Fig. 2 Comparison of the Community Temperature Index (CTI) of the paired dataset (Δ CTI, 3,839 pairs) for the different calibrated models (panels a, b and c). In the paired dataset, co-occurrence plots were matched with presence-only by subsampling the latter and selecting the closest available pseudo-plot that had been sampled. X-axis: Mean annual temperature (MAT) of the average of the five years preceding the respective assemblage survey of the co-occurrence dataset. Y-axis: Difference between the estimated CTI for the paired dataset (Δ CTI). A LOESS regression line is shown in each panel. Panel d: Temporal thermophilization trend. Thermophilization is estimated using the projection of the combined calibrated model using the paired dataset. Blue solid line: co-occurrence dataset, green dashed line: presence-only dataset.

Ad-hoc analyses of the CTI deviations

Across all transfer functions, there was a tendency for CTI overestimation in the presence-only data, particularly at lower temperatures (Fig. 2 and S7 ). This overestimation remained within the credibility interval when utilizing the transfer function based on the combined dataset, but deviations from the credibility interval were noted when using the transfer functions derived solely from either co-occurrence data or presence-only data (Fig 2. and S7 ). Because the areas of colder temperatures tend to be placed in areas of higher topographical reliefs in Norway, we suspected that these deviations were due to elevational biases in the presence only data towards more accessible and warmer valley bottoms compared to the co-occurrence plots. To evaluate this, we assessed elevation differences between the co-occurrence plots and their correspondingly paired presence-only pseudo-plots. Elevations were extracted from DEM (Digital Elevation Model) with 25 m resolution (Copernicus, 2016) using original coordinates of the co-occurrence data and coordinates provided for the presence-only data. Subsequently, we computed the average elevation of all observations of all species occurring in each co-occurrence plot and pseudo-plot. When multiple observations existed for a single species within a pseudo-plot, we randomly chose one record per species. Our findings indicate a general tendency that observations in the presence-only pseudo-plots were situated at lower elevations compared to the co-occurrence plots with a mean difference of -35.3 meters (std. error of the mean = 1.41, Fig. S8 ). Furthermore, it is evident that larger disparities in elevations correspond to greater variations in CTI values. The observed trend was quantified using linear models, yielding a slope of -1.36 × 10-4 °C/m for the co-occurrence calibration model (std. error =2 × 10-4, p = 0.645) and -1.07 × 10-3 °C/m for the presence-only calibration model (std. error = 4 × 10-4, p = 0.005), with a slope of -1.07 × 10-3 °C/m (std. error = 3 × 10-4, p = 0.003) the combined model (Fig.3 ).