1.1 A global land cover map produced through integrating multi-source datasets

Problem:
1.global land cover datasets:low accuracies(especially in forest and cropland domains)
solution:fusing the existing datasets
keywords:decision-fuse method;fuzzy logic
case:5global land cover datasets+3tree cover and croplands->1-km global land cover map(SYNLCover)
assessment:1)inter-comparison:SYNLCover had higher consistency; 2) quality assessment using the human-interpreted reference dataset:SYNLCover 71.1% while the others 48.6% and 68.9%
Details:
data:8 coarse-resolution(250m~1km) satalllte imagery
data processing:
1)uniform the geospatial reference sysytem:MODIS Sinusoidal projection;180°W ~ 180°E and 55°S ~ 90°N;250m->1km(the dominated land cover type);GeoTIFF
2)translate semantical variables:Classification scheme(target);Affinity scores(input and target);
3)integrate;
tree:S>=30;cropland:S>=43;
otherwise:

for life form: choose the maximun; if >1 then random
for leaf attributes:choose the maximun; if >1,decision matrix;else if=0,mixed

needleleaf=broadleaf>mixed
4)evaluate
1)Consistency analysis:
2)Accuracy assessment: