Climate, topography and productivity data
We extracted different sets of current climate data for two purposes: (i) to test the prediction that climate variability and drought would influence the use of shelters by arthropods at the experiment scale, and (ii) to predict the effects of future climate variability on refuge usage (Fig. 1). Climate data and topographic data were extracted from WorldClim (Fick and Hijmans, 2017) version 2 (http://www.worldclim.org/) and ENVIREM (Title and Bemmels, 2017) (https://envirem.github.io/). For each site we chose four variables for temperature (bio1, bio2, bio4, bio7), four variables for precipitation (bio12, bio14, bio15, Aridity), two variables related to topography (TRI and topoWET) and one variable denoting site productivity (AnnualPET). A detailed description of the variables is presented in the Table S2. WorldClim variables were extracted at 30 arc-second, 2.5 arc-minute and 10 arc-minute resolutions. Since the above variables were very strongly correlated among these different resolutions (Pearson correlation, r ≥ 0.97), we decided to use only data on a 10 arc-minute resolution. The variables TRI and topoWET from ENVIREM were available only for 30 arc-second resolutions. The variables Aridity and AnnualPET were extracted at 30 arc-second and 10 arc-minute resolutions; since these variables were strongly correlated between the two resolutions (Pearson correlation, r ≥ 0.98), we used only data on 30 arc-second resolutions.
Future climate database at local scale was extracted from WorldClim version 2 using MIROC5 (RCP8.5) and CCSM4 (RCP8.5) as representative concentration pathways of CO2 emission projected for 2070 (Romero et al ., 2018). Future bioclimatic variables are not available in ENVIREM databases. Since the bioclimatic variables were very strongly correlated between these two predictive climatic models (Pearson correlation, r ≥ 0.98), we focused our analyses only on MIROC5 (RCP8.5) database.
We measured average local temperature during the 10-day experiment. For sites missing such data, we extracted the mean near-surface daily temperature from the RNCEP database (Kemp et al., 2012). This was done for all the 10 days of the experiment, from which data we then calculated the average over the experiment. The RNCEP database has a spatial resolution of 2.5 x 2.5o and a temporal resolution of 6 h (Kemp et al., 2012).