storm tracking

The tracked object is a range-area-weighted reflectivity centroid computed on the lowest selected dual-polarization sweep: gates at or above 45 dBZ within 20 km of the previous target, each weighted by radar range multiplied by (reflectivity − 45 dBZ + 1) to reduce polar sampling-area bias.

The seven targets from 01:50–02:28 UTC were reviewed against reflectivity overlays and accepted. The 01:33, 01:38, and 01:44 UTC targets are an objective backward extension anchored to the accepted 01:50 target, with backward steps of 6.47–6.92 km, well within the 20 km search radius. The ten locations are therefore not all independently manually reviewed, and the three antecedent extensions should be read as such.

The reviewed candidates are shown in Figure 1.

Pseudo-RHI construction

For each volume the azimuth from KTLX to the tracked centroid is computed; at every dynamically selected elevation the nearest PPI ray to that azimuth is selected; the section is rejected if any nearest-ray offset exceeds 1°; and the section is built with pyart.util.cross_section_ppi, preserving native fields and masks in a CF/Radial NetCDF output. The largest observed nearest-ray offset is 0.484°.

The storm-centered horizontal coordinate is x_storm = √(x_radar² + y_radar²) − R_target, so zero marks the tracked core. Height is geometric gate height plus radar altitude, shown in kilometres MSL. Sections span ±20 km around the target and up to 12 km MSL.

Sensitivity

Alternative retrieval algorithms

The 02:02 UTC lowest selected sweep (0.483°) was processed with the same strict mask using Maesaka, Vulpiani, and Schneebeli retrievals. All three use S-band settings; Vulpiani and Schneebeli additionally use Py-ART's ρ_HV-aware phase prefilter. Only gates valid in each method and in Maesaka are compared.

Maesaka is the retrieval used everywhere else in this report; the other two were run only on this one sweep for comparison. The optional LP/Sobel phase product was also attempted and hit infeasible solver behaviour, so it produced no usable field here.

Strict, permissive, and unfiltered fields

Three K_DP fields are carried through the processing. All use the same Maesaka settings and differ only in which gates are admitted.

The permissive mask admits gates the strict mask rejects, and under it elevated K_DP appears earlier in the sequence. Those extra gates have lower ρ_HV, which occurs with mixed-phase or melting particles, hail, low signal-to-noise, sidelobes, or nonuniform beam filling.

The unfiltered diagnostic field drops the thresholds entirely, so the blank areas of the strict field fill in. What appears there can include mixed-phase and hail returns, low-SNR gates, and unstable phase-gradient artefacts, so it is not rainfall-quality K_DP and contributes to no column metric, lead time, or rainfall interpretation. It is shown beside the strict field so the extent and cause of the strict-field gaps are visible.

Where strict-mask K_DP is NaN, the gate was missing or excluded by QC. A blank region can also mean nothing was sampled there: the top elevation in these volumes is 19.51°, so over a core at 6.4 km range no beam reaches above roughly 2.5 km MSL, whatever the masking.

Reproducibility

Supplementary figures

Forty low-level KTLX reflectivity quick looks spanning 00:04–03:55 UTC are in figures/radar_quicklooks/; they were used for manual screening and storm-target refinement.