Comparison of three algorithms for estimating Snow Water Equivalent (SWE) over the La Grande River watershed using SSM/I data in the context of Hydro- Québec's hydraulic power management

Ongoing information on Snow Water Equivalent (SWE) is a crucial issue for Hydro-Québec in terms of reservoir management late in the winter or in the early months of spring. Three algorithms that use passive microwave data have been evaluated over the La Grande River Watershed in northeastern Quebec (Canada). The first and second algorithms that were tested consisted of a simple linear regression (LR) and piecewise linear regression (PLR) between a brightness temperature gradient (GTV) and SWE. The third was a multilayer feedforward artificial neural network (ANN). A self-organizing feature map (SOFM) was used to optimize the division of the input data into training, validation and testing subsets. The first simulation results show that it is possible to evaluate SWE with an acceptable accuracy in a taiga environment, with a mean absolute error (MAE) of 25% for the ANN algorithm and about 20% for the linear regression approach and piecewise linear algorithm.