Climatological Changes in Soil Moisture during the 21st Century over the Indian Region Using CMIP5 and Satellite Observations

Abstract

Climate data records of soil moisture (SM) are fundamental for improving our understanding of long-term dynamics in the coupled water, energy, and carbon cycles over land. However, many of these studies rely on models for which the errors are not yet fully understood over a region. This may have a considerable impact on the economic growth of the country if the model’s future predictions are used for studying long-term trends. Here we examined the spatial distribution of past, present, and future predictions of SM studied using the Coupled Model Intercomparison Project Phase5 (CMIP5) simulations for the historical period (1850–2005) and future climate projections (2006–2099) based on Representative Concentration Pathways (RCP-RCP2.6, RCP4.5, RCP6.0, and RCP8.5). Furthermore, the performance of modeled SM with the satellite AMSR-E (Advanced Microwave Scanning Radiometer-Earth observation system) was studied. The modeled SM variations of 38 Global Climate Models (GCMs) show discreteness but still we observed that CESM1-CM5, CSIRO-MK3-6-0, BCC-CSM1-1, and also BCC-CSM1-1-M, NorESM1-M models performed better spatially as well as temporally in all future scenarios. However, from the spatial perspective, a large deviation was observed in the interior peninsula during the monsoon season from model to model. In addition, the spatial distribution of trends was highly diversified from model to model, while the Taylor diagram presents a clear view of the model’s performance with observations over the region. Skill score statistics also give the accuracy of model predictions in comparison with observations. The time series was estimated for the future trend of the SM along with the past few decades, whereas the preindustrial and industrial period changes were involved. Significant positive anomaly trends are noticed in the whole time series of SM during the future projection period of 2021–2099 using CMIP5 SM model datasets.

1. Introduction

The hydrological cycle, flood/drought management, water storage, surface runoff, and other factors in addition to climate change play an important role in affecting soil moisture (SM) [1]. The key changes in the hydrological balance of the environment, alterations in seasonal distribution, magnitude, and duration of precipitation and evapotranspiration owe their fluctuations to SM variation [2]. Reichle et al. [3] described that sensible and latent heat fluxes’ variabilities are explored by root zone SM. The memory of climate change and atmospheric circulations has an impact on SM variations. Hence long-term variation of SM leads to advances in the investigation of predictions of climate changes, disaster management, and further solutions to agriculture challenges [4].

The soil moisture (SM) future predictions aid the analysis of distinctive research subjects of meteorology. Comparison of temperature responses in model simulations in which the SM is constrained to represent the present or future climate is very effective, whereas, in Europe, Seo et al. [5] observed that heatwave predictions in 2003 and 2010 were effective when there is an initialization of SM in model simulations. Low SM results in variations of net energy fluxes, raising the atmospheric temperature, which can probe the water vapor deficit and evaporative demand in the air. Seneviratne et al. [6] concluded that the feedback induced by soil drying explained nearly 20% of the mean temperature increase projected for southern Europe. In the technical summary of the Intergovernmental Panel on Climate Change (IPCC) 5th Assessment Report (AR5) [7], Stocker et al. [8] assessed that SM drying in the Mediterranean, South West United States of America (USA), and the South African region was consistent with projected changes in the higher circulation and increased surface temperatures. This concluded that there was higher confidence in surface drying in these regions by the end of this century under the RCP8.5 scenario. However, most recent assessments have highlighted uncertainties in dryness projections due to a range of factors including variations between the drought and dryness indices and the effects of enhanced CO₂ concentrations on plant water use efficiency [6,9,10]. SM context determines how tightly water is bound in the soil texture. The greater the moisture deficit in the root layer the more negative the SM potential against which water must be extracted by the plant. Water deficit in the root zone of a plant makes it harder to extract more SM, i.e., there will be negative SM potential [10]. The ability to absorb carbon dioxide for photosynthesis from the atmosphere is reduced by the change in SM values. Hence, SM variations lead to changes in agricultural productivity [10,11]. Agriculture is an economical asset for India; herewith, it is evident that calamity management and future analysis of the climate changes are necessary (which will be effectively progressive with SM inclusions) for India to be a productive and progressive country. Trenberth et al. [12] and Chen and Hu [13] also contributed their research to explain how SM variation in climate models, mostly in the first 2 m of soil, were involved in climate change. However, the root zone SM is a crucial factor in atmospheric circulations as well as in climate change. We found that so many surfaces SM studies [12,13,14,15] helped us to better understand regional studies over India even though several studies go through climate change analysis using remotely sensed SM datasets [16,17,18].

Satellite datasets are time-bound and available for a limited period of time, and in-situ measurements are restricted to certain regions. There is a gap between the model studies and long-term analysis of the Indian region. The CMIP5 model simulations replace the scarcity of data availability by providing multimodel global data sets of historical and future projections [1]. Zahid et al. [19] described that in the IPCC report, Coupled Model Intercomparison Project Phase3 (CMIP3) 15 global climate models projected the consistent summer dryness and winter wetness in merely a fraction of the northern middle and high latitudes. More than half of the models predicted constant wetness in central Eurasia and dryness in Siberia and in mid-latitudes in Northwest Asia. The decreases in SM are dominant in the tropics and subtropics. Seneviratne et al. [6] concluded that there is a strong impact of SM on regional climate using CMIP5 multi-model projections. SM comparison studies of 24 CMIP5 models with Global Land Data Assimilation System (GLDAS) reanalysis have been done over the South Asian region [19]. They found there is a slight increasing trend in most of the models over some areas of Pakistan, India, Iran, Afghanistan, and Bangladesh and a slight to extreme increase in SM has been noticed in Eastern India under RCP8.5 annually and seasonally. Yuan and Quiring [20] studied the Contiguous United States (CONUS) region using the CMIP5 data model’s ensemble. Their study exemplified greater monthly variations and found the most skillful simulation of deeper soil layers in the southern U.S. Ruosteenoja et al. [21] used 26 CMIP5 climate models of RCP4.5 and RCP8.5 and illustrated that in southern Europe, the long term mean SM is projected to decline substantially in all seasons.

Agriculture, droughts, and climate change were all influenced by SM, and it has to be comprehensively analyzed to get better results [22]. Al-Yaari et al. [1] confirmed that to better understand the water and energy fluxes at the land/atmosphere interface, surface temperature SM variations must be taken into account because they observed their great importance when using CMIP5 models over the CONUS. Regional analysis using CMIP5 assured their variability with individual models analyzed for mid and high latitudes rather than in the tropics. Therefore, SM variations, in terms of present and future scenarios, are essential to eliminate the ambiguity in choosing crops that suit a particular soil texture in agriculture yields. This study will predominantly focus on the performance of the model rather than its causality. We provide the seasonal variations SM datasets in historical as well as in RCP scenarios and uncertainties of the modeled SM for each model represented and the anomaly trends of the SM were showing an increasing trend towards 2099. We have focussed on the statistical significance of the modeled SM by comparing it with AMSR-E SM datasets from 2006 to 2016. Finally, performance scores of the models by using Taylor skill scores of the models were presented. Section 2 discusses the data and methodology involved in the study, and in Section 3, we provide a detailed description of the study’s results. Finally, Section 4, will include the prominent conclusions of the study.

2. Data and Methodology

Global general circulation models or GCMs of the CMIP5 generate SM datasets. All the GCMs are differently aligned in terms of spatial and temporal resolutions. The present study utilized simulations of the SM variables in total column SM (0–100 cm) which will be more prominent to analyze the deeper soil layer moisture [6,20,21,23]. To reduce the discrepancies in spatial and temporal resolutions, we regridded the datasets to uniform resolutions of 1° × 1° to avoid further complications in our comparisons. Herewith the spatial exposure of CMIP5 datasets is (−180° W and 90° N) and (180° E and −90° S). These datasets were developed by the Program Climate Model Diagnosis and Intercomparison (PCMDI) on behalf of the World Climate Research Program (WCRP), and this provided input for the IPCC on the AR5. We used bilinear interpolation for the uniform resolution. It is one of the common methods of interpolation to derive the required resolution, and we advanced it with some more unique parameters such as precipitation, etc. [18,20,24]. It is the appropriate method of regridding to match with satellite observations.

CMIP5 datasets were accessible in both traditional historical and future emission scenarios. Therefore, historical datasets were derived from the period 1850–2005 and the Representative Concentration Pathways (RCPs) from 2006 to 2100. Nevertheless, the projections are prevailing in four diverse scenarios RCP2.6, RCP4.5, RCP6.0, and RCP8.5, in consideration of their concentrations, which are representing a stabilization of radiative forcing at 2.6 W/m², 4.5 W/m², 6.0 W/m², and 8.5 W/m² respectively by 2100. Taylor et al. [25] conclude that the atmospheric resolution of 1/3 models is approximately 1.5 latitudes or less. The higher the resolution the greater the regional analysis over the globe. Although we have global SM datasets, greater amounts of data would still be desirable in regions with complex topography and coastal lines [19]. SM data has been downloaded from https://data.ceda.ac.uk/badc/cmip5/data/cmip5/ (accessed on 20 May 2021). The parameter Mrso indicates total soil moisture content (0–100 cm) in both CMIP5 scenarios of historical and RCPs. For the present study, we have taken all the available models of total SM from the Center for Environmental Data Analysis (CEDA). Hence, all 38 accessible models were used in the present study in historical and future projections which are shown in

Table 1. CMIP5 Historical and Representative Pathways along with indicating the existence and nonexistence of the models.

S. No		Model Name	Historical	RCP2.6	RCP4.5	RCP6.0	RCP8.5	References
1		BCC-csm-1-1	Y	Y	Y	Y	Y	Xin et al. [26]
2		BCC-csm-1-1-m	Y	Y	Y	Y	Y	Xin et al. [26]
3		BNU-ESM	Y	Y	Y	N	Y	Ji et al. [27]
4		CanCM4	Y	N	Y	N	N	Merryfield et al. [28]
5		CanESM2	Y	Y	N	N	Y	Arora et al. [29]
6		CMCC-CESM	Y	N	N	N	Y	Scoccimarro et al. [30]
7		CMCC-CM	Y	N	Y	N	Y	Scoccimarro et al. [30]
8		CMCC-CMS	Y	N	Y	N	Y	Scoccimarro et al. [30]
9		CNRM-CM5	Y	Y	Y	N	Y	Scoccimarro et al. [30]
10		CNRM-CM5-2	Y	N	N	N	N	Voldoire et al. [31]
11		ACCESS1-0	Y	N	Y	N	Y	Bi et al. [32]
12		ACCESS1-3	Y	N	Y	N	Y	Bi et al. [32]
13		CSIRO-MK3-6.0	Y	Y	Y	Y	Y	Rotstayn et al. [33]
14		INMCM4	Y	N	Y	N	Y	Volodin et al. [34]
15		IPSL-CM5A-LR	Y	N	N	Y	Y	Cheruy et al. [35]
16		IPSLCM5A-MR	Y	Y	Y	Y	Y	Cheruy et al. [35]
17		IPSLCM5B-LR	Y	N	Y	N	Y	Cheruy et al. [35]
18		FGOALS-G2	Y	Y	Y	N	Y	Li et al. [36]
19		MPI-ESM-LR	Y	Y	Y	N	Y	Zanchettin et al. [37]
20		MPI-ESM-MR	Y	Y	N	N	Y	Zanchettin et al. [37]
	21	MPI-ESM-P	Y	N	N	N	N	Raddatz et al. [38]
22		GISS-E2-H	N	Y	Y	Y	Y	Miller et al. [39]
23		GISS-E2-HCC	Y	N	Y	N	Y	Miller et al. [39]
24		GISS-E2-R	Y	Y	Y	Y	Y	Miller et al. [39]
25		GISS-E2-R-CC	Y	N	Y	N	Y	Miller et al. [39]
26		CCSM4	Y	Y	Y	Y	N	Gent et al. [40]
27		NorESM1-M	Y	Y	Y	Y	Y	Bentsen et al. [41]
28		NorESM1-ME	Y	Y	Y	Y	Y	Bentsen et al. [41]
29		HadGEM2-AO	N	Y	Y	Y	Y	Collins et al. [42]
30		HadGEM2-ES	N	Y	Y	Y	Y	Collins et al. [42]
31		GFDL-ESM-2G	Y	Y	N	Y	Y	Dunne et al. [43]
32		GFDL-ESM-2M	Y	N	Y	Y	Y	Dunne et al. [43]
33		GFDL-CM2p1	Y	N	Y	N	N	Dunne et al. [43]
34		CESM1-BGC	Y	N	Y	N	Y	Long et al. [44]
35		CESM1-CAM5	Y	Y	Y	Y	Y	Long et al. [44]
36		CESM1-CAM5-1-FV2	Y	N	N	N	N	Long et al. [44]
37		CESM1-FASTCHEM	Y	N	N	N	N	Lamarque et al. [45]
38		CESM1-WACCM	Y	N	N	N	N	Daniel et al. [46]

Y = Available, N = Not available.

presented below.

In recent years the reliable and less bias of SM estimates from microwave sensors (both active and passive) are available with a nearly daily temporal resolution. AMSR-E (Advanced Microwave Scanning Radiometer Earth observation system) provides the SM data a brightness temperature at the 6.9 GHz C-band. NASA (National Aeronautics and Space Administration), JAXA (Japan Aerospace Exploration Agency), and groups that are involved in developing data using several algorithms are tasked to retrieve the SM data from the brightness temperature with an accuracy goal of 0.06 m³/m³. Here the resolution of data for effectively 50 km during the period from 2006 to 2016 was used for the analysis. Draper et al. [47] validated the AMSR-E results with several ground-based stations and they observed the bais range of −0.01 to 0.19 m³/m³ and correlation values are greater than 0.7. Recently, Clara Draper et al. [47] compared AMSR-E SM compared to in-situ SM data from 12 locations in Southeast Australia and they found a strong association to ground-based SM, with typical correlations of greater than 0.8 and root mean squared deviations less than 0.03 m³/m³. Kishore et al. [18] demonstrated the long-term trends of SM using AMSR-E datasets, which were supportive of the point that investigating and monitoring global earth SM led to a better understanding of the water-energy balance. In order to compare the datasets CMIP5-SM and AMSR-E-SM, both were uniformly regridded to a resolution of 1° × 1° (lon × lat). For the study 38 individual GCMs in terms of monthly means over the period of more than a century, both historical and future projection scenarios were utilized. Hence, the relative comparison with the observations is done over the Indian region by using seasonal climatologies. The monthly means of the datasets were downloaded, and the spatial distribution of the seasonal means of monthly means was calculated at each grid point and also estimated the trends to detail the climatological changes in the atmosphere in past, present, and future scenarios of 38 individual GCMs as represented in Table 1. Additionally, the relative differences between the model and observational data have been derived along with monthly, seasonal, and annual means using daily satellite data. The monthly mean datasets that were obtained from the CMIP5 are in kg/m² units; however, to analyze the observational data, the units have to be uniform. Hence, Bai et al. [2] described the volumetric conversion of SM to get standardized units in m³/m³. The formula of volumetric conversion that was used is shown below:

$$V=\frac{S}{h{\rho }_w}$$

where, S = Soil moisture (in kg/m²)

• H = depth/height (in m)
• ρ_w = density of water (in kg/m³)
• V = Volumetric soil moisture (in m³/m³)

The relative trend of the model to the observational datasets over the Indian region is described by including the following statistical techniques: root mean square error, standard deviation, and Pearson correlation coefficient. The spatial and temporal distribution of the CMIP5 projections for more than a 100-year period will probe the predictions of the agriculture droughts assessments and flood/drought forecasting, in addition to the climate change analyses, which is an in-demand research topic of the world. Relative errors of the satellite observational SM data to model the simulated SM were displayed in a time series which was smoothened using a familiar technique: locally estimated scatter plot smoothing (Loess filter) [18,48]. A Taylor diagram is one of the conversant techniques to represent the standard deviation and the correlation coefficient together [17,49,50]. Herewith, by the use of this technique, all the future projection scenarios of CMIP5 are compared with AMSR-E. The skill of the models was analyzed by using the skill score formula shown below. The skill score is a statistic similar to the Nash-Sutcliffe such that the closer the score gets to one, the better the model prediction [51]. This function interprets model predictability using residual error and observed variability in data. Therefore, when the value of the skill score is 1, then it is in good agreement with the model prediction, when it is 0 or less than zero, this indicates the model predictions are worse than the observations.

Skill score = 1 − RMSE/STD_obs

where, RMSE = Root Mean Square Error

• STD_obs = Standard deviation of observations (AMSR-E)

3. Results

3.1. AMSR-E Seasonal and Annual Climatologies:

The seasonal and annual SM spatial distribution patterns of the AMSR-E from 2006 to 2016 over India were discussed in the below Figure 1. Seasonal means refer to subsequent months: Winter (December, January, and February), pre-monsoon (March–May), monsoon (June–September), and post-monsoon (October and November). The annual and seasonal variation of AMSR-E SM has a higher value during the monsoon compared to all other seasons since it has the maximum value of SM, i.e., 0.35 m³/m³ except in the Himalayan region. The Central Indian region has the highest values during the monsoon whereas, in the same region, the lowest values have also been observed during the pre-monsoon season, therefore low moisture is due to the motion of the heat belt further north and towards the north-western parts of India. Most of the Indian irrigation depends on monsoon moisture. Rainfall variations can lead to variabilities in the SM and evapotranspiration, meanwhile, the temperature is also a major driver of the SM. Tropical regions’ crop yields suitability and sustainability are mostly dependent on monsoon precipitation, which owes to SM variations as one of its aspects [18].

The investigation of the United Kingdom Meteorological Office (UKMO) datasets of SM was discussed by Unnikrishnan et al. [52], stating that there is a maximum SM during monsoon season. Furthermore, annual climatology follows a crucial pattern, indicating that there is a maximum wet condition in the North Central, North East, Western Himalayas, and the North West. This is due to having lower values in annual as well as seasonal climatology. Therefore, due to a reduction in the low-intensity rainfall, groundwater recharge has deteriorated in the North West and North Central parts of India during 1996–2016. Researchers have found that there is a 22% and 2% of rainfall deficiency in the North West and North Central regions respectively and 23% and 8% excess rainfall exists in the interior peninsula and Northeast India respectively [53]. The annual spatial patterns will be used to comprehend viable forms of drought or flood management and crop yielding.

3.2. CMIP5 Historical Seasonal Distributions

The spatial distribution of the CMIP5 historical models (total SM in m³/m³) in the monsoon season from 1850 to 2005 was analyzed using the 35 GCMs (Global Climate Models). From Figure 2, 10 GCM models with maximum wet conditions and 5 GCMs with moderate extreme wet conditions, and 5 GCMs with slightly dry to drier SM were observed, furthermore, all the remaining models exemplified moderate to dry conditions.

From the Figure above, GFDL-ESM models were having higher values of SM in the whole of India. There were higher wet conditions in the central Indian region in the BNU-ESM, CESM1-BGC, CESM1-CAM5-FV2, CESM1-FASTCHEM, CESM1-WACCM, NorESM1-ME, and NorESM1-M. INMCM4 had the highest values in the Himalayan region, and CSIRO-Mk3-6-0 also had the maximum values in the Himalayan region compared to other models as shown in Figure 2. CESM1-CAM5, GFDL-CMp1, and GISS models were having slightly lower SM values (dry) in the monsoon season. ACCESS, FGOALS-g2, and BCC-CSM were having moderate values of SM. Zahid et al. [19] also observed a similar pattern of increment in central India. Variability in the trend of more than half of the models lay between 0 to 0.6 m³/m³. An investigation by Dirmeyer et al. [54] also justifies that in the monsoon, there is a global drying trend in the high latitude and that there is a consensus for a wetter trend in central Asia and in the western Amazon basin during the monsoon. Zebaze et al. [55] detected that there is an overestimation of the CMIP5 models’ precipitation over the Gulf of Guinea and Eastern Africa caused by lateral boundary condition errors [56] and the existence of mountain ranges during 1975–2005.

3.3. RCP8.5 Seasonal Climatologies

Pre-industrial period of 1850 to the present, and the future projections of CMIP5 Representative Concentration pathways with max values of greenhouse gas concentration 8.5 W/m² are represented in the present section during the period of 2006–2099 in the monsoon seasonal distribution of all the 30 GCMs. Nevertheless, in Figure 3 the preindustrial period to the future projections, we found similarities in the GFDL-ESM models: they have maximum values, and the other BNU, CESM1 models, CSIRO-MK3-6-0, NorESM1 models are moderate wetter conditions in agreement with the historical datasets.

Hence, the slight to moderate changes in SM were very clear to the scale in ACESS models, FGOALS-g2, and BCC-CSM models, similar to the above results. HadGEM2-ES is the new model in the RCP8.5 as well it shows the slight to moderate changes of SM in India. Investigation of the RCP8.5 displays that more than half of the multi-models were in a consensus being below 0.8 m³/m³. Yuan and Quiring [20] demonstrated substantial variabilities of surface SM to root zone SM and found the greater uncertainties in surface SM over CONUS using 17 Earth System Model (ESM) models, and therefore Sheffield et al. [57] described precipitation overestimation of CMIP5 models over the western US is the reason for the wet biases and eastern US dry biases due to overestimated evapotranspiration by CMIP5 models. Soil water content in the deeper layer of 0–100 cm was more dominant in the comparison of surface layers in the satellite dataset. Yuan and Quiring [20] also demonstrated that the water content of the CMIP5 0–100 cm layer had greater values when compared to in-situ measurements, exclusively in drier months when the soil water content is <0.25 cm³/cm³ [20] over CONUS. Hence, it is clear that the maximum of the model’s performance is in good agreement with the previous studies. Even though spatial patterns are similar, we cannot make a conclusion in terms of the whole picture. To do that, ascertaining the performance of the model’s statistical methods is necessary.

3.4. Statistical Methods

3.4.1. Spatial Correlation Analysis between CMIP5 and AMSR-E

The CMIP5 model RCP8.5 has 22 different models for comprehensive comparison to the observational data, which will be executed well by including statistical analysis with the following statistical techniques: standard deviation, RMSE, and the correlation coefficient even though region to region data coarse existing models. The Pearson correlation coefficient (α = 0.05) between individual models to the satellite observations, i.e., monthly means during 2006–2016 indicated the spatial distribution pattern over India in Figure 4. Whereas most of the models (12) were having a (>0.7) consensus with the observational data sets, some models (5) were being (>0.6) in agreement with the observations except models (5) (<0.5) GFDL-ESM2G, IPSL-CM5A-LR, IPSL-CM5A-MR, IPSL-CM5B, INMCM4. Especially IPSL-CM5B-LR and INMCM4 were negatively correlated as the depth variations and also the country to country discrepancy existed from model to model [19,20]. Meanwhile, the changes in temperature can make amplifications in the atmospheric water cycle which may have increased the evapotranspiration and led to a decrease in SM [58]. Therefore, discrepancies exist in simulations from model to model and none of the models exhibited the same consistency from one country to another. Zahid et al. [19] also concluded that there is a difference in models in estimation and simulation. In the Indian region, over 12 models were exhibiting more than a 70% correlation with the AMSR-E satellite SM during the period 2006–2016. In addition, a maximum concentration of radiative forcing, i.e., RCP8.5 resulted in very high implications compared to other future projection scenarios. However, the model datasets had their own limitations in terms of depths as well as in other parameters from region to region [19,20]. It is clearly evident from observations in SM from past and future projections that the arid regions of the Northwest have the highest uncertainties and the Westcentral region displayed the least uncertainties [59]. Therefore, the results evidence surface feedback influences with SM variations. Hence, the uncertainty in temperature was also one of the influences of SM variation in the Indian region [59].

We estimated the root mean square error (RMSE), correlation, and standard deviations using satellite and CMIP5 models. These statistical results give the bias and relation between each grid point of the SM data. Figure 5 gives the correlation value for each grid and these results are implied in Figure 3. The differences are more the correlation value is less and standard deviations are more. Further, the number of models was reduced to 22 from 30. These 22 models are very less biased with less standard deviations compared to other models, and these are consistent with satellite measurements. These SM changes may cause regional changes in temperatures over a period of time, and these are not captured in some of the models. Singh and Achuta Rao [59] described the west-central region of India and they found a temperature rise of about 4.5 °C to 5 °C exhibited in consideration of the ensemble average of 20 years. In addition, they found a 5.5 °C rise in some regions of India during the winter season.

3.4.2. Soil Moisture Anomalies

The enumerated tendencies of SM based on the 38 unique GCM outputs from the CMIP5 with climate model scenarios of the RCP2.6, 4.5, 6.0, 8.5, and historical were described as having increasing tendencies towards 2099. Figure 6 displayed anomalies of the SM generated from the difference between CMIP5 RCPs and the base-period mean, i.e., the pre-industrial period mean during 1861–1900 using the Loess (Locally estimated scatter plot smoothing) filter with 20% smoothing. The anomalies differ with the concentrations of greenhouse gases (W/m²) from decade to decade. Whereas the base period mean is taken from the preindustrial period (1861–1900) of CMIP5 historical datasets and historical datasets were combined with RCP’s to generate a long-term analysis of (1861–2099), which expresses the preindustrial to industrial period variations along with future scenarios. The different radiative forcing as listed by the RCPs above have their own path of tendencies as per the concentrations; from Figure 6, we can observe the variations of different radiative forcing from the past to present and future scenarios. Every decade has its own variations from one to another. While we found 1861, 1871, 1901, 1911, 1921 had been slightly falling in the direction of negative trends, for 1941, 1951, 1961, 1971, 1981, 1991 we found that path continued towards negative trends. Whereas, SM has slight increasing trends towards a positive anomaly in 1891, 1931, 2001, and 2011. Thereafter, SM anomalies during the period 2021–2099 continuously increased towards a positive trend over India in all the radiative forcing scenarios (RCP2.6, RCP4.5, RCP6.0, and RCP8.5). Kishore et al. [18] also observed that there is a trend of the SM increasing in India at 0.0158 m³/m³/decade. Therefore, the decadal change of SM was in good agreement with satellite observations. Hence, the SM maximum positive trends were observed in terms of RCP8.5 with the highest greenhouse gases concentrated and RCP6.0 has maximum down drag towards negative trends. From Figure 6, we can observe that there is a similar trend for both RCP2.6 and RCP4.5 for nearly a century approximately. The land atmospheric feedbacks and their differences are major contributors to uncertainties in the models. Hence, the increasing trend in future projections may be due to the uncertainties in the models.

The Taylor diagram is one of the best ways of describing the multiple statistics together. Taylor diagrams illustrated RCP2.6, RCP4.5, RCP6.0, and RCP8.5 model’s standard deviation, correlation coefficient, and RMSE with AMSR-E in Figure 7 respectively. In addition, we observed that most of the model’s correlations were greater than 0.7 in all respected scenarios, while very few had below a 0.5 correlation with AMSR-E.

Meanwhile, In Figure 7, BNU-ESM, CESM1-CAM5, NorESM1, CSIRO-MK3-6-0, HadGEM2-AO, MPI-ESM, and BCC-CSM models in all RCP’s were clearly exhibiting >0.7 correlations with satellite observations over India. Therefore, we can observe that most of the models shown in all scenarios were in good agreement with the satellite observation during the period 2006–2016. These results represent the reference value of AMSR-E standard deviation, and it is clear that the models were in consent with the model datasets.

The scatter plot of the yearly mean and the uncertainty of the models are shown in Figure 8. Signifying the variability of the models in both historical and future projections. Uncertainty of the dataset over the region may be due to the epistemic uncertainty over India [59]. Uncertainty of the models was calculated by subtracting the overall average of the individual values and finally by taking the absolute mean of it. The total average mean during the period 2006 to 2099 was subtracted from each year’s mean of the model to know the uncertainty of the models. The HadGEM2 models in future projection displayed similar results in all the scenarios, the BCC models exhibited moderate values, and the BNU and NorESM1 models were exhibiting higher values in comparison to other models in terms of all future scenarios. The remaining models having uncertainties below 0.02 m³/m³ have moderate performance. Therefore, this allows for a clearer view of the model performance over the region, and the uncertainties will help to understand data evaluation. When we consider multiple models, the uncertainty changes and illustrates an increasing trend towards 2099. The results of Singh and Achuta Rao [59] have also evidenced the growth in the uncertainty of the precipitation and temperatures. However, it is a known fact that the changes in the temperature and precipitation are owing to influences in the SM, leading to the major uncertainties in the models. Knutti et al. [60] also discussed that CMIP5 was showing lesser warming conditions compared to CMIP3 due to the radiative forcing inclusions in CMIP5. This ascertains that the inclusions of more necessary corrective parameters make a model more reliable. Hence, the uncertainty and error corrections statistics are necessary for the best outcome of model performance.

Based on the monthly SM values from CMIP5, Figure 9 summarizes the overview of all the three statistical methods which are most effective to understand the model’s variability and performance over a region, representing S for Standard deviation, R for Root Mean square error, and C for Correlation coefficient evaluated with observational data of AMSR-E satellite SM which were shown in the checker-board indicating major variation of the model to model during 2006 to 2016. Therefore, maximum RMSE was exhibited by the CCSM4, CESM1, and GFDL models over all the scenarios even though well correlated. The BCC models were absolutely good enough over the Indian region showing lower error and better correlation. The INMCM4 model had the poorest performance due to a lack of data and more values that were NaN over a particular region. The remaining models had moderate performances. The impact of the uncertainty in models is increasing with the increasing concentration. Thus, the models are exhibiting different variations and no ideal model exists.

3.4.3. RCP’s Skill Scores

Skill score is one of the best statistics to analyze the model performances. Skill score is the difference between the ratio of RMSE and STD observation to one. From Figure 10, the skill of the models in terms of total column SM is similar to the surface level SM, but skill variability in intermodal is greater [20]. Here GFDL–ESM2G and GFDL–ESM2M are evidently showing their performance is very poor among the models in terms of all the scenarios. Therefore, when the value of the skill score is 1, it is in good agreement with the model prediction, when it is 0 or less than zero, the model predictions are poor than the observations.

Here all models were showing less than zero values, with the indication of poor predictions, but among all the scenarios, monthly means of SM of CMIP5 models and AMSR–E satellite observations HadGEM2, IPSL, MPI, and bcc models were comparatively better at predicting, by having values near to zero (<−10) during the period 2006–2016. Therefore, the other models CESM, BNU, and CESM provided only moderate predictions.

4. Conclusions

In this study, we analyzed the tendency in SM seasonal changes in India using CMIP5 models under different scenarios (RCP2.6, RCP4.5, RCP6.0, and RCP8.5) and satellite measurements. Initially, seasonal characteristics of AMSR-E SM show it will be drier in the pre-monsoon season and wetter in monsoon and post-monsoon seasons. The maximum SM is observed in the central and southern parts of India. The yearly means of nearly 11 years of satellite data varies between 0.1 to 0.3 m³/m³. The spatial seasonal mean indicates the increase or decrease of SM in India on regional scales, and these are essential for a better understanding of agricultural yields. The spatial seasonal correlations are estimated between AMSR-E and RCP8.5. The positive correlations are observed in the central and southern parts of India, and negative correlations are observed in the northern part of India. The positive correlations range from 0.6 to 0.8 and the negative correlations are from −0.2 to −0.4. The model performance differs significantly when being evaluated using several statistical metrics calculated between AMSR-E and CMIP5 RCPs, and we found a good correlation and fewer biases found in nearly 22 models. The year-to-year future normalized anomalies of SM of CMIP5 for different RCP scenarios during the period 1861–2099. The SM values are increasing over India from 2000 in all scenarios. Agreement between satellite and CMIP5 models SM is evaluated for each RCP by using the Taylor diagram for the 2006–2016 period. In all scenarios, the correlation is greater than 0.5 and standard deviations are less than 0.11. It indicates both satellite and CMIP5 model scenarios are in good agreement during the period 2006–2016. The discrepancies between satellite and CMIP5 models over India may be due to several factors involved in models and satellite datasets. Some factors include different spatial resolutions, model parameters, the land-atmosphere coupling algorithms, and model forcings could also be influential. These factors can introduce uncertainties into model-simulated soil moisture [20,61]. The model performance scores are scaled with respect to satellite data uncertainty values between 0 and 1, with 1 being perfect skill and 0 being no skill. The RCP2.6 and RCP4.5 performance are better than the other two scenarios.

Hence, among all the discrepancies we observed that CESM1-CM5, CSIRO-MK3-6-0, and BCC-CSM1-1, and also the BCC-CSM1-1-M, NorESM1-M models of CMIP5 performing as well as one another in all model pathway future scenarios over the region. Even though there exists uniqueness in projections, the RCP4.5 and RCP8.5 scenarios exhibited similar and better performance for BNU-ESM, HadGEM2-AO, FGOALS-g2, MPI-ESM-LR models, and they were in good agreement with satellite observations over the Indian region. In addition, we have observed MPI-ESM models are performing well in all RCP2.6, RCP4.5, and RCP8.5 scenarios respectively with a good correlation to the Indian regions spatially and temporally. We conclude that the study should be furthered in the future with the inclusions of the SM in relation to the other atmospheric parameters to better understand the future scenarios over the country. This study summarizes uncertainties and emphasizes that studies done in collaboration can also improve regional studies and along with correction inclusions, we will find the ideal models for the future predictions and regional analysis will be relevant and consistent.