Application of a Dynamic Clustered Bayesian Model Averaging (DCBA) Algorithm for Merging Multisatellite Precipitation Products over Pakistan

• Published in: Journal of hydrometeorology 2020-01, Vol.21 (1), p.17-37
• Main Authors: Rahman, Khalil Ur, Shang, Songhao, Shahid, Muhammad, Wen, Yeqiang, Khan, Zeeshan
• Format: Article
• Language: English
• Subjects: Algorithms; Arid regions; Artificial neural networks; Atmospheric precipitations; Bayesian analysis; Climate; Climate prediction; Climatic data; Correlation coefficient; Correlation coefficients; Datasets; Extreme weather; Gauges; Hydrologic studies; Hydrology; Monsoon precipitation; Monsoons; Morphing; Neural networks; Precipitation; Precipitation estimation; Precipitation gauges; Precipitation trends; Principal components analysis; Probability theory; Rain; Rainfall; Rainfall measurement; Remote sensing; Root-mean-square errors; Seasons; Spatial resolution; Studies; Topography; Tropical climate; Tropical rainfall; Tropical Rainfall Measuring Mission (TRMM); Weather forecasting
• ISSN: 1525-755X; 1525-7541
• DOI: 10.1175/JHM-D-19-0087.1

Merged multisatellite precipitation datasets (MMPDs) combine the advantages of individual satellite precipitation products (SPPs), have a tendency to reduce uncertainties, and provide higher potentials to hydrological applications. This study applied a dynamic clustered Bayesian model averaging (DCBA) algorithm to merge four SPPs across Pakistan. The DCBA algorithm produced dynamic weights to different SPPs varying both spatially and temporally to accommodate the spatiotemporal differences of SPP performances. The MMPD is developed at daily temporal scale from 2000 to 2015 with spatial resolution of 0.25° using extensively evaluated SPPs and a global atmospheric reanalysis–precipitation dataset: Tropical Rainfall Measurement Mission (TRMM) Multisatellite Precipitation Analysis (TMPA) 3B42V7, Precipitation Estimation from Remotely Sensed Information Using Artificial Neural Networks–Climate Data Record (PERSIANN-CDR), Climate Prediction Center morphing technique (CMORPH), and ERA-Interim. The DCBA algorithm is evaluated across four distinct climate regions of Pakistan over 102 ground precipitation gauges (GPGs). DCBA forecasting outperformed all four SPPs with average Theil’s U of 0.49, 0.38, 0.37, and 0.36 in glacial, humid, arid, and hyperarid regions, respectively. The average mean bias error (MBE), mean error (MAE), root-mean-square error (RMSE), correlation coefficient (CC), and standard deviation (SD) of DCBA over all of Pakistan are 0.54, 1.40, 4.94, 0.77, and 5.17 mm day −1, respectively. Seasonal evaluation revealed a dependency ofDCBAperformance on precipitation magnitude/intensity and elevation. Relatively poor DCBA performance is observed in premonsoon/monsoon seasons and at high/mild elevated regions. Average improvements of DCBA in comparison with TMPA are 59.56% (MBE), 49.37% (MAE), 45.89% (RMSE), 19.48% (CC), 46.7% (SD), and 18.66% (Theil’s U. Furthermore, DCBA efficiently captured extreme precipitation trends (premonsoon/monsoon seasons).