Effects of spatial aggregation on predictions of forest climate change response

We investigated the influence of spatial aggregation on modeled forest responses to climate change by applying the process-based Terrestrial Ecosystem Model (TEM) to a fine resolution spatial grid (100 km2) and to a coarse resolution spatial grid (2500 km2). Three climate scenarios were simulated: baseline (present) climate with ambient CO2 and 2 future climates derived from the general circulation models OSU and GFDL-Q with elevated atmospheric CO2. For baseline climate, the aggregation error of the national (U.S.) study area was very small, –0.4%. Forest-level aggregation error ranged from –1.6 to 11.8%, with the largest aggregation error occurring in boreal forest types. Coarse grid resolution inputs underestimated production for boreal and forested boreal wetland forests and overestimated net primary production (NPP) for temperate conifer, temperate deciduous, and temperate forested wetland forests. Aggregation error for coarse grid cells ranged between –25.6 and 27.3%. Aggregation errors were especially large in transition regions between temperate and boreal forest types. An analysis that homogenized inputs for the 10 km grid cells within a 50 km grid indicated that aggregation of forest types and air temperature from fine to coarse grid cells contributed most to the spatial aggregation error. The aggregation error for the OSU climate was similar to the GFDL-Q climate and both results were similar to the aggregation error of the baseline climate in magnitude, sign, and spatial pattern. While aggregation error was similar across the baseline, GFDL-Q and OSU scenarios, NPP response to the GFDL-Q and OSU climates increased 13 to 30% above the baseline NPP. Within each climate scenario, the estimated NPP response to climate change differed by less than 1% between the coarse and fine resolutions. Except for transition regions and regions with substantial variability in air temperature, our simulations indicate that the use of 0.5° resolution provides an acceptable level of aggregation error at the 3 scales of analysis in this study. Improvements could be made by focusing computational intensity in heterogeneous regions and avoid computational intensity in regions that are relatively homogeneous with respect to vegetation and air temperature.