ArboMAP 2.2

Models are now specified in the settings and are directly compared to one another. A variety of smaller changes to accommodate this, will be specified in the changelog.

ArboMAP 2.1

Includes fixed degree-of-freedom splines vs. thinplate splines, anomalized environmental and human data, and seasonally-varying vs. fixed splines. Various new outputs and visualizations.

ArboMAP 2.0

We have updated ArboMAP to use the most recent version of the mgcv library, which allows us to perform better calculations, faster with the bam(..., discrete=TRUE) method of Wood et al. This is a major change from previous versions, which should now be considered obsolete. It is unlikely that any estimates will change wildly, but the statistical methods we now use are more robust and in some cases might even be faster.

The bam(..., discrete=TRUE) method discretizes covariates before modeling, which allows bam to evaluate functions at far fewer points. This produces enormous increases in speed while changing estimates very little in most cases. Until recently, there was an error in the s(covariate, by=matrix) code we use to construct distributed lags on environmental data. This has been corrected, so ArboMAP now includes this method.

The benefits afforded by bam(..., discrete=TRUE) allow us to use thin-plate splines, which typically have better behaviors than cubic regression splines. The resulting non-/semiparametric models in ArboMAP outperform the parametric, cubic-spline models we published previously.

The js code to download weather data from Google Earth Engine has been streamlined. The previous code works, but occasionally and unpredictably ran into memory errors. We have reduced the footprint of the code so that these should be far less likely.