A “Neighborhood” Approach for Using Remotely Sensed Data to Post-Process Species Distribution Models for Conservation Assessments

Author

Bethany A. Johnson, CUNY City CollegeFollow

Date of Award

2023

Document Type

Thesis

Department

Biology

First Advisor

Robert P. Anderson

Second Advisor

Ana C. Carnaval

Third Advisor

Mary E. Blair

Keywords

species distribution modeling, conservation, IUCN, redlisting, Maxent, Handleyomys

Abstract

Species distribution models can be used to predict environmental suitability, estimate changes in species’ geographic ranges, and support conservation assessments. Masking suitability maps built from these models to remove areas currently lacking appropriate vegetation can refine predictions and aid conservation applications. However, the uncertainty around coordinates (even for recent records) can exceed the fine resolution of the additional data. Here, I present a novel yet simple methodological approach to reflect this reality of coordinate uncertainty of occurrence records and mismatch between climatic and habitat data resolution. I implement it for a forest-dwelling species currently considered threatened by the IUCN (Handleyomys chapmani). After building a Maxent species distribution model indicating climatically suitable areas, I determined deforestation tolerance thresholds (the lowest value of forest cover where the species has been observed recently) using two methods based on matching recent records with forest-cover data for the corresponding year: 1) extracting the exact pixel value where a record fell; and 2) using an aggregate measure of that and surrounding pixels (the “neighborhood” more likely within the radius of actual sampling that yielded the record). I applied the respective thresholds to mask the suitability prediction for H. chapmani, identifying areas of inferred habitat loss due to insufficient forest cover. For each method, from the post-processed map and records of the species, I calculated extent of occurrence (EOO) and area of occupancy (AOO), two metrics used by the IUCN for threat level categorization. The post-processing method had little effect on EOO estimates due to loss mostly within the range and not along the spatial margins, but for each method there was a vast difference when calculating EOO from occurrences versus from masked suitability maps. For AOO, the neighborhood method reduced the estimated values somewhat compared with the exact method, highlighting patchy loss throughout the range. However, the localized deforestation evident at a regional extent at the fine resolution of the forest-cover data largely disappeared when projecting to the 2 km x 2 km grid standardized by the IUCN. Overall, the values for EOO and AOO calculated here suggest removing the species from the threatened category, since the most plausible ones for each metric were above the respective thresholds (and substantially higher than those in the current IUCN account). The neighborhood method and associated code provided here may prove useful for many species, particularly those having occurrence records with high coordinate uncertainty in regions of low spatial autocorrelation of the environment, where small georeferencing errors can lead to great differences in habitat information. This research also highlights the importance of obtaining accurate georeferences (corresponding to the habitat sampled) when collecting field data and more generally of using habitat-refined distribution maps for conservation assessments.

Recommended Citation

Johnson, Bethany A., "A “Neighborhood” Approach for Using Remotely Sensed Data to Post-Process Species Distribution Models for Conservation Assessments" (2023). CUNY Academic Works.
https://academicworks.cuny.edu/cc_etds_theses/1115