The table below shows a summary of the occurrence records for the taxon. A record “with Lat/Long” is one that has a complete set of latitude and longitude values associated with it, which could be used for distribution modeling and the conservation gap analysis. In preparation for the conservation gap analyses, we classified each record based on whether it was an existing ex situ germplasm sample from a genebank, botanic garden, or other repository (labeled G, as most records were from genebanks), or a reference observation/voucher specimen (labeled H, as most of these records were from herbaria).
The map below shows the median, mean, and standard deviation (SD) of potential distribution model runs, as well as the predicted presence threshold distribution. The predicted presence threshold distribution was generated by applying a statistically determined threshold value to the median model distribution, and is used in the conservation gap analysis as the potential distribution model. The occurrences used in the modeling process are represented by dots, distinguishing between ex situ germplasm collections (G) and reference sightings/voucher specimens (H) records (as available), and the attribute data associated with the points can be viewed by clicking on the dots. The semi-transparent gray background reflects the native country-ecoregion area. Layers can be added to the map using the toggle buttons on the left side. The first layer turned on will be in the bottom position on the map.
For each taxon with at least 25 records with distinct coordinates, the predicted potential distribution was calculated as the median of ten MaxEnt model replicates (K = 10), using linear, quadratic, hinge and product features, with a regularization parameter ß=1.0. Taxa with very large numbers of records with distinct occurrences were modeled with the median of six rather than ten replicates due to computing constraints. For taxa with 12 to 24 records with distinct occurrences, the median of six replicates was calculated, and for those with eight to eleven records, the median of three replicates was calculated, both using the linear and product features. We evaluated the MaxEnt model output using the area under the receiver operating characteristic curve (AUC), the standard deviation of the AUC across replicates (SDAUC), and the proportion of the potential distribution model with a standard deviation of the replicates above 0.15 (ASD15). A robust model required an AUC mean ≥ 0.7; SDAUC < 0.15; and ASD15 ≤ 10%. Potential distribution models were thresholded using the maximum sum of sensitivity and specificity and clipped to the extent of the native country - ecoregion boundaries. When a robust model was not producible (generally for taxa with less than eight occurrences or where few occurrences were very close to one another), potential distributions were estimated by drawing a buffer of 50 km around each occurrence point and then clipping the total buffered area to the extent of the native country - ecoregion boundaries.
The table below shows the ecogeographic variables listed in order of importance based on the occurrence dataset, including whether or not they were used for distribution modeling, following the variable selection process. Bioclimatic predictors are from WorldClim2 . Topographic, wind speed, and solar radiation predictors were generated by CIAT DAPA and can be made available upon request.
The table below shows the ex situ conservation gap analysis summary. The Sampling Representativeness Score ex situ (SRS ex situ) calculates the ratio of germplasm accessions (G) available in ex situ repositories to reference/voucher (H) records for each taxon, making use of all compiled records, regardless of whether they include coordinates. The Geographical Representativeness Score ex situ (GRS ex situ) uses buffers of 50 km radius created around each G collection coordinate point to estimate geographic areas already well collected within the distribution models of each taxon, and then calculates the proportion of the distribution model covered by these buffers. The Ecological Representativeness Score ex situ (ERS ex situ) calculates the proportion of terrestrial ecoregions represented within the G buffered areas out of the total number of ecoregions occupied by the distribution model. A Final Conservation Score for ex situ (FCS ex situ) was derived by calculating the average of the three ex situ conservation metrics. All of the conservation metrics are on a scale from 0-100, with 0 = poor state of conservation and 100 = comprehensive conservation. The FCS ex situ is used to categorize species, with urgent priority (UP) for further conservation action assigned when FCS ex situ < 25, high priority (HP) where 25 ≤ FCS ex situ < 50, medium priority (MP) where 50 ≤ FCS ex situ < 75, and low priority (LP) for taxa whose FCS ex situ ≥75.
The GRS ex situ map feature shows the potential distribution model, with previous ex situ germplasm collection points surrounded by a 50 km buffer overlaid. Only germplasm (G) points are displayed on the map.
The ERS ex situ map feature shows ecoregions within the potential distribution model from which no ex situ germplasm collections have been made. The ecoregion extent has been clipped to be within the potential distribution model of the taxon. There are many cases where the ecoregion area within the potential distribution model extent is very small (e.g., less than 5 cells). Utilize the “All ecoregions” layer and the ERS ex situ legend to help locate these areas.