Fit random forest model

[Aariq]

• For each pseudo-genotype (WT, hotleaf, anthocyanin, dwarf)
• ecosystem as fixed effect/categorical/dummy variables
• predict NPP at new sites

[Aariq]

Can we use random forest to do timeseries forecasting?

https://www.r-bloggers.com/2019/09/time-series-forecasting-with-random-forest/

Other options besides using the whole timeseries:

• Just use the end date
• Just use summer months (still not iid probably)
• Don't use RF, use something like ARIMA or a GAM instead

[Aariq]

Using last year summer NPP as response variable, bioclim vars and ecosystem as predictors.

Would like to compare to some kind of linear regression model, as I suspect it will do better than RF

I should also consider including all "genotypes" in the model and treating genotype as a fixed effect. That will increase the power