Projects
Selected work in conservation science and spatial modeling. All of it is open source — repositories are linked from each write-up.
GAMMa
A web application that allows users to view the extent of their own collection against the data available on GBIF.
- R
- Shiny
- leaflet
- sf
- terra
- GBIF
The problem
Botanic gardens and genebanks hold living collections that are meant to safeguard a species' genetic diversity. But a collection of fifty plants gathered from one valley preserves far less than fifty plants gathered across a species' full range, and until recently there was no accessible way for a curator to tell the difference. The analysis existed — it had been developed over a decade of conservation research — but it lived in R scripts that assumed you were a spatial analyst.
The gap was capacity, not method. Institutions of different sizes needed to apply the same assessment without hiring one.
What GAMMa does
GAMMa compares a collection's accession records against wild occurrence records for the same taxon, then scores how much of the wild range the collection actually covers. Curators pull reference records directly from GBIF inside the tool or upload their own, add their accession data as a spreadsheet, clean the records interactively, and run the analysis.
The result is three scores, each bounded 0–100, averaged into a Final Conservation Score:
- Sampling — germplasm accessions relative to reference records.
- Geographic — the proportion of the estimated wild range falling inside buffers around collection points.
- Ecological — the share of the 814 terrestrial ecoregions the species occupies that the collection reaches.
Low scores localize: the maps show which parts of the range and which ecoregions are missing, which turns an abstract score into a collecting trip. Results export as a shareable HTML report.
The application also works across institutions. Metacollections — the combined holdings of several gardens, managed jointly — can be assessed as a single unit, which is where the redundancy and the real gaps in ex situ conservation tend to show up.
Uploaded data is never stored. Analyses run in a single session and time out, so institutions can assess sensitive collections without handing over records.
How it was built
An R and Shiny application, modularized by analysis stage, with leaflet for the interactive mapping, sf and terra for the spatial work, rgbif for occurrence retrieval against the GBIF Backbone Taxonomy, and R Markdown for report generation. The gap analysis itself extends the methodology of the GapAnalysis R package — the same lineage as the crop wild relatives work — reworked so that a collections manager, not a programmer, is the intended operator.
Outcome
GAMMa is in public beta. The project has run workshops for collections managers since 2024 and produced gap analyses supporting active collecting programs.
The work was initiated in 2023 by the Atlanta Botanical Garden, The Morton Arboretum, and the Montgomery Botanical Center, with a core team drawn from those institutions plus Colorado State University's Geospatial Centroid, the New York Botanical Garden, and BGCI-US. It is supported by the Institute of Museum and Library Services (award MG-252894-OMS-23), BGCI-US, and the United States Botanic Garden.
GapAnalysis R
An R package that scores how well a species is conserved, ex situ and in situ, from occurrence records and a distribution model.
- R
- terra
- sf
- CRAN
The problem
Conservation gap analysis asks a simple question: how much of a species' diversity is already safeguarded, in genebanks and botanic gardens (ex situ) and in protected areas (in situ), and where are the gaps? The method had been applied to hundreds of crop wild relatives, but the code lived in project scripts. Every new study re-implemented it, and results were hard to compare or reproduce.
What GapAnalysis does
The package turns that method into a documented, tested R library. Given occurrence records, a modeled distribution, an ecoregion layer, and a protected-area layer, it computes three scores for each conservation strategy:
- Sampling representativeness — how many occurrences are backed by a conserved accession.
- Geographic representativeness — how much of the modeled range falls within buffers around conserved samples, or inside protected areas.
- Ecological representativeness — how many of the ecoregions the species occupies are represented in collections or protected areas.
The three combine into a final conservation score, ex situ and in situ, and a combined score with a priority category. Maps of the gaps come out alongside the numbers, so a score points at places to collect or protect.
Why it matters
The package is the shared foundation for the work that followed: the crop wild relatives assessment applied it across taxa, and GAMMa wraps the same scoring in a web application for collections managers. Publishing the method as a package means a result from any of those studies can be re-run, and the method can be applied by anyone with occurrence data.
Publication
Carver, D., Sosa, C.C., Khoury, C.K., Achicanoy, H.A., Diaz, M.V., Sotelo, S., Castañeda-Álvarez, N.P., & Ramírez-Villegas, J. (2021). GapAnalysis: an R package to calculate conservation indicators using spatial information. Ecography, 44(7), 1000–1009. doi:10.1111/ecog.05430
Crop wild relatives of the United States
Species distribution modeling and conservation gap analysis for all of the United States crop wild relatives.
- R
- terra
- sf
- targets
The problem
Wild relatives of crops carry genetic diversity that breeders depend on, but their conservation status is poorly documented. The 2020 PNAS assessment established a national baseline; new species needed the same treatment without re-inventing the pipeline.
What was built
A modernized modeling workflow covering occurrence data cleaning, species distribution modeling, and ex situ / in situ gap analysis, structured so that adding a taxon is a configuration change rather than a code change.
Outcome
Reusable methods applied to additional species groups, with a companion repository (cwrSDM) adapting the Aichi Target 13 codebase for ongoing work. The wild grapevine assessment was published in Plants, People, Planet, and the same methodology now underpins GAMMa, a web application bringing gap analysis to botanic garden collections.
Publication
Carver, D., Khoury, C.K., Frances, A., McCarry, N., Diaz-Garcia, L., Galarneau, E., Gora, S., Haidet, M., Heinitz, C., Knapp, W., Meyer, A., Miller, A., Mims, R., Sapkota, S., Spurrier, C., & Wen, J. (2026). Conservation gap analysis for wild grapevines (Vitis L.) of the Americas. Plants, People, Planet.