Predicting Observed Soil Moisture Using Statistical Modeling

Soil moisture is the amount of water contained in a volume of soil. Soil moisture is important to agriculture, coastal ecosystems, and environmental engineering. For example, the amount of water holding capacity for a particular soil type is essential for irrigation, drought and flooding potential in coastal ecosystems, and waste water applications. As critical a parameter as soil moisture is, only a select number of mesonets measure soil moisture. The North Carolina Environment and Climate Observing Network (ECONet) is one of the few mesonets that measures volumetric soil moisture at 20 cm depth at 37 locations across NC. The purpose of this study is to use these measurements to evaluate the accuracy of a soil moisture estimation technique. Observed hourly average soil moisture from each station is used to create a statistical model that can produce a predicted value for each station. Using 4-5 years of hourly data, this model analyzes temporal variations to account for the change in soil moisture. The model can predict values up to 24 hours in advance. With every new observed soil moisture value, the model adjusts itself to account for new information. The model captures the observed soil moisture with an average error of less than ± 0.07 meters³meters^-3 (m³m^-3) at every ECONet and U.S. Climate Reference Network (USCRN) station. The USCRN stations are used to validate that the ARIMA model can work across networks. One of the limitations of the ARIMA model is that the station of interest must measure soil moisture for at least a year to have minimal error. For future work, these limitations of ARIMA will be addressed. Results from these statistical modeling experiments will be used to improve automated quality control of soil moisture sensor data for both current and historical observations.