Using Redundancy Analysis to Identify Opportunities for Improved Accuracy of National Scale Forest Soil C Estimation in Canada.

Accurate initialization of soil C stocks in forest ecosystem models is challenging but critical to forest C accounting, assessing responses to climate change, and evaluation of mitigation strategies. We identified opportunities to improve the accuracy of initial soil C in the Carbon Budget of the Canadian Forest Sector (CBM-CFS3) – a model of forest C dynamics used to support reporting and accounting.  We used redundancy analysis (RDA) to compare the variance structures of soil C determined by variables used in the model in two different soil C datasets; one based on results from application of the CBM-CFS3 at the national scale, the other from a large (n = 2391) set of ground plots. Total variation in the ground plot data was twice that of modeled data.  Variables used in the model explained 24% of the total ground plot variation. Soil C stocks in the mineral and organic soil horizons were highly correlated in the model dataset but not in the plot dataset. Productivity explained a larger proportion of the total variation in the model dataset than in the plot dataset, whereas tree species explained more variation in the plot dataset than in the model. Soil taxonomy (currently not used in the CBM-CF3) explained an additional 4 to 14 percent of soil C variation.  A small dataset (n = 599) was used to calibrate soil C initialization parameters for combinations of tree species and soil taxonomic groups. When calibrated rates were used they explained a greater proportion of the total variation (58%) than the CBM-CFS3 default parameters (37%) and the correlation (r) between mineral and organic horizon soil C was reduced from 0.78 to 0.20. This result demonstrates the potential of using combination of tree species and soil taxonomy to improve soil C stocks initialized by forest C models.