Climate Projections Over Different Climatic Regions of Afghanistan for Shared Socioeconomic Scenarios

Mohammad Naser Sediqi (  mohammad.naser.sediqi.p4@dc.tohoku.ac.jp ) Tohoku University Graduate School of Environmental Studies: Tohoku Daigaku Daigakuin Kankyo Kagaku Kenkyuka https://orcid.org/0000-0002-3569-161X Vempi Satriya Adi Hendrawan Tohoku University Graduate School of Environmental Studies: Tohoku Daigaku Daigakuin Kankyo Kagaku Kenkyuka Daisuke Komori Tohoku University Graduate School of Environmental Studies: Tohoku Daigaku Daigakuin Kankyo Kagaku Kenkyuka


Introduction
Climate change is a major environmental concern globally, particularly in dry areas (Rao et  Afghanistan the 12th vulnerable country to climate risk. The average temperature of Afghanistan showed a rise by 0.13 to 0.29°C/decade in the last fty years (Savage et al., 2009). The country has also experienced the impacts of climate change, particularly prolonged droughts and severe water stress in recent years (Muhammad et al., 2017). In Afghanistan, 98% of water resource is used for agriculture, and more than 80% of the population get their income from agricultural practices (Sediqi et al., 2019). Therefore, a sharp reduction of available water resources has severely affected the agriculture and agriculture-dependent population and economy.
Understanding the possible climate changes is crucial to antedate forthcoming water strain and resulting consequences in agroeconomy. Climate change assessment in Afghanistan has several speci c challenges (1) lack of su ciently dense, long term and reliable historical meteorological records due to the four last decades of insecurity and civil war, and (2) geographical characteristics, complex topography, and different climatic regions (Savage et al., 2009). Therefore, these make it challenging to assess climate change implications in the country.
Several tools are available to reproduce historical climate or project future climate changes. Global Climate Models (GCMs) of Coupled Model Intercomparison Project (CMIP) is one of the fundamental tools to reproduce the climate pattern Salman et al., 2018). The GCMs simulate climate considering several scenarios of carbon emissions, and land use and socioeconomic alterations (Dessai & Hulme, 2007). The global carbon releases and socioeconomic changes are related to the national, regional, and global political strategies. Afghanistan has two distinct regions, mountains and deserts, with completely different climates on annual and seasonal scales. The present study attempted to analyze the historical and future climate simulations for different climatic regions in Afghanistan.  (Ghulami, 2018). However, those were based on GCMs of CMIP5 and representative concentration pathway (RCP). No studies have been conducted yet to assess historical and future climate over Afghanistan using CMIP6, the latest CMIP dataset phase with around 55 GCMs (Iqbal et al., 2021). Compared to CMIP5 and CMIP3 datasets based on radiative forcing and carbon releases, the CMIP6 employs a set of Shared Socioeconomic Pathways (SSP) to the updated version of coupled global climate models. CMIP6 GCMs consider different mitigation and climate change adaptation based on projected population changes, economic development, ecosystem, and social aspects. It also would reduce the bias in GCMs of the earlier phases of CMIP. Several recent studies have reported the improvement of CMIP6 compared to its earlier version (Hamed, 2021;Kamruzzaman et al., 2021;Song et al., 2021). For instance, (Zamani et al., 2020) reported that CMIP6 GCMs' ensemble outperformed CMIP5 and the relative bias of winter rainfall in all stations of northeastern Iran was much lower than CMIP5.
Other studies also revealed improvement of CMIP6 for simulation of summer precipitation over East Asian monsoon and China (Xin et al., 2020). For this reason, it's signi cant to assess the historical and future climate pattern over Afghanistan using CMIP6.
The use of many GCMs for climate projection is often not preferred due to the large uncertainty associated with each GCMs and human and computational constraints (Shiru, Shahid, Dewan, et al., 2020). Therefore, an ensemble set of GCM with a minimum range of uncertainty is chosen. There are usually two basic ways to select suitable GCM: past-performance where skilled GCMs are identi ed according to their capacity to simulate the present climate and envelopes where GCMs are chosen according to their agreement in projections (Salman et al., 2018). This study used the past-performance method, as it is most widely used in literature, for the GCM selection.
The GCMs skills are evaluated based on their ability to simulate precipitation and temperature climatology over Afghanistan at annual and seasonal scales. The selected GCMs were used for future climate projections over different climate regions of Afghanistan. The paper is organized as follows: Section 2 describes the study area and datasets. Section 3 outlines the method adopted. Section 4 presents the outcomes, Section 5 discusses the results, and Section 6 provides conclusions. The other two are the transitional periods between winter and summer. The yearly rainfall ranges between 1000 mm in the northeast (Pular tundra zone) and below 50 mm in the south (arid desert zone). The northeast region experiences the minimum average temperature (< -5°C), and the southwest (arid desert) region the highest (> 28°C). widely used for precipitation studies globally, including in Afghanistan. Figure 2 shows the maps of annual, summer, and winter average Pre, Tmx, and Tmn over Afghanistan. Figure 3 represents the seasonal precipitation and temperature of different climatic regions during the reference period   The procedure adopted in the present study for selecting, downscaling, and preparing multimodel ensemble (MME) of GCMs, and projection of Pre, Tmx, and Tmn using MME are outlined below:

Study Area And
1. The simulated historical Pre, Tmx, and Tmn of 19 GCMs for 1975-2014 were re-gridded to a common resolution of 1° × 1°, as it is near the mean resolution of the models used. The GPCC/CRU data was also aggregated to 1° × 1° resolution.
2. Compromise Programming (CP) was applied to compare GCMs with GPCC/CRU data to identify a GCM subset, according to their skill in replicating GPCC Pre and CRU Tmx and Tmn.
3. The selected GCMs were downscaled to the GPCC/CRU resolution of 0.5°. Bias in GCMs simulation was corrected using quantile delta mapping (QDM), considering the GPCC Pre and CRU Tmx and Tmn as a reference.

Changes in
where, f j is a value of statistical metric jof a GCM, f * j is the optimum value of the metric jand Pis a parameter that can have a value ≥ 1. In this study, P equal to 1 was considered. CI near to zero indicates better performance.
The CI was estimated using three statistical metrics in this study, (1) Kling-Gupta e ciency (KGE, Eq. 2), (2) coe cient (CC, Eq. The Pre, Tmx, and Tmn of the selected GCMs were interpolated to GPCC/CRU points and bias-corrected using GPCC/CRU data. The quantile delta mapping (QDM) was used for this purpose, where, Q m (t) and Q s (t) are ith bias-corrected and GCM simulated data, F s and F − 1 0 are Cumulative distribution function (CDF) of raw GCM and inverse CDF of GPCC/CRU gridded data, respectively. During the period 1975-2014, 70% of the data was used for developing the QDM model, and the rest was employed for QDM model evaluation.

Generation of GCM ensemble projection
The bias-corrected GCMs projections were used to prepare MME mean to lessen the uncertainty related to individual GCM. The MME was employed to estimate the percentage changes in Pre and the absolute changes in Tmx and Tmn in two future time horizons, near  Table 3 presents the GCMs' skill in replicating Pre, Tmx, and Tmn based on three statistics. The GCMs' skill based on each variable is shown in (Fig. 4). The results showed good performance of some models in simulating one variable but low for another. For example, EC-Earth3-Veg showed good skill in simulating Pre (blue) but low for both Tmx (red) and Tmn (green). Therefore, RM was employed for the nal ordering of the GCMs. Black colored bars in (Fig. 4) show the GCM ranking based on RM. The results showed MPI-ESM1-2-LR as the most skilled model in replicating Afghanistan's observed climate, followed by ACCESS-CM2 and FIO-ESM-2-0. Therefore, these three GCMs can be considered best suited for the climate projection of the country.  Figure 5 shows the bias in MME mean Pre, Tmx, and Tmn of selected GCMs against the observation GPCC rainfall and CRU temperature before and after the bias correction for the reference period . The raw ensemble GCMs show an overall bias of -40 to 55% in Pre, where Pre was overestimated in most regions except a small area in the north (Fig. 5a). On the other hand, Tmx and Tmn showed a high negative bias or underestimation in the highlands and a high positive bias in desert regions (Fig. 5c,e). The QDM approach reduced the bias associated with the selected GCMs. The results indicate bias in Pre after applying QDM was in ±20% (( Fig. 5b,d,f). However, in most regions, the bias was only ±10%. Similarly, biases in Tmx and Tmn were reduced to ±1.6ºC after applying QDM.

Projected change in climate
A multimodal mean ensemble (MME) was developed from the three highest-ranked models. Figure 6presents the geographical distribution of the projected changes in annual Pre for two future periods under different SSPs estimated using MME mean. The Figure   6 shows changes in Pre in the range of -40 to more than 60 % for different SSPs for the wo periods. Desert areas were projected with the highest decrease, while the highest increase was projected in the highlands, especially for the late period under SSP5-8.5. Figure 7 shows the changes of Tmx over Afghanistan. The results showed an increase in Tmx for all scenarios and time horizons by 1.3 to 5.3 ºC. The lowest increase was projected for SSP1-2.6 in the country's south in the early period, while the highest increase was projected for SSP5-8.5 in the late period in the northeast. The spatial pattern in the change in Tmn over Afghanistan revealed a similar increase to Tmx (Fig. 8). The projected increase was in the range of 1.3 to 5.4 ºC.

Climate projections over different climatic regions
The projection of seasonal Pre, Tmx, and Tmn for different scenarios in each climatic region during the far future are shown in (Fig. 9).
The results showed increased temperature and decreased precipitation in all climate zones for different scenarios, except zone 1 (Hindu Kush region). An increase in Pre is projected in zone 1 during the rainy seasons.
The future projection of Tmx and Tmn in different climate zones for three SSPs are shown in (Fig. 10) and (Fig. 11), respectively. The line in each gure shows the mean projection, and the band represents the projections with a 95% con dence interval. The gures show a gradual increase in both Tmx and Tmn for all scenarios in all regions. The highest increase in Tmx and Tmn was projected in the highland region between 5.3 and 5.4 ºC for SSP5-85 at the end of the century. This may lead to an increase in snow melting stored in high elevated mountains. Besides, a high increase in Tmx and Tmn was also projected in zone 5 (southern plateau).

Discussions
Performance . It is expected that comparison using mean resolution provided the best estimation of models' performance.
The GCMs skill is estimated considering the capability in simulating the mean, dispersion, and distribution of observed climatology. They should also be capable of replicating a similar climatological pattern (Xin et al., 2020). Therefore, multiple metrics are required to fairly assess GCM's skills. In this study, KGE, CC, and SS were employed to determine GCM's skill in reconstructing observed rainfall and temperature patterns with less bias and similar distribution over 1975-2014. The use of a robust algorithm (CP) to integrate performance metrics has helped to provide an unbiased assessment of GCM's performance.
GCMs should provide skill in simulating both rainfall and temperature (Tmx and Tmn) together, as these three variables are needed for climate change studies (Hamed, 2021). For example, assessing climate change implications on oods requires precipitation and temperature to estimate snowmelt. This study considers these three climate variables together for GCM ranking. Therefore, the selected GCMs can be recommended for Afghanistan's climate change effect evaluation and adaptation planning.
The present study showed a possible precipitation reduction in most of Afghanistan. A dry climate dominates the country, and therefore, water is a major hindrance to crop agriculture and food security. The decrease in rainfall would make the situation worse. The precipitation in the country is projected to reduce more in low rainfall regions and increase in the areas that receive the highest rainfall. This indicates the dry region would become drier and the wet region wetter in the future. This would increase aridity and crop failure risk in the dry region, which covers most of the country.
The results showed a higher rise in temperature in the Hindu Kush region for all SSPs and time horizons. The precipitation in the wet (summer) season is also projected to increase in the region. Increased temperature may lead to an increase in snow melting stored in high elevated mountains. This may cause a large increase in river discharge in the summer. Increased rainfall would aggravate the condition and cause a sharp rise in oods in the region. The northeastern Hindu Kush are is most susceptible to hydrological disasters like oods in Afghanistan (OCHA, 2016) Climate change may increase ood frequency and severity in the region in the future.

Conclusions
This study assessed the skill of 19 CMIP6 GCMs in simulating present precipitation and temperature over different climatic regions of Afghanistan. Compromise programming based on three robust statistical metrics and a rating metric was employed to judge GCMs' performance. The study identi ed three GCMs as most suitable for climate projections over Afghanistan. The MMEs of selected GCMs revealed a large rise in maximum and minimum temperature for all regions for SSPs. In contrast, both increase and decrease in precipitation over the country were projected. A higher increase in precipitation was projected in the highlands and a decrease in the