Used here

gjam vignette:

PBGJAM: Predicting Biodiversity with Generalized Joint Attribute Models

Our research objectives:

To acomplish these goals we are assimilating biodiversity data, including field-based abundance observations from the Breeding Bird Survey (BBS), Forest Inventory and Analysis (FIA), and the National Ecological Observatory Network (NEON). We developed functions for downloading remotely sensed imagery and other environmental data using Google Earth Engine’s python API. We are also integrating lidar and hyperspectral imagery at NEON sites to characterize vegetation structure and foliar traits, respectively. We then use gjam to create species distribution and abundance maps to show how communities are expected to reorganize with climate change. Our maps and gjam outputs will be available on a web app PBGJAM. Example code and tutorials will be available on github so researchers can use gjam with their own biodiversity data.

Figure 1: PBGJAM: Predicting Biodiversity with Generalized Joint Attribute Models.

Figure 1: PBGJAM: Predicting Biodiversity with Generalized Joint Attribute Models.



Correlative species distribution models (SDMs) determine the observed distribution of a species as a function of environmental covariates. gjam was motivated by the challenges of modeling the distributions and abundances of multiple species associated with multifarious data, multivariate response, and zero-inflation.

Multifarious data

Abundance data are multifarious, differing among species, across study sites, and in sampling effort. gjam combines multiple data types into a single model while avoiding non-linear link functions. Most SDMs limit species abundances to presence-absence, presence-pseudo-absence, or presence-only data, reporting only whether or not a species has been found at a certain point. Other SDMs rely on link functions to model the response.

gjam accommodates discrete and continuous data on the observed scale to allow for transparent interpretation.