2019-11-13
See for jags_RW and simplexEDM. NaiveArima added in 0.9.0
With switch to portalcasting 0.15.0
2019-03-21
In nbGARCH and then extended into nbsGARCH, the models fall back
to a Poisson distribution if the negative binomial fit fails. Previously
(with only nbGARCH) the Poisson fit always succeeded in those back-ups,
but now (with nbsGARCH) that sometimes isn't the case (because the predictor
model is more complex) and even the Poisson fit can fail. So now for both
models, if that fit fails, we follow what occurs in pevGARCH which is to
use the fcast0 forecast of 0s and an arbitrarily high AIC (1e6).
2019-03-20
pevGARCH() was not set up for hindcasting within the pipeline based off
of the historical covariate forecasts. There's now smooth integration
via a few additional elements in the metadata list (covariate_source
and covariate_date_made) which are only presently used for hindcasts
but may prove to be useful with future models in forecast mode as well.
2019-03-19
2018-12-12
With the migration of code to [portalcasting](https://github.com/weecology/portalcasting
2018-03-20
The Box-Cox transformation was removed from the AutoArima model to have all of the models be predicting the same data.
2018-03-15
Covariate (weather and NDVI) were changed from recent averages to predicted values. We are currently using downscaled ENSMEAN climate predictions and a local NDVI forecast (using a seasonal auto ARIMA).
2018-02-08
Associated with the shift to newmoons, the pevGARCH model set was expanded to include an intercept-only model.
2018-02-08
In Feb 2018, the models were edited to work on the newmoon numbers, thereby requiring acknowledgment of missing surveys (i.e. newmoons where a complete survey was not conducted). Because the models do not handle missing values, we decided to interpolate missing data for the time being.