Data is fundamental to hydrological modeling and water resource management; however, it remains a major challenge in many data-scarce regions. Hydrological studies require both spatially and temporally continuous datasets for model calibration and validation. In most cases, spatial inputs such as digital elevation models and soil data from global databases, combined with user-generated land-use information, can reasonably represent hydrological processes.

In contrast, climate variables from meteorological stations in many regions are frequently sparse, incomplete, or unreliable. Streamflow records are particularly limited and often outdated, resulting in a reliance on outdated observations. Recently, alternative variables such as evapotranspiration and soil moisture have been employed for multi-variable model calibration and validation; however, these are typically derived from global datasets that may not adequately capture local conditions. Advances in machine-learning techniques offer promising opportunities to address data scarcity by enhancing the performance of hydrological models and improving water resource management. Moreover, machine-learning approaches refine globally available datasets for reliable local-scale applications and enable robust prediction of hydrological parameters from climate, soil, and topography-related input features.

Due to the limited availability of observed climate and streamflow data in data-scarce regions, most hydrological studies are conducted in data-rich areas. In the absence of ground observations, satellite-based and reanalysis climate datasets are widely used; however, these products are subject to inherent uncertainties because they are not direct measurements. Among all data limitations, the lack of streamflow observations represents the most critical challenge for hydrological modeling and water resource management, often forcing studies to rely on outdated records for model calibration and validation. In this context, machine learning offers promising opportunities to address data scarcity. Machine learning models can be used to estimate recent streamflow records based on historical observations and climate and other biophysical related variables, thereby extending available datasets. In addition, they can be applied to bias-correct coarse-resolution climate variables and enhance globally available evapotranspiration and soil moisture products, improving their suitability for local-scale hydrological applications.

We welcome original research papers addressing, but not limited to, the following topics:

-Streamflow prediction using machine learning models

-Applications of machine learning for bias correction of hydrologic and climate data

-Applications of machine learning for improving evapotranspiration and soil moisture estimates

-Data reconstruction and uncertainty-aware augmentation

-Machine learning based drought and flood risk assessment

-Applications directly addressing missing/uncertain hydrometeorological inputs or outputs in data-scarce settings

Authors are encouraged to present case studies demonstrating the role of machine learning in hydrological modeling and water resource management, either independently or in combination with hydrological models. Review papers covering any of the above-mentioned themes are also welcome. Submissions must focus on data-scarce/ungauged or poorly gauged regions and demonstrate methods under limited observations.

Articles

Frontiers in Water

Water and Artificial Intelligence