Water Resources Assessment of China's Transboundary River Basins Using a Machine Learning Approach

Abstract

A comprehensive and reliable assessment of the water resources in China's transboundary river basins is vital for water resources management and peaceful development. In this study, we built machine learning (random forest, gradient boosting, and stacking) and traditional linear models to identify the relation between the runoff coefficient and its influencing factors, including topography, climate, land cover, and soil. The cross‐validation results show that the machine learning models greatly outperform the traditional linear model in predicting runoff coefficient. High‐resolution (0.1°) runoff coefficient and runoff maps for the China's transboundary river basins riparian countries were produced and compared with other estimates at the country level. The best water resources estimates achieved from the machine learning model are consistent with the Food and Agriculture Organization of the United Nations AQUASTAT database (root‐mean‐square error = 76.97 km3/year, normalized root‐mean‐square error = 12%) at the country level. This outperformed two currently available runoff products: the UNH/GRDC Global Composite Runoff Fields and the Global Streamflow Characteristics Dataset. The study also demonstrated that accurate precipitation data can improve runoff and water resources estimation accuracy and that climate and topographic factors have a controlling role in prediction, whereas the influences of land cover and soils are weak. Finally, China's transboundary water resources were calculated and thoroughly assessed at basin and country levels.

Regionalization of hydrological model parameters using gradient boosting machine

“…The final GBM model is a stagewise additive model of previous individual trees. The GBM has been proven successful across many domains, including classification problems (Lawrence, 2004;Xia et al, 2020) and regression problems (Liao et al, 2020;Xenochristou et al, 2020;Yan et al, 2019) which is the case for this study. 185…”

Nghiên cứu tính toán chỉ số đánh giá tài nguyên nước mặt phục vụ cảnh báo, dự báo tài nguyên nước: Thí điểm lưu vực sông Srê Pốk

“…Một số nước trên thế giới hiện nay đã và đang sử dụng các chỉ số đánh giá tài nguyên nước phù hợp với điều kiện phát triển và yêu cầu phát triển cụ thể của từng nước để đánh giá khả năng nguồn nước của các lưu vực sông [7][8][9][10][11][12]. Ở Việt Nam trong một số báo cáo như Đánh giá tổng quan ngành nước, Chiến lược Quốc Gia về tài nguyên nước có đưa ra các chỉ số đánh giá tài nguyên nước mặt.…”

Machine Learning Predictions of pH in the Glacial Aquifer System, Northern USA

“…Variable importance and partial dependence plots (PDPs) were generated to determine the relative importance and direction of influence of predictor variables, respectively. The "gbm" package calculates the relative importance of each variable in reducing the loss function -which for continuous response terms is based on variable selection counts during splitting and the reduction of squared error (over all trees) attributable to each predictor (Friedman 2001, Yan et al 2019.…”

“…Here we use boosted regression trees (BRT)-a form of gradient boosting-to predict pH conditions throughout the GLAC. Previous investigators have demonstrated that ensemble-tree methods provide more accurate predictions of groundwater quality as compared to ANN or BN (Nolan et al 2015), and other statistical methods such as linear or logistic regression (Nolan et al 2014;Wheeler et al 2015;Ayotte et al 2016;Ransom et al 2017;Yan et al 2019). Tree-based models use a series of IF-THEN statements (decision rules) to split the data multiple times until specified criteria are met (e.g., number of trees, number of splits, minimum number of observations per node).…”