This poster describes the objectives and methods of our Predicting Biodiversity with Generalized Joint Attribute Models (PBGJAM) project.
We propose to fully integrate key remote sensing variables with continental scale ecological data to provide broadly accessible ecological forecasts to a user community of ecologists and managers. The product will integrate biological data and remote sensing in a form that translates directly to decision-ready products. To determine which species and communities are most vulnerable to climate change and to forecast responses of entire communities, we propose a forecasting framework and software that will be capable of synthesizing National Ecological Observatory Network (NEON), other biodiversity networks, and related physical and biological data with remotely sensed information. Species distribution models (SDMs) used to anticipate community responses to climate change can be unreliable and imprecise—current estimates range from 0 to 50% species loss. SDMs fail to accommodate the joint relationships between species and the different scales of measurement—they cannot coherently synthesize data and thus cannot predict entire communities. We have developed a generative model of community response to climate change that accurately predicts distribution and abundance of each species jointly as well as their organization in communities. Satellite imagery characterizes habitat with temporal, spatial, multispectral detail that goes well beyond that available from interpolated climate data and land cover/use maps. Frequent imagery collection also provides the opportunity to incorporate phenology dynamics across seasons as well as year-to-year changes over the last 17 years. We will test our model for different communities with a range of remotely sensed products at different temporal and spatial scales, including MODIS Leaf Area Index, Vegetation Indices, Continuous Vegetation Fields, Gross Primary Productivity and Land Surface Temperature. Joint analysis will allow substantial improvement in community prediction by synthesizing directly with NEON abundance data such ecosystem properties as canopy density and structure, productivity, surface heat exchange, fragmentation and disturbance. The technical themes of our proposal are data-centric technology (primary) and computational technology (secondary).
As part of this analysis, we will extend the predictive modeling to explore fine scale ecosystem attributes from data currently being acquired for NEON sites by the airborne observation platform (AOP) which include waveform and discrete lidar and hyperspectral information from JPL’s airborne imaging spectrometer. This will allow us to determine how fine scale ecosystem attributes are represented at the regional scale with Landsat and MODIS products.
Application to taxonomically diverse communities monitored in NEON and other networks will allow us to forecast community change and reorganization. Predictive distributions, with full uncertainty, will be updated together with data availability. Predictive sensitivity analysis under climate change will anticipate community reorganization and the habitats that will be most critical for new species assemblages. This two-year project will move the current version of our tool from TRL 3/4 to 6. Our new semi-automated interface will enable revised forecasts as updated climate, remote sensing and biodiversity data come available. The tool will enable instant selection and retrieval of key remotely-sensed products (e.g. MODIS products at the LP-DAAC) via web application programming interface (API), selection of biodiversity data via API (e.g. from NEON) or upload, parameterization of the model and display and download of model output. NASA interests and goals will be advanced by linking the broad NEON data network to the rich time series of regional NASA products to improve predictions and understanding of spatial-temporal distributions of ecological communities.