Multi-model simulations of long-term effects of forest harvesting on ecosystem productivity and C/N cycling

Background/Question/Methods

Effects of harvesting on site productivity and soil C/N dynamics appear to vary by differences in species composition, soil conditions, and harvesting practices. Since the impacts may exist beyond the duration of a single rotation, or an observed change in one rotation may not necessarily continue in perpetuity, ecosystem modeling is one of the most feasible ways of assessing long-term dynamics. Yet there are limitations in applying a single model, such as data availability for calibration, and potential differences among different modeling approaches. An average of multiple models’ simulations may compare better to empirical observations than a single model, as in the approach taken with global climatic models. We used a multiple model approach to simulate effects of bole-only harvesting (leaving tops and branches on site) on productivity, C, and N cycling in an aspen (Populus tremuloides) forest ecosystem. Five models, Biome-BGC, Century, Forecast, LANDIS-II with Century-based soil C extension, and PnET-CN, were run for 350 years with seven harvesting events on nutrient-poor, sandy soil in northern Wisconsin. Twenty C/N state and flux variables were summarized from the model’s outputs, and statistically analyzed using ordination and variance analysis methods.

Results/Conclusions

The multiple models’ average showed that bole-only harvest would not significantly affect long-term site productivity, though slight decreasing trends were identified in the variables. Soil C/N storage decreased significantly through the rotations but the reduction did not affect aspen productivity. Along with direct N removal by harvesting, extensive leaching after harvesting before canopy closure was another major source of N depletion. Even though a negative net N balance in the forest ecosystem was identified, N depletion did not cause a significant decline in productivity. In terms of model performance, these five models were notably different in the 20 variables examined. Although some models did show similarities in certain variables, PnET-CN produced unique results with every variable, and Century showed fewer outliers and similar temporal patterns to the mean of all models. In general, this study demonstrated that a multiple model approach would not only supply more robust outputs for long-term simulations but also supply a way to evaluate and improve an individual model.