Understanding how carbon assimilation rates respond to water availability is crucial for diagnosing global carbon and water cycles. This study aims to investigate characteristics and drivers of gross primary productivity (GPP) responses to soil water availability using three parameters from a light-use-efficiency (LUE) model: WI, kW and αW , representing the inflection point, slope and lag effect of GPP response to soil water availability changes, respectively. We followed a hybrid modeling approach coupling an artificial neural network with the LUE model to derive model parameters and examine intricate relationships between these parameters and features characterizing climate, vegetation, nutrient deposition, soil properties and elevation across 196 eddy covariance sites. Relationships between the LUE model parameters and observed ecosystem properties were analyzed using partial dependence plots and Shapley additive explanation dependence plots. Our results revealed significant statistical differences in parameters across plant functional types. Specifically, forests exhibited lower inflection points, responding more steeply and immediately to water availability changes, contrasting with smoother and lagged responses from open shrubs. Vegetation seasonality, represented by variability of enhanced vegetation index (EVI) and seasonal EVI, was the most influential noncategorical factor, followed by soil properties. Notably, the relationships were predominantly nonlinear. Additionally, older forest ecosystems generally showed lower vulnerability while responding more steeply to relative soil water availability changes than younger forests. While aridity was less influential on parameter variability than anticipated, aridity seasonality was a primary driver for the inflection point. High temperatures and substantial diurnal and annual temperature ranges were linked to pronounced lag effects. Despite these findings, challenges remain regarding model accuracy on annual scales, parameter uncertainties and interactions between features. Overall, this study underscores the spatial heterogeneity of GPP responses to soil water availability and highlights the importance of considering variability in model parameters and GPP sensitivities across space and time.