Investigating spatial-temporal land cover changes using statistical methodology

摘要

Mid-latitude forests may play some role in carbon sequestration in the global C budget. The circumpolar boreal ecosystem is studied. The Boreal Ecosystems Atmosphere Study (BOREAS) is an international investigation with the goal of improving the understanding of the exchanges of radiative energy, sensible heat, water, CO/sub 2/, and trace gases between the boreal forest and the lower atmosphere. The study region is in Saskatchewan and Manitoba, Canada. Experimental measures span scales ranging from the whole region to local tree stands and below. The Southern Study Area (SSA), north of Prince Albert, Saskatchewan, was used for the current modeling effort. Previous satellite studies of boreal forest succession in the Superior National Forest of the Northern U.S. assumed that forest succession obeyed a Markov process. A pixel-by-pixel transition matrix was calculated using two Landsat images, one acquired in 1973 and the other in 1983. The study showed that the forest ecosystem was dynamic in the sense that half of the landscape elements changed state; at regional scales however, the area proportions occupied by the various successional states changed little over the 10 year period. That is, the changes are dynamic locally but overall stable at the regional scale. Another important result from this analysis was that the forest ecosystem was patchy, not spatially random. This latter finding is the basis for the current model. A follow on study performed in the same area examined the effects of the site factors on transition probabilities: elevation, slope, aspect, time since fire, fire frequency, land ownership, and land management. The study, using logistic regression models, showed that site factors significantly affect the transition probability. The current model assumes that there is a spatial effect, an effect from site factors as well as the Markov process. Conceptually, the model assigns the odds that a pixel takes on the class k verses a reference class, say 1, based on a linear combination of an intercept, the effect of site factors, the spatial effect, and the temporal effect (Markov process).