Investigating tropical deforestation using two-stage spatially misaligned regression models

Deforestation in the tropics has been a major concern in conservation science for more than two decades. A standard explanation is population pressure, argued through descriptive statistical summaries, but the connection between local population and forest exploitation has not been clearly addressed from a formal modeling perspective. We implement such modeling here using a two-stage specification. At the first stage, we provide a spatial model for population counts. At the second stage, we provide a conditional spatial model for land use given population. A critical problem is misalignment. The population counts are recorded at various administrative unit levels. In particular, we work with twon-level counts. The land-use classifications are from remotely sensed satellite images and are provided at a 1-km × 1-km pixel level. We propose a methodology to implement regressions in this situation. The motivating data are obtained for the tropical wet forest on the eastern coast of Madagascar. This is a designated hotspot rain forest featuring high species diversity and high endemism. A fairly detailed analysis connecting land use with population data for this region is presented.