Image classification products forWater, Built, Forest, Secondary Forest,
and Savanna were merged into single, multi-class products (for
2000 and 2010) using discrete decision rules, and cases where features
were included in two or more classes were handled by these decision
rules. A priority order was used, where classes with the highest confidence and lowest misclassification/overlap between classes in the
interim classification products were assigned first, and those pixels
were then taken out of consideration for classification by subsequent interim
classification products. The priority order from highest to lowest
was: 1) Water from Fmask and NDWI, 2) Built from NDVI and radar,
3) Forest from SVVI and Hansen et al. (2013) products, 4) Secondary
Forest fromSVVI, and 5) Savanna from SVVI MVC texture and SVVI texture
MVC. Pixels not assigned to any of these classes were assigned to
the Agriculture class, as this was the only other LCLU class not yet
accounted for. Manually digitized Commercial Agriculture and Mining
classes were then added into the maps. For the final LCLU map
production step, the 2000 and 2010 thematic maps were subjected to
a 3 × 3 focal majority filter to spatially smooth the class values by considering
the majority of neighbor values in the 3 × 3 area around each
pixel.
Image classification products forWater, Built, Forest, Secondary Forest,and Savanna were merged into single, multi-class products (for2000 and 2010) using discrete decision rules, and cases where featureswere included in two or more classes were handled by these decisionrules. A priority order was used, where classes with the highest confidence and lowest misclassification/overlap between classes in theinterim classification products were assigned first, and those pixelswere then taken out of consideration for classification by subsequent interimclassification products. The priority order from highest to lowestwas: 1) Water from Fmask and NDWI, 2) Built from NDVI and radar,3) Forest from SVVI and Hansen et al. (2013) products, 4) SecondaryForest fromSVVI, and 5) Savanna from SVVI MVC texture and SVVI textureMVC. Pixels not assigned to any of these classes were assigned tothe Agriculture class, as this was the only other LCLU class not yetaccounted for. Manually digitized Commercial Agriculture and Miningclasses were then added into the maps. For the final LCLU mapproduction step, the 2000 and 2010 thematic maps were subjected toa 3 × 3 focal majority filter to spatially smooth the class values by consideringthe majority of neighbor values in the 3 × 3 area around eachpixel.
การแปล กรุณารอสักครู่..
