Spatial variability of infiltration patterns related to site characteristics in a semi-arid watershed

Preferential flow may strongly affect hydrology at different scales. Measurement of preferential flow however remains very difficult. Tracer-infiltration profiles are often used to measure the degree of preferential flow at plot scale. These experiments are time-consuming, costly and destructive. As a result existing dye-tracer studies are often based on a limited number of profiles. The aim of this study is therefore to select a limited number of soil and landscape characteristics with high predictive value for the degree of preferential flow. 18 rainfall simulations with a dye-tracer and additional site measurements such as soil type, vegetation type and soil physical parameters were performed in a 1 km² catchment in the Dehesa (Extremadura, Spain). A stepwise multiple regression procedure was used to select variables with a high predictive value for the degree of preferential flow. The infiltration patterns show that preferential flow is potentially present throughout the area and is spatially variable. At first inspection the spatial variability seems to relate to the site specific variables. This is corroborated by the stepwise multiple regression results, with R² values of 56% to 67% for the different preferential flow parameters (uniform infiltration front, maximum infiltration depth, total stained area and preferential flow fraction of stained area). A bootstrapping procedure however indicated that the bias due to stepwise variable selection with a large number of independent input variables and a low number of cases was high. Pre-selection of a limited set of variables (vegetation, texture, slope and location) based on expert knowledge gives lower but more powerful R² values (50% to 66%), without the bias due to stepwise variable selection. As a large part of the spatial variability of infiltration patterns can be explained with the selected site characteristics, the regression equations were used in combination with detailed maps of the four selected input variables to calculate maps of catchment scale spatial variability in infiltration patterns, thereby delineating sub-areas within the catchment with different infiltration patterns.