Streamflow Spatial Correlation Length Increases During Storms and Its Application in Data Assimilation Improves Streamflow Predictions

Abstract

The spatial correlation of streamflow in a river network captures how flows at different locations relate to one another and influence the downstream response. In this study, we analyze the spatial correlation of streamflow and rainfall during storm events in three Texas watersheds: San Antonio (2015 and 2016 storms), Guadalupe (2018 storm), and Llano (2018 storm). We quantify spatial correlation using two semivariogram approaches: an isotropic semivariogram, which uses Euclidean distance, and a topological semivariogram, which uses hydrological distance that considers the nested catchment structure. We find that correlation lengths from topological semivariograms are longer than those from isotropic semivariograms, with storm‐period lengths higher than those for non‐storm periods. We incorporate the correlation length into a data assimilation approach that couples the output from the Conceptual Functional Equivalent (CFE) model with a state‐space Muskingum method. Using the National Water Model (NWM) as a benchmark for performance comparison, we find that the base CFE model shows better timing accuracy than NWM outputs. Integration of spatial correlation into data assimilation shows modest gains at one site and larger improvements at another site. While the CFE model with data assimilation outperforms the NWM at two of the three validation sites, the NWM performs better at the remaining site due to its use of gridded precipitation data, unlike the spatially averaged values in CFE. These findings show that integrating the spatial correlation length in data assimilation improves streamflow predictions in high‐flow scenarios, with performance strongly influenced by the quality and representation of precipitation inputs.