A novel approach for river discharge prediction in Lancang Mekong River Basin: Incorporation of multisource remote sensing and LSTM model

This study focuses on the Lancang Mekong River Basin (LMRB), a vital transboundary river system characterized by significant spatiotemporal variability in discharge. Monitoring in this region is inadequate due to sparse network of ground-based gauging stations, particularly in remote or cross-border segments, making it a critical area for remote sensing applications. To address the challenge of monitoring river discharge at ungauged locations, we developed a framework using a Long Short-Term Memory (LSTM) model that integrates multi-mission satellite data. The model utilizes altimetry-derived water levels combined with reflectance ratios from optical sensors to predict daily river discharge. The research evaluates the model’s predictive accuracy across multiple stations and assesses the spatial transferability of the LSTM model to ungauged reaches. The results show distinct spatial geomorphology of the Lancang-Mekong River controls our ability to monitor its flow from space. A distinct spatial observability gradient along the river continuum was identified. In the narrow, meandering upstream reaches, complex terrain physically limits satellite accuracy due to radar land contamination and optical mixed pixel effects. Conversely, the broader, flatter downstream floodplains provide stable hydraulic geometries that allow for highly coherent discharge monitoring. Furthermore, spatial transferability analysis demonstrates that flow dynamics in the Lancang Mekong are localized. Predictive capability decays as distance from the training reach increases, proving that large scale transboundary discharge estimation requires segment specific modeling to capture the river's changing physical channel geometry. Hydrologically, altimetry derived water levels and reflectance ratios from optical sensors provide essential complementarity, effectively compensating for the temporal gaps and spatial limitations of individual sensors. • Integrated altimetry derived water levels and MODIS reflectance in an LSTM for daily river discharge prediction in LMRB. • Achieved Nash-Sutcliffe Efficiency of 0.54–0.92 across LMRB reaches using multisource satellite data. • Provides reliable discharge predictions for ungauged reaches to address sparse monitoring in transboundary basins.