borealbirds / gnm Goto Github PK

What we have on borealbirds.ca

• data locations (0/1 of species)
• mean density map and table by BCR-province
• distribution maps
• habitat associations by LCC05 cover type in each BCR-province unit

What we should produce from the latest results

data locations

1. a static png which is doable but not very useful
2. a raster with 0/1 at 1 km^2 resolution.

BCR-province level density

Calculating mean density and pop size is important.

1. An interactive map can be done but I don't see high value other than showing off.
2. A queryable table (sort/filter) would make more sense with a static map showing where BCRs are. This can be combined with distribution map to make it species specific.
3. static png : would anyone ever use a map of this rather than the continuous distribution/density map?
4. vector shapefile : would anyone ever need the densities in a spatial format?

Distribution/density maps

The underlying model is the same.

1. Static map (with BCRs and jurisdiction lines overlaid) can be shown along side the table
2. Raster layers of continuous maps (unclear on this bit about links pointing to relevant raster layers)

Habitat associations

Based on post-stratification using Canadian land cover 2010 or North Am land cover.

1. Static bar graphs (png) for each BCR jurisdiction unit (because sometimes people want to be able to plug an image directly into their own work).
2. Interactive bar graphs.

Data download

All species info combined (i.e. not separate by species)

• SQLite database with all relevant tables + metadata (density & pop size estimates by region, habitat level density estimates)
• Excel / csv version for the spreadsheet enthusiasts
• metadata & docs / reports (i.e., the methods & interpretation)

Species specific outputs

• 0/1 (detection) raster
• density raster
• various static pngs?
• BRT model (for custom prediction purposes)
• BRT output as zip file (variable importance, xv, effect plots etc)

How to deal with uncertainty?

• we can present marginal variability (i.e. in a region or land cover type how much density varies due to other predictors)
• we can incorporate bootstrap based variation (possibly x.1 version)

Testing Slack integration

• Check that we have all the bird data (Atlas updates, BU/WildTrax, etc)
• Use 100 km^2 buffer around BCRs and compile all bird data with offsets, save as bundles
• Bundle predictors and raster stacks by buffered BCR for prediction and intersect for modeling
• Finalize predictor sets for BCRs
• Finalize BRT code (1 species in 1 BCR) [possibly extending by bootstrap]
• Develop HPC code to run on Graham (check timings and RAM usage)
• Deploy data and run models on Graham and collect results (species by BCR, no bootstrap for now)