The new Passive microwave Neural network Precipitation Retrieval (PNPR) algorithm for the cross-track scanning ATMS radiometer: description and verification study over Europe and Africa using GPM and TRMM spaceborne radars

摘要

The objective of this paper is to describe the development and evaluate theperformance of a completely new version of the Passive microwave Neural networkPrecipitation Retrieval (PNPR v2), an algorithm based on a neural networkapproach, designed to retrieve the instantaneous surface precipitation rateusing the cross-track Advanced Technology Microwave Sounder (ATMS) radiometer measurements. This algorithm, developedwithin the EUMETSAT H-SAF program, represents an evolution of the previousversion (PNPR v1), developed for AMSU/MHS radiometers (and used anddistributed operationally within H-SAF), with improvements aimed atexploiting the new precipitation-sensing capabilities of ATMS with respect toAMSU/MHS. In the design of the neural network the new ATMS channels comparedto AMSU/MHS, and their combinations, including the brightness temperaturedifferences in the water vapor absorption band, around 183 GHz, areconsidered. The algorithm is based on a single neural network, for all typesof surface background, trained using a large database based on 94cloud-resolving model simulations over the European and the African areas.The performance of PNPR v2 has been evaluated through an intercomparison ofthe instantaneous precipitation estimates with co-located estimates from theTRMM Precipitation Radar (TRMM-PR) and from the GPM Core Observatory Ku-bandPrecipitation Radar (GPM-KuPR). In the comparison with TRMM-PR, over theAfrican area the statistical analysis was carried out for a 2-year(2013–2014) dataset of coincident observations over a regular grid at0.5°  ×  0.5° resolution. The results have shown agood agreement between PNPR v2 and TRMM-PR for the different surface types.The correlation coefficient (CC) was equal to 0.69 over ocean and 0.71 overvegetated land (lower values were obtained over arid land and coast), and theroot mean squared error (RMSE) was equal to 1.30 mm h over ocean and1.11 mm h over vegetated land. The results showed a slight tendencyto underestimate moderate to high precipitation, mostly over land, andoverestimate moderate to light precipitation over ocean. Similar results wereobtained for the comparison with GPM-KuPR over the European area (15 months,from March 2014 to May 2015 of coincident overpasses) with slightly lower CC(0.59 over vegetated land and 0.57 over ocean) and RMSE (0.82 mm hover vegetated land and 0.71 mm h over ocean), confirming a goodagreement also between PNPR v2 and GPM-KuPR. The performance of PNPR v2 overthe African area was also compared to that of PNPR v1. PNPR v2 has higher over the different surfaces, with generally better estimation of lowprecipitation, mostly over ocean, thanks to improvements in the design of theneural network and also to the improved capabilities of ATMS compared toAMSU/MHS. Both versions of PNPR algorithm have shown a general consistencywith the TRMM-PR.