Lawrence Berkeley National Laboratory

Hillslope similarity is an active topic in hydrology because of its importance in improving our understanding of hydrologic processes and enabling comparisons and paired studies. In this study, we propose a holistic bottom-up hillslope clustering based on a region's integrative hydrodynamic response quantified by the seasonal changes in groundwater levels "P. The main advantage of the "P clustering is its ability to capture recharge and discharge processes. We test the performance of the "P clustering by comparing it to seven other common hillslope clustering approaches. These include clustering approaches based on the aridity index, topographic wetness index, elevation, land cover, and machine-learning that jointly integrate multiple data. We assess the ability of these clustering approaches to identify and categorize hillslopes with similar static characteristics, hydroclimate, land surface processes, and subsurface dynamics in a mountainous watershed - the East River - located in the headwaters of the Upper Colorado River Basin. The "P clustering performs very well in identifying hillslopes with six out of the nine characteristics studied. The variability among clusters as quantified by the coefficient of variation (0.2) is less in the "P and the machine learning approaches than in the others (>0.3 for TWI, elevation, and land cover). We further demonstrate the robustness of the "P clustering by testing its ability to predict hillslope responses to wet and dry hydrologic conditions, of which it performs well when based on average conditions.