Land change in central Mexico: Landscape heterogeneity, natural variability, and classification uncertainty

Landscapes across the world are experiencing unprecedented rates of change, and the quantification of change and identification of drivers is a grand challenge in land change science research. The Lerma-Chapala-Santiago watershed of central Mexico encompasses a diverse range of topographic conditions, vegetation types, and human activities, and exemplifies the conflation of discrete anthropogenic change with natural variability. The goal of detecting discrete, human-induced, land change is urgent and important, yet challenging for three major reasons: (1) The quantification of land change using coarse spatial resolution data over broad areas has high uncertainty and error. (2) The disaggregation of discrete land change from natural land cover variability is difficult. (3) The assessment of large spatial extents over time necessitates the integration data from multiple sources. New methods must be developed to evaluate uncertainty in the classification of coarse spatial resolution data, to identify and quantify the range of natural variability and its impact on land change, and to integrate land cover data for change analyses. This dissertation research addresses the challenge through the three research studies:;The first study evaluated land cover and land cover change across the Lerma-Chapala-Santiago watershed from 2001 to 2007 using a Mahalanobis distance algorithm to classify imagery from Moderate Resolution Imaging Spectroradiometer (MODIS). In addition to the change assessment, this process generated soft-classified typicality images, a metric of classification uncertainty to identify locations of ambiguity or vagueness, and an index of confusion to quantify the potential for spurious change between the two maps.;The second study utilized multiple linear regression to explain the variability of Enhanced Vegetation Index composites through independent variables of precipitation, temperature, and elevation from 2001 to 2007. Results of the interannual and annual models were compared to change assessments to assess the influence of variability on the identification of discrete land change and error, demonstrating over 58% of vegetation variability was attributable to these climate variables.;The third study identified potential error in data transformation for land change analyses and introduced a new method, utilizing a vector framework data to reproject and rescale raster categorical data. Compared to current methods, results demonstrate data transformation processes yield up to 30% spurious change between two maps, but that errors of both quantity and location could be avoided through appropriate processing.;This dissertation addresses several issues surrounding use of coarse spatial resolution categorical data for land change analysis. Through an understanding of the potential for errors introduced by the variability and uncertainty inherent to the creation and transformation of land cover data, land change analyses can be maximally useful for better understanding the dynamic processes of natural and anthropogenic change across broad areas.