## Abstract

The paper deals with the projected rainfall for eight rain gauge stations located in the upper Mahanadi catchment in Chhattisgarh state of India and corresponding changes on the water availability in few reservoirs of the catchment. Rescaled predictors obtained from NCEP were used and statistically tested for selection of best-fit set of predictors using percentage reduction methods. The calibrated and validated models were used to generate multiple series for early, mid and late century periods, i.e. for 2020–2035 (FP-1), 2046–2064 (FP-2) and 2081–2099 (FP-3) under CMIP5 climatic forcing conditions of RCP2.6, RCP4.5 and RCP8.5 using predictors data of CanESM2, Canadian GCM. The comparisons of future predicted rainfall with the base period (1981–2003) showed mixed trends, viz. declining trend at five stations, both declining and increasing trend at two stations, and increasing trend at one station. The predicted reduced rainfall during August and September attribute to a significant impact on paddy cultivation and industrial development. The analysis of future catchment rainfall on five important reservoirs in this region indicated a reduction of 12–29% seasonal rainfall with respect to the base period rainfall; while for one reservoir not much variation (–7 to 5%) in the rainfall was noted, possibly due to the large catchment area.

## INTRODUCTION

Different reports of the Intergovernmental Panel on Climate Changes and other independent researchers (IPCC 2007, 2014; Wu *et al.* 2015) confirmed the changes in climatic pattern, both on a global and a regional scale. The changing climate may likely affect not only runoff availability and supplies of water, but also irrigation water demand, groundwater availability, agriculture and livestock, and may attribute frequent floods, droughts, and soil erosion affecting many more areas of life (IPCC 2014; Qiu *et al.* 2016). The linkage of rainfall and runoff plays an important role for planning and management of water resources on a local and a regional scale (Kar *et al.* 2015). The impact of climate change on the distribution of rainfall and the corresponding changes on the catchment hydrology are of concern to hydrologists, water resources planners and managers.

India is an agrarian country with nearly 75% population depending directly on agriculture and agro-related jobs. Most parts of India receive precipitation in the form of rainfall from the south-west monsoon that occurs during June to October every year. The distribution of rainfall is heterogeneous both on a spatial and a temporal scale. This spatiotemporal variability of rainfall compels the use of stored water in reservoirs and groundwater as the main source of irrigation water during the lean season. The Chhattisgarh, a state formed by dividing Madhya Pradesh in the year 2000, is known as one of the rice bowls of India because of its large scale paddy production using irrigation water from a number of surface storage structures. Being a newly formed state, it has progressive industrial development. The farm production and industrial development in the region may face a major setback if significant changes in the rainfall pattern on the catchment of the reservoirs occur due to plausible climate change. This eventually necessitates an assessment of climate change impacts on rainfall distribution and corresponding changes in water availability for development of suitable adaptation measures.

For assessing the climate change impact on rainfall and water availability, normally outputs of generalized circulation models (GCMs) are used with appropriate downscaling techniques. Several organizations worldwide have developed different GCMs to forecast future climate variables for probable carbon emission/environmental scenarios. The data generated by GCMs are unable to produce significant sub-grid scale features including topography and land use, which are essential inputs for hydrologic modeling and impact assessment analysis. The coarse grid data from GCMs can be resolved by downscaling them to local and basin scale with the help of dynamic or statistical techniques that bridge the large scale atmospheric conditions with local scale climatic data (Willems *et al.* 2012). Both these downscaling techniques have their own advantages and disadvantages for future projection of climate (Tukimat & Harun 2013). For example, the dynamic downscaling techniques use a physically based model that has the major drawback of its complexity, high computational cost and propagation of systematic bias from GCM to regional climatic model (Anandhi *et al.* 2008); whereas statistical downscaling techniques are reasonably accurate, best in class for developing relationships between GCM predictors and regional/station climatic data (Fowler *et al.* 2007). Sachindra *et al.* (2015) applied a key-predictand and key-station approach for statistical downscaling of monthly maximum temperature, minimum temperature and evaporation by developing a simple linear regression equation between predictand at key station and projection of climatic variables for other stations. The method proved very effective in simulating the results of climatic variables in close agreement with the observed data. Asnaashari *et al.* (2015) used projected statistical downscaled climate data of the Mississippi river in a hydrological model for simulating streamflows and found an increase of 0.1% in total precipitation, 0.7% reduction in snowfall and 1.4% increase in water deficit. Tukimat & Harun (2013) also assessed the impact of climate change on rainfall and temperature in the Muda area of Malaysia using a statistical downscaling technique. Keeping the potential of a statistical downscaling model (SDSM-DC) in view, this technique is employed in the present study to forecast precipitation data using CanESM2 (Canadian Global Circulation Model CGCM) weather predictors for developing scenarios of RCP2.6, RCP4.5 and RCP8.5.

## STATISTICAL DOWNSCALING MODEL-DECISION CENTRIC (SDSM-DC)

The SDSM-DC is user friendly software developed for the prediction of future climate data sets by statistical downscaling of GCM's outputs. The SDSM-DC can be used as a decision support system to generate plausible daily weather series under manually guided trend conditions, regression with GCM predictors and other unusual variables such as tidal surge. The SDSM-DC is a transfer function based model that can be categorized as a hybrid of stochastic, weather generator and multiple linear regressions model between daily predictand and a set of predictors representing local weather through seven major steps, which include quality control and data transformation, screening of predictor variables, model calibration, weather generation, statistical analysis, graphical representation and scenario generation. The weather generator of SDSM can also be used to infill missing data using statistical characteristics of series in data sparse regions and understand regional climate system (Wilby *et al.* 2014). The details about SDSM are available in the literature (Goodess *et al.* 2003; Wilby *et al.* 2004; Gachon *et al.* 2005).

*W*) for the wet day occurrence on

_{i}*n*predictor variables,

*X*on

_{ij}*i*th day using the following equation (Wilby & Dawson 2013) under the constraint 0 ≤

*W*≤ 1:

_{i}*r*is a stochastic output of linear random generator. After confirming a wet day, the downscaled total precipitation,

_{i}*P*, can be computed using: where

_{i}*Β*and

_{0}*β*are the regression coefficients and

_{j}*e*is the model error. The details can be seen in Wilby

_{i}*et al.*(2014). The flow chart for prediction of future climate through SDSM is presented in Figure 1 (Wilby

*et al.*2014).

## STUDY AREA AND DATA USED

The upper Mahanadi catchment in Chhattisgarh state of India located between 20°N to 21°N latitude and 80° 30′E to 82°E longitude has several major water resource projects for irrigation, industrial, drinking and power generation. The location map of the study area is in Figure 2. This part of Chhattisgarh state is rich in water resources and has a number of important water resources projects namely, Ravishankar Sagar, Tandula, Maramsilli, Dudhawa, Sondur, Kharkhara and Gondli, which are operated to provide water for irrigation, industries and domestic purposes. The observed daily rainfall data of eight rain gauge stations, namely Ambagarh, Balod, Bhanpura, Chamra, Gondli, and Rudri for the period 1981–2014, and Kanker, and Maramsilli for the period 1961–2014, have been utilized for analysis together with NCEP reanalyzed predictors for the corresponding period for modeling (Kalnay *et al.* 1996) and RCP2.6, RCP4.5 and RCP8.5 predictors data of CanESM2 GCM from 2006 to 2099 for the generation of future scenarios. The description of data used for different analyses is provided in Table 1. The locations of the eight rain gauge stations along with important water resource structures are presented in Figure 3.

S.N. | Particulars of data | Data period | Calibration | Validation | Uncertainty analysis |
---|---|---|---|---|---|

1. | Rainfall data of Ambagarh, Balod, Bhanpura, Chamra, Gondli, Rudri | 1981–2014 | 1981–1995 | 1996–2003 | 2006–2014 |

2. | Rainfall data of Kanker, Maramsilli | 1961–2014 | 1961–1985 | 1986–2003 | 2006–2014 |

3. | NCEP rescaled and CMIP5 climate data Mean sea level pressure ( mslpgl), mean sea level temperature (tempgl), surface, 550 hpa, 850 hpa geo-potential height (pgl, p500gl, p850gl), surface, 500 hpa, 850 hPa zonal velocity (p_ugl, p_5ugl, p_8gl), surface, 500 hpa, 850hPa meridional velocity (p_vgl, p_5vgl, p_8gl), surface, 500 hpa, 850 hPa vorticity (p_zgl, p5_zgl, p8_zgl), surface, 500 and 850 hPa airflow strength (p_fgl, p_5fgl, p_8fgl), surface, 500 and 850 hPa divergence (p_zhgl, p5_zhgl, p_8zhgl), surface, 500 and 850 hPa specific humidity (pshum, ps500gl, ps850gl), near surface, 500 and 850 hPa wind direction (pthgl, p5thgl, p8thgl) | 1961–2003 (NCEP) and 2006–2099 (CanESM2 RCP2.6, RCP4.5 and RCP8.5 scenarios) | Used for model development, uncertainty analysis and generation of multiple precipitation series |

S.N. | Particulars of data | Data period | Calibration | Validation | Uncertainty analysis |
---|---|---|---|---|---|

1. | Rainfall data of Ambagarh, Balod, Bhanpura, Chamra, Gondli, Rudri | 1981–2014 | 1981–1995 | 1996–2003 | 2006–2014 |

2. | Rainfall data of Kanker, Maramsilli | 1961–2014 | 1961–1985 | 1986–2003 | 2006–2014 |

3. | NCEP rescaled and CMIP5 climate data Mean sea level pressure ( mslpgl), mean sea level temperature (tempgl), surface, 550 hpa, 850 hpa geo-potential height (pgl, p500gl, p850gl), surface, 500 hpa, 850 hPa zonal velocity (p_ugl, p_5ugl, p_8gl), surface, 500 hpa, 850hPa meridional velocity (p_vgl, p_5vgl, p_8gl), surface, 500 hpa, 850 hPa vorticity (p_zgl, p5_zgl, p8_zgl), surface, 500 and 850 hPa airflow strength (p_fgl, p_5fgl, p_8fgl), surface, 500 and 850 hPa divergence (p_zhgl, p5_zhgl, p_8zhgl), surface, 500 and 850 hPa specific humidity (pshum, ps500gl, ps850gl), near surface, 500 and 850 hPa wind direction (pthgl, p5thgl, p8thgl) | 1961–2003 (NCEP) and 2006–2099 (CanESM2 RCP2.6, RCP4.5 and RCP8.5 scenarios) | Used for model development, uncertainty analysis and generation of multiple precipitation series |

## METHODOLOGY

The steps involved in statistical downscaling consisted of quality control, selection of set of predictors, calibration and validation of model, uncertainty analysis, generation of data series and statistical analysis/comparison.

### Quality control

For statistical downscaling, predictands (rainfall of different rain-gauge stations) and predictors (NCEP reanalyzed 26 parameters, Table 1) of concurrent periods were analyzed to identify gaps, outliers, and statistics to use these data for further analysis.

### Selection of predictors

The selection of an appropriate set of predictors is an important task in the downscaling process for which understanding of physical process and sensible predictors for atmospheric circulation are essential (Huang *et al.* 2011). Several methods have been propagated by different researchers for the selection of an appropriate set of predictors (Shongwe *et al.* 2006; Tripathi *et al.* 2006; Benestad *et al.* 2007 etc.). The present study considers a percentage reduction method (Mahmood & Babel 2013) together with a scatter diagram; the step-by-step procedure is outlined here:

Step 1: The correlation coefficient between the predictand (rainfall) and 26 NCEP rescaled predictors on monthly, seasonal and annual basis have been computed using a conditional approach and the top 10 predictors have been considered for further analysis.

- Step 2: The predictor ranked the first was termed as super predictor (
*SP*) and using it, absolute correlation coefficient, absolute partial correlation and percentage reduction (*PR*) have been computed for the remaining nine predictors using the following equation: where*P*and_{r}*R*are the partial and absolute correlation coefficient, respectively. Step 3: All the predictors having a

*P*-value more than 0.05 and absolute correlation coefficient with super predictor more than 0.70 have been removed to avoid multi-co-linearity. The predictor having the lowest*PR*value has been recognized as the second super predictor.Step 4: The second super predictor along with remaining predictors were used to follow Steps 2 and 3 to obtain the third predictor. In general, 1–3 predictors are sufficient to model climatic variability (Wilby

*et al.*2002; Chu*et al.*2010).

### Calibration and validation of model

After selecting a set of appropriate predictors and by using appropriate transformation under conditional process, empirical relationships between the predictand and predictors have been developed. The monthly, seasonal or annual models have been developed in SDSM using a *K-fold* cross validation technique (Markatou *et al.* 2005; Bedia *et al.* 2013; Casanueva *et al.* 2014). To use this technique, whole data series have been divided into two parts; the first ((*K**−* 1)/*K*) part is taken for calibration and to develop statistical relationships by appropriate transformation and model type, and the second (1*/K*) part is for validation. The calibration and validation of the model followed the steps given below:

Step 1: 3-fold cross validation technique has been used by considering two-thirds of the whole data series for calibration, and the remaining one-third for validation.

Step 2: Developed monthly, seasonal and annual models and performed judgment based on variance (R-squared value), standard error and graphical representation in SDSM.

- Step 3: The observed and generated series of the rainfall were de-biased using linear scaling technique proposed by Lenderink
*et al.*(2007) and Fang*et al.*(2015) as follows: where and are the de-biased and generated precipitation of the*j*th day of the*i*th month, respectively; and and are the observed and generated mean daily rainfall for the*i*th month, respectively. Step 4: Various goodness of fit criteria, including statistical coefficient of correlation (

*C*), adjusted R_{c}^{2}(*adj R*), Nash–Sutcliffe efficiency (^{2}*ɳ*) for monthly rainfall of monsoon season, have been used for selection of the best-suited model.

### Uncertainty analysis

*S*1(

*X*) and

_{1}, X_{2}, X_{3}…… X_{n}*S*2 (

*Y*) be the two paired samples, the statistics (

_{1}, Y_{2}, Y_{3}…… Y_{n}*V*) of Wilcoxon signed rank test of these samples can be computed by: where

_{s}*R*is the rank of the pair and

_{i}*n*is the number of non-zero in the pairs. For the null hypothesis,

_{r}*V*follows a distribution with expected value and variance . For this test, the

_{s}*p-*value suggested by Lehman (1975), as given in Equation (6), has been used: where and

*c*are the distribution function and continuity correction, respectively. The null hypothesis (

*H*

_{0}) is described by no inequality in monthly difference between the observed and simulated precipitation at the critical

*p*-value of 0.05 (95% confidence level) (Pervez & Henebry 2014).

### Generation of series

The weather generator of SDSM has been used for generating multiple ensembles of three future periods namely, near century (2020–2035), mid-century (2046–2064) and far-century (2081–2099) periods. The calibrated model and projected climate predictors have been used to generate multiple series of precipitation, which were later de-biased using Equation (4).

### Statistical analysis and comparison

The computation and comparison of several statistical tests of the observed and predicted data have been performed in SDSM to examine the changes in projected climate. Several generic and conditional statistics viz., mean, maximum, minimum, variance, peak over threshold, percentile, inter quantile range etc., have been ascertained to examine the changes in future precipitation regimes.

## RESULTS AND DISCUSSION

The SDSM DC software has been used to generate multiple future ensembles for the rainfall of eight rain gauges stations under RCP2.6, RCP4.5 and RCP8.5 climate forcing conditions for three future periods namely, FP-1 (2020–2035), FP-2 (2046–2081) and FP-3 (2081–2099) for the Upper Mahanadi basin. Twenty-six NCEP reanalyzed predictors and rainfall of the eight rain gauge stations, namely Ambagarh, Balod, Bhanpura, Chamra, Gondli, Kanker, Maramsilli and Rudri, have been analyzed using quality control option in SDSM and no data gap has been found in any of the series.

### Calibration and validation of statistical models

A 3-fold cross validation approach has been used, in which nearly two-thirds of the data series were used for calibration, and the remaining one-third for validation (Table 1). The methodology of percentage reduction along with a scatter diagram have been used for selecting the set of most appropriate predictors. The annual, seasonal and monsoon monthly series and goodness of fit criteria including coefficient of correlation (*C _{c}*), adjusted R

^{2}(

*adj R*), Nash–Sutcliffe efficiency (

^{2}*ɳ*) for monthly rainfall of monsoon season (principal months for rainfall) have been chosen for selecting predictors. For non-satisfactory results, the combination and/or transformation have been changed to obtain appropriate results. The list of selected predictors, model type, transformation,

*C*,

_{c}*adjR*and

^{2}*ɳ*for different rain gauge stations during calibration and validation are presented in Table 2. The graphical representations of the observed, calibrated and validated series of rainfall for a few rain gauge stations are given in Figure 4. The analysis of goodness of fit measures (Table 2) and graphical representation (Figure 4) showed that the selected predictors and developed multiple linear regression were capable of predicting different predictands with reasonable accuracy and hence can be used for the generation of future projected series.

S.N. | Rain gauge station | Selected predictors | Model type/Transformation | Model performance during calibration | Model performance during validation | ||||
---|---|---|---|---|---|---|---|---|---|

C _{c} | Adj R ^{2} | ɳ | C _{c} | Adj R ^{2} | ɳ (%) | ||||

1. | Ambagarh | ncepp5_fgl, ncepp8_ugl, ncepp850gl | Seasonal/None | 0.78 | 0.41 | 55.7 | 0.79 | 0.40 | 56.54 |

2. | Balod | ncepp5_ugl, ncepp8_ugl, ncepp850gl | Monthly/None | 0.69 | 0.55 | 85.9 | 0.66 | 0.66 | 75.3 |

3. | Bhanpura | ncepp5_ugl, ncepp5_zgl, ncepp8_zgl | Seasonal/None | 0.66 | 0.58 | 64.6 | 0.88 | 0.45 | 49.6 |

4. | Chamra | ncepp5_ugl, ncepp8_ugl, ncepp850gl | Seasonal/None | 0.67 | 0.58 | 96.7 | 0.77 | 0.44 | 69.5 |

5. | Gondli | ncepp_fgl, ncepp5_ugl, ncepp8_ugl | Seasonal/None | 0.69 | 0.54 | 76.6 | 0.61 | 0.68 | 52.6 |

6. | Kanker | ncepp5_ugl, ncepp8_ugl, ncepp500gl | Monthly/None | 0.68 | 0.55 | 73.6 | 0.55 | 0.72 | 66.14 |

7. | Maramsili | ncepp_ugl, ncepp8_ugl, nceps500gl | Monthly/None | 0.66 | 0.57 | 66.9 | 0.64 | 0.61 | 65.5 |

8. | Rudri | ncepp_ugl, ncepp850gl, ncepmslpgl | Monthly/None | 0.66 | 0.58 | 61.3 | 0.61 | 0.65 | 55.7 |

S.N. | Rain gauge station | Selected predictors | Model type/Transformation | Model performance during calibration | Model performance during validation | ||||
---|---|---|---|---|---|---|---|---|---|

C _{c} | Adj R ^{2} | ɳ | C _{c} | Adj R ^{2} | ɳ (%) | ||||

1. | Ambagarh | ncepp5_fgl, ncepp8_ugl, ncepp850gl | Seasonal/None | 0.78 | 0.41 | 55.7 | 0.79 | 0.40 | 56.54 |

2. | Balod | ncepp5_ugl, ncepp8_ugl, ncepp850gl | Monthly/None | 0.69 | 0.55 | 85.9 | 0.66 | 0.66 | 75.3 |

3. | Bhanpura | ncepp5_ugl, ncepp5_zgl, ncepp8_zgl | Seasonal/None | 0.66 | 0.58 | 64.6 | 0.88 | 0.45 | 49.6 |

4. | Chamra | ncepp5_ugl, ncepp8_ugl, ncepp850gl | Seasonal/None | 0.67 | 0.58 | 96.7 | 0.77 | 0.44 | 69.5 |

5. | Gondli | ncepp_fgl, ncepp5_ugl, ncepp8_ugl | Seasonal/None | 0.69 | 0.54 | 76.6 | 0.61 | 0.68 | 52.6 |

6. | Kanker | ncepp5_ugl, ncepp8_ugl, ncepp500gl | Monthly/None | 0.68 | 0.55 | 73.6 | 0.55 | 0.72 | 66.14 |

7. | Maramsili | ncepp_ugl, ncepp8_ugl, nceps500gl | Monthly/None | 0.66 | 0.57 | 66.9 | 0.64 | 0.61 | 65.5 |

8. | Rudri | ncepp_ugl, ncepp850gl, ncepmslpgl | Monthly/None | 0.66 | 0.58 | 61.3 | 0.61 | 0.65 | 55.7 |

*C _{c}* = Coefficient of correlation,

*Adj R*= Adjusted coefficient of determination,

^{2}*ɳ*= Nash–Sutcliffe efficiency.

### Uncertainty analysis

The uncertainty analysis has been carried out to determine how well the proposed statistical model with GCM predictors predicts the actual rainfall series. The *p*-values obtained from the Wilcoxon test for different rain gauge stations are presented in Table 3. As the *p-values* were found to be greater than 0.05 for all the stations except RCP2.6 of Maramsilli, the null hypothesis of 5% significance level can be accepted, which assumed that there is no significant difference in the observed and simulated monthly series.

Station | RCP2.6 | RCP4.5 | RCP8.5 | |||
---|---|---|---|---|---|---|

V _{s} | P-value | V _{s} | P-value | V _{s} | P-value | |

Ambagarh | 559 | 0.459 | 559 | 0.745 | 0 | 1.00 |

Balod | 612 | 0.174 | 637 | 0.179 | 615 | 0.274 |

Bhanpura | 606 | 0.321 | 674 | 0.078 | 611 | 0.294 |

Gondli | 427 | 0.310 | 591 | 0.410 | 545 | 0.761 |

Chamra | 599 | 0.227 | 565 | 0.596 | 521 | 0.766 |

Kanker | 342 | 0.452 | 400 | 0.166 | 351 | 0.561 |

Maramsilli | 444 | 0.013 | 420 | 0.057 | 409 | 0.057 |

Rudri | 224 | 0.212 | 287 | 0.652 | 276 | 0.528 |

Station | RCP2.6 | RCP4.5 | RCP8.5 | |||
---|---|---|---|---|---|---|

V _{s} | P-value | V _{s} | P-value | V _{s} | P-value | |

Ambagarh | 559 | 0.459 | 559 | 0.745 | 0 | 1.00 |

Balod | 612 | 0.174 | 637 | 0.179 | 615 | 0.274 |

Bhanpura | 606 | 0.321 | 674 | 0.078 | 611 | 0.294 |

Gondli | 427 | 0.310 | 591 | 0.410 | 545 | 0.761 |

Chamra | 599 | 0.227 | 565 | 0.596 | 521 | 0.766 |

Kanker | 342 | 0.452 | 400 | 0.166 | 351 | 0.561 |

Maramsilli | 444 | 0.013 | 420 | 0.057 | 409 | 0.057 |

Rudri | 224 | 0.212 | 287 | 0.652 | 276 | 0.528 |

### Rainfall projection

The multiple rainfall series of three future periods viz., FP-1 (2020–2035), FP-2 (2046–2064) and FP-3 (2081–2099), for eight rain gauge stations in the upper Mahanadi basin have been generated using a weather generator of SDSM for climate forcing scenarios of RCP2.6, RCP4.5 and RCP8.5 The base period and projected average monthly rainfall for the monsoon period and annual rainfall of eight rain gauge stations are presented in Figure 5(a) and 5(b) and Table 4. The seasonal rainfall plays an important role for water resource planning in India. Significant deviation of predicted seasonal rainfall from the present may adversely affect the availability of water in the region. The rainfall anomalies of the rain gauge stations from the base period are presented in Figure 6.

Station | Base period (BP) | RCP2.6 | RCP4.5 | RCP8.5 | ||||||
---|---|---|---|---|---|---|---|---|---|---|

FP-1 2020–2035 | FP-2 2046–2064 | FP-3 2081–2099 | FP-1 2020–2035 | FP-2 2046–2064 | FP-3 2081–2099 | FP-1 2020–2035 | FP-2 2046–2064 | FP-3 2081–2099 | ||

Ambagarh | 1,153 | 1,222.1 | 1,061.1 | 1,050.0 | 1,106.4 | 1,034.8 | 1,067.9 | 1,068.8 | 943.7 | 839.1 |

Balod | 983 | 1,052.7 | 946.0 | 1,008.9 | 916.1 | 909.1 | 851.7 | 1,036.5 | 967.9 | 790.6 |

Bhanpura | 1,524 | 1,216.3 | 1,198.6 | 1,149.9 | 1,277.4 | 1,200.7 | 1,207.3 | 1,168.8 | 1,043.7 | 939.1 |

Chamra | 1,142 | 1,087.4 | 1,051.9 | 942.2 | 983.9 | 1,022.7 | 1,011.6 | 1,047.2 | 971.7 | 955.0 |

Gondli | 967 | 757.6 | 766.6 | 719.6 | 772.6 | 721.1 | 753.7 | 707.8 | 709.0 | 708.8 |

Kanker | 1,109 | 1,212.8 | 1,258.7 | 1,266.2 | 1,107.3 | 1,080.1 | 1,085.6 | 1,110.6 | 1,109.1 | 1,102.2 |

Maramsilli | 1,078 | 1,007.4 | 1,015.5 | 1,100.8 | 841.2 | 996.0 | 1,164.0 | 914.8 | 1,017.7 | 1,123.5 |

Rudri | 1,213 | 1,323.6 | 1,295.8 | 1,468.7 | 1,055.4 | 1,066.2 | 1,342.5 | 1,068.8 | 943.7 | 839.1 |

Station | Base period (BP) | RCP2.6 | RCP4.5 | RCP8.5 | ||||||
---|---|---|---|---|---|---|---|---|---|---|

FP-1 2020–2035 | FP-2 2046–2064 | FP-3 2081–2099 | FP-1 2020–2035 | FP-2 2046–2064 | FP-3 2081–2099 | FP-1 2020–2035 | FP-2 2046–2064 | FP-3 2081–2099 | ||

Ambagarh | 1,153 | 1,222.1 | 1,061.1 | 1,050.0 | 1,106.4 | 1,034.8 | 1,067.9 | 1,068.8 | 943.7 | 839.1 |

Balod | 983 | 1,052.7 | 946.0 | 1,008.9 | 916.1 | 909.1 | 851.7 | 1,036.5 | 967.9 | 790.6 |

Bhanpura | 1,524 | 1,216.3 | 1,198.6 | 1,149.9 | 1,277.4 | 1,200.7 | 1,207.3 | 1,168.8 | 1,043.7 | 939.1 |

Chamra | 1,142 | 1,087.4 | 1,051.9 | 942.2 | 983.9 | 1,022.7 | 1,011.6 | 1,047.2 | 971.7 | 955.0 |

Gondli | 967 | 757.6 | 766.6 | 719.6 | 772.6 | 721.1 | 753.7 | 707.8 | 709.0 | 708.8 |

Kanker | 1,109 | 1,212.8 | 1,258.7 | 1,266.2 | 1,107.3 | 1,080.1 | 1,085.6 | 1,110.6 | 1,109.1 | 1,102.2 |

Maramsilli | 1,078 | 1,007.4 | 1,015.5 | 1,100.8 | 841.2 | 996.0 | 1,164.0 | 914.8 | 1,017.7 | 1,123.5 |

Rudri | 1,213 | 1,323.6 | 1,295.8 | 1,468.7 | 1,055.4 | 1,066.2 | 1,342.5 | 1,068.8 | 943.7 | 839.1 |

The mean seasonal rainfall of 1,096 mm from the Ambagarh station for the base period (1986–2003) was found to be increased to 1159 mm for period FP-1 (2020–2035) under RCP2.5, while all other scenarios indicated decreased seasonal rainfall by 3–30%. The rainfall projection of Bhanpura station indicated a decreasing mean seasonal rainfall of 1,106–1,201, 984–1,083 and 919–1,078 mm for periods FP-1, FP-2 and FP-3, respectively, in comparison to the base period seasonal rainfall of 1,483 mm. The mean seasonal rainfall of 920 mm from Gondli station for the base period was found to be reduced to 720, 738 and 680 mm for the RCP2.6 scenario; 745, 683 and 707 mm for the RCP4.5 scenario; and 679, 682 and 663 mm for the RCP8.5 scenario for future assessment periods of FP-1, FP-2 and FP-3, respectively. The seasonal rainfall at Balod station under the RCP4.5 scenario indicated a significant decrease varying between 8 and 13% for the assessment periods of FP-1 and FP-2. The mean seasonal rainfall of 1,031 mm from Maramsilli station for the base period was reduced by 5–14% for RCP2.6; 1–29% for RCP4.5; and 8–21% for RCP8.5. Apart from other stations, Kanker RG stations showed an increase of projected rainfall in the range of 1,174–1,221 mm under RCP2.6 scenario, 1,057–1,065 mm under RCP4.5 scenario and 1,079–1,094 mm under RCP8.5 climate scenario from the base period mean seasonal rainfall of 1,053 mm.

From the analysis of rainfall anomalies, it has been observed that Ambagarh, Bhanpura, Chamra, Gondli and Maramsilli may receive 4–38% lower seasonal rainfall in all three future periods under all scenarios. Balod and Rudri RG stations indicated a mixed pattern and Kanker RG station may receive higher rainfall under different climatic forcing conditions. It has been observed that all RG stations may get less rainfall if subjected to RCP8.5 followed by RCP4.5 then RCP2.5 scenarios.

### Water availability in reservoirs

The availability of water in reservoirs depends mainly on rainfall in the catchment area. To ascertain impacts of predicted rainfall on water availability in reservoirs, some important reservoirs located in the upper Mahanadi catchment, viz., Ravishankar Sagar, Tandula, Kharkhara, Gondli, Maramsilli and Dudhawa, have been included in the analysis. Annual and seasonal rainfall on the catchment of the respective reservoir for three future periods have been computed using Thiessen weights of rain gauges and compared with the catchment's rainfall of the base period. The rain gauge stations showing impacts on water availability in different reservoirs and their corresponding weights are given in Table 5. The comparison plots of rainfall on the catchment of each reservoir for different climatic scenarios with that of the corresponding base period rainfall are shown in Figure 7(a) and 7(b). The analysis showed that Tandula, Kharkhara, Maramsilli and Gondli reservoir will have less water availability; while Ravishankar Sagar and Dudhawa reservoir showed not much affect by changing climate scenarios despite having a change of catchment rainfall of between –8 and +3% for Ravishankar Sagar, and between –10 and +10% for Dudhawa reservoir. The reduced seasonal rainfall that varied between 12 and 29% for Tandula, 12 and 27% for Kharkhara and 18 and 26% for Gondli reservoir will significantly affect the water availability in the Tandula, Kharkhara and Gondli reservoir.

S.N. | Name of reservoir | Rain gauge station | Weight |
---|---|---|---|

1 | Ravishankar Sagar | Kanker | 0.55 |

Chamra | 0.20 | ||

Maramsilli | 0.12 | ||

Bhanpura | 0.07 | ||

Rudri | 0.05 | ||

2 | Tandula | Bhanpura | 0.39 |

Chamra | 0.25 | ||

Gondli | 0.18 | ||

Balod | 0.18 | ||

3 | Kharkhara | Gondli | 0.66 |

Ambagarh | 0.33 | ||

Bhanpura | 0.02 | ||

4 | Gondli | Gondli | 1.00 |

5 | Maramsilli | Maramsilli | 1.00 |

6 | Dudhawa | Kanker | 0.50 |

Maramsilli | 0.50 |

S.N. | Name of reservoir | Rain gauge station | Weight |
---|---|---|---|

1 | Ravishankar Sagar | Kanker | 0.55 |

Chamra | 0.20 | ||

Maramsilli | 0.12 | ||

Bhanpura | 0.07 | ||

Rudri | 0.05 | ||

2 | Tandula | Bhanpura | 0.39 |

Chamra | 0.25 | ||

Gondli | 0.18 | ||

Balod | 0.18 | ||

3 | Kharkhara | Gondli | 0.66 |

Ambagarh | 0.33 | ||

Bhanpura | 0.02 | ||

4 | Gondli | Gondli | 1.00 |

5 | Maramsilli | Maramsilli | 1.00 |

6 | Dudhawa | Kanker | 0.50 |

Maramsilli | 0.50 |

## CONCLUSIONS

The changing climatic conditions advocate the need for water policy modifications and necessary adaptation strategies for the spatiotemporal management of water resources, both on a regional and a local scale. Using a CanESM2-Canadian Global Circulation Model together with its weather predictor for developing scenarios, rainfalls of eight rain gauge stations in the upper Mahanadi catchment in Chhattisgarh (India) for three future periods FP-1 (2020–2035), FP-2 (2046–2064) and FP-3 (2081–2099) were projected for RCP2.6, RCP4.5 and RCP8.5 climate forcing conditions. The statistical downscaling technique by employing SDSM 5.2 DC was used to select an appropriate set of climate predictors from the analysis of scatter diagrams and percentage reduction methods in a 3-fold cross validation process. The selected sets of predictors were used for generating multiple ensembles and their monthly means for comparison. The models were developed based on statistical downscaling of GCM outputs and by selecting a set of climate predictors could promisingly generate the rainfall series for future periods under different climate forcing conditions.

The analyzed results showed that all rain gauge stations except one will have lesser seasonal and annual rainfall for all the future periods under the climatic forcing conditions than the base period rainfall. Out of eight rain gauge stations, five stations showed reduced future annual rainfall with a variation of 4–38%. Under all projection scenarios, October was found to have higher rainfall than the base period rainfall, while August and September were found to have less rainfall in most of the stations. The upper part of the Mahanadi basin has several important reservoir projects, and water availability in these reservoirs depend mainly on rainfall of their respective catchments. The results further showed that due to plausible climate change, the changing rainfall regime in the upper Mahanadi basin may influence the availability of water in the rivers and reservoirs and may enhance sectoral demands in the region, which essentially insists on the requirement of adaptation measures for sustainable water resource management.

## REFERENCES

*.*

*Nonparametrics: Statistical Methods Based on Ranks*. Holden-Day Inc., San Francisco.

*.*