1 Method

1.0.1 Data sources and cleaning

I obtained species counts for ‘true’ tits (Great Tit, Willow Tit, Blue Tit and Marsh Tit) and Great Spotted Woodpecker via the National Biodiversity Atlas (NBN) plug in for QGIS.(The FSC Plugin for QGIS v3 (2016)) The NBN Atlas is a major source of species observation data for British species and draws on numerous primary sources - for example Willow Tit observations are drawn from more than 60 datasets.

I extracted data by decade and appended records to give a complete dataset covering the period 1970- 2019 (there is a download limit of 1 million records). This includes the period reported major decline in Willow Tit numbers.

I excluded records with missing dates, geographical information (lat-long) or species information and filtered the dataset to include only those observations classified as ‘accepted’. After de-duplication this gave a dataset with 2.13 million records with records for each Tit species and Great Spotted Woodpecker.

I calculated species counts per year, per year and decade to control for collection effects () and by year and osgr_100m grid to examine spatial effects. I estimated to I also obtained data from the British Breeding Bird Survey to look at losses and gains in Willow Tit distributions. I modelled Willow Tit counts as a function of time (year), counts of other species (Blue and Great Tit, and Great Spotted Woodpecker), stratified by 100km grid squares using Poisson Generalised Additive Modelling (GAM).()

I calculated Willow Tit range using kernel density estimation for each decade.

I also calcuated a species distribution model incorporating climate variables, landcover data for 1990, 2000, and 2015 and competitor and predator species. I calculated 2019 counts for Blue Tit and GReat Spotted Woodplecker per 10km2 and converted these data to a raster whcih I combined with climate data and landcover data rasters. The procedure for alging extent, resolution and CRS of these rasters is outlined in Annex 1. I obtained climate adn elevetion data using teh getData function in the raster R package - this accesse worldclim and strm data. (EXPAND AND CLARIFY). I converted xxx Willow Titi point data to a spatial data format required to teh sdm R packag which I used to calculate the model.

The sdm package takes spatial points as input, a raster stack as predictors, adn calculates background poijnt to convert from “presence-only” to “presence-background” - it randomly samples points from the background (CHECK) excluding known pints. (CLUMSY)

This data object is then passed to the modelleing funtion. sdm allows multiple runs of range of algorithms through bootstraping and/or cross-validation. For this project I ran maxent, glmnet, ranger, brt models with n-fld cross validation using a 70:30 test-train split.

I assessed model accuracy with AUC and ROC curves and extracted variable importance to assess whchi predictors impacted on Willow Tit distributions. I then predicted teh Willow Tit distribyion to cerate a range of maps of habitat suitability.

Figure 1.1: Willow Tit counts

Figure 1.2: Annual species counts

1.1 Annual counts

These are shown in Figure 1.1 and Figure 1.2.

There are annual patterns in observation counts across all species with a large peak in 2001 and fall in 2002,

Figure 1.3: Gains and losses in Willow Tit distribution

FIgure 1.3 shows the extent of losses and gains … Breeding has been lost from much of Eastern and Southher Egland with amall gains in the Sourh Wst and North East.

1.1.1 Models

1.1.1.1 Predicting willow tit counts

Counts by year per million hectares 100k * 100k. Figure 1.4

## # A tibble: 5 x 5
##   term                                   edf ref.df statistic p.value
##   <chr>                                <dbl>  <dbl>     <dbl>   <dbl>
## 1 s(year)                               6.87   7.00      550.       0
## 2 s(blue_tit)                           8.67   8.91      234.       0
## 3 s(great_tit)                          8.44   8.73      496.       0
## 4 s(great_spotted_woodpecker)           8.86   8.95      591.       0
## 5 s(blue_tit,great_spotted_woodpecker) 26.2   27        1450.       0

Figure 1.4: GAM output

Partial dependence plots - contribution of each predictor - per 100 square km
- Year: Shows large decline 1970 - 90 - then recovery but more recent decline
- Blue tits: Numbers increase to between 2-3000 then wt counts decline
- Similar for gsw
- But not Great Tit - predictor of presence.

R Notebook

Julian Flowers

2021-12-05

1 Method

1.0.1 Data sources and cleaning

1.1 Annual counts

1.1.1 Models

1.1.1.1 Predicting willow tit counts

References