Abstract
High-resolution climate model projections for a range of emission scenarios are needed for designing regional and local adaptation strategies and planning in the context of climate change. To this end, the future climate simulations of global circulation models (GCMs) are the main sources of critical information. However, these simulations are not only coarse in resolution but also associated with biases and high uncertainty. To make the simulations useful for impact modeling at regional and local level, we utilized the bias correction constructed analogues with quantile mapping reordering (BCCAQ) statistical downscaling technique to produce a 10 km spatial resolution climate change projections database based on 16 CMIP6 GCMs under three emission scenarios (SSP2-4.5, SSP3-7.0, and SSP5-8.5). The downscaling strategy was evaluated using a perfect sibling approach and detailed results are presented by taking two contrasting (the worst and best performing models) GCMs as a showcase. The evaluation results demonstrate that the downscaling approach substantially reduced model biases and generated higher resolution daily data compared to the original GCM outputs.
Background & Summary
The unavailability of high-resolution climate data is an important obstacle for local climate change impact studies on agriculture and biodiversity. The analysis of climate change impact requires forcing data at sufficiently high-resolution for crop and hydrological models, both temporally (daily) and spatially (~1–10 km). Climate projection data produced by GCMs only exist at a coarse spatial resolution (typically a horizontal grid spacing >70 km) and with a high spread between models1. This variability (i.e., the range between models) induces uncertainties that propagate to calculated impacts (e.g., model outputs), which makes the studies less relevant to decision-making2. In the recent decade, this high uncertainty has been addressed using multi-model ensemble approaches which heavily lean on the use of multiple GCMs in crop modeling3,4,5,6 and hydrological modeling7,8,9. These studies require multiple state variables, in general from different emission scenarios, to be bias-corrected and downscaled to a higher temporal and spatial resolution than the original GCMs10,11,12,13. Downscaling is mainly necessary since GCMs do not resolve small-scale climate-affecting land surface features such as smaller mountain ranges. If GCM outputs are not appropriately corrected and downscaled, the process leads to highly erroneous results14.
For a country like Ethiopia with very diverse climate regimes modulated by its complex topography15,16, local planning, and monitoring are unimaginable without high-resolution climate data. In this regard, downscaling bridges the mismatch between coarse resolution climate outputs and data requirements by impact models17,18. Efforts have been made globally and regionally, including by WorldClim, CCAFS, and ref. 19, to create high-resolution climate projection datasets encompassing Ethiopia. For example, WorldClim (https://www.worldclim.org/data/cmip6/cmip6climate.html) has publicly released high-resolution CMIP6 projection datasets for 23 GCMs and four Shared Socio-economic Pathways (SSPs) at resolutions of 10 minutes, 5 minutes, 2.5 minutes, and 30 seconds. However, these datasets are limited in that they are only available as monthly values averaged over 20-year periods, which may not meet the demands of ecological impact models that require daily data. Additionally, the datasets are produced to provide a global quick overview of the projected climate, and the methods (https://worldclim.org/data/downscaling.html) used ignored the temporal trends and extremes and hence limiting its application in impact assessment studies. Similarly, CCAFS datasets (http://www.ccafs-climate.org), which use a simple bias correction technique called the Delta method, also have a limited temporal resolution, with only 30-year averages available. Ref. 19 have produced similar datasets for regions of East Africa (including Ethiopia), but their datasets are based on station data from 211 stations, and the models they used were limited to two GCMs. Despite these limitations, these datasets can still provide valuable insights into climate trends and support impact assessment studies, but alternative approaches may be necessary to obtain more comprehensive and robust data for specific research needs.
A variety of techniques and approaches exist to downscale coarse resolution GCMs output to a higher resolution. The most frequently used approaches in hydrological and agricultural studies are statistical and dynamical downscaling methods. The one difference between the two approaches lies in the required computational resources, especially when daily data are to be produced, but also in the accuracy of the results20. Regional Climate Models (RCMs) are used to dynamically downscale GCM outputs to a finer resolution by using the boundary and initial conditions from a GCM as input21. However, dynamical downscaling through RCMs is computationally very expensive21. This explains why high-resolution data from sets of GCMs under different emission scenarios are rarely produced using RCMs22,23.
Statistical downscaling is a technique used to generate high-resolution climate data by relating large-scale climate variables, such as temperature and precipitation, to local-scale variables2,24,25. Bias correction (BC) statistical downscaling (SD) is a popular technique used to produce high-resolution climate data from General Circulation Model (GCM) outputs, and they are often computationally efficient compared to other methods. The approach involves first applying a statistical bias correction to the GCM outputs using observations at the GCM grid and then using spatial downscaling to generate a more fine-resolution result14,26. Bias correction is a statistical technique used to adjust data that has a bias or systematic error so that it reflects the true values. The bias correction process can be accomplished through several methods, such as applying a simple change factor (i.e., ‘delta’) between the GCM output and observation2,27, or by using quantile-mapping techniques to match the GCM output to observations28,29,30,31. Once the bias correction is complete, the downscaled data can be generated using spatial downscaling techniques14.
The objective of this study was to produce a database of high-resolution bias-corrected climate change projections for Ethiopia. To this end, we bias-corrected daily maximum and minimum temperatures from 16 GCMs and daily precipitation from 12 GCMs and downscaled it to a 10 km spatial grid covering all of Ethiopia using the bias correction constructed analogues with quantile mapping reordering (BCCAQ) statistical downscaling technique. GCM outputs under three emission scenarios, also called Shared Socioeconomic Pathways (SSPs), were taken from the Coupled Model Intercomparison Project Phase 6 (CMIP6). As reference data, we used 5 km resolution gridded data for the base period (1983–2012). Furthermore, a perfect sibling framework (see section 2) was employed to assess the value of the downscaling strategy.
Here, we produced high-resolution daily precipitation and temperature data through a gridded statistical downscaling method called BCCAQ32. The method significantly reduced the biases between the GCMs output and the observation data and minimized the errors in the projections. The data will help overcome limitations in climate change impact studies on crop growth33,34,35,36 and hydrology37,38,39 in Ethiopia resulting from the insufficient number of GCM used. Our evaluation allows a clear perspective on individual GCMs by means of correlations, biases, and temporal evolution. Our study will make a valuable contribution to future model comparison efforts in Ethiopia and beyond by revealing the differences between the various models. In particular, our findings indicate that the simulations produced by the two Norwegian GCMs (NorESM2-LM and NorESM2-MM) may be implausible despite their strong performance in East Africa40. Conversely, the consistent results observed across Ethiopia’s diverse regions suggest that for the purposes of our study, fewer homogeneous regions, similar to the approach taken by ref. 41 may be sufficient. Finally, the dataset will help researchers examine a wide range of scenarios, which can help inform more comprehensive and robust adaptation strategies.
Methods
Gridded observation data
Daily climate data is available for the study area (Supplementary Fig. 1a) from CHIRPS (Climate Hazards Group InfraRed Precipitation with Stations) for precipitation42 and from CHIRTS (Climate Hazards Group InfraRed Temperature with Stations) for temperature43 (Supplementary Fig. 1b–d). The climate hazards group database provides free daily climate data with a gridded 5 km spatial resolution with quasi-global coverage (50S–50 N) (ftp://ftp.chg.ucsb.edu/pub/org/chg/products/ or https://www.chc.ucsb.edu/products). The data were generated in several stages by blending satellite records and in situ station data. They are available for precipitation since 1981 and for temperature from 1983 to 2016. The data have been evaluated for their reliability to represent the climatology and major meteorological systems in Ethiopia44,45,46,47,48 and elsewhere49,50,51,52.
The 16 GCMs from CMIP6 used for downscaling are listed in Table 1. The three selected SSPs are SSP2-4.5, SSP3-7.0, and SSP5-8.5 representing the medium, medium-high, and high-forcing scenarios, which in turn were based on so-called middle-of-the-road, regional rivalry, and fossil-fueled socioeconomic development scenarios, respectively53,54. The data of precipitation, maximum and minimum temperature, at a horizontal grid spacing ranging from about 70 km to 400 km were accessed from the CMIP6 web data portal55 (https://esgf-node.llnl.gov/search/cmip6). To be consistent across the models, all downloaded GCM data belonged to the ‘r1i1p1f1’ variant (standing for 1st realization, 1st initialization, 1st physics, and 1st forcing). A 30-year period (1983–2012) was adopted as a baseline period for method validation and a period centered at the 2050 s (2036–2065) was adopted for the projected climate for the case studies.
Delineating homogeneous precipitation zones
Ethiopia is characterized by diverse climate regions modulated by complex topography (Supplementary Fig. 1a) where precipitation is a crucial component, particularly for the agricultural sector15,16. To acknowledge this heterogeneity and better describe the follow-up analysis, we map out the long-term mean daily precipitation into homogeneous precipitation clusters using the CLARA algorithm (Clustering for LARge Application)56. In CLARA, a sample is drawn from a large dataset and partitioned into k clusters based on the K-medoid approach. The remaining data are then assigned to the nearest k clusters. The clustering resulted in nine homogeneous precipitation regions (R1-R9) with distinct seasonal patterns (Fig. 1), which are roughly like ecozones in previous studies57,58,59. Most regions receive precipitation mainly in June-September (months JJAS) except R1 and R2 where the main season is March-May (MAM) with a short rain in October. These relatively homogeneous (i.e., in terms of annual precipitation cycles) sub regions were used to facilitate aggregated comparisons between the different outputs before and after downscaling.
The downscalinsg approach
In this study, we followed an approach called bias correction constructed analogues with quantile mapping reordering (BCCAQ) to downscale the coarser scale GCM outputs to a finer scale26,32,60,61. Compared to other statistical downscaling approaches, BCCAQ has shown to reach a better skill in representing extremes32,60,61. In the BCCAQ approach (Fig. 2a), two stage independent calculations, Climate Imprint (CI) and Constructed Analogue (CA)26,62 are calculated in parallel, which we describe in more detail below.
Step A (Climate Imprint, CI, Fig. 2a): Step A1:- Calculating monthly climatologies separately from the daily observation and raw historical GCM datasets for the reference period (1983–2012) for the variable of interest (e.g., temperature) as follows (i-v).
-
i.
Extract the data for the month of interest (e.g., January) for each year in the reference period.
-
ii.
Calculate the average of all the daily values for each year.
-
iii.
Repeat step ii for all the years in the reference period.
-
iv.
Calculate the average of all the yearly averages. This gives the monthly climatology for the variable of interest for the month of January.
-
v.
Repeat these steps for all the months in the reference period.
Step A2:- Prepare raw GCM daily anomalies (1983–2100): For each day of the year, subtract the corresponding monthly climatology value for that day from raw GCM daily values for that day. This will give the daily anomaly for each day of the year.
Step A3:- Interpolating the raw GCM daily anomalies (1983–2100) to the same spatial resolution as the observation data (fine resolution): This will result in finer resolution data with the same temporal length as the coarser-resolution GCM outputs.
Step A4:- Apply the monthly climatology of observation to the interpolated raw GCM daily anomalies (1983–2100). For each day of the year and for each grid cell, add the corresponding monthly climatology value of the observation for that month to the interpolated raw GCM daily anomaly value.
Step A5:- Bias-correcting the raw GCM from Step A4 with the observation using quantile delta mapping (QDM, see below for detail description) and the data then will be ready for the next stage of the downscaling process.
Step B (Constructed Analogue, CA, Illustrated in Fig. 2b):
Step B1:- Aggregating the finer scale daily observation data to the GCMs grid.
Step B2:- Bias-correcting the raw GCM with the observation using QDM (see below for detailed description).
Step B3:- Searching for a subset of analogues GCM: For a given day of the year in the bias corrected GCM weather pattern to be downscaled (i.e., target pattern), a search is conducted for observed steps (i.e., +/− 45 days of form that particular day) to find a subset of analogues GCM weather patterns with the target pattern. Here, the top 30 closest time steps (days) that have the closest similarity (analogues) to a given GCM daily weather pattern are selected based on minimum root mean square error (RMSE)62.
Step B4:- The CA weights (regression coefficients) are then determined via ridge regression of the CA time steps (i.e., between the target weather pattern and the selected top 30 closest fields)26.
Step B5:- Linear combinations: For the given timestep, these derived CA weights (regression coefficients) in Step B4 are then used to linearly combine the 30 corresponding bias-corrected CI outputs in Step A3 to create the spatially downscaled high-resolution data.
Step B6:- Reordering: The daily downscaled data are reordered within a given month based on the daily bias-corrected CA ranks. The reordering process ensures a better representation of event-scale spatial variability and a broadly consistent long-term trend between the downscaled outputs and bias-corrected CI32,61,63 (Step A3). Further details can be found in ref. 26,32,60,62.
Quantile delta mapping (QDM)
For a given meteorological state variable of interest, QDM28,29,30,31 starts by estimating the empirical cumulative distribution functions (CDFs) of the raw historical GCM series (Fh) and that of the observations data (Fo) during a reference period (1983–2012). Concurrently, it prepares the CDF of the raw GCM projected series (Fm (t)), which is estimated over a time t, using 30-year moving windows. The “30-year sliding window” here refers to a moving timeseries of 30 years. For instance, to perform QDM for the projection years (t) 2035, 2036, 2037, etc., timeseries of 30 years i.e., 2021–2049 (centred at 2035), 2022–2050 (centred at 2036), 2023–2051 (centred at 2037), etc., will be taken. Then, the non exceedance probability, ϵ(t) associated with the raw GCM projected value at time t is calculated within the projection period based on estimated CDF, Fm (t). Next, the inverse CDFs of the raw historical GCM series (\(\left({F}_{h}^{-1}\left[\epsilon \left(t\right)\right]\right)\)) and observations \(\left({F}_{o}^{-1}\left[\epsilon \left(t\right)\right]\right)\) during a reference period (1983–2012), and that of the raw GCM projected series \(\left({F}_{m}^{-1}\left[\epsilon \left(t\right)\right]\right)\) are all evaluated at the nonexceedance probability (ϵ(t)) associated with the modelled value at time t. Then, for precipitation (temperature), the relative (absolute) change in quantiles between the reference periods and project time t, Δt can be calculated as the ratio (difference) of the \(\left({F}_{m}^{-1}\left[\epsilon \left(t\right)\right]\right)\) and \(\left({F}_{h}^{-1}\left[\epsilon \left(t\right)\right]\right)\). Finally, the bias-correction results δ(t) in the future period at time t equals \({F}_{o}^{-1}\left[\epsilon \left(t\right)\right]\) × Δt for precipitation, and \({F}_{o}^{-1}\left[\epsilon \left(t\right)\right]\) + Δt for temperature.
Evaluation approach
For computational reasons, the observation data (daily minimum and maximum temperature, and precipitation) were first aggregated from 5 km to 10 km resolution using bilinear interpolation. Bilinear interpolation is a widely used technique in climate studies to re-grid data e.g.64,65,66. The final downscaled GCM outputs were produced for all three emission scenarios until 2100 (SSP2-4.5, SSP3-7.0, and SSP5-8.5).To evaluate the impact of the chosen downscaling strategy we used an approach called the perfect sibling (PS) framework2,27,67. This approach serves to evaluate the evolution of future simulation when no observations exist such as in the case of future climate. In the perfect sibling framework, a random GCM will be selected from the list of available GCMs as pseudo-observations for a certain reference period. The rest of the GCMs are used as a prediction of the pseudo-observations and, hence, comparisons can be made between the performance of the individual GCMs for the selected reference period. In this study, MPI-ESM1-2-HR GCM was selected as a reference simulation (mimicking observations) or perfect sibling. The remaining GCMs were compared to it for the base period (1983–2012) and the 2050 s periods (2036–2065). A two-stage evaluation was conducted i) for the base period and ii) for the projected climate. First, Pearson correlation coefficients (r) and root mean squared errors (RMSE) were calculated between the simulated and gridded observations of annual mean maximum and minimum temperature (JJAS total precipitation) for the base period (1983–2012) for each grid cell. This allowed us to compare the skill of the models in representing the base climate before and after the application for the downscaling. Next, for the future projection period, the RMSEs between the perfect sibling and the rest of the GCMs were calculated to assess the skill of the GCMs in reproducing the perfect sibling for the whole grid cells. Here, the primary objective of the evaluations was not to compare the performance of different models but rather to assess the impact of the downscaling strategy. However, in order to provide a more comprehensive analysis of the evaluation’s temporal and spatial aspects, we selected two contrasting Global Climate Models (GCMs): CMCC-CM2-SR5, which had relatively poor performance, and MPI-ESM1-2-HR, which had better performance in reproducing observations during the reference period (prior to downscaling). To demonstrate the differences before and after the application of the downscaling process, we compared the cumulative density functions (CDFs) of various datasets, including observations, historical raw GCMs data, future raw GCMs data, and downscaled data. The term “bias” in this study refers to the absolute difference between the values of two fields, specifically the modeled value minus the observed value. The downscaling was performed using the High-Performance Computing Cluster bwUniCluster 2.0 (https://wiki.bwhpc.de/e/Main_Page). All statistical and spatial analyses in this study were performed using R, an open-source statistical software68. Downscaling was conducted using the ClimDown69 package. The ncdf470 and cmsafops71 packages were used to read and edit the GCMs’ NetCDF files. Regionalization of the country into homogeneous precipitation zones was done using the cluster package72. All plots were produced using packages ggplot273, cowplot74, gridExtra75, raster76, rasterVis77 and rgdal78 packages.
Data Records
The database of the statistically downscaled daily precipitation (pr), maximum temperature (tasmax) and minimum temperature (tasmin) for the current climate (1975–2015) and future scenarios (2016–2100) under the three SSPs (SSP2-4.5, SSp3-7.0 and SSP5-8.5) are stored in the CIMMYT repository79. The total file size is ~400 GB where each file is stored as a self-describing NetCDF file format with the same naming as their original GCM names with ‘Eth_’ as a prefix. For instance, the file name “Eth_pr_ACCESS-CM2_ssp370.nc” refers to the downscaled daily precipitation (pr) data for the GCM (ACCESS-CM2) under the SSP3-7.0 (ssp370) emission scenario and “Eth_tasmax_EC-Earth3-Veg_ssp245.nc” refers to the downscaled daily maximum temperature (tasmax) data for the GCM (EC-Earth3-Veg) under the SSP2-4.5 (ssp245) emission scenario. The gridded observation data sets are provided by the Climate Hazards Group and are publicly available at https://www.chc.ucsb.edu/data. The original CMIP6 data were downloaded from the Earth System Grid Federation (ESGF) website (https://esgf-node.llnl.gov/search/cmip6/), with detailed information about the data, including the terms of use.
Technical Validation
For the purposes of the demonstrations, we show the comparisons for the SSP3-7.0 emission scenario for the mid of the century (2036–2065) centered on the 2050 s for precipitation and temperature. We also show the comparison of the CDFs for the annual mean maximum and minimum temperature and JJAS total precipitation under each homogeneous precipitation region. The respective spatial plot for the long-term mean values where the CDFs are calculated from are also presented for the two contrasting GCMs.
Precipitation
Table 2 presents the 5th percentile, the median, and the 95th percentile of the spatial distribution model biases aggregated across the country of the 30-year JJAS (June-July-August-September) total precipitation for the two selected contrasting GCMs (the least performed CMCC-CM2-SR5 and most performed, MPI-ESM1-2-HR) and the corresponding observations (see Supplementary Table 1 for model evaluation results for the rest of the models). Spatially aggregated values of the GCMs and the respective biases (defined as the difference between model and observation, defined by bias = modeled value – observed value) with the observation before and after the downscaling process were compared. Table 2 also presents the projected values for the 2050 s under the SSP3-7.0 emission scenario with the respective biases and relative percentage changes after the downscaling. The ranges in the remainder of this section always refer to the 5th and 95th quantile ranges, unless stated otherwise. The JJAS period is the main rainy season for the large parts of the country except for eastern and southeastern parts57,58,59, where the long term mean JJAS total precipitation ranges from 250 to 1200 mm (Table 2 and Fig. 3a). Figure 3 shows the spatial plots of the two GCMs before and after downscaling with the corresponding biases during the base period and mean JJAS total precipitation. The subplots in the figure show marked spatial differences between the two GCMs both before and after downscaling. Before downscaling, both models show largely wet biases with the JJAS climatology values ranging from 345–1100 mm for the CMCC-CM2-SR5 (i.e., least performed) and 80–1010 mm for the MPI-ESM1-2-HR (i.e., most performed) (Table 2). Both models also show dry biases up to 600 mm over the northern highland parts of the country. The evaluation of CMIP5 models for Ethiopia also shows a dry bias over the highlands, underrepresenting orographic uplift80. Whereas the corresponding countrywide area-averaged JJAS climatology values range from 260 to 1000 mm and 260 to 1200 mm by the CMCC-CM2-SR5 (i.e., least performed) and MPI-ESM1-2-HR (i.e., most performed), respectively, after the application of the downscaling. The respective biases range from −360 to 205 mm and from −370 to 145 mm before downscaling and the biases decreased to a range of −195 to 35 mm and −15 to 30 mm after downscaling for the CMCC-CM2-SR5 (i.e., least performed) and MPI-ESM1-2-HR (i.e., most performed), respectively, showing an overall improvement due to the downscaling. However, the CDF plots in Fig. 4 highlight the importance of visually inspecting the spatial variations of the zonal averages across the homogeneous precipitation regions. It is very clear that, before downscaling, the CMCC-CM2-SR5 (i.e., least performed) outputs could not reproduce the CDFs of the observations well in many the regions (Fig. 4). Figures 3, 4 also show that the MPI-ESM1-2-HR (i.e., most performed) model’s relatively better representation is only valid on a country scale. However, the improvement due to the downscaling still holds to some extent in each homogeneous zone.
Based on the downscaled data, the projected JJAS total precipitation quantiles range from 280 to 900 mm and from 350 to 1235 mm for the CMCC-CM2-SR5 (i.e., least performed) and MPI-ESM1-2-HR (i.e., most performed), respectively, for the 2050 s under the SSP3-7.0 emission scenario (Supplementary Fig. 2). The corresponding percentage of change ranges −30 to 40% and from −5 to 65% for the CMCC-CM2-SR5 (i.e., least performed) and MPI-ESM1-2-HR (i.e., most performed), respectively, where mostly a negative (positive) anomalous JJAS total precipitation (up to 500 mm) is projected by the CMCC-CM2-SR5 (MPI-ESM1-2-HR) in large parts of the country (Table 2). This is clearly shown in the CDF plots where large leftward (rightward) shifts are observed particularly for high precipitation receiving regions (i.e., R6 - R8) in the CMCC-CM2-SR5 (MPI-ESM1-2-HR) projected JJAS precipitation. As in the MPI-ESM1-2-HR (i.e., most performed) model, positive projected changes in JJAS total precipitation were also reported for Ethiopia80 and across East Africa81,82. Figure 7. depicts the spatial pattern of the correlation (Pearson r) and the RMSEs between the observation and the 12 GCM outputs for JJAS total precipitation in the base period (1983–2012) before and after downscaling. The spatial correlation plots reveal that the two models from the Norwegian climate modeling consortium, (i.e., NorESM2-LM and NorESM2-MM), were exceptionally different from the rest of the models with negative correlation with the observation and higher biases (RMSE > 150 mm) for the largest part of JJAS precipitation benefiting regions of the country (Fig. 7a,b). Despite positive spatial correlations, higher biases were also observed in MIROC6 and ACCESS-ESM1-5 models for a considerable part of the country. The application of the downscaling strategy reduced the biases (RMSE < 150 mm) across a large part of the country in all the GCMs except for those two GCMs (Fig. 7c,d). Despite downscaling, the two Norwegian climate models (NorESM2-LM and NorESM2-MM) still exhibit negative correlations and higher biases errors. Moreover, the downscaling process led to an increase in both the intensity and spatial extent of the root mean square error (RMSE). Ref. 83 also identified those two models as outliers compared to other models. Figure 8 illustrates a comparison between the RMSE errors in the projected JJAS (June-July-August-September) total precipitation for the 2050 s under the SSP3-7.0 emission scenario, before and after applying downscaling techniques. The comparison is made between a perfect sibling projection and the other GCMs during the same projected climate period. The evaluation with the perfect sibling framework (Fig. 8 & Supplementary Table 2) shows that the downscaling substantially minimized the bias in the projected climate outputs (the mean RMSE over the whole region denoted by E measured in mm/season are shown on the top of each figure) for all models except for the two Norwegian models. The reduction in the errors ranges from 125 mm/season (~40% lower) for ACCESS-CM2 to 340 mm/season (~80% lower) for MIROC6.
Temperature
Table 3 presents the 5th percentile, the median, and the 95th percentile of the spatial distribution across the country of the 30-year mean annual maximum and minimum temperature for the two selected contrasting GCMs (the least performed CMCC-CM2-SR5 and most performed, MPI-ESM1-2-HR) and the observation (see Supplementary Tables 2, 3 for model evaluation results for the rest of the models). Aggregated across the country, the observed mean maximum (minimum) temperature ranges from 23 to 37 °C (12 to 27 °C) (Fig. 5a & Supplementary Fig. 4a). For maximum temperature, the historical mean values range from 21 to 30 °C for the CMCC-CM2-SR5 (i.e., least performed) and from 26 to 36 °C for MPI-ESM1-2-HR (i.e., most performed), resulting in absolute biases (from the observation) between −8.7 and −3.4 °C and between −3.2 and 1.0 °C, respectively, before downscaling (Fig. 5). Similarly, for minimum temperature, the historical mean values ranged from 21 to 30 °C and from 15 to 27 °C with absolute biases between 1.3 and 8.9 °C and between −2.2 to 2.7 °C for the CMCC-CM2-SR5 (i.e., least performed) and MPI-ESM1-2-HR (i.e., most performed), respectively (Supplementary Fig. 4). The historic GCM outputs of the models were also largely consistence across the country in reproducing the observed CDFs (Fig. 6 & Supplementary Fig. 6). Both models underestimate the maximum temperature in all homogeneous regions except R4-R6 and R8 for MPI-ESM1-2-HR. On the other hand, both models overestimate the minimum temperature in all regions except R2 and R3 for MPI-ESM1-2-HR. By downscaling, the biases of both maximum and minimum temperature were substantial reduced to about 0.1 °C in all regions. After downscaling, the projected (the 2050 s) mean annual maximum (minimum) temperature quantiles range from 25 to 39 °C (from 14 to 29 °C) in both models under the SSP3-7.0 emission scenario. Both models project a mean change of annual maximum (minimum) temperature between 0.6 to 1.4 °C (0.6 to 2.0 °C) (Supplementary Figs. 3a,c, 5a,c). The projected changes of minimum temperature were higher by approximately 0.6 °C in MPI-ESM1-2-HR (i.e., most performed) than in the CMCC-CM2-SR5 (i.e., least performed) (Supplementary Figs. 3b,d, 5b,d).
Figure 9 and Supplementary Fig. 7 illustrate the spatial pattern of Pearson correlation coefficient and of RMSE between the observations and the 16 GCM outputs for maximum and minimum temperature, respectively, in the base period (1983–2012) before and after downscaling. As in precipitation, the correlation plots show that the two Norwegian models were strikingly different from the rest of the models, showing a negative correlation with observations (Fig. 9a,c and Supplementary Fig. 7a,c). For CMCC-CM2-SR5, the figures show also negative correlations across a substantially large part of the country for minimum temperature (Supplementary Fig. 7a). Before downscaling, substantially larger errors (RMSE > 10 °C) in minimum temperature across the peripheral part of the country were produced by all models (Supplementary Fig. 7b). Similarly, particularly in the eastern parts of the country, higher errors in maximum temperature were observed for MIROC6 and CMCC-CM2-SR5 (Fig. 9b). By downscaling, the errors of all models in both maximum and minimum temperatures were substantially decreased to below 2 °C across the country, except the two Norwegian models (Fig. 9d & Supplementary Fig. 7d). In case of CMCC-CM2-SR5, where simulated minimum temperature was mainly negatively correlated with observations, downscaling resulted in a positive correlation across a substantially large part of the country (Fig. 9c & Supplementary Fig. 7c). In case of the two Norwegian models, there was no improvement by downscaling, both in terms of pattern correlation and error minimization. Maximum and minimum temperatures produced by these two models remained negatively correlated with observations and also showed larger errors (RMSE up to 8 °C) in a larger part of the country, particularly in the northwestern and northeastern parts.
Figure 10 (Supplementary Fig. 8) illustrates a comparison between the RMSE errors in the projected annual maximum (minimum) temperature for the 2050 s under the SSP3-7.0 emission scenario, before and after applying downscaling techniques. As with the precipitation, the evaluation by the perfect sibling comparison showed that downscaling substantially decreased the bias (the mean RMSEs over the whole region are denoted by E measured in °C are shown on the top of each figure) in the projected annual temperature outputs (Fig. 10, Supplementary Fig. 8 & Supplementary Table 4). For maximum temperature, the minimum reduction in error was 0.9 °C (~70% lower) for EC-Earth3-Veg while the maximum reduction was 7.3 °C (~93% lower) for MIROC6 (Fig. 10). Whereas for minimum temperature, the minimum reduction in error was 0.8 °C (~60% lower) for GFDL-ESM4 while the maximum reduction was 4.2 °C (~93% lower) for CMCC-CM2-SR5 (Supplementary Fig. 8). Overall, the downscaling strategy reduced the errors in projected annual temperature approximately by about 51 to 94%. Using a simple delta bias correction strategy, comparable bias minimization was reported through the perfect sibling evaluation framework2. Our results confirm that better performance is found for temperature compared to precipitation in regions where climate systems are modulated by complex topography84.
Our downscaled data sets provide high-resolution climate projections that can be used to assess the potential impacts of climate change on various sectors such as agriculture, water resources, and energy. By using our data sets, researchers can gain valuable insights into the potential changes in climate patterns at a local or regional scale, which can inform decision-making and policy development. However, it is important to note that our data sets have some limitations that should be considered when interpreting the results of impact studies. For example, the downscaling technique used in this study assumes that the historic model bias is stationary and valid for the future while it is not warranted that this is the case. Additionally, we would like to note that the technique used here is merely a statistical post-processing and cannot compensate or eliminate the individual GCM’s structural deficiencies. Furthermore, these techniques are not always perfect and can introduce their own sources of bias, so it is important to carefully evaluate the quality of the data generated using these methods. Furthermore, the presented evaluation could only cover some central aspects and further evaluations, e.g., interannual variability or spell length distributions, are encouraged to help explain the uncertainties related to the methodology85,86. Therefore, we recommend that researchers carefully evaluate the assumptions and limitations of our data sets when interpreting the results of their impact studies. In terms of the downscaled data sets from the different models, we recommend using the data sets that have the highest spatial correlations and lowest bias and errors (see Supplementary Table 1), as these will likely provide the most accurate representation of the local climate. However, we encourage researchers to explore the performance of multiple downscaled data sets from different models to evaluate the uncertainty associated with the projections.
Code availability
The R packages used in this study are freely available and detailed in the methods section. The R codes utilized for the analysis, along with the necessary input data, are publicly available at https://doi.org/10.5281/zenodo.7950777. The code demonstrates how to implement the climate downscaling technique and compute the means and sums of climate variables for various periods on the netcdf files to replicate all the findings mentioned in the manuscript.
References
Murphy, J. M. et al. Quantification of modelling uncertainties in a large ensemble of climate change simulations. Nature 430, 768–772 (2004).
Navarro-Racines, C., Tarapues, J., Thornton, P., Jarvis, A. & Ramirez-Villegas, J. High-resolution and bias-corrected CMIP5 projections for climate change impact assessments. Scientific. Data 7, 1–14 (2020).
Challinor, A. J., Smith, M. S. & Thornton, P. Use of agro-climate ensembles for quantifying uncertainty and informing adaptation. Agricultural and Forest Meteorology 170, 2–7 (2013).
Martre, P. et al. Multimodel ensembles of wheat growth: Many models are better than one. Global change biology 21, 911–925 (2015).
Tao, F. et al. Contribution of crop model structure, parameters and climate projections to uncertainty in climate change impact assessments. Global Change Biology 24, 1291–1307 (2018).
Wallach, D. et al. Multimodel ensembles improve predictions of crop-environment-management interactions. Global change biology 24, 5072–5083 (2018).
Perra, E. et al. Multimodel assessment of climate change-induced hydrologic impacts for a Mediterranean catchment. Hydrology and Earth System Sciences 22, 4125–4143 (2018).
Dams, J., Nossent, J., Senbeta, T. B., Willems, P. & Batelaan, O. Multi-model approach to assess the impact of climate change on runoff. Journal of Hydrology 529, 1601–1616 (2015).
Schewe, J. et al. Multimodel assessment of water scarcity under climate change. Proceedings of the National Academy of Sciences of the United States of America 111, 3245–3250 (2014).
Tesfaye, K. et al. Climate variability and change in Bihar, India: Challenges and opportunities for sustainable crop production. Sustainability (Switzerland) 9 (2017).
Abera, K., Crespo, O., Seid, J. & Mequanent, F. Simulating the impact of climate change on maize production in Ethiopia, East Africa. Environmental Systems Research 7 (2018).
Tesfaye, K. et al. Assessing climate adaptation options for cereal-based systems in the eastern Indo-Gangetic Plains, South Asia. Journal of Agricultural Science 157, 189–210 (2019).
Rettie, F. M., Gayler, S., Weber, T. K. D., Tesfaye, K. & Streck, T. Climate change impact on wheat and maize growth in Ethiopia: A multi-model uncertainty analysis. PLoS ONE 17, 1–26 (2022).
Hamlet, A. F., Salathé, E. P. & Carrasco, P. Statistical Downscaling Techniques for Global Climate Model Simulations of Temperature and Precipitation with Application to Water Resources Planning Studies. 28 (2010).
Diro, G. T., Grimes, D. I. F. & Black, E. African Climate and Climate Change. vol. 43 (2011).
Van den Hende, C. et al. Analysis of rain-shadows in the Ethiopian Mountains using climatological model data. Climate Dynamics 56, 1663–1679 (2021).
Chen, J. & Zhang, X. J. Challenges and potential solutions in statistical downscaling of precipitation. Climatic Change 165, 1–19 (2021).
Ahmed, K. F. et al. Statistical downscaling and bias correction of climate model outputs for climate change impact assessment in the U.S. northeast. Global and Planetary Change 100, 320–332 (2013).
Gebrechorkos, S. H., Hülsmann, S. & Bernhofer, C. Statistically downscaled climate dataset for East Africa. Scientific. Data 6, 2–9 (2019).
Chokkavarapu, N. & Mandla, V. R. Comparative study of GCMs, RCMs, downscaling and hydrological models: a review toward future climate change impact estimation. SN Applied Sciences 1, 1–15 (2019).
Rummukainen, M. State-of-the-art with regional. Clim Change 1, 82–96 (2010).
Kendon, E. J., Jones, R. G., Kjellström, E. & Murphy, J. M. Using and designing GCM-RCM ensemble regional climate projections. Journal of Climate 23, 6485–6503 (2010).
Sørland, S. L., Schär, C., Lüthi, D. & Kjellström, E. Bias patterns and climate change signals in GCM-RCM model chains. Environmental Research Letters 13 (2018).
Noël, T., Loukos, H., Defrance, D., Vrac, M. & Levavasseur, G. A high-resolution downscaled CMIP5 projections dataset of essential surface climate variables over the globe coherent with the ERA5 reanalysis for climate change impact assessments. Data in Brief 35 (2021).
Chakraborty, D., Dobor, L., Zolles, A., Hlásny, T. & Schueler, S. High-resolution gridded climate data for Europe based on bias-corrected EURO-CORDEX: The ECLIPS dataset. vol. 8 121–131 (2021).
Maurer, E. P., Hidalgo, H. G., Das, T., Dettinger, M. D. & Cayan, D. R. The utility of daily large-scale climate data in the assessment of climate change impacts on daily streamflow in California. Hydrology and Earth System Sciences 14, 1125–1138 (2010).
Hawkins, E., Osborne, T. M., Ho, C. K. & Challinor, A. J. Calibration and bias correction of climate projections for crop modelling: An idealised case study over Europe. Agricultural and Forest Meteorology 170, 19–31 (2013).
Lorenz, D. J., Nieto-Lugilde, D., Blois, J. L., Fitzpatrick, M. C. & Williams, J. W. Downscaled and debiased climate simulations for North America from 21,000 years ago to 2100AD. Scientific. Data 3, 1–19 (2016).
Gudmundsson, L., Bremnes, J. B., Haugen, J. E. & Engen Skaugen, T. Technical Note: Downscaling RCM precipitation to the station scale using quantile mapping – a comparison of methods. Hydrology and Earth System Sciences Discussions 9, 6185–6201 (2012).
Tian, J. et al. Projections of precipitation over China based on CMIP6 models. Stochastic Environmental Research and Risk Assessment 35, 831–848 (2021).
Maraun, D. Bias Correcting Climate Change Simulations - a Critical Review. Current Climate Change Reports 2, 211–220 (2016).
Werner, A. T. & Cannon, A. J. Hydrologic extremes - An intercomparison of multiple gridded statistical downscaling methods. Hydrology and Earth System Sciences 20, 1483–1508 (2016).
Tesfaye, K. et al. Maize systems under climate change in sub-Saharan Africa. International Journal of Climate Change Strategies and Management 7, 247–271 (2015).
Kassie, B. T. et al. Exploring climate change impacts and adaptation options for maize production in the Central Rift Valley of Ethiopia using different climate change scenarios and crop models. Climatic Change 129, 145–158 (2015).
Muluneh, A., Biazin, B., Stroosnijder, L., Bewket, W. & Keesstra, S. Impact of predicted changes in rainfall and atmospheric carbon dioxide on maize and wheat yields in the Central Rift Valley of Ethiopia. Regional Environmental Change 15, 1105–1119 (2015).
Muluneh, A., Stroosnijder, L., Keesstra, S. & Biazin, B. Adapting to climate change for food security in the Rift Valley dry lands of Ethiopia: Supplemental irrigation, plant density and sowing date. The Journal of Agricultural Science 155, 703–724 (2017).
Tesfalem, A., Brook, A., Abraham, W. & Alemayehu, M. Impacts of Climate Change Under CMIP5 RCP Scenarios on the Hydrology of Lake Ziway Catchment, Central Rift Valley of. Journal of Environment and Earth Science 8 (2018).
Emiru, N. C. et al. Impact of Climate Change on the Hydrology of the Upper Awash River Basin, Ethiopia. Hydrology 9 (2022).
Tarekegn, N., Abate, B., Muluneh, A. & Dile, Y. Modeling the impact of climate change on the hydrology of Andasa watershed. Modeling Earth Systems and Environment 8, 103–119 (2022).
Ayugi, B. et al. Comparison of CMIP6 and CMIP5 models in simulating mean and extreme precipitation over East Africa. vol. 41 6474–6496 (2021).
Ware, M. B. et al. Climate regionalization using objective multivariate clustering methods and characterization of climatic regions in ethiopia. Meteorologische Zeitschrift https://doi.org/10.1127/metz/2022/1093 (2022).
Funk, C. et al. The climate hazards infrared precipitation with stations - A new environmental record for monitoring extremes. Scientific Data 2, 1–21 (2015).
Verdin, A. et al. Development and validation of the CHIRTS-daily quasi-global high-resolution daily temperature data set. Scientific Data 7, 1–14 (2020).
Dinku, T. et al. Validation of the CHIRPS satellite rainfall estimates over eastern Africa. Quarterly Journal of the Royal Meteorological Society 144, 292–312 (2018).
Kabite Wedajo, G., Kebede Muleta, M. & Gessesse Awoke, B. Performance evaluation of multiple satellite rainfall products for Dhidhessa River Basin (DRB), Ethiopia. Atmospheric Measurement Techniques 14, 2299–2316 (2021).
Belete, M. et al. Evaluation of satellite rainfall products for modeling water yield over the source region of Blue Nile Basin. Science of The Total Environment 708, 134834 (2020).
Taye, M., Sahlu, D., Zaitchik, B. F. & Neka, M. Evaluation of satellite rainfall estimates for meteorological drought analysis over the upper blue nile basin. Ethiopia. Geosciences (Switzerland) 10, 1–22 (2020).
Malede, D. A., Agumassie, T. A., Kosgei, J. R., Pham, Q. B. & Andualem, T. G. Evaluation of Satellite Rainfall Estimates in a Rugged Topographical Basin Over South Gojjam Basin, Ethiopia. Journal of the Indian Society of Remote Sensing 0123456789 (2022).
Muthoni, F. K. et al. Long-term spatial-temporal trends and variability of rainfall over Eastern and Southern Africa. Theoretical and Applied Climatology 137, 1869–1882 (2019).
Muthoni, F. Spatial-temporal trends of rainfall, maximum and minimum temperatures over West Africa. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 13, 2960–2973 (2020).
Saeidizand, R., Sabetghadam, S., Tarnavsky, E. & Pierleoni, A. Evaluation of CHIRPS rainfall estimates over Iran. Quarterly Journal of the Royal Meteorological Society 144, 282–291 (2018).
Zambrano-Bigiarini, M., Nauditt, A., Birkel, C., Verbist, K. & Ribbe, L. Temporal and spatial evaluation of satellite-based rainfall estimates across the complex topographical and climatic gradients of Chile. Hydrology and Earth System Sciences 21, 1295–1320 (2017).
O’Neill, B. C. et al. The roads ahead: Narratives for shared socioeconomic pathways describing world futures in the 21st century. Global Environmental Change 42, 169–180 (2017).
Meinshausen, M. et al. The shared socio-economic pathway (SSP) greenhouse gas concentrations and their extensions to 2500. Geoscientific Model Development 13, 3571–3605 (2020).
Eyring, V. et al. Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization. Geoscientific Model Development 9, 1937–1958 (2016).
Clustering large applications (program CLARA). in Finding groups in data (eds. Kaufman, L. & Rousseeuw, P. J.) 126–163, https://doi.org/10.1002/9780470316801.ch3 (John Wiley & Sons, Inc, 1990).
Diro, G. T., Grimes, D. I. F. & Black, E. Teleconnections between Ethiopian summer rainfall and sea surface temperature: Part II. Seasonal forecasting. Climate Dynamics 37, 121–131 (2011).
Tsidu, G. M. High-resolution monthly rainfall database for Ethiopia: Homogenization, reconstruction, and gridding. Journal of Climate 25, 8422–8443 (2012).
Alhamshry, A., Ayele, F. A., Yasuda, H., Kimura, R. & Shimizu, K. Seasonal Rainfall Variability in Ethiopia and Its. Water 12, 1–19 (2020).
Cannon, A. J., Sobie, S. R. & Murdock, T. Q. Bias correction of GCM precipitation by quantile mapping: How well do methods preserve changes in quantiles and extremes? Journal of Climate 28, 6938–6959 (2015).
Masud, B. et al. Means and extremes: Evaluation of a CMIP6 multi-model ensemble in reproducing historical climate characteristics across Alberta, Canada. Water (Switzerland) 13 (2021).
Hidalgo, H. G., Dettinger, M. D. & Cayan, D. C. Downscaling with constructed analogues: Daily precipitation and temperature fields over the United States. California Energy Commission, PIER Energy-Related Environmental Research. CEC5002007123 (2008).
Maraun, D. Bias correction, quantile mapping, and downscaling: Revisiting the inflation issue. Journal of Climate 26, 2137–2143 (2013).
Thorarinsdottir, T. L., Sillmann, J., Haugen, M., Gissibl, N. & Sandstad, M. Evaluation of CMIP5 and CMIP6 simulations of historical surface air temperature extremes using proper evaluation methods. Environmental Research Letters 15, (2019).
Almazroui, M., Saeed, S., Saeed, F., Islam, M. N. & Ismail, M. Projections of Precipitation and Temperature over the South Asian Countries in CMIP6. Earth Systems and Environment 4, 297–320 (2020).
Wu, H., Lei, H., Lu, W. & Liu, Z. Future changes in precipitation over the upper Yangtze River basin based on bias correction spatial downscaling of models from CMIP6. Environmental Research Communications 4, 45002 (2022).
Hawkins, E. et al. Increasing influence of heat stress on French maize yields from the 1960s to the 2030 s. Global Change Biology 19, 937–947 (2013).
R Core Team. R: A language and environment for statistical computing. (R Foundation for Statistical Computing, 2021).
A Cannon, J Hiebert, A Werner & S Sobie. ClimDown: Climate downscaling library for daily climate model output. (2016).
Pierce, D. ncdf4: Interface to unidata netCDF (version 4 or earlier) format data files. (2019).
Kothe, S. Cmsafops: Tools for CM SAF NetCDF data. (2021).
Maechler, M., Rousseeuw, P., Struyf, A., Hubert, M. & Hornik, K. Cluster: Cluster analysis basics and extensions. (2022).
Wickham, H. ggplot2: Elegant graphics for data analysis. (Springer-Verlag New York, 2016).
Wilke, C. O. Cowplot: Streamlined plot theme and plot annotations for’ggplot2’. (2020).
Auguie, B. gridExtra: Miscellaneous functions for “grid” graphics. (2017).
Hijmans, R. J. Raster: Geographic data analysis and modeling. (2022).
Perpiñán, O. & Hijmans, R. rasterVis. (2022).
Bivand, R., Keitt, T. & Rowlingson, B. Rgdal: Bindings for the’geospatial’ data abstraction library. (2021).
Rettie, F. M., Gayler, S., Weber, T. K. D., Tesfaye, K. & Streck, T. High-resolution CMIP6 climate projections for Ethiopia. CIMMYT Research Data Team https://doi.org/11529/10548895 (2023).
Jury, M. R. Statistical evaluation of CMIP5 climate change model simulations for the Ethiopian highlands. International Journal of Climatology 35, 37–44 (2015).
Li, J. et al. Evaluation of CMIP6 Global Climate Models for Simulating Land Surface Energy and Water Fluxes During 1979–2014. Journal of Advances in Modeling Earth Systems 13, 1–32 (2021).
Akinsanola, A. A., Ongoma, V. & Kooperman, G. J. Evaluation of CMIP6 models in simulating the statistics of extreme precipitation over eastern africa. Atmospheric Research 254, 105509 (2021).
Liersch, S. et al. Are we using the right fuel to drive hydrological models? A climate impact study in the Upper Blue Nile. Hydrology and Earth System Sciences 22, 2163–2185 (2018).
Lun, Y. et al. Assessment of GCMs simulation performance for precipitation and temperature from CMIP5 to CMIP6 over the Tibetan Plateau. 41, 3994–4018 (2021).
Gutiérrez, J. M. et al. An intercomparison of a large ensemble of statistical downscaling methods over Europe: Results from the VALUE perfect predictor cross-validation experiment. International Journal of Climatology 39, 3750–3785 (2019).
Mendoza Paz, S. & Willems, P. Uncovering the strengths and weaknesses of an ensemble of quantile mapping methods for downscaling precipitation change in Southern Africa. Journal of Hydrology: Regional Studies 41, 101104 (2022).
Dix, M. et al. CSIRO-ARCCSS ACCESS-CM2 model output prepared for CMIP6 CMIP historical. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.4271 (2019).
Dix, M. et al. CSIRO-ARCCSS ACCESS-CM2 model output prepared for CMIP6 ScenarioMIP ssp370. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.4323 (2019).
Dix, M. et al. CSIRO-ARCCSS ACCESS-CM2 model output prepared for CMIP6 ScenarioMIP ssp585. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.4332 (2019).
Dix, M. et al. CSIRO-ARCCSS ACCESS-CM2 model output prepared for CMIP6 ScenarioMIP ssp245. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.4321 (2019).
Ziehn, T. et al. CSIRO ACCESS-ESM1.5 model output prepared for CMIP6 CMIP historical. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.4272 (2019).
Ziehn, T. et al. CSIRO ACCESS-ESM1.5 model output prepared for CMIP6 ScenarioMIP ssp370. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.4324 (2019).
Ziehn, T. et al. CSIRO ACCESS-ESM1.5 model output prepared for CMIP6 ScenarioMIP ssp585. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.4333 (2019).
Ziehn, T. et al. CSIRO ACCESS-ESM1.5 model output prepared for CMIP6 ScenarioMIP ssp245. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.4322 (2019).
Semmler, T. et al. AWI AWI-CM1.1MR model output prepared for CMIP6 CMIP historical. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.2686 (2018).
Semmler, T. et al. AWI AWI-CM1.1MR model output prepared for CMIP6 ScenarioMIP ssp370. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.2803 (2019).
Semmler, T. et al. AWI AWI-CM1.1MR model output prepared for CMIP6 ScenarioMIP ssp585. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.2817 (2019).
Semmler, T. et al. AWI AWI-CM1.1MR model output prepared for CMIP6 ScenarioMIP ssp245. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.2800 (2018).
Lovato, T. & Peano, D. CMCC CMCC-CM2-SR5 model output prepared for CMIP6 CMIP historical. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.3825 (2020).
Fogli, P. G., Iovino, D. & Lovato, T. CMCC CMCC-CM2-SR5 model output prepared for CMIP6 OMIP omip1. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.13230 (2020).
Lovato, T. & Peano, D. CMCC CMCC-CM2-SR5 model output prepared for CMIP6 ScenarioMIP ssp585. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.3896 (2020).
Lovato, T. & Peano, D. CMCC CMCC-CM2-SR5 model output prepared for CMIP6 ScenarioMIP ssp245. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.3889 (2020).
(EC-Earth), E.-E. C. EC-Earth-Consortium EC-Earth3-Veg model output prepared for CMIP6 CMIP historical. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.4706 (2019).
(EC-Earth), E.-E. C. EC-Earth-Consortium EC-Earth3-Veg model output prepared for CMIP6 ScenarioMIP ssp370. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.4886 (2019).
(EC-Earth), E.-E. C. EC-earth-consortium EC-Earth3-veg model output prepared for CMIP6 ScenarioMIP ssp585. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.4914 (2019).
(EC-Earth), E.-E. C. EC-Earth-Consortium EC-Earth3-Veg model output prepared for CMIP6 ScenarioMIP ssp245. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.4882 (2019).
(EC-Earth), E.-E. C. EC-Earth-Consortium EC-Earth3 model output prepared for CMIP6 CMIP historical. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.4700 (2019).
(EC-Earth), E.-E. C. EC-Earth-Consortium EC-Earth3 model output prepared for CMIP6 ScenarioMIP ssp370. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.4884 (2019).
(EC-Earth), E.-E. C. EC-earth-consortium EC-Earth3 model output prepared for CMIP6 ScenarioMIP ssp585. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.4912 (2019).
(EC-Earth), E.-E. C. EC-Earth-Consortium EC-Earth3 model output prepared for CMIP6 ScenarioMIP ssp245. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.4880 (2019).
Krasting, J. P. et al. NOAA-GFDL GFDL-ESM4 model output prepared for CMIP6 CMIP historical. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.8597 (2018).
John, J. G. et al. NOAA-GFDL GFDL-ESM4 model output prepared for CMIP6 ScenarioMIP ssp370. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.8691 (2018).
John, J. G. et al. NOAA-GFDL GFDL-ESM4 model output prepared for CMIP6 ScenarioMIP ssp585. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.8706 (2018).
John, J. G. et al. NOAA-GFDL GFDL-ESM4 model output prepared for CMIP6 ScenarioMIP ssp245. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.8686 (2018).
Volodin, E. et al. INM INM-CM4-8 model output prepared for CMIP6 CMIP historical. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.5069 (2019).
Volodin, E. et al. INM INM-CM4-8 model output prepared for CMIP6 ScenarioMIP ssp370. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.12329 (2019).
Volodin, E. et al. INM INM-CM4-8 model output prepared for CMIP6 ScenarioMIP ssp585. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.12337 (2019).
Volodin, E. et al. INM INM-CM4-8 model output prepared for CMIP6 ScenarioMIP ssp245. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.12327 (2019).
Volodin, E. et al. INM INM-CM5-0 model output prepared for CMIP6 CMIP historical. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.5070 (2019).
Volodin, E. et al. INM INM-CM5-0 model output prepared for CMIP6 ScenarioMIP ssp370. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.12330 (2019).
Volodin, E. et al. INM INM-CM5-0 model output prepared for CMIP6 ScenarioMIP ssp585. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.12338 (2019).
Volodin, E. et al. INM INM-CM5-0 model output prepared for CMIP6 ScenarioMIP ssp245. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.12328 (2019).
Boucher, O. et al. IPSL IPSL-CM6A-LR model output prepared for CMIP6 CMIP historical. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.5195 (2018).
Boucher, O. et al. IPSL IPSL-CM6A-LR model output prepared for CMIP6 ScenarioMIP ssp370. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.5265 (2019).
Boucher, O. et al. IPSL IPSL-CM6A-LR model output prepared for CMIP6 ScenarioMIP ssp585. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.5271 (2019).
Boucher, O. et al. IPSL IPSL-CM6A-LR model output prepared for CMIP6 ScenarioMIP ssp245. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.5264 (2019).
Tatebe, H. & Watanabe, M. MIROC MIROC6 model output prepared for CMIP6 CMIP historical. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.5603 (2018).
Shiogama, H., Abe, M. & Tatebe, H. MIROC MIROC6 model output prepared for CMIP6 ScenarioMIP ssp370. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.5752 (2019).
Shiogama, H., Abe, M. & Tatebe, H. MIROC MIROC6 model output prepared for CMIP6 ScenarioMIP ssp585. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.5771 (2019).
Shiogama, H., Abe, M. & Tatebe, H. MIROC MIROC6 model output prepared for CMIP6 ScenarioMIP ssp245. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.5746 (2019).
Jungclaus, J. et al. MPI-M MPI-ESM1.2-HR model output prepared for CMIP6 CMIP historical. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.6594 (2019).
Schupfner, M. et al. DKRZ MPI-ESM1.2-HR model output prepared for CMIP6 ScenarioMIP ssp370. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.4399 (2019).
Schupfner, M. et al. DKRZ MPI-ESM1.2-HR model output prepared for CMIP6 ScenarioMIP ssp585. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.4403 (2019).
Schupfner, M. et al. DKRZ MPI-ESM1.2-HR model output prepared for CMIP6 ScenarioMIP ssp245. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.4398 (2019).
Wieners, K.-H. et al. MPI-M MPI-ESM1.2-LR model output prepared for CMIP6 CMIP historical. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.6595 (2019).
Wieners, K.-H. et al. MPI-M MPI-ESM1.2-LR model output prepared for CMIP6 ScenarioMIP ssp370. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.6695 (2019).
Wieners, K.-H. et al. MPI-m MPI-ESM1.2-LR model output prepared for CMIP6 ScenarioMIP ssp585. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.6705 (2019).
Wieners, K.-H. et al. MPI-M MPI-ESM1.2-LR model output prepared for CMIP6 ScenarioMIP ssp245. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.6693 (2019).
Yukimoto, S. et al. MRI MRI-ESM2.0 model output prepared for CMIP6 CMIP historical. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.6842 (2019).
Yukimoto, S. et al. MRI MRI-ESM2.0 model output prepared for CMIP6 ScenarioMIP ssp370. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.6915 (2019).
Yukimoto, S. et al. MRI MRI-ESM2.0 model output prepared for CMIP6 ScenarioMIP ssp585. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.6929 (2019).
Yukimoto, S. et al. MRI MRI-ESM2.0 model output prepared for CMIP6 ScenarioMIP ssp245. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.6910 (2019).
Seland, Ø. et al. NCC NorESM2-LM model output prepared for CMIP6 CMIP historical. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.8036 (2019).
Seland, Ø. et al. NCC NorESM2-LM model output prepared for CMIP6 ScenarioMIP ssp370. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.8268 (2019).
Seland, Ø. et al. NCC NorESM2-LM model output prepared for CMIP6 ScenarioMIP ssp585. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.8319 (2019).
Seland, Ø. et al. NCC NorESM2-LM model output prepared for CMIP6 ScenarioMIP ssp245. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.8253 (2019).
Bentsen, M. et al. NCC NorESM2-MM model output prepared for CMIP6 CMIP historical. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.8040 (2019).
Bentsen, M. et al. NCC NorESM2-MM model output prepared for CMIP6 ScenarioMIP ssp370. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.8270 (2019).
Bentsen, M. et al. NCC NorESM2-MM model output prepared for CMIP6 ScenarioMIP ssp585. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.8321 (2019).
Bentsen, M. et al. NCC NorESM2-MM model output prepared for CMIP6 ScenarioMIP ssp245. Earth System Grid Federation. https://doi.org/10.22033/ESGF/CMIP6.8255 (2019).
Acknowledgements
This publication is an output of a PhD scholarship at the University of Hohenheim in the framework of the project “German-Ethiopian SDG Graduate School: Climate Change Effects on Food Security (CLIFOOD)” between the University of Hohenheim (Germany) and the Hawassa University (Ethiopia), supported by the DAAD with funds from the Federal Ministry for Economic Cooperation and Development (BMZ), funding code 57316245. The authors acknowledge support by the state of Baden-Württemberg through bwHPC. We would like to thank Dr. Kirsten Warrach-Sagi, Institute of Physics and Meteorology, Hohenheim University, for her valuable comments and suggestions. The authors also acknowledge the World Climate Research Programme, which, through its Working Group on Coupled Modeling, coordinated and promoted CMIP6. The authors thank the climate modeling groups for producing and making available their model output, the Earth System Grid Federation (ESGF) for archiving the data and providing access, and the multiple funding agencies who support CMIP6 and ESGF. Furthermore, the authors are grateful to the Climate Hazards Group for producing the CHIRPS and CHRPT datasets. Finally, we extend our sincere gratitude to CIMMYT for hosting the datasets on their server, and we would like to give special thanks to Enrique Garza Sánchez for his exceptional efforts in managing the meta data, uploading the datasets to Dataverse, and ensuring their public accessibility.
Funding
Open Access funding enabled and organized by Projekt DEAL.
Author information
Authors and Affiliations
Contributions
Conceptualization: F.M.R., S.G. and T.S.; methodology, formal analysis, visualization, software and writing—original draft preparation: F.M.R.; supervision, investigation and resources: S.G., T.D.K. and T.S.; writing–review and editing: F.M.R., S.G., T.D.K., K.T., K.W.S. and T.S.; project administration and funding acquisition: S.G. and T.S.; All authors have read and agreed to the published version of the manuscript.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Rettie, F.M., Gayler, S., Weber, T.K.D. et al. High-resolution CMIP6 climate projections for Ethiopia using the gridded statistical downscaling method. Sci Data 10, 442 (2023). https://doi.org/10.1038/s41597-023-02337-2
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41597-023-02337-2