The quantitative analysis of threats to the species provided by this report must be combined with expert evaluation. In particular estimating the extent of occurrence and area of occupancy of species with partially known distributions is extremely challenging. A range of methods have been applied in this report in order to guide the evaluation.
The key areas suggested by the IUCN guidelines are.
| Category | EOO | AOO | |
|---|---|---|---|
| 1 | Critically endangered | < 100 km2 | < 10 km2 |
| 2 | Endangered | < 5,000 km2 | < 500 km2 |
| 3 | Vulnerable | < 20,000 km2 | < 2,000 km2 |
The IUCN guidlines for redlisting suggest that the extent of occurrence can be estimated by calculating the area of a convex hull drawn around known observations of the species. There are well known problems with this approach.
The results of fitting a convex hull and an extended convex hull with a 50 km buffer are shown below.
A concave hull can be thought of as the geometry you get by “vacuum sealing”“ a set of geometries. The se should avoid some of the artefacts of concave hulls. However unlike a convex hull, there is no single form for a concave hull.
The IUCN guidelines suggest the use of Alpha hulls. These are generalisations of a convex hulls. They are produced by first finding the Delauny triangulation of the points. The average line length is measured and all lines above a multiple (alpha) of this value are deleted. The IUCN guidelines state that
"The value of α can be chosen with a required level of resolution in mind. The smaller the value of α, the finer the resolution of the hull. Experience has shown that an α value of 2 is a good starting point for some species (however, the value to use for specific cases of assessing reductions in EOO should be based on a compromise between minimizing the potential bias associated with incomplete sampling of outlying occurrences and minimizing the departure from a convex hull).”
This approach can be implemented easily in R using the alpha hull package. However there are some challenges assocated with the use of Alpha hulls.
It is difficult to apply this approach consistently for all species. The area of the hull is highly sensitive to the value of alpha. Alhtough a range of values can be used, the interpretation of the meaning of the alpha parameter is non intuitive and cannot be directly related to the observed distribution of the species as it is sensitive to the mean arc length of the Delauny triangulation. In some cases the area of the alpha hull produced by the alpha hull package is ill defined as it does not form a simple topology.
A more intuitive way of calculating a convex hull is though polygon inflation and deflation. Points within an expanded buffer radius are first united to form a polygon. The buffer is then reduced by the same extent to leave the points united as a hull. The distance used to merge the points has to be determined. However this parameter has an intuitive interpretation. If the distance between points is over twice the buffer distance the hull will show disjunction. If the buffer is dissolved by slightly less than the radius used for expansion the result is a convex hull which extends slightly beyond the limits of the collection points.
The figure below shows the results of
These areas will typically be similar to the results of fitting an alpha hull with alpha values around 2 and 4 respectively. However the results of fitting an alpha hull are not predictable.
The area of the simple convex hull is 1259.7 thousand square km.The area of a convex hull extended by 50 km around the points is 1574.9 thousand square km.The area of a concave hull based on an expansion of 500 km around the collection points is 307 thousand square km. The area of a convex hull produced by uniting points within a 50km radius and extending a buffer around the united points of 10km is 51.6 thousand km2
These areas are compared to the limits used by the IUCN graphically below. The red line represents the EOO that would lead to the species being classified as critically endangered. The orange line is the upper limit for endangered. If the AOO is below the green line the species may be considered vulnerable under the B2 criteria. Note that a logarithmic scale is used in order to allow a wide range of areas to be compared.
Species distribution models that use climatic variables as input may be useful as an alternative means of estimating Extent of Occurrence. They represent areas of potential distribution, so are not suitable for determining area of occupancy. The IUCN state that models of habitat suitability cannot be used directly to estimate a taxon’s AOO because “they often map an area that is larger than the occupied habitat (i.e., they also map areas of suitable habitat that may presently be unoccupied). However, they may be a useful means of estimating AOO indirectly, provided the three following conditions are met. i) Maps must be justified as accurate representations of the habitat requirements of the species and validated by a means that is independent of the data used to construct them. ii) The mapped area of suitable habitat must be interpreted to produce an estimate of the area of occupied habitat. iii) The estimated area of occupied habitat derived from the map must be scaled to the grid size that is appropriate for AOO of the species.”
The output from the climate model can be combined with a convex hull. The intersect between them is one representation of the extent of potential occurrence.
The predictive ability of a climate niche model based on soil moisture values dynamics, annual temperature and annual temperature range was tested by splitting the data by latitude. The AUC values were 0.77, and 0.48.
The best AUC value of 0.77 suggests that the model has some ability to predict new data.
Cover has been extracted from the landsat resolution (28.5m x 28.5m) coverage for the year 2000 produced by Sexton et al (2013)-
Statistics calculated for 10 km x 10 km grid cells within the area representing the intercept between the niche model predictions and buffered convex hull. The overall mean forest cover was calculated to be 44 %. The classification is based on NDVI so the mean forest cover will typically underestimate the percentage of dry forest. An alternative to using the overall mean forest cover is to calculate the percentage of landsat pixels with an estimated cover of at least 30%. This value is 66 %.
The model based estimates of EOO can be adjusted downwards to represent the proportion of the area with some forest cover.
A more complete analysis would use cover change maps at this resolution. These are not yet available for the complete study area.
Area of occupancy is extremely difficult to estimate from incomplete data. The IUCN suggest that a 4km squared (2km x 2m) grid be placed over obervations of known presence and the total area calculated from that. It is simpler and rather more accurate to use a buffer around points. In order to be comparable to the IUCN method the buffer can be made used with sqrt(4/pi) = 1.128 km radius.The area of occupancy under this very strict IUCN criteria is 574 square kilometers..
This estimate of AOO assumes that the species is only found around the sites from which it has been collected and that the sites are defined by a very narrow radius. It may be credible in a few cases for species that are known to be extremely localised. If the radius is extended from 1km to 10km the area increases as shown below.
Because most collections have been made from accessible areas the procedures laid out in the IUCN guidelines are unlikely to provide accurate estimates of the true area of occupancy when botanical collections are used as input. Expert judgement will usually be needed in order to suggest the area occupied. The mapping tool interface has been designed to aid this process.
The number of collections deposited in herbaria has fallen dramatically iin recent years as a result of changes in the funding and scientific culture. Therefore the temporal pattern cannot be used as a proxy estimate of abundance. The pattern must be interpreted in the light of expert knowledge of the species concerned.
A species can only be registered as present within an area in which collections are being made. The figures below provide a guide to the spatial density of collections
Projections using climate niche models are often used to suggest threats to species as a result of climate change. These methods are only likely to be reliable if the species climate niche can be accurately paramaterised. As disgnostic tests reveal that models may only have moderate predicitive ability empirical evidence of realised climate change over the last fifty years has been included as a measure of threat, rather than extrapolation. The National Center for Atmospheric Research provide a comparison of avaliable empirical data sets. Gridded coverages derived from climate records draw on the same underlying data sources, but have slightly different characterisitics. An extrapolated historical data set is provided by Willmotta and Matsuura of the University of Delaware at a 0.5 degree resolution. This is the finest resolution global coverage currently avaliable. In 2012 it was updated to include years up to 2010. Changes in precipitation patterns and temperature within the suggested area of distribution of the species can thererefore be analysed by extracting grid cells that intercept the species distribution.
The fitted spline model can be used as a visual guide to any trend.
##
## Family: gaussian
## Link function: identity
##
## Formula:
## avg ~ s(yr)
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(yr) 1.10 1.19 1.13 0.29
## Analysis of Variance Table
##
## Response: avg
## Df Sum Sq Mean Sq F value Pr(>F)
## yr 1 77915 77915 1.5 0.23
## Residuals 49 2537210 51780
The trend is a change of 2.66 mm per year. The p-value under a linear model is 0.23. This is not statistically significant
##
## Family: gaussian
## Link function: identity
##
## Formula:
## avg ~ s(yr)
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(yr) 1 1 60.2 6e-11
## Analysis of Variance Table
##
## Response: avg
## Df Sum Sq Mean Sq F value Pr(>F)
## yr 1 3.31 3.31 60.2 4.5e-10 ***
## Residuals 49 2.70 0.06
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
The trend is a change of 0.17 degrees per decade. The p-value under a linear model is 0.00000. This is highly statistically significant
Most fires in the study area are anthropogenic in origin and are indicative of agricultural activity. Fires, as detected by Modis sensors, increase in frequency in dry years. A high density of fires in the buffer around the collection points is suggestive of a high level of disturbance. An increasing trend in freuquency may suggest and increase in agricultural activity and forest clearance.
The query that extracts the data can take several minutes to run if the modelled area is larger than around 100,000 km2. In this case the threat is likely to be low, so the analysis is not run.
The trend is a change of 0.19 fires per 100 km2 per year. The p-value under a linear model is 0.36. This is not statistically significant.However a linear model may not be appropriate in this case.
IUCN Standards and Petitions Subcommittee 2013 Guidelines for Using the IUCN Red List Categories and Criteria. Version 10.1 Prepared by the Standards and Petitions Subcommittee Downloadable from http://www.iucnredlist.org/documents/RedListGuidelines.pdf
National Center for Atmospheric Research Staff (Eds). Last modified 07 Nov 2013. “The Climate Data Guide: Global (land) precipitation and temperature: Willmott & Matsuura, University of Delaware.” Retrieved from https://climatedataguide.ucar.edu/climate-data/global-land-precipitation-and-temperature-willmott-matsuura-university-delaware.
Sexton, J. O., Song, X.-P., Feng, M., Noojipady, P., Anand, A., Huang, C., Kim, D.-H., Collins, K.M., Channan, S., DiMiceli, C., Townshend, J.R.G. (2013). Global, 30-m resolution continuous fields of tree cover: Landsat-based rescaling of MODIS Vegetation Continuous Fields with lidar-based estimates of error. International Journal of Digital Earth, 130321031236007. doi:10.1080/17538947.2013.786146.
Willmott, C.J. and S.M. Robeson (1995). Climatologically aided interpolation (CAI) of terrestrial air temperature. International Journal of Climatology, 15(2), 221-229.
Willmott, C. J. and K. Matsuura (2012) Terrestrial Air Temperature and Precipitation: Monthly and Annual Time Series (1900 - 2010), http://climate.geog.udel.edu/~climate/html_pages/README.ghcn_ts2.html.