Forecasting #
Previous sections cover the original M4MD model fitting pipeline, developed by L. Zachmann. This section documents the forecasting extension developed by the Eric and Wendy Schmidt Center for Data Science & Environment in collaboration with the National Park Service’s Inventory & Monitoring Division. The forecasting pipeline takes in a fitted model (as described in the previous sections of this guide) and produces posterior predictive forecasts under user-specified future climate scenarios.
In context of this repository, “model fitting” refers to running analysis-pipeline.R, which fits a Bayesian hierarchical model to historical data via MCMC. “Forecasting” refers to running forecast-pipeline.R, which reads a fitted model’s posterior draws and projects the response variable forward under future climate scenarios.
Two subsections exist for the forecasting pipeline:
Run a forecast
- Start here for your first forecast with a step-by-step walkthrough and mock data OR
- Bring your own data + fitted model and learn how to run a forecast
Understand a forecast
- Go here to understand the underlying posterior predictive framework OR
- Explore what each step of the forecast pipeline is doing