Background & Summary

Current ecological research deals with the subject of analysing climate-induced changes in species distribution. In recent years the research interest concerning various ecological topics greatly increased1,2,3. Common knowledge is that the distribution of vegetation highly depends on the climate conditions4,5. To quantify this climate induced forcing, direct gradients in the form of bioclimatic variables are widely used as predictors as they have physiological relevance to species6. A vast amount of global bioclimatic variables are already available with various spatial and temporal resolutions, based on different source data. Data from historic periods are often generated using direct observations such as station data in conjunction with indirect observations, for instance remote sensing data (WorldClim7,8) or based on remote sensing data alone (MERRAclim9). Others were derived from the before mentioned WorldClim in combination with CRU10 data (CliMond11,12) or are either based on ERA-Interim13 in combination with station data (CHELSA14) or ERA5 reanalysis data15 (BIOCLIMATE16,17). A very different approach is used to derive past (Holocene), present and future projections (ecoClimate18) through simulation data from CMIP5 (Coupled Modeling Intercomparison Projects)19 and PMIP3 (Paleoclimate Modeling Intercomparison Projects)20. Similarly, simulation data from GCMs (Global Circulation Models) are utilised to derive past conditions for the late Quaternary21 (last 120,000 years).

Nearly all data sources mentioned provide historic data but also future projections (WorldClim, CliMond, BIOCLIMATE, ecoClimate). However, solely ecoClimate uses identical source data for all time periods. Especially WoldClim data7,8 can be considered as standard predictor variables for species distribution models. The data basis for WoldClim’s future and historic periods are different. For the historic period 1970 to 2000 it is gridded station data in conjunction with remote sensing products. For the future projections it is CMIP622 GCMs (General Circulation Models or Global Circulation Models). This can introduce a climate model bias, which can be minimised through bias-adjustment in order to correct the statistics of the projections during a historic time period. Nevertheless, the provided data of the historic time periods differ in methodology from the projection data. This introduces additional bias as both datasets exhibit a different characteristic which will be passed on to following applications. The CMIP6 GCMs carry a spatial resolution of 20 km to 260 km23 depending on the model used. On the other hand, CMIP5 simulations with RCM for Europe in the framework of EURO-CORDEX24,25,26 offer a resolution of ~12.5 km and are to date regional climate projections with the highest resolution in large ensembles.

We present 26 bioclimatic variables with an European extent for the historic 1971 to 2000 and the projection period 2021 to 2098, both based on regionalised CMIP5 simulation data from EURO-CORDEX and ReKlies-DE27 from within the Helmholtz-Initiative Climate Adaptation and Mitigation (HI-CAM) initiative28. The input climate variables are bias-adjusted using E-OBS data (version v20.0e)29 and spatially disaggregated. The extracted historic bioclimatic data are validated against equivalent E-OBS based variables. Our variables allow training and projection on the same simulation data and thus same characteristics, which lessens the bias for further applications such as species distribution modelling.

Methods

Data source

Regional Climate simulations serve as input for the calculation of bioclimatic variables. These climate simulations were bias-corrected and disaggregated within the Helmholtz-Initiative (HI-CAM)28 and consist of regionalised CMIP5 simulations from EURO-CORDEX and ReKlies-DE27. They encompass data based on 11 Regional Climate Model (RCM) downscaling simulations as output from 10 different Global Climate Models (GCM). These simulations consist of 88 realisations that can be assigned to 3 different Representative Concentration Pathways (RCP) 2.6, 4.5 and 8.5 defined by Intergovernmental Panel on Climate Change (IPCC)30. Realisations with GCM CNRM-CM5 and/or RCM HIRHAM5 involved were omitted according to errata31 and (https://www.euro-cordex.net/078730/index.php.en, last access: 20th June 2024) at the time of simulation creation. This leads to omitting 18 realisations and therefore leading to 70 valid realisations, (final realisations listed under Table 1). Thereof 18 realisations are counted as RCP 2.6, 13 realisations as RCP 4.5 and 39 realisations as RCP 8.5.

Table 1 Overview of all considered realisations within HI-CAM28 EURO-CORDEX24,25,26 CMIP519 project (Information and data can be retrieved from https://aims2.llnl.gov/search/cmip5/https://aims2.llnl.gov/search/cmip5/).
Full size table

The simulation data have a spatial resolution of 0.11° (12.5 km) and are bias-adjusted using the trend-preserving ISIMIP3BASD methodology32. An ensemble version (v20.0e)29 of gridded E-OBS data33 with a different gridding method and improved uncertainty estimate served as reference data for the bias-adjustment. During bias-adjustment the realisation data (HI-CAM) are aligned with the statistics of the reference data. The time period considered here is 1971 to 2000. They include daily minimum, maximum and average temperature as well as daily precipitation. The E-OBS data33 were derived based on climate stations across Europe and tested thoroughly for their quality29,34. The underlying climate station data vary in their temporal and spatial coverage across Europe and carry some inhomogeneities. Consequently, the gridded E-OBS data also exhibit inhomogeneities34. E-OBS data users of any version have to be aware of these limitations. The ensemble version v20.0e was remapped from their original resolution 0.25° (25 km) to 0.11° using a conservative remapping method to fit the realisation data (HI-CAM) resolution.

The bias-adjusted realisation data are then spatially disaggregated from 0.11° to 0.03125° (3 km) using External Drift Kriging (v2.0.0)35 with elevation36 as external drift factor. Realisation variables with daily temporal resolution considered for the bioclimatic calculation are as follows: daily average air temperature (tas) [°C]37, maximum air temperature (tasmax) [°C]38, minimum air temperature (tasmin) [°C]39 and precipitation (pr) [mm d−1]40. UFZ mHM-Hydrological model output41,42,43 daily potential evapotranspiration (pet) [mm d−1]44 and weekly soil water content for 6 different soil layers (swc) [mm] serve as additional input. The realisation data and mHM model output are available in the WGS84 reference frame (EPSG: 4326) with a spatial resolution of 0.03125° for the extent of the European continent excluding Turkey, Iceland and Russia (lat: 35° lon: −11°, lat: 71.5° lon: 41°) and temporal range from 1971 to 2098, Fig. 1. We chose to use biogeographical regions of Europe (EBR)45 to facilitate the presentation of bioclimatic variables in the following.

Fig. 1
figure 1

HI-CAM data availability for Europe. Dividing the area in eight biogeographical regions of Europe (EBR)45 facilitates the presentation of bioclimatic variables in selected subsequent figures.

Full size image

Calculation of bioclimatic variables

Figure 2 illustrates the process of bioclimatic variable computation using capital letters within the following subsections. A total of 26 bioclimatic variables were selected following these of ANUCLIM46 and WorldClim variables7,8 as well as literature review, see Table 2. The variables tas, tasmax, tasmin, pr, pet and swc serve as input to the calculation. For swc the sum over all soil layers per time step per realisation was calculated beforehand. We utilised anuclim and atmos indicators from Python package xclim47 as well as algorithms of our own for the calculation. We calculated the bioclimatic variables for each realisation with a temporal sampling period of 1 year for the time range 1971 to 2000 serving as historic and 2021 to 2098 serving as projection time period. This accounts for 26 bioclimatic variables for 70 realisations [A]. The following descriptions apply pixel-wise to each bioclimatic variable.

Fig. 2
figure 2

The process of bioclimatic variable calculation for historical 1971 to 2000 and projection period 2021 to 2098 outlined by a flowchart. Capital letters indicate the steps and are referred to in the following descriptions. The total of 70 HI-CAM28 EURO-CORDEX24,25,26 CMIP519 climate realisations serve as input [A]. We calculated the historical period based on the median of all realisations [B] and validated them against E-OBS version (v20.0e)29 [C]. In addition, the realisations can be assigned to three different RCPs (Representative Concentration Pathways) − 2.6, 4.5, 8.5 [D]. Here, we extracted the 5th, 50th and 95th percentile within each RCP [E] to account for the variance within. From all periods (historic and projection) we have then extracted the temporal median [F].

Full size image
Table 2 26 calculated bioclimatic variables with short and full bioclimatic variable names as well as units the variables are provided in.
Full size table

Calculation of ensemble statistics

The time range 1971 to 2000 serves as historic time period due to data availability not before 1971. Over the course of the period all realisations run quite similar [B]. We extracted the temporal 5th, 50th and 95th percentile per realisation to receive three values representing the time period [F]. This serves as input to our data validation [C].

For the projection time period, we selected 4 time periods with a similar range to the WorldClim8 data, which are common predictor variables in species distribution modelling studies. The periods are as follows 2021 to 2040, 2041 to 2060, 2061 to 2080 and 2079 to 2098 due to data availability until 2098 and to keep the length of 20 years consistent. Here we extracted two versions for the bioclimatic variables.

  1. (1)

    As we can assign each realisation to an RCP [D], Table 1, we extracted the 50th but also 5th and 95th percentile from all realisations belonging to the same RCP [E] to obtain the variance within. The annual 50th percentile development of selected variables per EBR is shown in Figs. 3, 4, 5 and 6. In addition, Figs. 7, 8, 9 and 10 reveal the annual course of the 5th, 50th and 95th percentile for the continental EBR as an example. Subsequently, we extracted the temporal 5th, 50th and 95th percentile for each RCP and period similar to the historic data [F]. This results in 4 periods and 3 RCPs each accompanied by 9 percentiles (within RCP and temporal 5th, 50th, 95th).

    Fig. 3
    figure 3

    Annual course of BioVar1 (Mean Temperature) 50th percentile for eight EBRs45 during the period 2020 to 2098. Here the mean (black line) and standard deviation (grey background) is extracted per EBR45. Subfigures show annual mean temperature for RCP 2.6 (a), 4.5 (b) and 8.5 (c).

    Full size image
    Fig. 4
    figure 4

    Annual course of BioVar12 (Precipitation Sum) 50th percentile for eight EBRs45 during the period 2020 to 2098. Here the mean (black line) and standard deviation (grey background) is extracted per EBR45. Subfigures show BioVar12 for RCP 2.6 (a), 4.5 (b) and 8.5 (c).

    Full size image
    Fig. 5
    figure 5

    Annual course of BioVar13 (Index of Aridity) 50th percentile for eight EBRs45 during the period 2020 to 2098. Here the mean (black line) and standard deviation (grey background) is extracted per EBR45. Subfigures show BioVar13 for RCP 2.6 (a), 4.5 (b) and 8.5 (c).

    Full size image
    Fig. 6
    figure 6

    Annual course of BioVar20 (Last Spring Frost) 50th percentile for eight EBRs45 during the period 2020 to 2098. Here the mean (black line) and standard deviation (grey background) is extracted per EBR45. Subfigures show BioVar20 for RCP 2.6 (a), 4.5 (b) and 8.5 (c).

    Full size image
    Fig. 7
    figure 7

    BioVar1 (Mean Temperature) for all RCPs (left), RCP 2.6, RCP 4.5 and RCP 8.5. Figures of single RCPs include the 5th, 50th an 95th percentiles. For all subfigures the mean (thick line) and standard deviation (shaded background) for continental EBR are extracted.

    Full size image
    Fig. 8
    figure 8

    BioVar12 (Precipitation Sum) for all RCPs (left), RCP 2.6, RCP 4.5 and RCP 8.5. Figures of single RCPs include the 5th, 50th an 95th percentiles. For all subfigures the mean (thick line) and standard deviation (shaded background) for continental EBR are extracted.

    Full size image
    Fig. 9
    figure 9

    BioVar13 (Index of Aridity) for all RCPs (left), RCP 2.6, RCP 4.5 and RCP 8.5. Figures of single RCPs include the 5th, 50th an 95th percentiles. For all subfigures the mean (thick line) and standard deviation (shaded background) for continental EBR are extracted.

    Full size image
    Fig. 10
    figure 10

    BioVar20 (Last Spring Frost) for all RCPs (left), RCP 2.6, RCP 4.5 and RCP 8.5. Figures of single RCPs include the 5th, 50th an 95th percentiles. For all subfigures the mean (thick line) and standard deviation (shaded background) for continental EBR are extracted.

    Full size image
  2. (2)

    In addition, we have extracted the temporal 5th, 50th and 95th percentile for each period (realisation and projection). This results in 4 periods and 3 percentiles, for all 70 realisations and allows the user to chose only realisations of interest for further analysis. An example of the temporal course of BioVar1 (Mean Temperature) for the realisation (met) 46 is included in Fig. 11.

    Fig. 11
    figure 11

    Temporal course over all 5 time steps (historic and projections) of BioVar1 - Mean Temperature - for realisation (met) 46, which is defined within RCP 4.5. Subfigure (a) shows the 5th, (b) the 50th and (c) the 95th temporal percentile per time step. The blue line within the legend indicates the areal mean.

    Full size image

All described calculations were performed using Python package xarray48. We re-projected the final data products to ETRS89 Lambert Azimuthal Equal Area (EPSG: 3035) with resampling method nearest neighbour using rioxarray49 and set the spatial resolution to 3 km.

Data Records

The dataset can be obtained from a Zenodo repository50 which is comprised of ten tar.gz files.

Each file inside the tar.gz contains all 26 bioclimatic variables with order and names according to Table 2 column Variable name. The filenames for the realisations are described as follows:

bioVars_1971–2000_met.tar.gz → Projections per realisations for period 1971–2000

bioVars_2021–2040_met.tar.gz → Projections per realisations for period 2021–2040

bioVars_2041–2060_met.tar.gz → Projections per realisations for period 2041–2060

bioVars_2061–2080_met.tar.gz → Projections per realisations for period 2061–2080

bioVars_2079–2098_met.tar.gz → Projections per realisations for period 2079–2098

Regarding the reference and projection periods the filenames inside the tar.gz files are

biovars_met01_1971–2000_5.tif

with 01 as realisation number, 1971–2000 as period and 5 as temporal percentile within the period.

For bioVars separated into RCPs the filenames are described as follows:

bioVars_2021–2040_rcp.tar.gz → Projections per RCP for period 2021–2040

bioVars_2041–2060_rcp.tar.gz → Projections per RCP for period 2041–2060

bioVars_2061–2080_rcp.tar.gz → Projections per RCP for period 2061–2080

bioVars_2079–2098_rcp.tar.gz → Projections per RCP for period 2079–2098

Regarding the RCP for the projection periods the filenames inside the tar.gz files are

biovars_rcp2.6_50perc_2061–2080_5.tif

with 2.6 as RCP, 50 as percentile within the RCP, 2061–2080 as period and 5 as temporal percentile within the period.

In addition, the validation results are part of the Zenodo repository50. The folders eobs and worldclim each contain scatter plots of all possible validation variables (folder plots) as well as statistics (folder stats) for R2 and RMSE.

Technical Validation

We applied the bioclimatic calculation including the temporal 5th, 50th and 95th percentile on the E-OBS data to compare them with our historical bioclimatic variables based on HI-CAM data. Similarly, these E-OBS bioclimatic variables were reprojected to ETRS89 Lambert Azimuthal Equal Area (EPSG: 3035) using nearest neighbour as resampling method and the spatial resolution was set to 3 km. As the variables potential evaporation and soil water content are not available from E-OBS, we omitted bioclimatic variables that rely on these variables, which are Mean Soil Water Content and Water Deficit. We compared our historic variables with E-OBS variables based on the period percentiles resolution (selection of scatter plots, Figs. 12, 13). A quantitative analysis was run on the period percentiles resolution. In this quantitative analysis, we randomly sampled 10 pixel values per entity which is variable, temporal percentile and EBR from E-OBS as well as variable, realisation, temporal percentile and EBR from HI-CAM bioclimatic variables. For each of these entities we calculated the R2 and RMSE, see tables in Zenodo repository50. This process was iterated 20 times and statistics were averaged among them. We see highest R2 values for bioclimatic variables without precipitation involved (e.g. Mean Temperature - BioVar1, Diurnal Temperature Range - BioVar2, Mean Temperature of the Warmest Quarter - BioVar10, Mean Temperature of the Coldest Quarter - BioVar11 and Growing Degree Days above 5 °C - BioVar21 - please see Zenodo repository50 for these plots). Two rather extreme event bioclimatic variables Dry Days - BioVar23 and Max Consecutive Dry Days - BioVar24 - please see Zenodo repository50 for these plots) show lowest agreement across all EBRs. The three smallest EBRs (Arctic, Pannonian and Steppic) show lower agreement with E-OBS variables than other EBRs. The origin of this lower agreement might evolve from lower station density in these areas (Spain, the Balkans and Northern Scandinavia) that were used for generating the E-OBS data34. The lower density in Spain, which is largely part of EBR Mediterranean, affects the agreement less as it does in the low station density EBRs that are much smaller (Pannonian and Steppic in the Balkans as well as Arctic in Northern Scandinavia).

Fig. 12
figure 12

Scatterplots of E-OBS version (v20.0e)29 vs. historical bioclimatic BioVar13 (Index of Aridity) for the time period 1971 to 2000 and each realisation separately. The comparison was iterated 20 times for 10 randomly sampled pixel per EBR45. These plots include 5 randomly sampled pixel per realisation from within the comparison data pool. Subplots indicate the temporal (a) 5th, (b) 50th and (c) 95th percentile.

Full size image
Fig. 13
figure 13

Scatterplots of E-OBS version (v20.0e)29 vs. historical bioclimatic BioVar20 (Last Spring Frost) for the time period 1971 to 2000 and each realisation separately. The comparison was iterated 20 times for 10 randomly sampled pixel per EBR45. These plots include 5 randomly sampled pixel per realisation from within the comparison data pool. Subplots indicate the temporal (a) 5th, (b) 50th and (c) 95th percentile.

Full size image

As the bias-correction of the HICAM variables was done using the E-OBS v20.029 data, we are confident, that a change to the newest E-OBS v30.029 data for the validation might not improve the data to a significant amount. E-OBS v20.0 data were used in recent publications and proofs to be a well accepted standard (among other51,52). The newest E-OBS version (v30.0)29 offers a longer time-frame, more variables at higher spatial resolution as well as an enlarged station density in comparison to E-OBS v20.029,53. Merely, the enlarged station density might improve our bioclimatic variables from the process of bias-correction. However, our study was designed and conducted using E-OBS v20.029 for bias-correction, ensuring consistency in data analysis, validation and results interpretation. Changing to E-OBS v30.029 might introduce discrepancies in results due to changes in data handling, resolution or underlying corrections.

To rule out lower qualitative biolimatic variables due to lower station density in E-OBS v20.029 and overoptimistic validation using the same data for bias-correction as well as validation, we enlarged the validation using  WorldClim data version 2.18. We utilised the data with 2.5 minutes spatial resolution, re-projected and resampled these data to EPSG:3035 and 3 km to match our bioclimatic variables. As WorldClim8 offers the temporal mean per time period, we used our 50th temporal percentile data for validation. We were able to validate 17 of our 26 bioclimatic variables, that are available from WorldClim8 as well.

Similar to the validation using E-OBS29 data, we randomly sampled 20 pixel values per variable and EBR from WorldClim8 as well as variable, realisation and EBR from HI-CAM bioclimatic variables. The amount was increased to account for slightly shifted pixel locations between both inputs. Again, this process was iterated 20 times and statistics were averaged among them.

The results for BioVar1 (Mean Temperature) and BioVar4 (Temperature Seasonality), see Fig. 14, as well as BioVar12 (Precipitation Sum) and BioVar15 (Precipitation Seasonality), Fig. 15, remain very promising, see statistics in the Zenodo repository50. Some variables from WorldClim version 2.18 show a slightly different methodology of extraction (Temperature Seasonality – BioVar4, Diurnal Temperature Range - BioVar2), which results in higher RMSE values, see plots and statistics in the Zenodo repository50. Partly unfavourable R2 values (BioVar2) e.g. for Boreal EBR might evolve from the fact, that WorldClim8 variables are averaged for the time period 1970–2000 whereas for BioVars the 50th percentile is utilised for the time period 1971–2000. Especially for extreme variables with extreme seasonal climatic differences (e.g. EBR Boreal) this can lead to unfavourable R2 results.

Fig. 14
figure 14

Scatterplots of WorldClim version (2.1)8 vs. historical bioclimatic variables for the time period 1971 to 2000 for each realisation separately. The comparison was iterated 20 times for 20 randomly sampled pixel per EBR45. These plots include 5 randomly sampled pixel per realisation from within the comparison data pool. Subplots indicate the (a) BioVar1 (Mean Temperature) and (b) BioVar4 (Temperature Seasonality). The axes limits vary in result of different calculation of WorldClim and historical Temperature Seasonality.

Full size image
Fig. 15
figure 15

Scatterplots of WorldClim version (2.1)8 vs. historical bioclimatic variables for the time period 1971 to 2000 for each realisation separately. The comparison was iterated 20 times for 20 randomly sampled pixel per EBR45. These plots include 5 randomly sampled pixel per realisation from within the comparison data pool. Subplots indicate the (a) BioVar12 (Precipitation Sum) and (b) BioVar15 (Precipitation Seasonality). The axes limits vary in result of different calculation of WorldClim and historical Precipitation Seasonality.

Full size image

Usage Notes

The bioclimatic data are provided here in three parts. (1) The historic time period, which is derived for the time period 1971 to 2000 per realisation. Whereas for the projection period (2) the realisations are then separated into three RCPs and combined within. Here the combining encompasses the 5th, 50th, 95th percentile which allows to account for the variance among the realisations within the RCP. If only specific realisations are to be taken into account (3) we provide all realisations separately. All three parts are supplied as the temporal 5th, 50th, 95th percentile for predefined periods as described in the Method section. We are encouraged to define similar temporal periods (historic as well as projection time periods) as in WorldClim8. Providing the historical and the projection periods derived from the same simulation data enables the usage of these bioclimatic variables with low shifting bias among. Use cases for these variables might be species distribution modelling. The variables from the historic period act as predictor variables during the model training process and the variables from the projection periods (RCP or singular realisations) serve as predictor variables for the model projection. A principal component analysis (PCA) or a correlation analysis of all variables within the historic period reveals a high correlation of multiple variables and should be considered for further use of the data. Despite the utilisation of E-OBS v20.029 (released October 2019) instead of E-OBS v30.029 (released September 2024) for bias-correction and validation, we are confident about the quality of our bioclimatic variables. Nevertheless, areas of more sparse station coverage between both E-OBS versions might introduce some slight uncertainties in our bioclimatic variables in those regions.