Abstract The diagnostic soil moisture equation (DSME), which estimates soil moisture from antecedent precipitation, can be used to gap‐fill and extend existing in situ records. We train and evaluate the DSME at 157 Soil Climate Analysis Network and 113 US Climate Reference Network (USCRN) stations across the contiguous United States. All possible calibration periods of one to five calendar years are tested (10,841 model runs) and validated over the remaining data for the 5‐cm depth at each station. A minimum of 2 years is required to train the DSME with a median root mean square error (RMSE) for all stations equal to 0.046 m 3 m −3 ; this is reduced to 0.040 m 3 m −3 if the stations are filtered to just those where the model is eventually successful (RMSE less than 0.06 m 3 m −3 achieved). Extending the period of record to 4 years reduces the median RMSE to 0.043 m 3 m −3 for all stations and 0.038 m 3 m −3 for the filtered subset. The distribution of RMSE across in situ soil moisture percentile bins reveals that the model error results from underestimation of extremes, particularly on the wet end. The DSME is capable of modeling soil moisture for column depths up to 1 m (the deepest extent of the testing) although it performs better at depth when there is more than one in situ sensor collecting validation data at each location, as is the case across the USCRN. There are distinct regional patterns in model performance, which are likely a result of both hydrometeorological regimes and circumstances that affect in situ data quality.
Walker et al. (Sun,) studied this question.