Keywords
remote sensing, land cover, species distribution modelling, mexico
Start Date
1-7-2012 12:00 AM
Abstract
Current changes of biodiversity result almost exclusively from human activities. As a consequence, spatially continuous estimates of species distributions are needed to support biodiversity evaluation and management. In the last two decades, species distribution models (SDMs) have been established as important tools for extrapolating in situ (point) observations. To account for current habitat loss, climate data used as predictors in SDMs need to be complemented by measures of current land surface characteristics. For this purpose, two alternative data sources are available, namely categorical land cover and continuous remote sensing data, each with their advantages and drawbacks. The objective of this study was therefore to directly compare the suitability of an existing land cover classification and remote sensing time series for the delineation of current biotope availability. The analysis used the Maximum Entropy algorithm to model the distributions of twelve tree species representative of the major Mexican forest types. Model results were evaluated based on AUC (area under curve) and statistical model deviance and revealed that land cover-based models overestimated species distributions and that the suitability of land cover data was dependent on species characteristics. The findings of this study support the selection of predictors in species distribution modelling in the future.
Suitability of land cover and remote sensing data for modelling species distributions
Current changes of biodiversity result almost exclusively from human activities. As a consequence, spatially continuous estimates of species distributions are needed to support biodiversity evaluation and management. In the last two decades, species distribution models (SDMs) have been established as important tools for extrapolating in situ (point) observations. To account for current habitat loss, climate data used as predictors in SDMs need to be complemented by measures of current land surface characteristics. For this purpose, two alternative data sources are available, namely categorical land cover and continuous remote sensing data, each with their advantages and drawbacks. The objective of this study was therefore to directly compare the suitability of an existing land cover classification and remote sensing time series for the delineation of current biotope availability. The analysis used the Maximum Entropy algorithm to model the distributions of twelve tree species representative of the major Mexican forest types. Model results were evaluated based on AUC (area under curve) and statistical model deviance and revealed that land cover-based models overestimated species distributions and that the suitability of land cover data was dependent on species characteristics. The findings of this study support the selection of predictors in species distribution modelling in the future.
