LB 9-81 - Modelling curlew sightings based on habitat preferences and climatic variables on both sides of the Irish Sea

Background/Question/Methods:

ECHOES is an inter-regional project looking at the effect of climate change on bird habitats around the Irish Sea. One question is how habitat preferences of curlew (Numenius arquata) vary seasonally across the whole of Britain and Ireland. We used CORINE satellite imaging (2006, 2012 and 2018 data) at 1km x 1km resolution to establish the proportion of each land cover class in each grid square.



For 200 months (January 2003 to August 2019), curlew observation records, from both systematic studies and citizen science data are available (from both Britain and Ireland). The response variable is binary – did at least one curlew sighting take place in a particular grid square in a given month?



In addition to the CORINE land cover variables, we used five monthly climatic variables (including frost days, dry days and total rainfall) and variables derived from digital terrain models.



We used penalised generalized linear models (with lasso regression) to reduce the number of land cover class explanatory variables in the models. Ten-fold cross-validation and an established method for selecting the penalty term led to a substantial reduction from the initial 37 land class variables to between 3 and 12 depending on the time of year.

Results/Conclusions:

The distribution of curlew sightings across Britain and Ireland varies considerably throughout the year, with predominantly coastal distributions in the winter months and a dispersion that includes the interior of the islands in summer months. Our principle focus is overwintering curlew. Two winter months offer a brief overview of our findings.



In November 2006, the land cover classes positively related to log-odds of a curlew sighting are estuaries, intertidal flats, salt marshes, beaches, dunes and sands and port areas. Frost days have a strong, negative impact on the log-odds of a sighting. In January 2007, frost days again negatively impact the log-odds of curlew sightings and estuaries, intertidal flats and salt marshes have positive impact. In both cases, the predictive probability map of the model shows broad agreement with the observed distribution of sightings.



A detailed analysis will enable us to build a good understanding of where, when and under what environmental conditions curlew are likely to be seen, as well as illustrating how habitat preferences change through the cycle of the year. The next step is to introduce future climate predictions to assess how climate change will impact the distribution of curlew around the Irish Sea in coming decades.