Canadian Forest Service Publications
Characterizing the spatial pattern of soil, carbon and nitrogen pools in the Turkey Lakes Watershed: A comparison of regression techniques. 2002. Creed, I.F.; Trick, C.G.; Band, L.E.; Morrison, I.K. Water, Air, and Soil Pollution: Focus 2: 81-102.
Issued by: Great Lakes Forestry Centre
Catalog ID: 21505
Availability: Order paper copy (free)
There is considerable spatial heterogeneity in organic carbon (C), total nitrogen (N), and potentially mineralizable nitrogen (PMN) pools in the soils of the Turkey Lakes Watershed. We hypothesized that topography regulates the spatial pattern of these pools through a combination of static factors (slope, aspect and elevation), which influence radiation, temperature and moisture conditions, and dynamic factors (catenary position, profile and planar curvature), which influence the transport of materials downslope. We used multiple linear regression (MLR) and tree regression (TR) models as exploratory techniques to determine if there was a topographic basis for the spatial pattern of the C, N and PMN pools. The MLR and TR models predicted similar integrated totals (i.e., within 5% of each other) but dissimilar spatial patterns of the pools. For the combined litter, fibric and hemic layer, the MLR models explained a significant portion of the variance (R2 = 0.38, 0.23 and 0.28 for C, N and PMN, respectively), however, the residuals were large and biased (the smallest contents were over-predicted and the largest contents were under-predicted). The TR models (9-branch), in contrast, explained a greater portion of the variance (R2 = 0.75, 0.67 and 0.62 for C, N and PMN, respectively) and the residuals were smaller and unbiased. Based on our sampling strategy, the models suggested that static factors were most important in predicting the spatial pattern of the nutrient pools. However, a nested sampling strategy that included scales where both static (among hillslopes) and dynamic (within hillslope) factors result in a systematic variation in soil nutrient pools may have improved the predictive ability of the models.