project team: John Smith, David Durden, Emma Mendelsohn, Carl Boettiger
This repo holds code for a spatially explicit forecasting challenge pipeline to benchmark spatial models using MODIS leaf index data. In this example we focus on locations of wildfire burns and recovery.
Our goal is to select locations across a variety of environments and burn conditions. Currently we have two sites selected from Monitoring Trends in Burn Severity (MTBS). These shapefiles are available in the /shp directory.
- California August complex fire
- Colorado East Troublesome
Functions are stored in the R/ directory.
fire_bbox()reads in a fire boundary shapefile and determines a bounding box for grabbing MODIS data with a padding option.ingest_planetary_data()downloads data from Microsoft planetary comuputer and returns agdalcubedata cube proxy object.create_target_file()subsets the data cube, pulls data for a given data and serializes target geotiff to disk.spat_climatology()creates climatology predictions and serializes prediction geotiff to disk. Predictions are created using an ensemble of historical data within a given month. If historical data is missing, values are treated asNAand bootstrap re-sampling is performed using previous monthly data.scoring_spat_ensemble()assigns CRPS (Continuous Ranked Probability Scores) and Logarithmic Scores for a given target file and ensemble forecast. Serializes scored geotiff to disk.na_bootstrap_fun()is used internally for re-sampling during creation of climatological forecasts. The function takes a vectorxof (possibly missing) data and fillsNAvalues using a bootstrap re-sampling of non-NAvalues.
This project uses renv for package management. Use renv::restore() to load project packages.
- Ingest additional fire sites. Potential locations
- NEON GRSM: https://www.neonscience.org/
- NEON SOAP: https://www.neonscience.org/field-sites/soap
- Arizona rapid burn/recovery
- Eastern canada fires
- Ingest addition data streams (e.g., burn intensity from MTBS)
- Deployment for submissions