search

James-Thorson-NOAA / VAST

Spatio-temporal analysis of univariate or multivariate data, e.g., standardizing data for multiple species or stages
http://www.FishStats.org
GNU General Public License v3.0
124 stars 53 forks source link

Xconfig_zcp in release 2.1.0 #155

Closed N-DucharmeBarth closed 4 years ago

N-DucharmeBarth commented 5 years ago

Hi Jim, I just saw that you released a new feature Xconfig_zcp which should provide greater control on estimating the effect of environmental covariates if I'm interpreting your source code correctly. Do you have any further documentation on the use of this feature? Best, Nicholas

agruss2 commented 5 years ago

Until now, VAST was able to account for the effect of environmental covariates on density (or any other dependent variable) through the gamma_p X_p term, where gamma_p is the effect of the environmental covariates and X_p is a vector of environmental covariates. The new feature Xconfig_zcp was introduced to have a means to estimate a spatially varying coefficient for indices like the PDO or the cold pool extent. The new feature Xconfig_zcp is such that the gamma_p X_p term is now substituted with a (gamma_p + sigma_p epsilon_p) X_p term where epsilon_p is a zero-mean Gaussian Markov random field with unit variance for each covariate p. When estimating a spatially varying coefficient, VAST estimates its variance sigma_p^2 as a fixed effect, and estimating sigma_p^2 = 0 results in the reduced, linear effect of the covariate of interest. VAST users must specify that gamma_p = 0 for any covariate p that is identical across space whenever intercepts are estimated as fixed effects; this will need to be done, for example, when fitting to cold-pool extent.

I am just taking a crack at responding to you from the material that Jim kindly shared with me, so that VAST users can have some first information on this useful new feature :)

agruss2 commented 4 years ago

@James-Thorson-NOAA @N-DucharmeBarth I suggest closing the present issue, now that the paper reporting the above-mentioned developments has been published (https://aslopubs.onlinelibrary.wiley.com/doi/abs/10.1002/lno.11238).

James-Thorson commented 4 years ago

Thanks Arnaud for looking through these, and Nicholas for asking originally. Yes, that paper documents the benefits of allowing spatially-varying coefficients. Xconfig_zcp can also be used to facilitate backwards model selection, by starting all elements at 1 and then changing some to 0 as they drop out.