LAUSR

Abstract Urban sprawl and the accompanying development is set to increase pressure on the hydrological system. In turn, the hydrology of an area affects the infrastructure. Hydropedology, can assist to sustainably manage the infrastructure-hydrology interaction. Even though the hillslope scale is accepted as ideal for hydropedological assessments, regional soil maps have been used when hydropedology assessments were undertaken for larger areas. This study uses a hillslope based approach within a digital soil mapping method to conduct a hydropedological assessment for a large (12,000 ha) urban area under developmental pressure within Johannesburg, South Africa. One hundred and thirty-three hillslopes within the study area were delineated and mean and standard deviation of terrain attributes were calculated for each hillslope. Based on these values, the conditioned Latin hypercube sampling (cLHS) method was used to select 30 hillslopes on which soil observations would be made, in a transect, at the surveyor's discretion. A hydropedological soil map was created with the multinomial logistic regression algorithm and the 142 soil observations made. The soil map attained an acceptable 69% validation point accuracy and a Kappa value of 0.59, and was used to create a hillslope conceptual hydrological response map. The chi-square test and QQ-plots indicated that the cLHS selected hillslopes represented the hillslopes of the study site well, but the observations only represented some of the terrain attributes well. Despite the range of the observation dataset covering at least 98.7% of the pixels for the multi resolution index of valley bottom flatness (MRVBF), topographic wetness index (TWI) and altitude above channel network layers, their site distribution differed statistically from that of the observation distribution. The hillslope based survey approach allowed for an acceptable soil map to be created, while allowing for hillslope based hydropedological interpretations to be made. The “cost” for the hillslope based approach was that the distribution of some covariates for the observations differed from their distributions over the entire site.