Linking biological field data and remote sensing for decision making – Examples from Kakamega Forest, Kenya

Remote sensing in combination with GIS-based modelling is widely used for landscape-scale assessments including changes over time. Such assessments can be complemented by modelling scenarios of possible future land use changes to evaluate potential ecological consequences. The resulting visualizations represent a valuable means of finally communicating the results to the various stakeholders involved in decision making. Today, effective forest management planning for nature conservation should ideally consider biodiversity per se as well as ecosystem service provision of different forest management types. Situated in western Kenya, Kakamega Forest is of high conservation value due to its exceptionally high biodiversity and its unique mix of species. Besides being an attraction to tourists from round the world, the forest is threatened due to the fact that it is rather small in size (with a gazetted area of 240 km²) and located in one of the most densely populated rural areas of Kenya. Not only are the people living adjacent to the forest making use of its resources, the Kakamega Forest has been commercially logged and otherwise exploited for a very long time. Today’s forest disturbance levels and forest type composition are therefore a result of its complex forest use history, which has been thoroughly investigated and mapped with spatial data covering the past ca 100 years. In comparison to other forest areas, Kakamega Forest is special in regard to the mix of various forest types which have resulted as a consequence of disturbance: besides near-natural forest and secondary forest, forest plantations of mixed indigenous species, and monocultures of indigenous or of exotic species can be found. Both the existence of many different forest types and the detailed and available spatio-temporal data, in many instances derived from remotely sensed data, provide an ideal basis for the spatially explicit modelling and assessment of forest biodiversity and ecosystem services and their change, thereby supplementing the forest management planning efforts undertaken on the ground. By employing a long-term land cover time series, including supervised multispectral classification results based on Landsat imagery (2008, 2003, 1984), visual interpretation of historical aerial photography (1965/67, 1948/(52)) and digitized forest fills as marked on old topographic maps (1913; Schaab et al., 2010), as well as different land use scenarios (ambitious nature conservation, realistic approach, revenue-driven forest management) derived by implementing a rule-based allocation procedure in GIS, the effects of land cover change on bird diversity and carbon stock are estimated. Here, the question is if tree plantations can compensate for natural forest loss, because plantations may support lower bird species diversity and different species composition. Regarding carbon stocks, questions arise concerning the carbon sequestration potential of the degraded areas in Kakamega Forest. Field data on the diversity of 115 bird species (Farwig et al., 2008) was tested against forest type and the GIS-derived spatial measures of forest patch size, patch shape, patch connectivity und distance to farmland via statistical analyses. Here, birds were classified according to their forest dependencies as specialists, generalists and visitors. The statistical analysis revealed a direct relationship between bird richness and forest type, while the distinct pattern of differences in PC1 values for the bird communities per forest type could be explained by forest type in combination with patch shape (perimeter-area ratio). The amount of carbon stored in the forest was modeled by a re-stratification of field data-based calculations of above ground biomass (94 plots; Glenday, 2006) to the land cover classification by distinguishing between six forest classes, applying an age-adjustment to account for tree growth and soil transmission, and considering all five carbon pools. A particular challenge concerned adjusting the change matrix to account for impossible subsequent land cover classes due to inaccuracies in the multispectral classification; this obstacle was addressed by applying decision rules. The results suggest a 47% decline in total number of bird individuals for Kakamega Forest from 1913 to 2003, which translating to a loss of 880,000 birds, with the changes in forest type composition strongly affecting bird communities. Although the 61% decrease in near-natural forest in Kakamega Forest (incl. the forest patches Kisere and Malava) until 1984 has been almost regained through plantation and forest regrowth by 2008 (i.e. forest cover being back to ca 90% of 1913), carbon stock remained low from 1984 onwards, making up 52-61% of the amount estimated for 1913. The results for the five time steps applied further served as reference points in time when comparing the different levels reached by the five scenarios: With the realistic scenario a total number of birds only slightly higher than in the two latest time steps is suggested, while the conservation-focused scenarios suggest potential bird levels comparable to those of 1913. Regarding carbon stocks, the sequestration amount modelled in the realistic scenario already approaches that modelled for 1913. The two conservation-focused scenarios result in even higher stocks, with a higher gain based on the replanting of mixed indigenous species in lieu of waiting for natural succession to climax. In conclusion, the modeling results for the land use scenarios clearly show the potential of conservation actions for Kakamega Forest for improving both bird diversity and carbon storage. The land cover change time series reveal value by comparing various scenarios to estimated bird diversity and carbon stock at particular times and forest states during the last 100 years. Here, accordant visualizations help to successfully communicate important information about nature conservation.