Remote-sensing and biodiversity in a changing climate Catherine Graham SUNY-Stony Brook Robert Hijmans, UC-Berkeley Lianrong Zhai, SUNY-Stony Brook Sassan Saatchi, JPL/UCLA Tom Smith, UCLA
Research Program (NIP) Integrate remote-sensing data into species distributional modeling Determine remote-sensing correlates of species richness across multiple taxonomic groups and spatial scales Integrate remote-sensing data with patterns of evolutionary diversification Predict future species distributions Train Latin America and US scientists and conservationists
Research Program (NIP) Integrate remote-sensing data into species distributional modeling Determine remote-sensing correlates of species richness across multiple taxonomic groups and spatial scales Integrate remote-sensing data with patterns of evolutionary diversification Predict future species distributions Train Latin America and US scientists and conservationists
1) Extract environmental data for point localities; Annual Temperature Annual Rainfall 2) Make statistical model describing distribution in envirnomental space; Species Distributional Models 3) Project this model in geographic space to create a map.
Possible environmental datasets Remote sensing Indirect measurements High resolution Global coverage Recent coverage Climate Direct measurements Low resolution Extrapolations Global coverage Long term coverage
Remote-sensing data: issues for species distributional modeling Age of point locality data Spatial accuracy of point locality data
Remote-sensing data: issues for species distributional modeling Climate Remote- sensing
A solution Use all point locality data with climate surfaces (museum and accurate recent survey data) Use only “accurate” point locality data with remote-sensing layers (modis tree)
CLIMATE ONLY
CLIMATE & REMOTE-SENSING
CLIMATE & REMOTE-SENSING WITH ACCURATE POINTS
Research program Integrating remote-sensing data into species distributional modeling Determining remote-sensing correlates of species richness across multiple taxonomic groups and spatial scales Integrating remote-sensing data with patterns of evolutionary diversification Predicting future species distributions Training programs in Latin America and the US
Extinction Risk from Climate Change (Thomas et al. 2004; Nature) Predict 18 to 35% of species ’committed to extinction’ by 2050 Global warming major threat to biodiversity
Potential Problem with Niche Modeling and Climate Change Future climates will not be completely analogous to current. => Will models predict lower probabilities (model artifact)? => Validity of models should be tested using experimental approaches, historical evidence, physiological models and internal consistency.
Environmental space and climate change Current Future Species environmental requirements
Approach Compare results from physiology-based models (mechanistic models) with species distribution models Assume mechanistic models are “correct”
Experimental Design Compare
Species distribution (niche models) used BIOCLIM – envelop (boxcar) method DOMAIN – based on similarity statistics GAM – Non-linear regression MAXENT – machine learning/maximum entropy
Variation in range size and location predicted by models
Errors
Environmental space and climate change Current Future Species environmental requirements BIOCLIM GAMS MAXENT DOMAIN
Modeling species distributions across climates Species distributional modeling can provide similar results to mechanistic models. Performance of species distributional models varies Next? Incorporate climate change with land use patterns to evaluate extinction risk for a suite of species