For now model building is performed on only the LULC data (Mapbiomas 6). Also as a first example only considers the timesteps 1985 and 1986. The current process is as follows:
- Remaps the original 40+ mapbiomas classes to 6 main ones (forest, non-forest vegetation, farming, non-vegetated, water, none).
- Overlays the remapped mapbiomas rasters on a coarse reference grid (900m, so 100*100 of the original 30m pixels) and calculated the % of the previous classes contained in each coarse 900m cell.
- Specifies a bayesian network in which the observations are the coarse 900m pixels. The explanatory variables are the % of the pixel covered by farming and by the non-vegetated classes in 1985. The target variable is a binary variable where 1=some amount of forest (1985) was transformed to either farming or non-vegatated (1986), 0=in any other case.
