Combining long-term land cover time series and field observations for spatially explicit predictions on changes in tropical forest biodiversity

摘要

Combining spatially explicit land cover data from remote-sensing and faunal data from field observations is increasingly applied for landscape-scale habitat and biodiversity assessments, but without modelling changes quantitatively over time. In a novel approach, we used a long-term time series including historical map data to predict the influence of one century of tropical forest change on keystone species or indicator groups in the Kakamega-Nandi forests, western Kenya. Four time steps of land cover data between 1912/13 and 2003, derived from Landsat satellite imagery, aerial photography and old topographic maps, formed the basis for extrapolating species abundance data on the army ant Dorylus wilverthi, the guild of ant-following birds and three habitat guilds of birds differing in forest dependency. To predict the species' spatio-temporal distribution, we combined spatially explicit geographical information system (GlS)-based modelling with statistical modelling, that is, ordinary least square (OLS) regression models for D. wilverthi and simultaneous autoregressive (SAR) models for ant-following birds. We directly related bird habitat guilds to five forest classes as distinguished in the land cover time series. Extrapolation results over time predict dramatic losses in abundance for D. wilverthi (56%), ant-following birds (58%) and forest bird individuals in general (47%) due to a forest loss of 31% and small-scale fragmentation within the past century. Extrapolations based on a scenario of further deforestation revealed the negative consequences of clearing and splitting up continuous forest into isolated patches, whereas a reforestation scenario suggests the positive impact of natural forest regrowth and indigenous-tree planting. This study demonstrates the high potential of integrating remotesensing and field-based faunal data for landscape-scale quantitative assessments over time. In addition, it shows the suitability of extrapolation studies for evaluating measures of forest conservation.