A lag-aware and interpretable two-stage framework for identifying dominant precipitation-related predictors in the data-scarce Mekong River Basin

Study area The Mekong River Basin is a data-scarce transboundary basin where sparse, unevenly distributed, and temporally discontinuous gauge observations hinder reliable precipitation estimation, particularly under strong subregional hydroclimatic heterogeneity. Research focus This study developed a lag-aware and interpretable two-stage framework to identify dominant precipitation-related predictors across the basin. In the first stage, transfer entropy (TE) was used to preliminarily screen candidate predictors by assessing directional lagged dependence from each candidate variable to precipitation over 1–7-day lags. In the second stage, tree-based machine learning and SHapley Additive exPlanations (SHAP) were used to quantify predictor contributions within the fitted models and diagnose predictor structures across subregions and rainfall regimes. New hydrological insights The framework reduced a high-dimensional candidate predictor space to compact sets of 10–15 inputs. Under a common sample-based split, the compact models achieved a mean Nash–Sutcliffe efficiency of 0.232, compared with 0.275 for models using the complete preliminarily screened sets, and performed better overall than conventional feature-selection alternatives. Across subregions, cloud microphysical variables aloft and lower-atmospheric moisture variables were recurring contributors to modelled precipitation. Convective available potential energy (CAPE) and directional transport-related predictors also played important roles. In the heavy-precipitation subset (≥90 mm day−1), transport-, flux-, and energy-related predictors became more prominent in the fitted-model SHAP rankings. This rainfall-regime contrast provides a useful diagnostic of regional precipitation processes, although the heavy-precipitation results remain exploratory because of the limited number of gauge-day samples.

Authors

Institutions

Publication Details

Journal
Journal of Hydrology Regional Studies
Published
2026-11-18
DOI
https://doi.org/10.1016/j.ejrh.2026.104207
Primary Topic
Precipitation Measurement and Analysis
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A lag-aware and interpretable two-stage framework for identifying dominant precipitation-related predictors in the data-scarce Mekong River Basin

Yiqun Sun, Tianshan Zeng, Xi Weng, Dongdong Gao et al.
1 links
Journal of Hydrology Regional Studies
Precipitation Measurement and Analysis
article

A lag-aware and interpretable two-stage framework for identifying dominant precipitation-related predictors in the data-scarce Mekong River Basin

Yiqun Sun, Tianshan Zeng, Xi Weng, Dongdong Gao, Nengmin Zhu, Hongle Shi, Qiongfang Li, Xiaogang Tian
article en

Abstract

Study area The Mekong River Basin is a data-scarce transboundary basin where sparse, unevenly distributed, and temporally discontinuous gauge observations hinder reliable precipitation estimation, particularly under strong subregional hydroclimatic heterogeneity. Research focus This study developed a lag-aware and interpretable two-stage framework to identify dominant precipitation-related predictors across the basin. In the first stage, transfer entropy (TE) was used to preliminarily screen candidate predictors by assessing directional lagged dependence from each candidate variable to precipitation over 1–7-day lags. In the second stage, tree-based machine learning and SHapley Additive exPlanations (SHAP) were used to quantify predictor contributions within the fitted models and diagnose predictor structures across subregions and rainfall regimes. New hydrological insights The framework reduced a high-dimensional candidate predictor space to compact sets of 10–15 inputs. Under a common sample-based split, the compact models achieved a mean Nash–Sutcliffe efficiency of 0.232, compared with 0.275 for models using the complete preliminarily screened sets, and performed better overall than conventional feature-selection alternatives. Across subregions, cloud microphysical variables aloft and lower-atmospheric moisture variables were recurring contributors to modelled precipitation. Convective available potential energy (CAPE) and directional transport-related predictors also played important roles. In the heavy-precipitation subset (≥90 mm day−1), transport-, flux-, and energy-related predictors became more prominent in the fitted-model SHAP rankings. This rainfall-regime contrast provides a useful diagnostic of regional precipitation processes, although the heavy-precipitation results remain exploratory because of the limited number of gauge-day samples.

Journal of Hydrology Regional StudiesVol. 68
Hohai University (CN), Biogas Institute of Ministry of Agriculture (CN), Ministry of Agriculture and Rural Affairs (CN)
National Natural Science Foundation of China
Clean water and sanitation, Climate action
Precipitation Measurement and Analysis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon