These data on bird species abundance and environmental variables were used in testing and comparing two different species distribution model validation methods that are applied to models which are used to predict the effects of climate change on species' distributions. The aim of the study was to investigate whether different validation methods give different results of the model's predictive performance and to demonstrate that validation methods based on measuring and validating a “static” pattern in distribution can assess model performance over-optimistically compared to methods based on measuring and validating a “change” in the distribution, which can assess the predictive performance more critically.

Bird data come from two types of surveys. Most data come from systematic national land bird monitoring surveys where volunteers survey either line transects (Finland) or a combination of line transects and point count stations (Sweden and Norway) once a year during the breeding season. Survey routes were 6 km (Finland and Norway) or 8 km (Sweden) in total length, each typically a square or rectangle, and covered the countries at approximately 20-40 km distance of each other. A smaller fraction of the data comes from more randomly located point count stations (Sweden and Norway) and line transects (Finland).

The total number of individuals (Sweden) or pairs (Norway and Finland) detected was recorded per species for each route. Surveys were carried out early morning and in good weather conditions. Birds were detected by both auditory and visual cues by trained observers using standardized field protocols. If a species is not detected when slowly walking along the transect line, it is reported as absent (its abundance is zero).

Not all study sites were visited each year. Records with insufficient accuracy or associated information were cleaned out. In total, the data consists of 2,591 bird survey sites (Norway 486, Sweden 1,197, Finland 908). We considered each visit to each survey site (line transect or point count survey) as one sampling unit (19,753 visits in total). In Finland, waterbirds were not monitored in surveys until 2006 onwards.

General information on the systematic national bird surveys and their detailed methods can be found online; Sweden: www.fageltaxering.lu.se, Norway: https://tov-e.nina.no/, Finland: www.luomus.fi/en/bird-monitoring.

Climate data come from the E-OBS weather data by the European Climate Assessment & Dataset project, ECA&D: https://www.ecad.eu/download/ensembles/download.php#datafiles. E-OBS gridded daily mean temperature data (version 15, 0.25 degree resolution) was used to calculate the mean temperatures of winter, spring, and summer periods for each study year. For each study site, the climate data was obtained by overlapping the study sites (line transect or point count routes) with the climate grid.

Land cover data comes from the Corine land cover data (100-meter resolution) by the European Environment Agency:  https://land.copernicus.eu/pan-european/corine-land-cover. To describe the habitat where birds were observed, we drew a 300-metre-wide buffer around each line transect and group of points in point count route and determined the land cover (44 classes) inside the buffer from the Corine land cover data. We then aggregated the land cover types into six classes and calculated their proportions (value range 0–1) on the buffer.

The original bird survey data set includes 63 species whose occurrence and location information are classified as “sensitive” according to Scandinavian laws and regulations, and thus cannot be openly shared as such. Therefore, in the openly shared data set the coordinates for the survey sites have been made coarser. The original, accurate data are available from the corresponding author upon reasonable request.

We acknowledge that in the earliest data from the 1970s to 1990s there is a maximum of 25 years of mismatch between land cover and bird data because land cover information was not available until the year 2000 onwards.

The shared R-code subsets the processed survey data and performs species distribution modeling and validation, and further analysis (it does not include processing of the variables). The R-code does not work directly with the openly shared 'observations_coarse.csv' where the true coordinates are made coarser and their corresponding variables are named x_coarse and y_coarse. The shared code works properly only with the original 'observation.csv' -file which can not be openly shared.

The bird observation data has been originally gathered from multiple excel files which are not shared here. Climate and land cover variables have been preprocessed using separate R-codes or QuantumGIS-program.