Water Budget Closure in the Upper Chao Phraya River Basin, Thailand Using Multisource Data

Accurate quantification of the terrestrial water cycle relies on combinations of multisource datasets. This analysis uses data from remotely sensed, in-situ, and reanalysis records to quantify the terrestrial water budget/balance and component uncertainties in the upper Chao Phraya River Basin from May 2002 to April 2020. Three closure techniques are applied to merge independent records of water budget components, creating up to 72 probabilistic realizations of the monthly water budget for the upper Chao Phraya River Basin. An artificial neural network (ANN) model is used to gap-fill data in and between GRACE and GRACE-FO-based terrestrial water storage anomalies. The ANN model performed well with r & GE; 0.95, NRMSE = 0.24 - 0.37, and NSE & GE; 0.89 during the calibration and validation phases. The cumulative residual error in the water budget ensemble mean accounts for ~15% of the ensemble mean for both the precipitation and evapotranspiration. An increasing trend of 0.03 mm month(-1) in the residual errors may be partially attributable to increases in human activity and the relative redistribution of biases among other water budget variables. All three closure techniques show similar directions of constraints (i.e., wet or dry bias) in water budget variables with slightly different magnitudes. Our quantification of water budget residual errors may help benchmark regional hydroclimate models for understanding the past, present, and future status of water budget components and effectively manage regional water resources, especially during hydroclimate extremes.

Automated summary

Accurate quantification of the terrestrial water cycle in the upper Chao Phraya River Basin is achieved through the combination of multisource datasets. The analysis considers remotely sensed, in-situ, and reanalysis records to calculate the terrestrial water budget and uncertainties of its components. Different closure techniques are applied, resulting in multiple probabilistic realizations of the monthly water budget. An artificial neural network model is used to fill data gaps and improve accuracy. The cumulative residual error in the water budget ensemble mean accounts for approximately 15% of the ensemble mean for precipitation and evapotranspiration. The study findings provide important insights for regional hydroclimate modeling and water resource management.