Methods

Overview

Our design brief was to develop an accurate remote sensing classification system that avoided the difficult technical details typical of satellite remote sensing. Here you can learn about how remap works. Remap relies on the incredible data storage and geospatial analysis capacities of the Google Earth Engine.

Remap requires only georeferenced training points that identify the different classes of interest, which most often representative ecosystems or land cover classes. Each training point is then intersected with a range of satellite datasets (predictors) to train a random forest classifier. Once the random forest is trained, remap classifies all of the pixels present in a study area into one of the classes present in the training set.

Remap uses freely available, high resolution, global-scale, and publicly available biophysical ( slope, elevation), spectral ( Landsat Archive imagery) and climatic ( precipitation, temperature) datasets to inform the classification. Each dataset is automatically processed into relevant indices, such as the Normalized Difference Vegetation Index, before training and running the classifier.

Predictors
Type Short Name Long Name Earth Engine ID
Index NDVI Normalised Difference Vegetation index LANDSAT/LC8_SR
Index NDWI Normalised Difference Water index LANDSAT/LC8_SR
Index WBI Water Band Index LANDSAT/LC8_SR
Index BR Blue band minus Red band LANDSAT/LC8_SR
Index BG Normalised Difference Blue Green LANDSAT/LC8_SR
Band Value Blue Blue band LANDSAT/LC8_SR
Band Value Green Green band LANDSAT/LC8_SR
Band Value Red Red band LANDSAT/LC8_SR
Band Value NIR Near Infrared band LANDSAT/LC8_SR
Elevation Elevation SRTM Digital Elevation Data 30m USGS/SRTMGL1_003
Elevation Slope SRTM Slope USGS/SRTMGL1_003
BIOCLIM Mean Annual Temperature Mean Annual Temperature WORLDCLIM/V1/BIO
BIOCLIM Annual Precipitation Annual Precipitation WORLDCLIM/V1/BIO
See the About page or read our paper for more information.
Supported and funded by: