With the increasing demand for machine learning application in hydrometeorological forecast, we face the urge to demystify the black-box of machine learning as the lack of interpretability hampers adaptation of machine learning.
Here, taking soil moisture (SM) prediction of one FLUXNET site (Haibei,China, named as CH-Ha2) as an example, we used air forcing variables, timekeeping, energy processing, net ecosystem exchange and partitioning, and sundown as input data. We aimed to predict the daily SM via historial dataset. We aimed to interpret the model via ExplainAI toolbox, including MSE-based feature importance, permutation importance, partial dependence plot, individual conditional expectation, accumulated local effect, Shapley values, local interpretable model-agnostic explanations.