Farm-level models of spatial patterns of land use and land cover dynamics in the Ecuadorian Amazon

Longitudinal studies examining socio-demographic and other contextual factors are vital to understanding landscape change. Landscape structure, function, and change are assessed for the northern Ecuadorian Amazon by examining the composition and spatial organization of deforestation, agricultural extensification, and secondary plant succession at the farm level in 1990 and 1999 through the integration of data from a satellite time-series, a longitudinal household survey, and GIS coverages. Pattern metrics were calculated at the farm level through the generation of a hybrid land use and land cover (LULC) digital classification of Landsat Thematic Mapper (TM) data. Population, labor, and other household variables were generated from a scientific sample of survey farms or fincas interviewed in 1990 and resurveyed in 1999. Topography, soils, and distance and geographic accessibility measures were derived for sample farms through a GIS as well as qualitative assessments from household surveys. Generalized linear mixed models (GLMMs) were generated for 155 and 157 fincas in 1990 and 1999, respectively, using pattern metrics at the landscape level as dependent variables, and biophysical, geographical, and socio-economic/demographic variables as independent variables. The models were derived to explore the changing nature of LULC at the finca level by assessing the variation in the spatial structure or organization of farm landscapes in 1990 and 1999, and the extent to which this variation could be explained by the available data. Results indicate rapid population growth causing substantial subdivision of plots, which in turn has created a more complex and fragmented landscape in 1999 than in 1990. Key factors predicting landscape complexity are population size and composition, plot fragmentation through subdivision, expansion of the road and electrical networks, age of the plot (1990 only), and topography. The research demonstrates that the process of combining data from household surveys, satellite time-series images, and GIS coverages provide an ideal framework to examine population–environment interactions and that the statistical models presented are powerful tools to combine such data in an integrated way.