How to make a biogeographic province
If these notes are too brief, please write airwin at mta dot ca for more information.
Assemble data
We use nLw 443 and nLw 555 or 551 from SeaWiFS and MODIS/Aqua available at OceanColor. SST is obtained from OceanColor or Pathfinder/AVHRR.
We used monthly and annual maps, processed with SeaDAS and downsampled to 36km resolution.
Cluster data
The full data set too large to cluster, so we subsample the full dataset down to 32,000 points, sampled uniformly in time, but biased slightly towards coastal regions. Before clustering, we scale each variable by subtracting the sample means and dividing by the sample variance. We produce files with 32,000 rows with entries for: latitude, longitude, nLw443, nLw555, SST, K-means cluster number, Ward's cluster number, a combined cluster number, and a colour for the combined cluster. Here are the files for
- SeaWiFS/AVHRR scaling and clustering
- MODIS/Aqua scaling and clustering
Create high resolution maps
We use a nearest-neighbor (Euclidean distance in scaled nLw443, nLw555, SST) matching algorithm to assign every pixel with good data in a high resolution map to one of the 32,000 rows (and thus one of about 72 clusters).