PRISM and Radar Estimation for Precipitation (PREP): PRISM enhancement through ANN and radar data integration in complex terrain

Abstract

Site-specific estimates of precipitation can be used to assess crop productivity and identify areas vulnerable to crop damages caused by extreme weather events such as droughts and floods. Spatial interpolation of precipitation such as Parameter-elevation Regressions on Independent Slopes Model (PRISM) has been used to estimate precipitation in an area of interest. However, the reliability of spatial interpolation is often affected by the availability of precipitation measurements from weather stations in a given region especially under complex terrain conditions. Here we propose an alternative approach for site-specific estimation of precipitation using both radar reflectivity data and topographic features. At first, radar reflectivity data are used as inputs to an artificial neural network (ANN) for estimation of precipitation. These radar precipitations at each grid cell are used to represent the observations at virtual weather stations for spatial interpolation using PRISM. Furthermore, the radar precipitations are compared with the observations at actual weather stations for their bias correction. This approach is referred to as PRISM and Radar Estimation for Precipitation (PREP). A case study was conducted in Jeollabuk-do, South Korea to compare the degree of agreement between PREP and PRISM. It was found that PREP had higher degree of agreement for the daily estimates of precipitation than PRISM in the given region with a complex terrain including coast and mountains. For example, the root mean square error (RMSE) of precipitation estimates for PREP was 22.1% less than that for PRISM in 2020. PREP also had greater value of the critical success index (CSI) than PRISM especially under heavy precipitation events, e.g.,>180 mm, and no rainfall conditions. These findings indicate that the PREP would improve the reliability of site-specific estimates of precipitation, which would facilitate decision-making in agriculture and early warning of extreme weather events.