Results

Comparison of predicted values for co-occurrence and presence-only data

The average Δ CTI in the paired dataset (difference between pseudo and co-occurrence assemblages ) for the study period (1905-2007) was low independent of which calibration model was used, but it was statistically significantly different when calibrating the transfer function with the co-occurrence data only (Fig.2a-c; paired t-test average difference = 0.062, std.error 0.016, t = 3.7068, df = 3,838, p-value <= 0.001). The model calibrated with presence-only data showed an average difference of 0.029 (std.error 0.021, t = 1.35, df = 3,838, p-value = 0.17), and the model calibrated with the combined dataset showed an average difference of 0.014 (std.error 0.021, t = 0.68, df = 3,838, p-value = 0.49). In all cases the presence-only CTI were slightly higher than those of the co-occurrence dataset.
Since the transfer function calibrated with the combined dataset displayed the lowest difference in paired CTI (Fig. 2c ), we relied on this calibration model as the basis for illustrating the temporal pattern of thermophilization for both datasets (Fig. 2d ). This revealed that there was no statistically significant difference between thermophilization trends base on the co-occurrence and presence-only datasets, and that the thermophilization values increased steadily around the same time (Fig. 2d ).