Premium
Integrating precipitation zoning with random forest regression for the spatial downscaling of satellite‐based precipitation: A case study of the Lancang–Mekong River basin
Author(s) -
Zhang Jing,
Fan Hui,
He Daming,
Chen Jiwei
Publication year - 2019
Publication title -
international journal of climatology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.58
H-Index - 166
eISSN - 1097-0088
pISSN - 0899-8418
DOI - 10.1002/joc.6050
Subject(s) - downscaling , precipitation , environmental science , climatology , satellite , empirical orthogonal functions , mean squared error , normalized difference vegetation index , longitude , regression , structural basin , latitude , climate change , meteorology , geology , geography , statistics , mathematics , oceanography , aerospace engineering , engineering , paleontology , geodesy
Downscaling satellite‐based precipitation to fine scales is crucial for deepening our understanding of global hydrologic cycles and water‐related issues. In this study, a novel approach that integrates precipitation zoning with random forest regression is proposed for the spatial downscaling of satellite‐based precipitation. Precipitation zoning is delineated through iterative rotated empirical orthogonal function (REOF) analyses of ground‐ and satellite‐based precipitation observations. Random forest regression is applied to link the satellite‐based precipitation to 1‐km‐resolution predictors such as latitude (Lat), longitude (Lon), elevation, aspect, slope, and the normalized difference vegetation index (NDVI). The accuracy of the resultant downscaled precipitation is evaluated based on five statistical evaluation indices. The performance of the proposed approach is exemplified in the Lancang–Mekong River basin, taking Tropical Rainfall Measuring Mission (TRMM) 3B43 and Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks–Climate Data Record (PERSIANN–CDR) products acquired in 2001 (wet year), 2005 (normal year), and 2009 (dry year) as the databases. The results show that seven precipitation subregions can be roughly distinguished in the study basin. Zoning‐based downscaling outperforms non‐zoning‐based downscaling in terms of accuracy, resulting in statistically significant reductions in root mean square error (RMSE) and mean absolute error (MAE) of 1–17% across the entire basin. Among the selected predictors, the variables Lat and Lon are the most important for precipitation estimation, whereas the remaining variables have lesser and subregion‐dependent importance. The proposed approach is promising for generating high‐spatial‐resolution precipitation data in regions with sparse ground‐based observations and differentiated climatic regimes.