Identification of Soil Moisture–Precipitation Feedback Based on Temporal Information Partitioning Networks

Abstract

The underlying mechanisms by which soil moisture might affect precipitation are essential for weather prediction and climate change projection. Process-based models that are capable of providing physical explanation are often tripped in concerns about model uncertainty in structure, data, and parameters, etc. The information theory has provided a new paradigm to explain the feedback system. This study focused on to identify the soil moisture–precipitation feedback based on direct long-term observations using an information-based method, Temporal Information Partitioning Networks (TIPNets). Normalized transfer entropy was proposed to describe feedback strength. In comparison with Granger causality and correlation, the TIPNets method was independent of parameterizations and spatiotemporal resolution effects, also elucidate information-theoretic insights. The results in the case study of Illinois, indicated that soil moisture–precipitation feedback existed and was significant during May to mid-July. The bidirectional dominant lags of 3–10 and 6–17 days were both shorter in summer. Areas with strong feedback were located in south-central Illinois. Besides, the wooded grassland coverage was the most related land use type to feedback strength, which could be one kind of the positive feedback mechanism in Illinois.