A Method to Efficiently Apply a Biogeochemical Model to a Landscape

Biogeochemical models offer an important means of understanding carbon dynamics, but the computational complexity of many models means that modeling all grid cells on a large landscape is computationally burdensome. Because most biogeochemical models ignore adjacency effects between cells, however, a more efficient approach is possible. Recognizing that spatial variation in model outputs is solely a function of spatial variation in input driver variables such as climate, we developed a method to sample the model outputs in input variable space rather than geographic space, and to then use simple interpolation in input variable space to estimate values for the remainder of the landscape. We tested the method in a 100 km×260 km area of western Oregon, U.S.A. , comparing interpolated maps of net primary production (NPP) and net ecosystem production (NEP) with maps from an exhaustive, wall-to-wall run of the model. The interpolation method can match spatial patterns of model behavior well (correlations>0.8) using samples of only 5 t o 15% of the landscape. Compression of temporal variation in input drivers is a key step in the process, with choice of input variables for compression largely determining the upper bounds on the degree of match between interpolated and original maps. The method is applicable to any model that does not consider adjacency effects, and could free up computational expense for a variety of other computational burdens, including spatial sensitivity analyses, alternative scenario testing, or finer grain-size mapping.