Core algorithm review

Previous implementation

Given shapefile, outputs a new shapefile and raster

For each vector feature class:

• reproject to 3857 if needed
• if point data, buffer -> output_data feature class
• if polygon data, CopyFeatures(reprojected_data) -> output_data feature class
• add sap field to output_data
• for each feature in output_data
  • calculate sap and update feature
    • get area of feature - (bracket lower at 10_000 min)
    • sap = weight / area (bracket lower at 500 min why?)
  • for each row in output_data
    • create fishing_lyr in-memory FeatureLayer
    • RepairGeometry(fishing_lyr)
    • create raster with sap cell value
    • add to list of rasters
  • merge results into outCellStats - CellStatistics - calculate sum per-cell across entire list of raster rasters
  • Clip outCellStats to user-defined clipping bounds

Questions

• Why reproject to 3857 Web Mercator up-front? Is the varied area of each cell a problem?
  • Units are meters, allowing user to specify cell size in meters.
  • Web Mercator is ultimately used to display the rasters
  • As long as the study area is relatively small (how small?), the difference in area per cell across the extent of that study area will be minimal (how minimal?).
• Why is area set to min of 10_000?
• Why is sap set to min of 500?
• Why is the user-defined clipping bounds needed?

Performance improvement opportunities

• don't generate all the intermediate rasters, one per feature, use burn-in algorithm
• separate core module that kicks off the actual heatmap jobs, one per processor (local) or one per Lambda
• use MemoryFile if within memory limits
• bucket features and use windowed process to parallelize (or create separate rasters and merge)