The relative importance of rainfall intensity versus saturated hydraulic conductivity for runoff modeling of semi-arid watersheds

Abstract

When using distributed rainfall runoff models in order to simulate the runoff volume and time distribution, one faces the problem of how to represent the spatial distribution of rainfall intensity and soil characteristics when the actual continuous distributions are unknown. There are two objectives for this thesis. The first is to investigate, for semi-arid regions, how the scale of rainfall intensity and soil features affects the simulation of rainfall excess and which is of more importance. The second is to utilize probability distribution theory to develop a scheme which represents both the spatial distribution of rainfall intensi ty and the saturated hydraulic conductivity (Ks) in order to accurately simulate the true runoff for semi-arid regions when knowing limited statistical information (mean and variance) on each feature (rainfall intensity and Ks). All conclusions made are assuming that they hold for semi-arid regions only and the· assumed true watershed output is the model simulation which uses the finest resolution for soil features, vegetation, and rainfall intensity. Furthermore, the model used for this study provided very accurate simulations of the actual streamflow for the watershed used (Walnut Gulch). It was found that, when using real data for rainfall and soils, the spatial distribution for rainfall intensity is more important to represent than that of soil features with respect to accurately reproducing the assumed true streamflow. However, when using synthetic data generated from probability distributions, it was found that, for semi-arid regions, the spatial distribution of Ks was of more importance. Hence, certain conclusions concerning which is more important to spatially characterize with respect to accurately simulating streamflow can be different, depending upon if one uses synthetic data versus real data. The lognormal distribution was found to produce an excellent fit to the Ks data and the exponential distribution was found to produce an excellent fit to the spatial rainfall intensity distribution. However, this goodness-of-fit (for rainfall) can be dependent upon time and/or the amount of localization which the storm possesses. The rainfall parameters for the probability distributions of rainfall intensity were assumed to change with time but were not related at all to location within the watershed. When using the probability distributions to characterize the spatial rainfall intensity and Ks distributions, it was found that characterizing the Ks distribution provided more accurate simulations than characterizing the rainfall intensity distribution. It was also found that spatially characterizing both rainfall intensity and Ks simultaneously provided more accurate simulations than just representing one and not the other.