MACHINE LEARNING APPROACHES OF WIVERN L2 PRODUCTS OF LWP AND IWP AND CONSEQUENCES ON PATH INTEGRATED ATTENUATION

P. Sanò¹, M. Montopoli^1,2, D. Casella¹, M. Coppola³, G. Panegrossi¹, A. Battaglia³

¹Consiglio Nazionale delle ricerche, Istituto di Scienze dell’Atmosfera e del Clima (CNR-ISAC), ²Centro di Eccellenza in Telerilevamento E Modellistica Previsionale di eventi Severi (CETEMPS) Università Dell’Aquila, ³Dipartimento di Ingegneria dell’Ambiente, del Territorio, Politecnico di Torino,

The undisputed success of NASA’s Cloudsat mission (which carried a W-band radar pointing at the nadir), lasted 17 years, demonstrated how the measurement of cloud profiles on a global scale was one of the key factors in improving the quantification of the planet’s energy and water balance. Building on this momentum, the EarthCARE satellite was launched in May 2024, carrying, among other sensors, a nadir-pointing W-band radar with Doppler capability. The radar on the ESA-JAXA EarthCARE mission, with its high sensitivity (-35dBZ), promises to provide a much more accurate picture of global cloud cover, and with its Doppler capability, it will provide information on cloud dynamics and precipitation from a satellite for the first time. EarthCARE’s nominal lifetime is three years from launch, with possible extensions. The future scenario, although fraught with uncertainties, including geopolitical ones, sees, among other initiatives, the ESA-WIVERN mission which is in its B1 phase, moving towards a full implementation. WIVERN is a never-seen-before concept mission carrying a W-band conical scanning Doppler radar with a simultaneous 89GHZ radiometric channel.

This work covers two aspects aspect of the WIVERN L2 products: i) retrieval performances of Ice Water content (IWC) and Liquid water path (LWP) exploiting collocated WIVERN radar reflectivity profiles (Z), brightness temperature (Tb) and ancillary information (e.g. Temperature, surface type and freezing level altitude); ii) how LWP uncertainty maps into total path attenuation retrieval that is the main driver to compensate the extinction of Z profiles in rain.

The analysis proposed uses WIVERN radar and radiometer simulations from CAPTIVATE dataset (a set of microphysical retrievals from 1199 polar sunsynchronous orbits of the A-Train 1D of multiple instruments, radar, lidar and IR), and a two-step multi model machine learning approach to retrieve IWP and LWP. The path attenuation compensation techniques implemented are those within the Hitschfeld-Bordan suite.

The results are very promising, showing a BIAS and Fractional Standard Error (worst case) of the order of 20% and 40% for IWC, respectively, whereas for LPW the scores reduce to 5% and 10% respectively. About path integrated attenuation (PIA) compensation on Z, results suggest an RMSE of 4dB in compensating Z when total PIA is known with an uncertainty at worst of 5 dB.