Towards Better Future Climate Information over Central Asia

Proceedings of the Royal Academy for Overseas Sciences 3 (2025 – 1): 63-77 DOI: 10.5281/zenodo.16925602 Towards Better Future Climate Information over Central Asia* by Sara Top** Keywords. — Regional Climate Modelling; Climate Projections; Climate Change. Summary. — Unprecedented, widespread and rapid changes in the climate system have been observed in every region across the globe. Moreover, the increase in global surface temperature is expected to continue until at least 2050, which will lead to more extreme events. Regional climate models (RCMs) are used to downscale the information of global circulation models (GCMs) over particular regions of interest to study the effect of global warming at higher spatial and temporal resolutions. Such highresolution climate information is still scarce for Central Asia, while this region covers a wide range of climatic zones and includes some densely-populated cities. High-resolution climate projections over Central Asia are presented under 1.5° C, 2° C and 3° C global warming. Trefwoorden. — Regionale klimaatmodellering; Klimaatprojecties; Klimaatverandering. Samenvatting. — Op naar betere, toekomstige klimaatinformatie over Centraal-Azië. — Ongeziene, uitgebreide en snelle veranderingen in het klimaat zijn waargenomen in elke regio van de wereld. Bovendien wordt tot minstens 2050 een verdere toename van de globale oppervlaktetemperatuur verwacht, wat zal leiden tot meer extreme gebeurtenissen. Regionale klimaatmodellen worden gebruikt om informatie van globale modellen om te zetten naar data met een hogere ruimtelijke en temporele resolutie over een beperkt studiegebied om zo het effect van klimaatverandering gedetailleerder te kunnen bestuderen. Dergelijke hoge resolutie klimaatinformatie over Centraal-Azië is echter schaars, terwijl deze regio een groot aantal klimaatzones vertegenwoordigt en enkele steden met een zeer hoge bevolking omvat. Hoge resolutie klimaatprojecties over Centraal-Azië worden toegelicht bij een globale klimaatopwarming van 1,5° C, 2° C en 3° C. Mots-clés. — Modélisation climatique régionale; Projections climatiques; Changement climatique. Résumé. — Vers une meilleure information du climat futur de l’Asie centrale. — Des changements inédits, étendus et rapides du système climatique ont été observés dans toutes les régions du monde. De plus, il est prévu que la température de surface mondiale continue d’augmenter au moins jusque 2050, ce qui entraînera une hausse des phénomènes extrêmes. Les modèles climatiques régionaux sont utilisés pour améliorer l’information des modèles globaux en se focalisant sur des régions d’intérêt particulier afin d’étudier l’effet du réchauffement climatique à des résolutions spatiales et temporelles plus élevées. Ces informations climatiques à haute résolution sont encore rares pour l’Asie centrale. En outre, cette région couvre beaucoup de zones climatiques et comprend plusieurs villes densément peuplées. Des projections climatiques à haute résolution sur l’Asie centrale sont ici présentées dans le cadre d’un réchauffement planétaire de 1,5° C, 2° C et 3° C. * Paper presented at the meeting of the Section of Natural and Medical Sciences held on 17 May 2022. Text received on 6 September 2022 and submitted to peer review. Final version, approved by the review ers, received on 10 July 2023. ** Postdoctoral researcher, Department of Atmospheric Physics, Ghent University, Krijgslaan 281 S9, B-9000 Ghent (Belgium). — 64 — Introduction Extreme events such as heatwaves, heavy precipitation, droughts, and tropical cyclones become more intense and occur more frequently (IPCC, 2021). These observations have been linked to the unprecedented and rapid changes in the climate system (IPCC, 2021). In 2020 a global nearsurface temperature rise of 1.26° C (1.12° C - 1.37° C) relative to the baseline period 1850-1900 was reached (IPCC, 2021, p. 191). This stands for the approximate temperature rise since the pre-industrial period. Parties all over the world pledged in 2016 in the Paris Agreement to pursue a limitation of global warming to 1.5° C or to stay well below 2° C with respect to the preindustrial temperature levels. The IPCC special report on the impacts of global warming of 1.5° C above pre-industrial level (IPCC, 2018) states with high confidence that global warming will reach 1.5° C between 2030 and 2052 if greenhouse gas concentrations continue to increase at the same rate. The increase in global temperature is moreover expected to continue at least until 2050 under the socio-economic scenarios that are considered as feasible (IPCC, 2021). A global warming of 2° C will therefore be exceeded during the 21st century unless greenhouse gas emissions are drastically reduced in the coming decades. A further ongoing global warming will lead towards even more extreme events and will for example induce more frequent exceedance of human health heat thresholds, causing reduced well-being, labour productivity and an excess in mortalit y (Casanueva et al., 2020; IPCC, 2021). In order to study the potential future climate pathways and their impacts, future climate simulations are necessary. Global circulation models (GCMs) make such simulations possible by including physical laws to describe the atmosphere and its interactions. They simulate global climate information at a coarse resolution of typically ~100 km by ~100 km, with 25 km being the finest resolution currently reached (Demory et al., 2020). Regional climate models (RCMs) consist similarly to GCMs out of physical laws, but the physical laws of RCMs describe smallerscale processes that cannot be resolved by the coarse resolution of GCMs. RCMs are therefore used to downscale the data of GCMs over particular regions of interest to study the effect of global warming at higher spatial and temporal resolutions. The spatial resolution is typically 25 km by 25 km or finer. Such high-resolution climate information is still scarce for Central Asia (Kotova et al., 2018). Additionally, the Central Asia region covers a wide range of climatic zones and comprises some densely-populated cities, which increases the potential value of highresolution climate information when it would be available for this region (IPCC, 2021). Moreover, high-resolution climate information is needed to study the possible impact of climate change on, for example, crop yields (Vannoppen et al., 2020), ecosystems (Cai et al., 2020), urban climate (Cugnon et al., 2019), building materials (Hedayatnia et al., 2021), etc. The following sections describe therefore the method and outcomes of the future climate projections that have been produced with the RCMs ALARO (Giot et al., 2016) and REMO (Jacob & Podzun, 1997). ALARO was developed and maintained by the Royal Meteorological Institute of Belgium (RMIB) (Termonia et al., 2018), while REMO was developed at the Max Planck Institute for Meteorology and is maintained and used by the Helmholtz-Zentrum Hereon Climate Service Center Germany (HZH-GERICS) (Jacob et al., 2012). Before presenting the outcomes of climate projections, the first section explains how regional climate modelling works, why we can trust the results and which methods were applied to obtain the results described in the following sections. The second and third sections present the future changes in respectively temperature and precipitation over Central Asia, together with their uncer- — 65 — tainties. In this way we have reduced the gap in high-resolution climate data and knowledge of the possible future climate evolutions under 1.5° C, 2° C and 3° C over this region. Regional Climate Modelling over Central Asia The model domain applied over Central Asia, as depicted in figure 1, was defined by the Coordinated Regional Climate Downscaling Experiment (CORDEX). This international initiative was launched with the aim of designing and conducting several high-resolution experiments on prescribed spatial domains across the globe. Furthermore, CORDEX creates a framework to perform downscaling towards high-resolution climate data, to evaluate these downscaled data, and to characterize uncertainties of regional climate change projections by combining data of different RCMs into ensemble projections (Giorgi & Gutowski, 2015). Although this initiative resulted in an added value since the joint agreements and modelling efforts led towards large multi-model ensembles over several regions produced by different modelling groups, some regions such as Central Asia are until today sparsely covered by high-resolution climate data. Therefore, climate data with the unprecedented spatial resolution of 25 by 25 km were produced over this region in the course of the AFTER-project (Kotova et al., 2018). The evaluation data, historical runs and future projections can be retrieved from the ESGF data nodes (website: http:// esgf.llnl.gov/; last access: 31 August 2022) and can be freely used for further research. Before RCMs can be trusted for producing climate projections and their data applied in impact models, a model evaluation is needed to retrieve possible model errors or biases and to gain confidence in the RCM downscaling procedure. Top et al. (2021) evaluated both ALARO and REMO over the Central Asia CORDEX domain (CAS-CORDEX) (see fig. 1) by comparing their downscaled results to observed and re-analysis data. Top et al. (2021) concluded that the RCMs ALARO and REMO are able to capture the general climate, but both models have some deficiencies for a certain variable over particular parts of the modelling domain. For example, ALARO is subject to significant positive temperature biases in winter, followed by large negative biases in spring for the northern part of the CAS-CORDEX domain and REMO simulated excessive precipitation amounts over the Tibetan Plateau during all seasons. These shortcomings have to be taken into account when using and interpreting the data of these RCMs for climate projections. To downscale future climate data of a GCM to a finer grid the dynamical downscaling procedure was applied. Hereby data produced by a GCM of the fifth Coupled Model Intercomparison Project (CMIP5) cycle was used as the lateral boundary condition to force an RCM that produces climate data at higher spatial and temporal resolutions. The RCM ALARO was forced by the GCM CNRM-CM5, while REMO was forced by three GCMs, namely: HadGEM2-ES, MPIESM-LR and NorESM1-M. In this way a small ensemble of four GCM-RCM members was created to reduce biased conclusions. The multi-model mean of the ensemble is obtained by ta king an equal average over the four members. The variation between the outcomes of the different GCM-RCM members was quantified by calculating the standard deviation between the different members to see whether the general outcome is supported by each GCM-RCM combination and whether a particular change in temperature or precipitation is likely to occur or not under a 1.5° C, 2° C and 3° C warmer world. A larger spread between the GCM-RCM members in the ensemble means that there is a higher uncertainty in the future prevalent temperature or precipitation at that location. Since the ensemble only consists of four members, one should keep in mind that only a small part of the full uncertainty range will be covered. — 66 — Fig. 1. — IPCC6 subregions in the CORDEX Central Asia (CAS-CORDEX) domain (based on Top, 2022). et al. — 67 — Climate projections are always conditional on the considered future socio-economic scenario. To determine possible future climate outcomes different shared socio-economic pathways (SSPs) are combined with representative concentration pathways (RCPs) to include time series of emissions and concentrations of all greenhouse gases, aerosols and chemically active gases, as well as land-use/land-cover changes. RCPs prescribe these possible future changes in time by using radiative forcing values. For example, in the RCP 2.6 scenario the radiative forcing peaks to 3 W m-2 before 2100 and then declines to 2.6 W m-2 by the end of the century (van Vuuren et al., 2011). Future simulations were run from 2006 to 2100 and are for all GCM-RCM members performed for a low and very high emission scenario, RCP 2.6 and RCP 8.5 respectively. RCP 8.5 represents rather an unlikely high-risk future scenario since it assumes a fivefold increase in coal use by 2100, while coal use peaked in 2013 (Hausfather & Peters, 2020). This scenario is, however, taken into account to identify all potential risks and to be prepared for extreme cases since it includes, for example, the uncertainty of important feedback effects that might be underestimated by current climate models, such as the carbon cycle feedbacks. By using the low and very high RCPs as boundary condition for climate projections, both the optimistic and worst case climate change scenarios are taken into account. To investigate how the temperature and precipitation patterns will evolve over time, the period 1976-2005 is selected as historical modelling period. The climate change signal with respect to the near past is then obtained by subtracting the historical thirty-year mean from a future thirty-year mean. To present the spatial variation in climate change under a global warming of 1.5° C, 2° C or 3° C with respect to the pre-industrial era, the 1850-1900 period has to be used as reference period as done by the IPCC (IPCC, 2021). To obtain the results under a 1.5° C, 2° C or 3° C warmer world, the warming from 1850-1900 until the historical period 1976-2005 has to be taken into account and amounts 0.69° C (0.52° C – 0.82° C) (Jacob et al., 2018; IPCC, 2021). The central years and thirty-year periods when the different GCMs coupled to the RCMs reach 1.5° C, 2° C and 3° C were determined by using the method of Vautard et al. (2014) and are presented in tables 1, 2 and 3 respectively. The 1.5° C, 2° C and 3° C pe riods differ for each RCP scenario because human-induced emissions evolve differently under these scenarios (tabs. 1, 2, 3). These years in tables 1, 2 and 3 make it possible to trace back when changes in temperature and precipitation patterns, described in the following sections, are likely to take place given a certain emission scenario. Some GCMs do not reach the 2° C and/or 3° C global warming threshold by 2085 under RCP 2.6, which makes it impossible to define the thirty-year period for these GCMs, indicated by “inf” in tables 2 and 3. Table 1 Central year and thirty-year period when global warming reaches +1.5° C for general circulation models (GCMs) used under different representative concentration pathways (RCPs) (adopted from Top, 2022) RCP 2.6 RCP 8.5 GCM +1.5° C central year +1.5° C period +1.5° C central year +1.5° C period CNRM-CM5-r1i1p1 2040 2026-2055 2029 2015-2044 HadGEM2-ES-r1i1p1 2019 2005-2034 2018 2004-2033 MPI-ESM-LR-r1i1p1 2049 2035-2064 2026 2012-2041 NorESM1-M-r1i1p1 2063 2049-2078 2031 2017-2046 — 68 — Table 2 Central year and thirty-year period when global warming reaches +2° C for general circulation models (GCMs) used under different representative concentration pathways (RCPs) (adopted from Top, 2022) – “inf” indicates that the warming level was not reached by 2085. RCP 2.6 RCP 8.5 GCM +2° C central year +2° C period +2° C central year +2° C period CNRM-CM5-r1i1p1 inf inf 2044 2030-2059 HadGEM2-ES-r1i1p1 2037 2023-2052 2030 2016-2045 MPI-ESM-LR-r1i1p1 inf inf 2044 2030-2059 NorESM1-M-r1i1p1 inf inf 2046 2032-2061 Table 3 Central year and thirty-year period when global warming reaches +3° C for general circulation models (GCMs) used under different representative concentration pathways (RCPs) (adopted from Top, 2022) – “inf” indicates that the warming level was not reached by 2085. RCP 2.6 RCP 8.5 GCM +3° C central year +3° C period +3° C central year +3° C period CNRM-CM5-r1i1p1 inf inf 2067 2053-2082 HadGEM2-ES-r1i1p1 inf inf 2051 2037-2066 MPI-ESM-LR-r1i1p1 inf inf 2067 2053-2082 NorESM1-M-r1i1p1 inf inf 2072 2058-2087 Under RCP 8.5 1.5° C, 2° C and 3° C global warming is reached by all four GCMs (tabs. 1, 2, 3). The fact that this is the only scenario that makes it possible to investigate which regional changes will take place under an extreme global warming of 3° C is an additional reason to take this scenario into account. In table 2, the 2° C level is not reached under RCP 2.6, except for the GCM HadGEM2ES. The central year when 1.5° C is reached for the GCMs CNRM-CM5 and MPI-ESM-LR lies between 2030 and 2052, which is according to the IPCC the period when it is very likely that 1.5° C global warming will be reached, while NorESM1-M reaches 1.5° C global warming later (tab. 1). According to the observations within the framework of the IPCC (IPCC, 2021), the steep warming trend of HadGEM2-ES, reaching 1.5° C in the central year 2019 (tab. 1), is too fast compared to the observed global warming. Different to NorESM1-M, CNRM-CM5 and MPI-ESM-LR, the GCM HadGEM2-ES simulates a positive shortwave cloud radiative forcing which can partly explain the strong climate sensitivity of HadGEM2-ES (Andrews, Gregory, Webb & Taylor, 2012). In the following sections the spatial differences in future and historical climate are presented based on the changes in the multi-model mean of temperature and precipitation. The multi-model mean of the thirty-year periods representing a 1.5° C, 2° C and 3° C global warming since the preindustrial period is compared to the historical period 1976-2005. Since the 1976-2005 period already experienced a warming of 0.69° C, the differences between the 1.5° C, 2° C and 3° C scenarios show the effect of 0.81° C, 1.31° C and 2.31° C global warming when compared to the historical 1976-2005 period. To define whether changes in the distribution of the simulated variables under 0.81° C, 1.31° C and 2.31° C global warming are significant, a Mann-Whitney U-test, also named the Wilcoxon rank sum test, at a 95 % significance level is performed. This test puts the values of both historical and future periods into one sample with an indication to which period — 69 — the values belong. When the historical and future values are randomly ranked, then the values of the two periods are not different and no increase or decrease is found and vice versa (Corder & Foreman, 2011). A one-sided test is applied for temperature since a warming trend is expected over the CAS-CORDEX domain (IPCC, 2021) and a two-sided test is used for precipitation because an increasing or decreasing precipitation trend is expected depending on the region (IPCC, 2021). Future Change in Temperature over Central Asia Figure 2 shows the spatial distribution of the multi-model mean for temperature during the historical period 1976-2005 and presents the regional warming under different global warming levels. The Wilcoxon signed-rank test at a 95 % significance level shows at annual level and for each season that the complete CAS-CORDEX region is expected to undergo significant warming under all three investigated global warming levels when compared to the temperature distribution during the historical 1975-2005 period. Furthermore, it can be derived from figure 2 that a more severe regional warming will take place under more severe global warming. Moreover, the regional warming trend is steeper than the global warming trend since an increase in 0.81° C globally leads to a change of 1 to 3° C over most of the CAS-CORDEX domain. The northern region of the CASCORDEX domain will undergo a faster temperature increase than the mean warming trend over the CAS-CORDEX domain, while Southeast China experiences a slower warming. When seasons are compared, the warming trend is more prominent during autumn and winter over most parts of the domain. Figure 3 presents the spread between the different model members of the ensemble for the three warming scenarios. The very similar patterns in spread that appear for each warming scenario (columns in fig. 3) show that the dissimilarities between the models remain the same over time. This implies that the dissimilarities between the models are mainly due to the inter-model uncertainty on regional processes that do not change over time. Over the northeastern region of the CAS-CORDEX domain there is a large standard deviation up to 5° C, indicating a large uncertainty at annual level (see fig. 3). This corresponds to the region for which a large deviation with the reference datasets was found during the evaluation of RCMs and for which ALARO had a cold bias, while REMO had a warm bias (Top et al., 2021). Large uncertainties are also found during winter over the full domain, with the highest uncertainty in the eastern part of the domain up to a standard deviation of 10° C in northern China, Mongolia and eastern Russia. Also during spring there is a large model spread in the northern part of the region, with the largest standard deviation being up to 7° C over eastern Russia. These large uncertainties can again be linked to the deficiencies that were reported in the evaluation study of RCMs. The variation between the models is smaller during summer with values up to 3° C, except for some parts in East Europe and Lake Baikal in southeastern Russia. In autumn the RCMs had the best performance during the evaluation study (Top et al., 2021) and this results in a small standard deviation up to 2° C between the models over most of the region, except for northern China, Mongolia and eastern Russia. This makes the resulting significant regional warming during autumn very robust and likely to occur. The variation between the model outcomes of the different GCM-RCM members is small during the seasons and over the regions for which the two RCMs performed well (Top et al., 2021). This is remarkable since the different ensemble members could project different future climates. The deficiencies that were reported during the evaluation of RCMs lead to a larger spread between the model outcomes for the future projections, which causes an increase in uncertainty. Therefore, the RCMs should be improved or a bias adjustment should be carried out to reduce the uncertainty in the future projections. — 70 — Fig. 2. — Spatial variability in the annual and seasonal temperature means for the historical period (1976-2005) and temperature change under 1.5° C, 2° C and 3° C global warming showing the effect of 0.81° C, 1.31° C and 2.31° C global warming with respect to the historical period shows that additional global warming causes a significant additional warming over the Central Asia domain (based on Top, 2022). — 71 — Fig. 3. — Spatial variability in the annual and seasonal spread, quantified by the standard deviation, between temperatures of the different ensemble members and RCP scenarios under global warming of 0.81° C, 1.31° C and 2.31° C with respect to the historical period (1976-2005) or 1.5° C, 2° C and 3° C global warming with respect to the pre-industrial baseline period (based on Top, 2022).