Added gains of soil moisture content observations for streamflow predictions using neural networks

Soil moisture content is an important hydrologic component regulating watershed streamflow production and atmospheric processes in the near surface. Its potential benefits as inputs to streamflow models must be investigated in anticipation of greater availability of such observations in the future. Added gains in streamflow prediction performance were demonstrated from including soil moisture content observations in neural networks, which are a category of models with efficient and practical applicability comparable to that of usual black box and conceptual hydrologic models. Soil moisture index time series, derived from a simple conceptual model, were also tested as inputs and were also beneficial to the neural network models, although to a lesser extent. Issues related to the calibration of neural network models such as the choice of inputs and the use of the stacking method were also addressed.