Soil moisture revamps the temperature extremes in a warming climate over India

Soil moisture (SM) plays a crucial role in altering climate extremes through complex land-atmosphere feedback processes. In the present study, we investigated the impact of SM perturbations on temperature extremes (ExT) over India for the historical period (1951–2010) and future climate projection (2051–2100) under 4 K warming scenario. We note that more than 70% area of the Indian landmass has experienced significant changes in characteristics of ExT due to SM perturbations. In particular, we see larger impact of SM perturbations on ExT over the north-central India (NCI), which is a hotspot of strong SM-temperature coupling. Over NCI, a 20% departure in SM significantly revamps frequency, duration and intensity of ExT by 2–5 events/year, 1-2 days/event and 0.5–2.1 °C, respectively, through modulating surface energy partitioning, evapotranspiration and SM memory. Importantly, the impact of SM perturbations on frequency and duration of ExT events becomes less prominent with intensification of global warming.


Introduction
Major land regions of the world have been exhibiting a severe rise in temperature extremes (ExT) during the past few decades (Perkins et al., 2012). Modelling studies for the future climate also reported a prominent increase in the extreme temperature characteristics all around the world (Brown, 2020;IPCC, 2018;Sanjay et al., 2020). The Indian subcontinent is highlighted as one of the hotspots for exacerbated ExT conditions during the last few decades (Perkins-Kirkpatrick and Lewis, 2020;Satyanarayana and Rao, 2020). Recent studies have projected that a spate of red-hot extremes is likely to become more common over the Indian region at the end of the 21 st century (Das and Umamahesh, 2021;Krishnan et al., 2020;Mizuta et al., 2017;Murari et al., 2015;Rohini et al., 2019). Hot weather conditions over India generally persist during the pre-monsoon months of March, April, and May (De et al., 2005;Kothawale et al., 2010). These extreme heat conditions may extend during the Indian summer monsoon months as well as the post-monsoon season due to the prolonged monsoon break like conditions and land-surface relevance (Raghavan, 1966;Dimri, 2019;Ganeshi et al., 2020). With large spatio-temporal variability over the Indian region, warm extremes exert a serious impact on the ecosystem, human health, agriculture and economy (Im et al., 2017;Patz et al., 2005).
A primary cause for enhancing the frequency, duration and intensity of ExT is anticipated to the global warming, as a response to anthropogenic climate forcing (Baldwin et al., 2019;IPCC, 2021IPCC, , 2014Krishnan et al., 2020). The physical mechanism involved in the formation of such ExT is linked to large-scale atmospheric dynamics as well as regional-scale land-atmosphere interactions (De and Mukhopadhyay, 1998;Ganeshi et al., 2020;Ghatak et al., 2017;Joshi et al., 2020;Ratnam et al., 2016). The atmospherically driven phenomenon such as quasi-stationary Rossby wave-train and El Niño Southern Oscillation originated from the North-Atlantic Ocean and tropical Pacific, respectively, in-turn maintains the warm and dry air thermodynamically at the surface, through the high-pressure system. On the other hand, the land-atmosphere coupling can be a dominant aspect for explaining the processes underlying extreme heat magnitude, duration and severity (Fischer et al., 2007;Ganeshi et al., 2020;Gevaert et al., 2018).
Soil moisture (SM) has received a great deal of attention in weather and climate sciences.
As a crucial component of the land-atmosphere interaction, SM also acts as temporal storage of atmospheric anomalies (Delworth and Manabe, 1988;Entin et al., 2000;Koster and Suarez, 2001;Wu and Dickinson, 2004a). This behaviour of the soil to persist the moist and dry conditions caused by atmospheric forcing is called soil moisture memory (SMM) or soil moisture persistence (Delworth and Manabe, 1988;. This long-term persisting nature or memory of SM is associated with its low-frequency variability (Manabe and Delworth, 1990). Moreover, the low-frequency SM variability has the potential to induce the most pronounced impact on nearsurface temperature and precipitation variability (D'Andrea et al., 2006). Therefore, as an essential variable of the climate system, the study of soil moisture variability provides key insights into the observational and modelling aspects of land-atmosphere feedback processes (Ganeshi et al., 2020;Miralles et al., 2012;Mujumdar et al., 2021;Seneviratne et al., 2010).
The influence of SM on near-surface temperature is mainly determined through limiting surface energy partitioning and evapotranspiration (Jaeger and Seneviratne, 2011). Several methods were proposed earlier to find the role of land-atmosphere coupling on ExT (Gevaert et al., 2018;Miralles et al., 2012;Seneviratne et al., 2010). One way of exploring the SM impact is by analysing composites of wet and dry land surface states associated with ExT (Rohini et al., 2016). The second approach uses the Generalized Extreme Value (GEV) theory for assessing changes in ExT, by fitting SM as covariate (Ganeshi et al., 2020;Whan et al., 2015). Such a stationary or non-stationary GEV method estimates return values of temperature maxima with or without considering SM as a covariate. The third way of understanding the SM impact is through performing the sensitivity experiment using the climate model run (Erdenebat and Sato, 2018;Seneviratne et al., 2006). These simulations allow us to investigate the one-way influence of SM on the atmosphere since SM feedback to the atmosphere is removed by controlling SM variability.
Here, we utilize the novel method of SM sensitivity experiments over the Indian region to assess the role of wet and dry SM conditions on extreme temperature variability. The model experiments are carried out using the high-resolution (~60 km) version of MRI-AGCM3.2 (sub-section 2.1.1).
Both wet and dry experiments are performed in this study as they depict the potential to reinforce the climate extremes (Liu et al., 2014).
Detailed understanding of causes for hot extremes and its increasing trend over India in past and future climate has been a topic of concern since the 1980s (Dimri, 2019;Im et al., 2017;Patz et al., 2005). The existing knowledge conveys that ExT occurrences over India have been discussed in the context of large-scale atmospheric dynamics and regional-scale land-atmosphere interactions, where the associated dry SM conditions are also highlighted as one of the contributing factors (Ratnam et al., 2016;Rohini et al., 2016;U S De and Mukhopadhyay, 1998).
However, the impact of SM-T coupling on ExT is not exclusively understood due to insufficient spatio-temporal SM observations and, the difficulty in filtering the impact of climate extremes on SM with reanalysis and observational datasets. State-of-the-art model sensitivity experiments are being used to overcome the aforementioned issues and investigate the sensitivity of climate extremes to the SM variability over major areas of the world (Erdenebat and Sato, 2018;Seneviratne et al., 2006). In a review of SM sensitivity experiments over the Indian subcontinent, researchers mostly explored the soil moisture-precipitation feedback mechanism using the model simulations (Asharaf et al., 2012;Raman et al., 1998;Shukla and Mintz, 1982). However, indepth investigation of SM-T coupling and its role on the ExT over the Indian region is still warranted. Therefore, the present study aims to unravel the underlying role of SM-T coupling on ExT by performing high-resolution (~60 km) SM sensitivity experiments from the MRI-AGCM3.2 model (refer to sub-section 2.1.1 for more details).
The outline of this paper is as follows: Section 2 presents the model simulation experiments, data and methodology used in this study. Section 3 includes a detailed description of strong land-atmosphere coupling regions, ExT, and the impact of SM on ExT over the Indian region analyzed using model simulations for historical  and future climate (2051-2100). Finally, section 4 summarizes the main findings of the study.

2.
Data and methodology  (Kanamitsu et al., 1983). For time integration, a two-time level semi-implicit Semi-Lagrangian scheme with the increase in computational stability is used in MRI-AGCM3.2 (Yoshimura and Takayuki, 2005). A new cumulus parameterization scheme (Yoshimura et al., 2015) is introduced in MRI-AGCM3.2 based on a scheme by Tiedtke (1989).
The radiation scheme used in the model is similar to the Japan Meteorological Agency's operational model, except with the treatment of aerosols. MRI-AGCM3.2 version includes absorption due to water vapour (line and continuum absorption), carbon dioxide (in the 15 mm band, near-infrared region, etc.), and ozone (in the 9.6 mm band, visible and ultraviolet region).
The model also deals with absorption due to methane (CH4), dinitrogen monoxide (N2O), and chlorofluorocarbons (CFCs) in the longwave scheme, for considering their greenhouse effect.  (Hirai et al., 2007). For the boundary level mixing scheme, the Mellor-Yamada level 2 turbulence closure scheme (Mellor and Yamada, 1974) is implemented in a model. For more details of MRI-AGCM3.2, please refer to Mizuta et al., (2012).
MRI-AGCM3.2 is extensively used in the investigation of numerous studies such as extreme events, simulating global precipitation and its change in future climate, and reproduction of tropical cyclones (Kusunoki, 2017;Mizuta et al., 2012). Furthermore, model validation with IMD observations and various assimilation products illustrate the least bias in the simulations corresponding to the SM-T coupling characteristics (described in subsection 3.1). As the model biases are not critical, the study further performs the soil moisture sensitivity experiments to understand the impact of SM on ExT.

Experimental setups
In the present study, we have conducted six long-term high-resolution MRI-AGCM3.  The sensitivity experiment initialized by decreasing the soil moisture on 1 st day of each month by 20% in FUT.

2051-2100 (50 years)
WET-SM (HIST+20) The sensitivity experiment initialized by increasing the soil moisture on 1 st of each month by 20% in HIST and FUT.

1951-2010 (60 years)
WET-SM (FUT+20) The sensitivity experiment initialized by increasing the soil moisture on 1 st of each month by 20% in FUT.

Other data products
With the focus on the ExT and land-atmosphere interaction, we have validated and corrected the model simulations using the observational and data assimilation products over the Indian region.
Observational data include mean temperature, maximum temperature and precipitation data sets prepared by India Meteorological Department (Pai et al., 2015;Srivastava et al., 2009). Two data assimilation products are used in the present study: 1) Global Land Data Assimilation System shown that GLDAS outputs are in good agreement with observations over the Indian region (Ganeshi et al., 2020;Mujumdar et al., 2021b;Sathyanadh et al., 2016). Out of four landsurface models driven by GLDAS, the present study uses the Noah land Surface Model version3.3 (Noah LSM 3.3) forced by the global meteorological forcing dataset (Sheffield et al., 2006).
Furthermore, the high resolution (~ 4 km) data generated by using LDAS (version 3.4.1) is explored here to validate and remove the bias in the SM over the Indian region (Nayak et al., 2018).
The LDAS version is mainly designed for both coupled and uncoupled modes within the Weather Research and Forecasting model. More details of the LDAS high-resolution SM data product can be found in Nayak et al. (2018).

Soil moisture-temperature (SM-T) coupling
Determining the SM-T coupling is an important factor for assessing the impact of SM on ExT (Seneviratne et al., 2006). Based on the model experiments and analytical techniques several methods were discussed earlier to evaluate the coupling strength between SM and surface temperature (Dong and Crow, 2018;Miralles et al., 2012;Seneviratne et al., 2013). In the present study, quantification of surface temperature sensitivity to the SM change (SM-T coupling strength) has been carried out using the method suggested by Dirmeyer (2011). Here, we are using a similar method for surface temperature sensitivity instead of surface energy fluxes. The method described by Dirmeyer (2011) overcomes the shortcomings in the correlation method by considering the variance of SM at each grid point. SM-T coupling metric used in this study is given in equation 1.
In this method, we estimate the linear regression slope (ℛ ) of temperature anomaly on the SM anomaly. Furthermore, the coupling metric is determined by multiplying the negative value of soil moisture standard deviation ( ) to regression slope (ℛ ).

Extreme temperature indices
Extreme temperature conditions can be detected using various criteria depending upon the different climate zones (Nairn and Fawcett, 2013 for each year at each grid point.

Generalized Extreme Value (GEV) distribution
The quantitative description of the impact of SM on ExT is carried out in the present study using the statistical approach of GEV theory (Whan et al., 2015). Block maxima perspective is demonstrated to fit GEV to the yearly maximum temperature (ExTI) at each grid point, with and without considering SM as a covariate. The extReme software package of R-programming to perform the GEV fit is used in this study (R Core Team, 2016;Gilleland and Katz, 2016). The GEV analysis is carried out in two important steps. The first step of analysis consists of stationary GEV fit for block maxima of each year with no covariate (given below in equation 2). Whereas, the second step consists of a non-stationary GEV model fit to ExTI with the inclusion of SM as a covariate. The GEV analysis is dependent upon the three important parameters: 1) scale parameter ( ), 2) location parameter (μ), and 3) shape parameter (ξ). Scale parameter ( ) explains the variability in the dataset, mean of the fitted distribution is represented in terms of location parameter(μ) and the shape of the fitted distribution is shape parameter (ξ) (i. e. Gumbel:ξ = 0, Frechet:ξ> 0, Weibull:ξ< 0). The GEV analysis is carried out on the area-averaged value of ExTI and SM over the hotspot of strong land-atmosphere coupling (NCI).
where, y is standardized annual mean SM and, 0 and 1 are fitting constants for location parameters.
The negative value of the estimated shape parameter for the HIST and FUT experiment indicates the ExTI follows the Weibull distribution (Table 2 and 4). The significance for perfect stationery and non-stationary GEV fit using the likelihood-ratio-test (LRT) showed a good fit for ExTI distribution and inclusion of SM as covariate dominantly improves the model fit. In the GEV analysis impact of SM on extremes is documented by analyzing differences between 50-year return levels of ExTI from DRY-SM and WET-SM sensitivity experiments for historical and future climate.

Soil moisture memory (SMM)
Property of the soil to remember wet or dry anomalies caused by atmospheric forcing is generally termed as soil moisture memory (Delworth and Manabe, 1988;Wu and Dickinson, 2004b). The present study measures the SMM in terms of time-scale lag at which the autocorrelation drops to 1/e (e-folding time scale) of its value (Ganeshi et al., 2020). The method of e-folding time scale is based on the 30-day lag autocorrelation values of soil moisture anomalies considering the exponential decay of SM autocorrelation function (equation 4).
where r (τ) is the autocorrelation function, τ is the lag and λ is called decay time-scale. SMM analysis is also extended to the WET-SM and DRY-SM experiments to explore the impact of SMM on ExT.

Validation of model simulations with observational and assimilation data sets
The present study uses three different types of data sets (see section 2.2) to validate six variables than the interior parts of India. An important point to be noted from Fig. 1h is that the model is having maximum cold bias over the north (bias < -5 °C) and north-east (bias< -2 °C) Indian region.
However, relatively less cold bias over the north-central and southern peninsular India shows the reliability of model data sets over the regions.
The present study uses LDAS data set to validate the SM model output. The spatial distribution of SM over India indicates a direct association with PR. Wet SM conditions are mostly observed over the Western Ghats, north-east India and north Indian region (Fig. 1o).  (Fig. r). On average, MRI-AGCM3.2 underestimates ET over the Indian region ( Fig. 1 l). The Western Ghats, north-central, north and north-east Indian regions indicate the larger negative bias of -1 to -2 mm/day. Model simulation mostly shows larger values of ET over the transitional region where a sufficient amount of SM strongly constrains ET variability due to available radiational energy (Fig. 1r). In agreement with a previous study by Ganeshi et al. (2020), the transitional regions are mostly located over central India. On the other hand, less ET over the north-west and north Indian region is observed due to the SM and energy limiting conditions, respectively.
Describing other two fluxes, in which latent heat flux (LHF) is closely associated with the ET process and the SHF has the dominant influence from radiational energy (Fig. 1 p and q). Spatial distribution of biases SHF and LHF showing the contrast behaviour with respect to each other ( Fig.1 j & k).
The present study uses a monthly bias-corrected method by Soriano et al. (2019) to correct bias in the model outputs. Bias corrected climatological features of all the variables are shown in Fig. 1.
The result indicates that the method is able to reduce the bias induced in the model. Comparison of bias-corrected fields from MRI-AGCM3.2 with observational and reanalysis data sets is carried out with the help of Taylor statistics (Taylor, 2001). Three statistical scores (correlation, standard deviation and root mean square error) are used for the Taylor method to analyze corrected outputs ( Fig. 2). All the corrected model outputs are standardized and averaged over the Indian region before representing through the Taylor statistics. The result shows a significant improvement (with correlation > 0.7 and RMSE < 1) in the model performance using the monthly correction method of Soriano et al. (2019) (Fig. 2). It can be observed that SM and SHF have the highest correlation values greater than 0.9 and least RMSE (< 0.5) with respect to LDAS and GLDAS, respectively.
Whereas, the lowest correlation (0.7 to 0.8) is observed in the PR and LHF with RMSE close to 0.65.

Mean features of hydro-meteorological variables for future simulation
This section describes the mean features of six different hydro-meteorological variables (PR, Tmax, RF, SHF, LHF and ET) over India for future climate (FUT) and its change with respect to the historical simulations (HIST). FUT simulation indicates a similar spatial distribution of all hydro-meteorological variables as that of HIST simulations ( Fig. 1 and 3  Climatological spatial distribution of SM over the Indian region in future climate indicating to have a direct association with PR (Fig. 3c). With an increase in PR in future climate, the model indicates an increase of SM almost everywhere over the Indian region. FUT model simulation shows that, with an average 1 mm/day increase in PR over the Indian region, SM will likely to be increased by at least 2% as compared to HIST simulations. Wet conditions (increase in PR and SM) and warming over the Indian landmass are expected to reinforce the surface energy partitioning and ET rate in the future climate. The model result indicates that an increase in SM and temperature will be expected to lose more energy to the atmosphere in terms of LHF, through the increase in the process of ET ( Fig. 3j and l). Unlike the LHF, overall, FUT simulation resulted in less amount of heat transfer to the atmosphere by the sensible heating process as the dominance of energy released accounted during the latent cooling process (Fig.3h). The relation between SM, Tmax, SHF and LHF is highly non-linear, which can be defined using the process of landatmosphere coupling (Ganeshi et al., 2020;Miralles et al., 2012;Seneviratne et al., 2013). the results from historical climate, the coupling index underlines the NCI region as a hotspot of strong soil moisture-temperature interaction, with an increase in coupling magnitude (Fig. 4b).

Soil moisture-temperature (SM-T) coupling over the Indian region
From Fig. 4a and b, it is to be noted that the area of strong SM-T coupling is likely to expand under future warming scenarios. The expansion or shrinking of strong soil moisture-temperature coupling regions can be an important factor in the backdrop of climate change . Figure 5 shows the spatial distribution of the long-term mean extreme frequency (ExTF), duration (ExTD) and intensity (ExTI) over the Indian region for historical    Fig. S3).

Impact of soil moisture on temperature extremes in historical and future climate
In the present study, the influence of soil moisture on ExT is diagnosed using sensitivity experiments (WET-SM and DRY-SM experiments). These experiments are carried out for historical and future climates over the Indian region (see sub-section 2.1.2). In the WET-SM experiment, SM at each grid point is increased by 20% on the 1 st day of each month. Whereas, for the DRY-SM experiment, SM is decreased by 20% on the 1 st day of each month at each grid point. year (ExTF), 0-1 days per event and long-term mean intensity ~1 o C than that of FUT ( Fig. 7 b, e & h). Unlike the HIST+20 experiment, the FUT+20 simulation exerted a higher impact of wet SM for reducing ExT over the Indian region. FUT+20 result indicates that wet conditions in future reduce the ExTF by 3-4 events per decade, ExTD by 3-4 days per event and long-term mean (for period 2051-2100) intensity by ~2 o C (Fig. 7 c, f &i).
The main aim of this study is to understand the role of SM on ExT over the hotspot of SM-T coupling. Therefore, further sensitivity analysis is carried out over the strong SM-T coupling regions of north-central India (NCI). Historical results over the NCI suggest that DRY-SM (HIST-20) leads to an increase in the ExTF ~5 events per year, ExTD ~ 1.8 days per event and long-term mean ExTI ~ 0.71 o C, as compared to HIST (Fig. 8) Analysis of extremes over the NCI is also supported by the Generalized Extreme Value (GEV) theory and probability distribution approach. The yearly block maxima approach is applied to the non-stationary GEV model fitting of extreme temperature intensity (ExTI) index considering soil moisture as a covariate. Further, the SM influence on extremes is quantified using the difference between 50-years return values of DRY-SM and WET-SM sensitivity experiments. Fig. 9 shows the return level plot for yearly block maxima fit to non-stationary GEV model in the case of WET-SM (Hist+20/Fut+20) and DRY-SM (Hist-20/Fut-20) experiments. Fig. 9a  has a maximum impact on extremes in the post-monsoon season than that of pre-monsoon and monsoon seasons (Fig. 11 and Fig. 12). For historical climate, the probability of extremes with intensity greater than the 90 th percentile of PDF distribution is increased by a minimum of 0.5 o C (Fig. 11). In DRY-SM simulation (HIST-20) during the post-monsoon season, the probability of ExTI over NCI is increased by nearly 3.5 o C compared to the control run (HIST). However, the likelihood of future post-monsoonal ExTI over NCI is expected to rise by at least 4 o C (Fig. 12). In Soil moisture memory (SMM) is another crucial aspect of the climate system that affects landatmosphere interactions (Ganeshi et al., 2020;Mujumdar et al., 2021;. Therefore, in the present study, we have evaluated the SMM characteristics over the Indian region with the help of an e-folding time scale ( Delworth and Manabe (1993) has linked the SMM time-scale with the persistence of atmospheric variability and thus consecutively on near-surface temperature. In comparison to wet and dry SM regimes, moderate SM zones will experience faster evaporative damping of SM anomalies due to available radiational energy, and so have the potential to influence near-surface temperature variability. The results pointed out the decrease of the SMM time scale almost everywhere over an Indian region in the DRY-SM sensitivity experiment (Fig. 13 k). The results also highlight that the SMM time scale over the weak coupling regions will not be modified by the change in SM content due to lower sensitivity.
On the other hand, SMM time-scale behaviour is highly non-linear in WET-SM experiments over the Indian region (Fig. 13 l). Model output shows that SMM can increase or decrease with the WET-SM experiment (HIST+20). Over the north-central Indian region, dry soil moisture (HIST-20) conditions lead to reducing the persistence time scale by 1 week and wet soil moisture (HIST+20) intensifies the SMM by a few days (< 1 week). In summary, drier SM conditions over the strongly coupled region will cause the SM anomalies to dissipate faster due to evaporative damping. In principle, faster dissipation of SM anomalies favours near-surface temperature warming by reducing evapotranspiration and increasing sensible heat flux across the strongly coupled region of NCI. On the other hand, further investigation need to be carried out to study the factors responsible for non-linear behavior between SMM and WET-SM over the weak coupling zone.

Summary and conclusion
The present study evaluates the impact of soil moisture (SM) variability on long-term changes in annual extreme temperature characteristics (frequency, duration and intensity) over India for historical   SM-T coupling strength in the MRI-AGCM3.2 is investigated here using the regression method.
The spatial pattern of SM-T coupling indicates a hotspot located over the north-central Indian (NCI) region. The result from SM-T coupling is similar to the previous study by Ganeshi et al. (2020). This analysis pointed out that higher coupling strength over the region of NCI and indicates the dominant role of SM on temperature by controlling the surface energy partitioning. Finally, the linkage between SM and ExT is shown in this study using the four important factors related to the water and energy cycle. These factors cover the evapotranspiration (ET), soil moisture memory (SMM), sensible and latent heat flux (SHF and LHF). DRY-SM and WET-SM sensitivity experiments also identified that influence of SM on ExT is closely associated with four important factors of energy and water cycle. In wet conditions, the amount of energy available at the surface is used to cool the near-surface atmosphere by increasing the LHF through the ET process. As a result, the near-surface air temperature is remaining below normal conditions, and high temperature occurrences are gradually slowed. While, in case of dry sensitivity run, below normal SM conditions cause the sensible heating process to entrain more energy back into the environment by reducing the ET rate. Drier SM conditions also diminish long term SM memory time-scale (SMM) of remembering the positive anomalies caused by the atmospheric forcing (mainly the precipitation). Consecutively, dry sensitivity experiments suggest the higher impact of soil moisture to increase the extreme temperature conditions by changing the ET, SMM and surface energy partitioning.    (3 rd column), SHF in W/m 2 (4 th column), LHF in W/m 2 (5 th column), ET in mm (6 th column) for the FUT experiment (1 st row), and the difference between FUT and HIST experiment (2 nd row).        ExTI (3 rd row) for FUT (1 st column), difference between FUT-20 and FUT experiment (2 nd column), difference between FUT+20 and FUT experiment (3 rd column) during the future climate 2051-2100.