Future projections of seasonal temperature and precipitation for India

Ninety climate models, from four consortiums—CMIP5, CMIP6, NEX-GDDP, and CORDEX—are evaluated for the simulation of seasonal temperature and precipitation over India, and subsequently, using the best ones, their future projections are made for the country. NEX-GDDP is found to be the best performer for the simulation of surface air temperature for all the four seasons. For the simulation of precipitation, CMIP6 performs the best in DJF and MAM seasons, while NEX-GDDP performs the best in JJAS and ON seasons. The selected models suggest that temperature will increase over the entire Indian landmass, relatively more over the north-western part of the country. Furthermore, the rate of warming will be more in winter than in summer. The models also suggest that precipitation will increase over central eastern and north-eastern India in the monsoon season, and over peninsular India during post-monsoon months.


Introduction
Climate change is the greatest threat facing humanity. The unabated and unchecked rise in global temperature, mainly due to human influence, will render severe and lethal extremes in the form of blazing heatwaves, frequent and pervasive droughts, record floodings, intense storms, and among others (Hirabayashi et al., 2013;Lee and Marotzke, 2021;IPCC, 2022). India, where the indication of an impending climate change is already apparent and is expected to get even worse in an ever-warming world, is among the most vulnerable regions where a large population depends on climate sensitive sectors like agriculture, fisheries, animal husbandry, human health, tourism etc (Krishnan et al., 2020). Given the grave vulnerability of the region to climate change's consequences, it is highly imperative to have credible records of future changes to aid effective adaptation and mitigation efforts.
Global climate models (GCMs) are essential for understanding various climatic behaviors and providing future climate predictions. Since the inception of climate modeling, several improvements have been incorporated progressively, which have enhanced our understanding of the existing climate over time and provided more reliable future predictions. However, despite huge advances, GCMs are often faced with difficulties when it comes to simulating the local and regional climatic features and as a result hinders the regional impact assessments (Mishra et al., 2018a;Rana et al., 2020) imperative for formulating informed adaptation and mitigation efforts. Several studies showed an unsatisfactory performance of global models over India (Mishra et al., 2018a;Jain et al., 2019a), especially for the precipitation, and reported the presence of large biases in their simulations. These biases in global models are often attributed to poorly resolved physical processes, coarser resolutions, oversimplified parameterization of complex climate-relevant processes (Wei and Qiao, 2017;Mishra et al., 2018a), among others. To overcome this difficulty and to make impact assessments at local and regional scales more promising, one of the approaches is the downscaling of output from global models to generate a higher resolution picture. There are two downscaling methods such as statistical and dynamical downscaling. Dynamical downscaling uses regional models often driven by global model output to generate region-specific climate information (Giorgi et al., 2009), whereas statistical downscaling uses statistical techniques to produce high resolution climate information. One of the international efforts to obtain dynamically downscaled products is the Coordinated Regional Climate Downscaling Experiment (CORDEX, Giorgi et al., 2009), which provides finer-scale regional projections. The CORDEX datasets showed promising improvement in simulating several climatic features for many countries (Endris et al., 2013), but for India, no significant improvements are exhibited, although efforts have been made by several authors (Mishra et al., 2018b;Jain et al., 2019a). Previous regional climate modeling studies showed useful results at the levels of homogenous zones and cities (Dash et al., , 2015. Based on some previous studies, one can infer that some of the dynamically downscaled climate products are useful, even at the city level (Dash et al., , 2015Pattnayak et al., 2013). However, there are questions about the robustness of those results. Keeping this in view, the single model studies or multi-model mean (MMM) from a single set of experiments such as CIMP5 and CMIP6 (Menon et al., 2013;Sharmila et al., 2015;Moon and Ha, 2020) are not adequate to be used to provide reliable future projections, particularly at the regional scales. Therefore, taking the MMM based on models from different consortiums, such as CMIP5, CMIP6, CORDEX, and NEX-GDDP, is one solution, which not only facilitates intercomparison but also facilitates quantification of uncertainties in a way to produce reliable projections. Considering the heterogeneity of the Indian climate, it is very essential to examine the state-of-the-art downscaled climate products for their societal use at the regional and local levels. Similarly, several statistical downscaling methods have been used to generate high-resolution features. One such product includes the National Aeronautics and Space Administration (NASA) Earth Exchange Global Daily Downscaled Projections (NEX-GDDP, Thrasher et al., 2012). The dataset has been featured in several studies and reported to have alleviated major biases in temperature and precipitation simulations, including over India (Jain et al., 2019b;Sahany et al., 2019). More details of these datasets are given in the Data and Methodology section.
Overall, each dataset has its own advantages and disadvantages; hence, while generating future projections, it is imperative to assess the capacity of different datasets to reproduce the observed features. This study is one such work in that direction. The objectives of this study are (i) to evaluate the performance of CMIP6, CMIP5, NEX-GDDP, and CORDEX in simulating the seasonal surface air temperature and precipitation over India with respect to observations during four different seasons, namely, December to February (DJF), March to May (MAM), June to September (JJAS), and October and November (ON), and to select the most suitable product that shows better performance, among others, in simulating surface air temperature and precipitation during each of the seasons, and (ii) to generate future projections for each of the seasons from the selected models under the most aggressive emission scenarios.
The National Aeronautics and Space Administration (NASA) Earth Exchange Global Daily Downscaled Projections (NEX-GDDP) constitute a set of global, high-resolution, bias-corrected data that has been statistically downscaled from the CMIP5 model outputs. The bias correction is done by utilizing the global climate data from the Global Meteorological Forcing Dataset (GMFD) provided by the Terrestrial Hydrology Research Group at Princeton University (Sheffield et al., 2006) using the bias-corrected spatial disaggregation (BCSD) method (Wood et al., 2002(Wood et al., , 2004Thrasher et al., 2012). The CMIP5 model data are first compared with the GMFD data for the period 1950-2005 using quantile mapping, and the corrections obtained are used to adjust the future projections. The NEX-GDDP daily data from the same 21 CMIP5 models are used for 1975-2005 (historical) and -2099Thrasher et al., 2012). We also used (Tables 1, 2) the CORDEX (10 models), CMIP5 (21 models, the same models as NEX-GDDP), CMIP6 (38 models), and NEX-GDDP (21 models) to compute the Multi-model ensemble means (MMMs). The surface air temperatures from IMD, CORDEX, CMIP5, and CMIP6 are re-gridded to NEX-GDDP (at 0.25 • × 0.25 • ) resolution for comparison. The statistical significance of Pearson's correlation is determined by a two-tailed Student's t-test that is used to test the null hypothesis. The trends in temperature and precipitation are calculated using the Theil-Sen estimator (Theil, 1950;Sen, 1968), and the significance is estimated using the Mann-Kendall test (Mann, 1945;Kendall, 1975).

Results
The performance of models in simulating area-averaged surface air temperature (hereafter temperature) and precipitation over the four seasons, their root mean square error (RMSE) and pattern correlation coefficient (    For precipitation ( Figures 1E-H), CMIP5 and CMIP6 exhibit higher and comparable PCC and RMSE values during DJF and MAM seasons; however, given the sizeable amount of improvement incorporated in CMIP6 (Volodin and Gritsun, 2018;Tatebe et al., 2019;Wu et al., 2019;Bi et al., 2020;Boucher et al., 2020) and several studies reporting the improved skill of CMIP6 over India (Gusain et al., 2020;Choudhury et al., 2022), we thereby found that CMIP6 is a better option for both seasons. During JJAS and ON, it is clear that NEX outperforms the other MMMs in terms of PCC and RMSE values, providing it with a better choice for generating future projections during these two seasons.
After selecting suitable MMMs for each of the seasons in simulating the historical mean temperature and precipitation, we go ahead to present the future projections over India. Figure 2 presents the spatial pattern of trends in temperature and precipitation during all the four seasons for the period 2015-2099 under the RCP8.5 emission scenario for CMIP5 products and SSP5-8.5 for CMIP6 products. Stippling denotes trends significant at 99% level.
The temperature trends for 2015-2099 (RCP8.5 scenario) using the NEX data are shown in Figures 2A-D. A pervasive warming trend is shown during all the four seasons, increasing progressively in the higher latitudes. The maximum warming is noted over the north and northwestern part of India during all the seasons. It can be seen that the maximum warming trends over the northern Himalayas (>0.7 • C decade −1 ) and minimum warming trends (0.4-0.55 • C decade −1 ) over the peninsular India during all seasons. During DJF and MAM, relatively maximum warming trends (>0.6 • C decade −1 ) are observed over the northwest and parts of central northeast regions, with a moderate warming rate (0.55-0.65 • C decade −1 ). During JJAS, relatively minimum warming trends are observed compared to other seasons, with a higher magnitude of trends over the northern Himalayas (>0.7 • C decade −1 ), moderate over northwestern, central northeast, and parts of Indian peninsular regions (0.4-0.6 • C decade −1 ). During ON, the maximum rate of warming is seen over the northern Himalayas and parts of north-western regions (>0.6 • C decade −1 ), moderate warming over remaining parts of the north-western region, central northeast regions (0.5-0.6 • C decade −1 ), and a minimum rate over the peninsular region (<0.5 • C decade −1 ).
The future trends in precipitation under SSP5-8.5 and RCP8.5 scenarios are shown in Figures 2E-H; for DJF and MAM, we have selected CMIP6 as a better candidate (justified aforementioned), whereas for JJAS and ON, NEX MMM is selected for credible future projections. The monsoon precipitation is expected to increase significantly almost over the entire Indian landmass except for parts of north-western regions and parts of the northern Himalayas, whereas, for other seasons, a significant increase in precipitation is expected over the limited parts of India. A uniformly increasing precipitation trend (0.1 mm day −1 decade −1 ) over most parts of the country during DJF is also noted, accompanied by a significantly increasing trend over the northeast (0.1-0.3 mm day −1 decade −1 ) during MAM. During JJAS, the north-eastern region exhibits a higher rate of increase (>0.5 mm day −1 decade −1 ), central northeast regions and Western Ghats a moderate rate of increase (0.3-0.5 mm day −1 decade −1 ), and peninsular regions a lower rate of increase (0.1-0.2 mm day −1 decade −1 ). During ON, significantly increasing trends are observed over the peninsular region (0.3-0.5 mm day −1 decade −1 ) and parts of west central and central northeast regions (0.2-0.3 mm day −1 decade −1 ). The remaining regions show weak insignificantly increasing trends. Figure 3 presents the time series of weighted area averaged temperature and precipitation projections during the four seasons from their corresponding suitable MMMs for the period 2015-2099. The solid lines indicate the time series from MMMs, and the shading represents the intermodel standard deviation.
Temperature ( Figures 3A-D) will steadily increase in the future during all seasons. The rate of warming is by and large the same during all the seasons except for JJAS which exhibits a relatively lower rate. The warming level at end of the twenty-first century is ∼5 • C during all the seasons except for monsoon season, which shows a relatively lower warming (4 • C).
Unlike temperature, precipitation exhibits a higher intermodal standard deviation ( Figures 3E-H), hence a higher uncertainty in the projected values. Precipitation increases during all seasons, projecting a higher precipitation by the end of the twenty-first century. During DJF and MAM, CMIP6 projects an increase in precipitation (0.2-0.3 mm day −1 ); however, during DJF, the rate of increase in precipitation is relatively low. Among all the seasons, JJAS exhibits a higher increase by ∼2 mm day −1 . During ON, a large interannual variability can be seen, indicating that the region would encounter more erratic rainfall during that season. The projected increase for that season would be ∼0.5 mm day −1 by the end of the century.

Discussion and conclusions
The performance of major modeling consortiums (CMIP5, CMIP6, NEX-GDDP, and CORDEX) to represent surface air temperature and precipitation across the Indian landmass over four seasons (DJF, MAM, JJAS, and ON) has been examined. Following that, the products that best represent temperature and precipitation for each season were selected, and future projections were created using the models selected. We find that for temperature, the NEX-GDDP showed a better agreement with the observations in simulating the historical  temperatures during all the four seasons. For precipitation, CMIP6 had the best performance during DJF and MAM, but during JJAS and ON, NEX-GDDP showed far better agreement with the observations. We find that the Indian subcontinent will experience widespread warming with stronger warming over the higher latitudes. Maximum warming will be over the northwestern parts of the country, whilst the peninsular region will experience relatively lesser warming. Among all seasons, the monsoon season will witness the least warming. Similarly, future precipitation projections reveal that monsoon precipitation is expected to increase significantly over most parts of the Indian landmass.
We find that the temperature will increase steadily in future during all the seasons, with maximum warming of ∼4-5 • C by the end of the current century. Similarly, future precipitation projections show an increasing trend during all four seasons, with the monsoon season experiencing the maximum increase in precipitation by ∼2 mm day −1 . Rising mean temperatures may lead to an increase in high-temperature extremes in future and an increase in both mean and variability of precipitation might have a strong influence on changes in high-precipitation extremes. These findings would be useful for future planning.
The IMD data can be found at this link (http://www.imd.gov. in/).

Author contributions
PS and SM conceived the idea and designed the work. PS processed the data and created the figures. PS and NK performed .
/fclim. . the analysis and wrote the manuscript with the inputs from all the co-authors. All authors contributed to the article and approved the submitted version.

Funding
The research is partly supported by the DST Center of Excellence in Climate Modeling at IIT Delhi and the YES Foundation.