Use Sentinel-2 to: (a) track variability in individual lakes' surface area, and (b) produce a lake occurrence mask across the Arctic.
Part 1: LakeScan | Scan the entire landscape to identify lakes while filtering out rivers, shadows, burn scars, etc.
Part 2: LakeTrack | Identify precise locations of individual lakes and record their changes through time.
LakeScan tells us where to point our water classification and also provides the basis for tracking lakes as objects rather than unrelated pixels. The output of LakeScan is a shapefile of buffered lakes, which is in turn the input for LakeTrack.
S2bgrnExport.ipynb OR S2maskedmosaic.ipynb
Rapidly filter, mask, visualize, and export Sentinel-2 images from GEE.
S2download.ipynb
Batch download Sentinel-2 images
S2imStack.ipynb
Combine the R,G,B,NIR Sentinel-2 downloads into 4-band images
Predictor.ipynb
Use the UNET model to predict lake locations
S2vectorize.ipynb (not yet committed)
Combine multiple UNET predictions into binary lake images, buffer lake extent, and vectorize. \
Based on all available Sentinel-2 images, LakeTrack provides a final lake occurrence mask for customizable areas and timeframes as well as a timeseries of each lake's surface area. \
LakeTimeSeriesv1.ipynb
Process images in Earth Engine and export near-daily observations of water extent
LakeTimeSeriesv2visualize.ipynb
Process images in Earth Engine and export a lake occurrence composite image
TimeSeriesConcatv1.ipynb
Combine Earth Engine exports into indexed timeseries for each lake.