NCEAS / msai4earth-esa

MSAI4Earth Ecosystem assessment
2 stars 0 forks source link

Incorrect classification on water #5

Open carmengg opened 2 years ago

carmengg commented 2 years ago

While looking at outputs from the spectral+lidar model, I noticed some points on the water were classified as ice plant. These points were correctly classified as non-ice plant in the spectral-only model.

Investigating the source of this error, I found that the way I downloaded the CFO canopy height datasets somehow added a buffer around the coast where the canopy height value was 0, instead of having the correct value of "no data" or -9999. In the image below, the green is the Santa Barbara shape file, the white is the buffer in the canopy height raster where all values are 0, and the black are the -9999 values in the raster.

Screen Shot 2022-08-18 at 12 32 56 PM

To try to solve this I downloaded the CFO canopy height using their suggested method. Initially I did not do it like that because I was unable to import gdal (it's from osgeo import gdal). The buffer around the coast disappeared by downloading the raster in this way.

Screen Shot 2022-08-18 at 12 41 56 PM

I will update the canopy height rasters (2016, 2018, and 2020) and check how this alters the training set and downstream tasks.