prepare_data should check for polygons with 0 pop

[timcdlucas]

In the Nigeria example, with reduction factor of 2, one of the polygons (Bakassi) has a total population of zero, which is what causes the problem. Probably want to check for this before fitting the model.

[simon-smart88]

Aha nice catch. Should be easy to implement.

[simon-smart88]

I had a look into this and adding weights = TRUE to terra::extract does result in some population being recorded at lower resolutions. You're right that it has a performance penalty, but it's not huge (~ 50 % increase) so if that's a concern, could we run extract once with weights = FALSE and then run it again on a subset where population == 0 with weights = TRUE?

[simon-smart88]

Got this for checking for zero aggregation per polygon:

  # Check for polygons with zero aggregation
  aggregation_sum <- terra::extract(aggregation_raster, polygon_shapefile, cells=TRUE, na.rm=TRUE, ID=TRUE, weights = weights, fun = "sum")
  zero_aggregation <- aggregation_sum[aggregation_sum[, 2] == 0,]
  if (nrow(zero_aggregation) > 0){
    zero_polygons <- polygon_shapefile[zero_aggregation$ID,]
    if (na.action){
      for (p in zero_polygons[[id_var]]){
        warning(paste0(p, "has zero aggregation values and has been removed from the response data"))
      }
      polygon_shapefile <- polygon_shapefile[-zero_aggregation$ID,]
    } else {
      for (p in zero_polygons[[id_var]]){
        warning(paste0(p, "has zero aggregation values"))
      }
      stop("Please remove the polygons from the response data or set na.action = TRUE or weights = TRUE")
    }
  }

I realised though that setting weights = TRUE makes things more complicated down the line for the covariates though as you end up with partial pixels. weight x aggregation works fine to give the estimated aggregation for each pixel but that's not going to work for covariates. So it's probably easiest for now just to remove the polygons and forget about weights = TRUE but we could consider including it as option only for handling the aggregation raster.