Testing the hybrid-3D Hillslope Hydrological Model in a Real-World Controlled Environment

Abstract:

Hillslopes play an important role for converting rainfall into runoff, and as such, influence the
terrestrial dynamics of the Earth's climate system. Recently, we have developed a hybrid-3D (h3D)
hillslope hydrological model that couples a 1D vertical soil column model with a lateral pseudo-2D
saturated zone and overland flow model. The h3D model gives similar results as the CATchment
HYdrological model (CATHY), which simulates the subsurface movement of water with the 3D Richards
equation, though the runtime efficiency of the h3D model is about 2-3 orders of magnitude faster.

In the current work, the ability of the h3D model to predict real-world hydrological dynamics is
assessed using a number of recharge-drainage experiments within the Landscape Evolution
Observatory (LEO) at the Biosphere 2 near Tucson, Arizona, USA. LEO offers accurate and high-
resolution (both temporally and spatially) observations of the inputs, outputs and storage dynamics of
several hillslopes. The level of detail of these observations is generally not possible with real-world
hillslope studies. Therefore, LEO offers an optimal environment to test the h3D model.

The h3D model captures the observed storage, baseflow, and overland flow dynamics of both a larger
and a smaller hillslope. Furthermore, it simulates overland flow better than CATHY. The h3D model
has difficulties correctly representing the height of the saturated zone close to the seepage face of the
smaller hillslope, though. There is a gravel layer near this seepage face, and the numerical boundary
condition of the h3D model is insufficient to capture the hydrological dynamics within this region.

In addition, the h3D model is used to test the hypothesis that model parameters change through time
due to the migration of soil particles during the recharge-drainage experiments. An in depth calibration
of the h3D model parameters reveals that the best results are obtained by applying an event-based
optimization procedure as compared to using a single parameter set for all events. However no clear
temporal evolution of the model parameters is observed between events.