Closed ethanwhite closed 7 years ago
I implemented this for a single exogenous predictor (annual NDVI) before my talk in Wyoming. Still need to explore how to do it for multiple exogenous predictors. We won't be able to dump all of the exogenous predictors we have into these simple models because we only have 20 time points to fit them to, so we'd either need to include limited subsets and do model selection at each site, or perhaps pick the most important features from the spatial models.
At our meeting last week, the consensus seemed to be that we'd use the 3 most "important" variables according to some metric (if I remember correctly).
How does elevation, bio_5, and ndvi_sum sound (based on @sdtaylor's figure)? That gives us climate, topography, and greenness.
Sounds good to me.
This happens in
auto.arima
using the optionalxreg
argument.