SPECIALIZED ALGORITHMS FOR TROPICAL CYCLONE WIND RETRIEVALS WITH SATELLITE MW RADIOMETERS

L. Ricciardulli¹, T. Meissner¹

¹Remote Sensing Systems

Space-based wind measurements in extreme environments, as those seen in Tropical Cyclones (TC), have historically been challenging because of the presence of rain, which impacts the observed signal and is hard to discern from the wind signal. Recent years saw important advancements in extreme wind measurements in storms with microwave (MW) radiometers, brought first by L-band sensors, such as SMAP or SMOS, which retain very good sensitivity to winds (10-70 m/s) without being affected by rain. Other radiometers such as WindSat and the AMSR series also allow extreme measurements in rain by taking advantage of the different sensitivity to rain of C- and X-band channels, but they require statistical algorithms specially trained for extreme wind speeds and in intense rain.

While performing statistically well for intense wind speeds (above 33 m/s, the Category 1 lower threshold), where the wind-induced signals are stronger than the rain ones, the C/X-band empirical algorithms present residual biases, particularly significant in the 20-30 m/s storm-wind regime. An examination of many such biased cases in the AMSR2 TC-wind retrievals revealed that they were associated with two types of situations: (1) wind retrievals in the TC rain bands, or (2) in regions in the core of the storm with unusually cold brightness temperatures (TB) in the 89 GHz (W-band) channels (lower than 200 K), indicating areas of very deep convective towers. These biases can result in overestimates of the storm intensity and size. This poses a problem for the operational forecast community, when ingesting these parameters into their automated forecast systems.

In order to mitigate these biases, we developed a new empirical linear regression algorithm for AMSR2 that utilizes all frequency channels from C- to W-band, and it is trained specifically in TC conditions. It turns out that including the W-band channels is essential to improve the performance of the AMSR TC-wind algorithm. The new algorithm has been trained using an extensive set of storm scenes from the Hurricane Weather Research and Forecast (HWRF) model. The HWRF wind fields were resampled at the AMSR2 spatial resolution (25 km grid), interpolated at the satellite time and shifted so that the storm centers of HWRF and satellite passes are matching. The training dataset was based on 38 TCs in the period 2018-2023 in all basins (~200 storm scenes; ~150000 colocations). The new C to W-band algorithm significantly mitigates the residual spurious rain/cloud ice bias. The same algorithm methodology can be easily extended to other radiometers with the same frequency range (i.e., AMSR3). We also plan to explore the potential skill of all-weather wind algorithms for radiometers without the C-band (for instance GMI or WSF-M).

The presentation will cover the basic methodology for the new C-W-band wind algorithm, it will summarizes some validation activity of the AMSR2 TC-wind fields, which were processed with the new algorithm for recent storms in 2023-25. The validation includes comparisons of the AMSR TC-winds with wind fields from a testing dataset of HWRF storms, as well as comparison of satellite-derived storm features such as intensity and size with operational products such as the Best Track data and the satellite-consensus dataset SatCon.