Herons E011
2050707
11/28/2021
Introduction
The Little Egret (Egretta garzetta) has recently colonised Great Britain. (Lock and Cook (1998)) Having been assessed as a rarity by the British Rarities Committee for decades a general increase in numbers was noted through the 1980s and after an influx in 1989 and the early 1990s it became clear that that the Little Egrets was becoming established in Britain. They were were first noted to have successfully bread on Brownsea Island in 1996. By 1999, there were an estimated 30 breeding pairs, (@musgrove2002). This had grown to 718 pairs in 2012 (Hiley et al. (2013)) and by 2018 there were an estimated 1,100 pairs, with perhaps 12,000 individuals over-wintering.(Ornithology (2020))
Little Egrets are wetland birds preferring lakes. marshes, floods and estuaries. They have a wide diet from large insects to small fish and amphibians. They are classified as GREEN in the UK and of Least Concern in Europe.
The recent influxes are thought to be in part due to the northward movement of the species in mainland Europe and post-breeding dispersion. (Little egret (egretta garzetta) - BirdLife species factsheet (n.d.)) Although wetland degradation is a threat globally to Little Egrets, it is thought that wetland protection is contributing to its success in Britain rather than climate change per se.(Ornithology (2010))
Nevertheless, in common with many taxa, Little Egrets are progressing northwards.(Hickling et al., 2006; Thomas and Lennon, 1999a) Thomas and Chris compared the range limit across a range of British breeding bird species between the 1968-72 and 1988-91 atlases and found that ‘southern’ species had moved north by an average 19km in that time. The southern limit of ‘northern’ species also moved north. They postulated climate warming as a probable explanation. (Thomas and Lennon (1999b))
This finding has been replicated across many bird species and taxa. Hickling et al found that of the 16 taxonomic group they analysed, most taxa had advanced northwards at varying rates. On average the northern margin of the range for the birds species included had moved by 29km over a 20 year period.@hickling2006a Pearce-Higgins et al found that birds in particular are susceptible to climate change impacts.(Pearce-Higgins et al., 2017; Pearce-Higgins and Crick, 2019) More recently Gillings et al have examined the shift in distribution of 122 British breeding bird species and confirmed northward shifts in ‘southerly’ species with a mean expansion of 13.5km, but there were individualist range shifts in other direction. Additionally there was no single predictive climate variable.( Gillings, Balmer and Fuller (2015))
This study has 2 objectives:
- To examine the trends in observations and the range of the Little Egret in Britain
- To estimate the extent to which climate variables may be responsible for range shifts.
Methods
Data sources
I obtained species observation data for Egretta garzetta from the NBN (Atlas@nbnatla) for 1990 to 2019 (most up to date year) via the NBN atlas plug-in to QGIS.@thefsc2016a Boundary data obtained from the ONS open geography portal.
Climatic variables - present and future (clim5) - were obtained from Worldclim (WorldClim (n.d.)) which I extracted directly via the raster package in R.( Hijmans et al. (2021))
Trends
I calculated annual observation counts per hectad (10 km2) per year from the NBN data- mapped to GB 10km grids to estimate trends visualise and spatial distribution. I used QGIS to obtain kernel density estimates (kde) to identify “hotpsots” and summarise the home range of Little Egrets by decade. These along with hectad ount maps give a visual impression the distribution of Little Egret populations over time.
To estimate the northern most edge of the distribution a commonly method is to take the mean centroid of the 10 most northerly occupied hectads per unit time. (Thomas and Lennon (1999b)) However for this study I calculatedthe 90th centile of the distribution of latitudes of annual observations to which I fitted a linear model to allow estimation of the rate of progression, - the slope of the fit - after converting to metres.
Species distribution modelling
Finally, to estimate the impact on climate I fitted a species distribution model (SDM) to summarise the the association of the range with climate, and forecast future range. SDMs are widely used to assess the impact of current and future climate on distributions and habitat suitability. (Elith and Leathwick (2009); Santini et al. (2021))
SDMs attempt to predict the spatial occurrence of a species of interest based on abiotic (climate, elevation) and biotic covariates (other species, vegetation). The choice of predictors will depend on the precise question to be answered. The conduct of SDMs draws heavily on the principles of statistical learning, which underpins modern modelling and machine learning practice.(Hastie, Tibshirani and Friedman (n.d.)) SDM has greatly developed and current applications often apply multiple classification algorithms of which logistic regression, glmnet, random forest, boosted regression trees, maxent and generalised additive modelling (GAM) are commonly used.()
Because binary classification algorithms need ‘yes’-‘no’ data and observation data is “presence-only”, it is usually necessary to generate background or pseudo-absence data by randomly sampling spatial points from the background.
The climate data is a raster with 19 bands for various metrics of temperature and precipitation at 2.5 degree resolution across the globe. Some of the metrics are highly correlated and I used a variance inflation factor (VIF) approach to remove correlated metrics. For future climate I used the AC, 85 2.5 model to 2050. I used glmnet, ranger (a fast implementation of random forest), generalised additive models (GAM) and boosted regression trees (BRT) as classification algorithms - these are available in the sdm R package.
All analyses were conducted in QGIS (Welcome to the QGIS project! (n.d.)) and R Software. Species modelling was conducted using the sdm R package.(Sdm package - RDocumentation (n.d.))
Results
Trends and changes in distribution
The annual trend in observation is shown in Fig 1 and the geographical extent of spread in Fig 2.
## Reading layer `Countries_(December_2017)_Boundaries' from data source
## `https://opendata.arcgis.com/datasets/f2c2211ff185418484566b2b7a5e1300_3.geojson'
## using driver `GeoJSON'
## Simple feature collection with 3 features and 10 fields
## Geometry type: MULTIPOLYGON
## Dimension: XY
## Bounding box: xmin: -8.649996 ymin: 49.86478 xmax: 1.763571 ymax: 60.86078
## Geodetic CRS: WGS 84## Reading layer `10km_grid_region' from data source
## `/Volumes/LaCie/E011/shp/gb-grids/10km_grid_region.shp' using driver `ESRI Shapefile'
## Simple feature collection with 2860 features and 1 field
## Geometry type: POLYGON
## Dimension: XY
## Bounding box: xmin: 0 ymin: 0 xmax: 660000 ymax: 1230000
## Projected CRS: OSGB 1936 / British National Grid
Add a comment
| decade | occupied |
|---|---|
| 1990 | 3656 |
| 2000 | 96034 |
| 2010 | 361190 |
## Simple feature collection with 3264 features and 3 fields
## Geometry type: GEOMETRY
## Dimension: XY
## Bounding box: xmin: -7.658296 ymin: 49.86222 xmax: 2.216353 ymax: 60.81259
## Geodetic CRS: WGS 84
## # A tibble: 3,264 × 4
## decade osgr_10km n geometry
## * <dbl> <chr> <int> <GEOMETRY [°]>
## 1 1990 NJ92 2 MULTIPOINT ((-2.084653 57.31554), (-2.009951 57.33802…
## 2 1990 NK05 1 POINT (-1.918037 57.58502)
## 3 1990 NM46 1 POINT (-6.167934 56.70685)
## 4 1990 NT19 2 POINT (-3.369429 56.14011)
## 5 1990 NT67 1 POINT (-2.562269 55.96672)
## 6 1990 NU13 3 MULTIPOINT ((-1.763461 55.60838), (-1.75544 55.62184)…
## 7 1990 NU20 1 POINT (-1.552494 55.31576)
## 8 1990 NX96 1 POINT (-3.641708 54.96852)
## 9 1990 NY06 1 POINT (-3.52477 54.97454)
## 10 1990 NZ29 2 MULTIPOINT ((-1.553301 55.24388), (-1.6083 55.24856))
## # … with 3,254 more rows## Simple feature collection with 6 features and 3 fields
## Geometry type: GEOMETRY
## Dimension: XY
## Bounding box: xmin: 145000 ymin: 635000 xmax: 415500 ymax: 855000
## Projected CRS: OSGB 1936 / British National Grid
## # A tibble: 6 × 4
## decade osgr_10km n geometry
## <dbl> <chr> <int> <GEOMETRY [m]>
## 1 1990 NJ92 2 MULTIPOINT ((395000 825000), (399500 827500))
## 2 1990 NK05 1 POINT (405000 855000)
## 3 1990 NM46 1 POINT (145000 765000.1)
## 4 1990 NT19 2 POINT (315000 695000)
## 5 1990 NT67 1 POINT (365000 675000)
## 6 1990 NU13 3 MULTIPOINT ((415000 635000), (415500 636500), (415500 …
Comment
Increasing frequency of observation
Northward movement evident
Northward
\[ \operatorname{latitude\_90} = \alpha + \beta_{1}(\operatorname{year}) + \epsilon \]
| Characteristic | Beta | 95% CI1 | p-value |
|---|---|---|---|
| year | 2,360 | 12, 4,707 | 0.049 |
|
1
CI = Confidence Interval
|
|||
Climate variables
## [[1]]
## NULL
##
## [[2]]
## NULL
The performance of each model is shown as a series of ROC (Receiver Operating Curves) in fig xxx. These curves visualise how well the models discriminate between presence and absence. A perfect model would have an area under the curve (AUC) of 1, and a completely random model 0.5. All the models performed well with the ranger algorithm giving the best performance with a mean AUC of 0.96.
Variable importance
##
## The variable importance for all the models are combined (averaged)...
Having established that the distribution of Little Egrets can be reasonably accurately predicted by climate variables, the question is are there any which are more predictive. The sdm package calculates variable importance ((vimp) metrics which are averaged across all models - these can be thought of as the proportion of different model runs in which each variable is a significant predictor of outcome. The mean importance per variable is shown in figure xxx and shows that mean temperature is a strong predictor, followed by the maximum temperature in the warmest month and isothermality (mean daily temperature range /annual temperature range)
We can now use the model to predict the current habitat suitability for Little Egrets.
Discussion
Little egrets have successfully colonised Britain and continue to move northward at an annual rate of x km per year increasing both their range and abundance. Species distribution modelling suggests that climate variables - especially mean temperature and summer warmth are predictive of the range. Forecasting to 2050 suggests that Little Egrets will continue their progress northward and continue to colonise much of the rest of Britain.
I could have improved the model by adding additional predictors such as elevation and a measure of habitat such as RAMSAR or waterway data, but this is beyond the scope of this work.
Other heron and large wading bird species appear to be following suit. For example the Great White Egret (Ardea alba) is now commonplace (Great white egret no longer rare bird as numbers boom across UK and europe (2021)), Cattle Egrets (Bubuculus ibis) and Spoonbills (Platalea… have established breeding colonies and Glossy Ibis is being seen in increasing numbers.
Conclusion
Like the European Bee-eater (Merops ,) which is considered to be “an iconic species of climate change”, ….