Improving land surface model accuracy in soil moisture simulations using parametric schemes and machine learning

Soil moisture (SM) controls the interaction of water and energy between the land surface and the atmosphere. Accurate knowledge of SM is vital for effective water resources planning and management, agricultural production, climate prediction, and drought disaster monitoring. In this study, we chose three parameterization processes-canopy stomatal resistance, beta-factor, and runoff-and explored 24 different parameterization schemes (S1-S24) for SM simulation using the Noah land surface model with multiparameterization (Noah-MP) in Asia during 2000-2022, and we systematically evaluated their simulation performance. S9 has the best simulation performance, with the highest correlation coefficient (R = 0.69) and Nash-Sutcliffe efficiency coefficient (NSE = 0.59), while S20 has the lowest R (0.23) and NSE (0.13). After removing irrigated land, S9 still maintains the best simulation performance (R = 0.68). The simulations are better in Central Asia, West Asia, and coastal areas but worse on the Tibetan Plateau. Seasonal simulation performance is best in summer (R = 0.68, NSE = 0.58) and worst in spring (R = 0.54, NSE = 0.43). We found that the long short-term memory (LSTM) model improves the simulation accuracy of Noah-MP by 20 %, showing a higher coefficient of determination (R2) and NSE during training and testing periods. In terms of spatial patterns, the LSTM model improved SM simulation by Noah-MP in areas with poor simulation performance, especially on the Tibetan Plateau. SM simulation accuracy was improved most notably in spring, with an increase of 27 %, which may be related to the seasonal simulation performance of Noah-MP. This study demonstrates the potential of combining traditional land surface models like Noah-MP with advanced machine learning techniques such as LSTM to significantly enhance SM simulations, particularly in complex and data-scarce regions like the Tibetan Plateau. These advancements can support better water resource management, agricultural planning, and climate adaptation strategies, especially in regions vulnerable to climate change and extreme weather events.