Oral_Present_Omid_zandi

 

UofA-HIPAA V1: University of Arizona High Latitude Infrared-based Precipitation retrieval Algorithm using AVHRR sensor Version 1

Omid Zandi1, K. K. Kumah1, Ali Behrangi1

1Department of Hydrology and Atmospheric Sciences, University of Arizona, Tucson, AZ

 

As global warming accelerates, high-latitude regions experience significant changes in precipitation patterns that impact the hydrologic cycle, energy balance, and climate feedback mechanisms. Accurate long-term precipitation estimates are essential for understanding these changes. While geostationary infrared (IR) observations enable the collection of long-term precipitation records within ~60° N/S, their utility diminishes at higher latitudes due to oblique viewing angles. Passive Microwave (PMW) products face retrieval challenges over snow- and ice-covered surfaces. Currently, the Global Precipitation Climatology Project (GPCP) relies on outdated retrievals from the TIROS Operational Vertical Sounder (TOVS) and Atmospheric Infrared Sounder (AIRS) in these regions.

 

We present a novel, double Machine Learning (ML)-based approach for precipitation retrieval that utilizes brightness temperature and cloud properties from AVHRR (PATMOS-X), environmental data from the MERRA-2 reanalysis model, and surface type data from the AutoSnow global product. Coincident PMW-based precipitation estimates, CloudSat, and ERA5 were employed for training. A post-processing step refines the estimated precipitation regions by using characteristics of the brightness temperature of precipitating cloud patches.

 

Leveraging AVHRR’s four-decade-long record, this product provides a consistent high-latitude precipitation dataset, improving upon AIRS retrievals. Independent evaluation with IMERG in 2010 shows Kling-Gupta Efficiency (KGE) scores of 0.62 (55°N–90°N) and 0.54 (55°S–90°S), compared to AIRS’ 0.26 and -0.3.

 

This is the first version of our product, and efforts are ongoing to explore deep learning methods, such as Convolutional UNet models, to enhance retrieval accuracy and spatial resolution in future versions. Our work builds upon the legacy of PERSIANN precipitation production, originally developed at UofA, to further advance IR-based precipitation retrieval algorithms.