Outline

[admahood]

Keep this updated, reference scripts and s3 folders used

• [ ] Training data
  • [ ] use NAIP?
• [ ] Classification method
  • [ ] class labels - binary, multiple classes, fuzzy set method
    • I tried running the latest models using both the old SC threshold method and the new manual binary split. the manual classes seemed to produce better results but the variability is still there so we should probably consider other methods going forward - D, 5/22/19
• [ ] modelling method
• [ ] incorporating climate
  • [ ] test efficacy of precipitation anomaly as a predictor variable (and possibly other precip measures)
    • [x] downsample prism data to match ls5 resolution
      • [ ] script for downscaling: prism_resample_script.R
    • [x] create GB-wide precip anomaly rasters and attach to training data points
    • [x] apply model using precip anomaly as a predictor
    • [x] explore results; especially effect on year-to-year variability
      • year-to-year variability of unrealistic magnitude still present in model results.. D, 5/22/19
• [ ] incorporating year-to-year variability

[durphy1227]

• downsampling prism data, creating precip anomaly rasters for model input, and creating precip anomaly as a training variable all takes place in _prism_resamplescript.R.

• the latest (much cleaner & more efficient) model training and model application scripts are: model training: model_training_withprecip_current_May21.R model application: model_application_withprecip_current_May21.R