Methods:

Data Downloading: The process begins with setting up a Conda environment and installing necessary packages, including eumetsat::eumdac for downloading data from the EUMETSAT data store. The data is downloaded using a Jupyter notebook (EUMDAC_downloadSST.ipynb shows an example of using the Data Store's API). Note that two approaches were used when developing this dataset. The first approach contained only the highest quality data: (applied filtration: SST > 0 & quality_level == 5 & NOT (l2p_flags.river) & NOT (l2p_flags.lake)) and the second approach (relaxed) contains data from lower quality classes, and has higher uncertainty, but better coverage (applied filtration: SST > 0 & quality_level > 1). The results in this dataset are from the relaxed approach with greater coverage.

Data Extraction: After downloading daily NetCDF (NC) files, the data is extracted. This involves two steps: first, extracting the NC files from the downloaded data using a command-line script that calls the gpt.exe tool from the SNAP software with a specific XML processing script (SNAPprocessCollocate.xml). The second step involves extracting the SST data from the NC files, again using gpt.exe, but this time with a different XML processing script (SNAPextractSST.xml).

Data Merging: After extraction, the SST files are merged together using another Jupyter notebook (MergeSST2.ipynb shows an example of how this is done). This creates monthly SST GeoTIFFs and a single combined SST dataset from all the individual files.

Data Validation: Finally, the processed SST data is validated using the SNAP Desktop software. This involves extracting pixel values from the processed SST data and comparing them to reference data from a text file (lighthouse2017_wTemp.txt). The extracted pixel values are saved in a separate folder for further analysis.

The GDAL utility was used, specifically gdalwarp to match the extent with existing rasters, gdalinfo to calculate statistics for each band, and gdal_translate to export as single band GeoTIFF files.

The final layers provided here were extrapolated from the process above by applying a circular neighbourhood focal mean to match the extent of other environmental predictor values.