Cellular automata with non-linear transitio rules for simulating land cover change
1. Cellular automata with non-linear transitio rules for simulating land cover change Katarzyna OSTAPOWICZ [email_address] T he 2009 Annual International Conference of the Royal Geographical Society , 2 6-28 August 2009 , Manchester Department of GIS, Cartography and Remote Sensing Institute of Geography and Spatial Management Jagiellonian University
11. Workflow Input data Change probabilities Transition rates Natural and antropogenical variable CALIBRATION PART SIMULATION PART ANN/SVM Land cover maps Management plans cross-tabulation Land cover change simulation
12. Input data Forest/non-forest: 1987, 2000, 2006 ( source: Landsat images, supervised, hierarchcal approach combining image segmentation, knowledge-based rules and likelihood decision rule ) Elevation and slope ( source: STRM DEM, spatial resolution 90 m ) Distance to artificial areas (source: land cover map 2006; distance operation) Migration, NUTS type (urban/rural), distace to urban NUTS (source: GUS) Ownership: state/private forest (source: state forest)
13. Training plots (800, 200 per class) forest aforestation and natural succession non-forest deforestation
14. Transition rules f(P ij , N i , R ij ) CHANGE PROBABILITIES (P ij ) NEIGHBOURHOOD (N i ) Σ n i > 6 (i – land cover type) forest f (elevation, slope, migration, NUTS type, ownership, distance to artificial areas) TRANSITION RATES (R ij ) e.g. for forest 0.25% per year TRAINING: forest change between 1987-2000-2006 SCENARIOS: 2006-2056
15. Maximum accuracy for transsition rules ANN: 75% SVM: 79% 57,06 2050 53,27 2020 55,79 2040 52,16 2010 54,54 2030 50,98 2000 forest cover [%] year forest cover [%] year