Identification of time-varying parameters of a monthly Budyko function in US MOPEX catchments and its implications

Model Parameter Estimation Experiment (MOPEX) catchments across the continental United States. The Budyko framework provides a simple yet useful method for estimating evapotranspiration (ET), traditionally used at the mean annual scale due to the assumption of constant catchment water storage. Under changing conditions, time-invariant model parameters may not fully capture variations in the underlying surface. Several studies have extended the Budyko framework to the monthly scale; however, few studies have identified monthly time-varying Budyko model parameters and explored the factors driving their variability. This study develops a method for identifying time-varying parameters κ and y 0 of a two-parameter Budyko model using the Ensemble Kalman Filter. Additionally, Extreme Gradient Boosting is used to explore the factors influencing time-varying parameters. Monthly time-varying parameters are effectively identified through simultaneous assimilation of ET and runoff observations, achieving a Kling-Gupta Efficiency of 0.94 in ET simulation. (2) 11 and 2 catchments display change points and trends in parameter κ , respectively, while 64 and 48 catchments exhibit change points and trends in y 0 , respectively. (3) Parameter κ is mainly influenced by climatic factors in low-latitude, storage-related factors in high-latitude, and normalized difference vegetation index in central U.S. catchments, respectively. y 0 is mainly influenced by total soil moisture, along with vegetation and agricultural land cover changes in the central region. • High accuracy of monthly ET simulation via data assimilation and machine learning. • Reveal distinct spatial patterns in parameter trends, change points, and periodicities. • Key drivers of time-varying parameters are climatic, storage-related factors, and NDVI. • Provide insights into ET simulation under changing conditions.