Climatologies at high resolution for the earth's land surface areas

Abstract

High-resolution information on climatic conditions is essential to many applications in environmental and ecological sciences. Here we present the CHELSA (Climatologies at high resolution for the earth's land surface areas) data of downscaled model output temperature and precipitation estimates of the ERA-Interim climatic reanalysis to a high resolution of 30 arc sec. The temperature algorithm is based on statistical downscaling of atmospheric temperatures. The precipitation algorithm incorporates orographic predictors including wind fields, valley exposition, and boundary layer height, with a subsequent bias correction. The resulting data consist of a monthly temperature and precipitation climatology for the years 1979-2013. We compare the data derived from the CHELSA algorithm with other standard gridded products and station data from the Global Historical Climate Network. We compare the performance of the new climatologies in species distribution modelling and show that we can increase the accuracy of species range predictions. We further show that CHELSA climatological data has a similar accuracy as other products for temperature, but that its predictions of precipitation patterns are better.

Figure 1. Workflow of the CHELSA model output statistics and interpolation algorithm for precipitation data.

Resulting raster datasets (parallelograms) from each calculation step (arrows) are shown for each step of the algorithm. Predictor variables are indicated in yellow, raster datasets of the dependent variable (precipitation) are indicated in blue, and bias correction raster datasets are indicated in green. The final climatological product is indicated in orange.

Figure 2. Small scale comparison of model fit with station data from GHCN^, for annual precipitation sums derived from six different models.

Models have been calculated for each station separately, including a minimum of 16 surrounding stations in a 2° search radius. This illustrates how well modelled precipitation corresponds to measured stations at the scale <=2°. The figure shows the mean R² value for 2° grid cells. The upper row additionally illustrates different steps of the CHELSA algorithm, with ERA-Interim performing worse than GPCC, and CHELSA showing the highest fit after including orographic effects.

Figure 3. Temporal precision of the CHELSA reanalysis which forms the basis for the climatologies in a small region of South Africa.

Precipitation anomalies are shown as deviations from the mean precipitation in the respective time period. Grey=station data, red=CHELSA, green=ERA-Interim, blue=CRU, orange=CHIRPS.

Figure 4. Bias ratio comparison of annual precipitation sums for six different climatologies with the CRU climatology at the global scale.

All models have a substantial dry bias over Greenland when compared to CRU (TRMM/TMPA (3B43) does not include Greenland). Large differences in bias ratios can be observed in the Atacama, where CHELSA, and GPCC are drier than CRU, and WorldClim, CHPclim, TRMM/TMPA (3B43), and ERA-Interim are wetter. Also on the Himalaya plateau large differences are visible, with CHELSA, GPCC, and TRMM (3B43) being drier, and WorldClim and ERA-Interim being wetter.

Figure 5. Comparison of precipitation patterns in the complex terrain of Bhutan (country boundaries in black) between TRMM/TMPA (3B43), WorldClim, CHELSA, the statistical downscaling approach of Böhner, the topography from GMTED2010, and the cloud cover climatology from Wilson & Jetz.

In this region, most precipitation falls during the SW-monsoon in the northern summer, when wet air masses from the SW are lifted at the south face of the Himalayas and dry until reaching the Tibetan high plateau. While the mesoscale patterns are in congruence between models, there are clear differences at the microscale. WorldClim predicts wet valleys and dry mountain faces, whereas CHELSA and Böhner predict dry valleys and wet windward exposed mountain faces due to the inclusion of orographic predictors. CHELSA and Böhner are also in closer congruence with the observed distribution of cloud in the area, which shows lower cloud cover in the isolated mountain valleys compared to the wind exposed mountain faces in the south.

Figure 6. Temporal comparison of the CHELSA algorithm with GHCN Version 3 (temperature), and MODIS (MOD11C3).

Coefficients of determination give the global correlation between products for a specific month. CHELSA temperatures show significantly higher correlations with GHCN (Wilcoxon Test: W=23,254, P<0.001). Correlations between CHELSA and MODIS for temperature (mean R²=0.99).

Figure 7. Comparison of the performance of 67 species distribution models using generalized linear models with WorldClim (blue) and CHELSA (red) precipitation and temperatures in Switzerland as predictors.

Models are compared using paired t tests on model performance statistics (adjusted D², the area under the receiver operation statistics curve (AUC), Kappa statistics, and true skill statistics (TSS)). All performance statistics indicate a higher mean performance of CHELSA over WorldClim, with only the TSS statistics not being significant. Red lines indicate models in which SDMs based on CHELSA performed better then SDMs based on WorldClim, blue indicates the opposite relationship. T-values are given for the paired t-test comparing WorldClim with CHELSA.

Figure 8. Comparison of species distribution models based and climate data from WorldClim and CHELSA of Astragalus monspessulanus for Switzerland.

Models were calculated using generalized linear models with mean annual precipitation from WorldClim (a) and CHELSA (c), as well as the mean annual temperature. The distributions based on WorldClim (b) and CHELSA (d) represent a binary distribution (red) with a threshold of the maximum kappa (WorldClim: threshold=0.675, AUC=0.89, D²=0.36, Kappa=0.60, TSS=0.66, CHELSA: threshold=0.7155, AUC=0.91, D²=0.47, Kappa=0.66, TSS=0.76). Occurrences are marked in green. The circle indicates the area of the dry Rhine valley in eastern Switzerland in which WorldClim overestimates precipitation and therefore does not predict the range of the species correctly in that region (compare with d).