Estimating and improving the rain/no-rain detection skill of remotely sensed and reanalyzed precipitation products

Authors: DONG, J. - US Department Of Agriculture (USDA); Crow, Wade; REICHLE, R. - Goddard Space Flight Center

Submitted to: Journal of Hydrometeorology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/1/2020
Publication Date: 10/1/2020
Citation: Dong, J., Crow, W.T., Reichle, R. 2021. Estimating and improving the rain/no-rain detection skill of remotely sensed and reanalyzed precipitation products. Journal of Hydrometeorology. 21:2419:2429. https://doi.org/10.1175/JHM-D-20-0097.1.
DOI: https://doi.org/10.1175/JHM-D-20-0097.1

Interpretive Summary: Rainfall products derived from satellite observations commonly overpredict the occurrence of rain. As a result, inaccurate rain/no rain determination is a key source of error in soil moisture and streamflow forecasts that respond to these products. Poor rain/no rain determination is typically corrected using rain gauge observations, which are subject to large uncertainties in the agricultural areas of the world that lack high-quality, ground-based rainfall instrumentation. Using new statistical tools and maximum likelihood theory, this paper develops a binary rain/no-rain product that accurately estimates global rainfall occurrence using only satellite-based observations. Since it requires no ground-based inputs, this new product can be applied globally to correct rain/no-rain determination errors present in existing satellite-based rainfall products. Such enhanced products will eventually provide improved input for global agricultural drought and water resource forecasting systems

Technical Abstract: Rain/no-rain detection error is a key source of uncertainty in regional and global precipitation products that propagates forward into on-line hydrological and land surface modeling simulations. Such detection error is difficult to evaluate and/or filter without access to high-quality reference precipitation datasets. For cases where such access is not available, this study investigates the feasibility of estimating rain/no-rain detection error using categorical triple collocation (CTC) and three noisy, but independent, precipitation datasets. Through well-controlled numerical experiments, and validation using spatially dense precipitation gauge networks, we demonstrate that CTC can robustly estimate the rain/no-rain detection error of a product without reliance on gauge-based reference data. Based on the CTC-derived error estimates, a weighted merging algorithm (CTC-M) is proposed to optimally combine remote sensing and reanalysis precipitation products into a rain/no-rain time series. Both analytical solutions and numerical tests are used to demonstrate the superiority of CTC-M estimates in rain/no-rain detection relative to its parent precipitation products. As a result, CTC-M is expected to benefit global precipitation estimation by improving the estimation of precipitation occurrence for gauge-based and multi-source precipitation products. Rainfall products derived from satellite observations commonly overpredict the occurrence of rain. As a result, inaccurate rain/no rain determination is a key source of error in soil moisture and streamflow forecasts that reply on these products. Poor rain/no rain determination is typically corrected using rain gauge observations, which are subject to large uncertainties in agricultural areas of the world lacking high-quality, ground-based rainfall instrumentation. However, using new statistical tools and maximum likelihood theory, this paper develops a binary rain/no-rain product that accurately estimates global rainfall occurrence using only satellite-based observations. Since it requires no ground-based inputs, this new product can be applied globally to correct rain/no-rain determination errors present in existing satellite-based rainfall products. Such enhanced products will eventually provide improved input for global agricultural drought and water resource forecasting systems