Forest cover change in Miombo Woodlands: modeling land cover of African dry tropical forests with linear spectral mixture analysis

Abstract Dry tropical forests are experiencing some of the highest rates of change among the globe's forests. In sub-Saharan Africa, gross (loss, gain) and net changes in dry tropical forest areas are difficult to quantify at sub-national scales because of high spatio-temporal variability in land cover conditions due to vegetation phenology and land use practices. In this project, we developed new, field-validated remote sensing characterizations of dry season surface components to separate forest from non-forest land cover, and assessed forest changes from the 1990s–2010s in a Tanzanian Miombo Woodland landscape. Using a linear spectral mixture analysis (LSMA) approach with Landsat 5–8 data, we examined the hypothesis that higher proportions of substrate and non-photosynthetic vegetation (NPV) at non-forest regions distinguished them from forest cover against seasonally variable land cover conditions. Subsequently we evaluated the efficacy of multi-temporal classification and single-date image thresholding for identifying forest from non-forest cover. We found significantly greater proportions of substrate and NPV over non-forest compared to forest areas that enabled identification of forest cover across dry season images. Single-date, forest/non-forest maps based on an LSMA-derived metric attained overall accuracies of 81.0–85.3%, which approached multi-temporal unsupervised classifications (86.5% for forest/non-forest maps). Applying the LSMA-derived metric to study forest changes, our study region experienced a net 15.0% loss of 1995 forest area, and a 7.0% overall reduction in the total forest-occupied land cover from 1995–2011. Areas of gross forest gain were substantial, totaling 13.6% of the 1995 forest area. We found differing patterns in gross forest losses and gains among sub-regions and through time in our Tabora study area, which provide bases for testable hypotheses in future research on regional and localized drivers affecting forest cover. Our finding that non-green surface components distinguished forest from non-forest via an LSMA approach may be widely applicable to studying forest conversions in Miombo Woodlands and other dry tropical forests. This approach may also be useful for evaluating how land cover conditions change in response to potential land use or climate driving variables, or the impact of land changes for carbon balance and other ecosystem processes.