On the Potential of the Advance Very High-Resolution Radiometer for Precipitation Retrieval in High Latitudes Using Machine Learning

Abstract:

Precipitation retrieval has been an exigent topic, especially in high latitudes (HL) (i.e. poleward of 50°) and over frozen surfaces. Given the importance of quantifying HL precipitation, and ample challenges that the current precipitation products face over these regions, the present research investigates the potential of the Advanced Very High-Resolution Radiometer (AVHRR) for precipitation retrieval in HL using CloudSat (CS) and Machine Learning (ML). CS provides direct observations of snow over land and ocean and light rainfall over the ocean in HL with unprecedented signal sensitivity and can be considered as the baseline for precipitation rate in HL. AVHRR may also be suitable for HL precipitation retrieval because 1) AVHRR data have been collected continuously since 1981 from multiple platforms, and 2) PMW precipitation products have several issues over snow and ice surfaces and may not have sufficient sensitivity to detect and estimate light rainfall and snowfall. ML can help to establish a robust relationship between CS precipitation and brightness temperature and cloud properties from AVHRR, providing a valuable alternative to mitigate precipitation retrieval challenges in HL with broad applications in precipitation products that are extended to the poles. AVHRR precipitation rate is retrieved from brightness temperature together with cloud probability and auxiliary variables such as 2-meter wet-bulb temperature and total precipitable water (TPW). The results indicate that the AVHRR-ML algorithm is capable of both detection and estimation of precipitation with a relatively acceptable accuracy considering the limitations of CS and AVHRR. Also, it was found that the addition of TPW as a feature to the ML algorithm is more effective for estimation than it is for detection. A comparison of AVHRR-ML product with Atmospheric Infrared Sounder (AIRS) product indicates that the AVHRR-ML can provide precipitation estimates in HL comparable to AIRS. AVHRR-ML and AIRS are also compared with other products in HL, suggesting that AVHRR-ML provides reasonable estimates that are spatially and temporally comparable to other products such as reanalysis and those that employ in situ data.