Machine learning modeling of soil moisture time series for a hillside at Sulmachun watershed, South Korea

Time series of soil moisture is an important status variable for understanding hillslope hydrological processes at the mountain hillside because soil moisture plays a critical role in regulating water retention, generating runoff and controlling vegetation dynamics. In order to explore the ultimate interpretation based on past hydrologic information (e.g., precipitation and soil moisture history) for contemporary soil moisture status, several machine learning models had been applied using systematically collected soil moisture measurements along transects in a hillslope. The fitness of models was evaluated in terms of coefficient of determination, mean absolute error and root mean square error. Appropriate lag extent for parsimonious modeling of soil moisture was explored and determined through heuristic approaches which can be explained by historic gain and loss and uncertainty contribution. Modeling results indicate that the vertical infiltration to weather rock as primary hydrological process for most measurement points. Two distinct modeling performances in soil moisture modeling at top hill and streamside points indicate the degree of hydrologic process complexity can be identified through delineated AI modeling results. This study highlights the potential of machine learning based time series modeling for prediction of soil moisture and corresponding hydrologic process configuration in the mountain hillslope.