Abstract: Soil moisture (SM) regimes are important to identify when studying land–atmosphere (L–A) coupling, yet are not often considered in assessing forecast skill. Various studies have documented the relationship between SM and evaporative fraction (EF), which is used to define distinct hydroclimatic regimes that provide a process-based diagnostic of L–A interactions. Here, we first identify regimes in the Community Earth System Model version 2 subseasonal-to-seasonal (CESM S2S) forecasts, using six potential segmented regression models of EF as a function of SM at each grid cell. After identifying SM regimes in the CESM S2S forecasts, we compare them with observational regimes from the Global Observationally-Based Land–Atmosphere Coupling Metrics (GOLAM; v1.0) dataset. We then derive SM–EF regimes from long-running CESM2 simulations and compare them with observational regimes from GOLAM v1.0. However, the extent to which model–observation biases are associated with mismatches between modeled and observed SM–EF regimes remains unexplored. Therefore, we use logistic regression, a supervised learning algorithm, to identify which model–observation biases are most strongly associated with global patterns of SM–EF regime mismatch. To train the logistic regression, we use climatological mean biases in several predictor variables, including meteorological data, vegetation cover fraction, soil characteristics, and topographic features, while the target variables represent the SM–EF regime matches/mismatches between models and observations. After training the logistic regression, we evaluate model performance using 5-fold stratified cross-validation and cross-entropy. Global maps of dominant SHAP contributors with the direction of their effects are used to identify which predictors and model biases most strongly drive regime mismatches. Ultimately, this work aims to improve our understanding of L–A interactions and the role of land-surface processes in S2S prediction, while guiding improvements in CESM land-surface parameterizations.