Eleven general circulation models/global climate models (GCMs) – BCCR-BCCM2.0, INGV-ECHAM4, GFDL2.0, GFDL2.1, GISS, IPSL-CM4, MIROC3, MRI-CGCM2, NCAR-PCMI, UKMO-HADCM3 and UKMO-HADGEM1 – are evaluated for Indian climate conditions using the performance indicator, skill score (SS). Two climate variables, temperature T (at three levels, i.e. 500, 700, 850 mb) and precipitation rate (Pr) are considered resulting in four SS-based evaluation criteria (T500, T700, T850, Pr). The multicriterion decision-making method, technique for order preference by similarity to an ideal solution, is applied to rank 11 GCMs. Efforts are made to rank GCMs for the Upper Malaprabha catchment and two river basins, namely, Krishna and Mahanadi (covered by 17 and 15 grids of size 2.5° × 2.5°, respectively). Similar efforts are also made for India (covered by 73 grid points of size 2.5° × 2.5°) for which an ensemble of GFDL2.0, INGV-ECHAM4, UKMO-HADCM3, MIROC3, BCCR-BCCM2.0 and GFDL2.1 is found to be suitable. It is concluded that the proposed methodology can be applied to similar situations with ease.

INTRODUCTION

General circulation models/global climate models (GCMs) are three-dimensional mathematical models based on principles of fluid dynamics, thermodynamics and radiative heat transfer. These are easily capable of simulating/forecasting present/future values of various climatic parameters. Output of GCMs can be used as the basis to analyze impacts on hydrologic systems (Smith & Chiew 2010). However, uncertainties in developing and applying GCMs, initial condition, boundary condition, model structure and emission scenarios significantly affect their output (Wilby & Harris 2006). These necessitate selection of a suitable GCM/ensemble of GCMs (as sometimes no single model is found to be uniformly superior) so that downscaling of various parameters can be explored with confidence for adaptation at regional/local level for drinking water supply, irrigation, floods and droughts situations. Anandhi et al. (2011) discussed three general approaches for selecting GCMs: namely, use all available GCMs and use the multi-model ensemble mean or choose a subset of GCMs. They also cautioned that ‘neither good performance across an arbitrary suite of measures of observed climate, nor agreement in output across a collection of models, provides a rigorous basis for assessing the accuracy of a future prediction’. Knutti et al. (2010) suggested suitable metric/measures to assess the performance of GCMs, whereas Guilyardi et al. (2009) emphasized that metrics ‘should be concise, physically informative, societally relevant and easy to understand, compute and compare’.

Keeping these points in view, 11 GCMs – BCCR-BCCM2.0, INGV-ECHAM4, GFDL2.0, GFDL2.1, GISS, IPSL-CM4, MIROC3, MRI-CGCM2, NCAR-PCMI, UKMO-HADCM3 and UKMO-HADGEM1 – are chosen from the CMIP3 data base. There is no specific basis for choosing two GFDL variations and only NCAR-PCMI. However, the two variations of GFDL are considered in order to study how they respond to the chosen problem. Necessary data for these models over India have been extracted from the IPCC data distribution center website.

Chosen GCMs are evaluated for: (a) Upper Malaprabha catchment, Karnataka, India; (b) two Indian river basins, Krishna and Mahanadi, covered by 17 and 15 grids of size 2.5° × 2.5°, respectively; and (c) India (covered by 73 grids of size 2.5° × 2.5°) using the performance indicator, skill score (SS) for the climate variable, temperature (three levels, i.e. 500, 700, 850 mb and referred from now as T500, T700, T850) and precipitation rate (Pr). The multicriterion decision-making method (MCDM), technique for order preference by similarity to an ideal solution (TOPSIS), is used to rank the GCMs. Table 1 presents information about the GCMs used in the study. For convenience, only the acronyms mentioned in column 4 are used further.

Table 1

Details of GCMs considered and their acronyms

S. No. (1) GCM (2) Organisation (3) Acronym (4) 
BCCR-BCCM 2.0 Bjerknes Centre for Climate Research, Norway BCCR 
INGV-ECHAM 4 Istituto Nazionale Di Geofisica E Vulcanologia, Italy ECHAM 
GFDL2.0 Geophysical Fluid Dynamic Laboratory, USA GFDL2.0 
GFDL2.1 Geophysical Fluid Dynamic Laboratory, USA GFDL2.1 
GISS Goddard Institute for Space Studies, USA GISS 
IPSL-CM 4 Institut Pierre Simon Laplace, France IPSL 
MIROC3 Centre for Climate Research, Japan MIROC3 
MRI-CGCM2 Meteorological Research Institute, Japan CGCM2 
NCAR-PCMI Parallel Climate Models, NCAR, USA PCMI 
10 UKMO-HADCM3 UK Met Office, UK HADCM3 
11 UKMO-HADGEM1 UK Met Office, UK HADGEM1 
S. No. (1) GCM (2) Organisation (3) Acronym (4) 
BCCR-BCCM 2.0 Bjerknes Centre for Climate Research, Norway BCCR 
INGV-ECHAM 4 Istituto Nazionale Di Geofisica E Vulcanologia, Italy ECHAM 
GFDL2.0 Geophysical Fluid Dynamic Laboratory, USA GFDL2.0 
GFDL2.1 Geophysical Fluid Dynamic Laboratory, USA GFDL2.1 
GISS Goddard Institute for Space Studies, USA GISS 
IPSL-CM 4 Institut Pierre Simon Laplace, France IPSL 
MIROC3 Centre for Climate Research, Japan MIROC3 
MRI-CGCM2 Meteorological Research Institute, Japan CGCM2 
NCAR-PCMI Parallel Climate Models, NCAR, USA PCMI 
10 UKMO-HADCM3 UK Met Office, UK HADCM3 
11 UKMO-HADGEM1 UK Met Office, UK HADGEM1 

LITERATURE REVIEW

Different researchers used various metrics for evaluating GCMs which are explained in brief.

Perkins et al. (2007) assessed 14 GCMs for daily minimum and maximum temperatures and daily rainfall for 12 regions of Australia. The evaluation was based on SS. GCMs were ranked based on each variable SS as well as on average SS. Similar studies were reported in Suppiah et al. (2007), Maximo et al. (2008), Perkins et al. (2009, 2013) and Perkins & Pitman (2009). Model climate performance index and model variability index were developed by Gleckler et al. (2008) for evaluating GCMs. GCM evaluation aspects are also discussed in detail in Mujumdar & Ghosh (2008), Pierce et al. (2009), Macadam et al. (2010), Fordham et al. (2011) and Johnson et al. (2011). Intercomparison of GCMs for hydrologic predictability are discussed by Reshmi Devi & Nagesh Kumar (2010). Johnson & Sharma (2009) and Ojha et al. (2014) applied the variable convergence score method, respectively, for case studies of Australia and India which is used to rank climatic variables based on coefficient of variation of an ensemble of GCMs. Both studies concluded that there is no single widely acceptable metric for assessing climate models. Fu et al. (2013) applied a multicriteria score-based method to assess GCM performance at the regional scale over the southeastern Australia region. Monthly mean sea level pressure, monthly air temperature, monthly and annual rainfall are considered for evaluation of 25 GCMs. They also compared the CMIP5 and CMIP3 based GCMs. It is concluded that results can be used for better regional climate change impact analysis. Su et al. (2013) evaluated performance of 24 GCMs in CMIP5 environment over the eastern Tibetan Plateau by comparing the model outputs with ground observations for precipitation and temperature. It is observed that most GCMs reasonably capture the climatological patterns and spatial variations of the observed climate for temperature. Hughes et al. (2014) assessed the skill of nine GCMs for 15 catchments in five regions of South Africa with the objective of testing whether GCMs are able to reproduce precipitation distribution statistics and patterns of seasonality, persistence and extremes. Differences in the GCMs' skill across the different regions and in the skill ranking between coastal areas and inland regions are observed. Grose et al. (2014) assessed a set of 27 GCMs from the CMIP5 ensemble for their performance for the western tropical Pacific and differences from CMIP3. It is concluded that careful interpretation and consideration of biases is required when using CMIP5 outputs for generating regional climate projections.

Limitations of the above-mentioned studies are as follows:

  • The SS approach was not used for assessing suitability of GCMs and adaptability for Indian climate conditions.

  • No study has used the MCDM process where all criteria are simultaneously considered for ranking GCMs.

The objectives of the present study, addressing the above limitations, are summarized below:

  • To analyze the capability of chosen GCMs to simulate monthly precipitation and temperature (at three levels) using a probability based SS.

  • To develop a methodology for ranking GCMs that can be used as the basis for hydrological modelling applications.

  • To evaluate the suitability of GCMs for Indian conditions.

STUDY AREA

Upper Malaprabha catchment

Upper Malaprabha catchment, Karnataka State, India is located between latitudes 15°00′ and 16°12′ N and longitudes 74°14′ and 76°05′ E. The catchment area of the river up to the dam site is 2,564 km2. The area of the reservoir at full reservoir level is 13,578 ha. The reservoir has live storage capacity of 870 Mm3. The mean annual rainfall in the command area is 576 mm. Figure 1 presents a location map of the Malaprabha reservoir. There are two main canals in this project. The left bank canal serves a command area of 53,137 ha and the right bank canal serves 128,634 ha.

Figure 1

Location map of Upper Malaprabha catchment, Krishna and Mahanadi river basins.

Figure 1

Location map of Upper Malaprabha catchment, Krishna and Mahanadi river basins.

Krishna and Mahanadi river basins

Figure 1 also presents a location map of the Krishna and Mahanadi river basins. Krishna basin lies between latitude 13°07′ and 19°25′ N and longitude 73°21′ and 81°09′ E. The river rises in the western ghats at an altitude of 1,337 m just north of Mahabaleshwar, about 64 km from the Arabian Sea. The catchment area is 258,948 km2 spread over the states of Maharashtra (26.8%), Karnataka (43.7%) and Andhra Pradesh (29.5%). Mahanadi basin lies between latitude 19°20′ and 23°35′ N and longitude 80°30′ and 86°50′ E. The catchment area is 141,600 km2 spread over the states of Chhattisgarh, Orissa, Madhya Pradesh, Jharkhand and Maharashtra, of which more than 99% is in Chhattisgarh and Orissa.

India

India lies between longitudes 68°7′ and 97°25′ E and latitudes 8°4′ and 37°6′ N. The average annual rainfall is 125 cm, but with significant spatial variations. In addition, there is significant regional and temporal variation in the rainfall distribution. Around 80% of the annual rainfall is received in the four rainy months of June to September.

SS AND ANALYSIS

SS (Perkins et al. 2007) measures the similarity between two probability density functions (PDFs) related to GCM-based simulation and observation. It measures the amount of overlap between GCM-based PDF and observed PDF. The SS approach estimates the cumulative minimum value of GCM-based and observation-based distributions of each binned value. This results in the measurement of the common area between the two PDFs expressed as the SS. If a model simulates the observed conditions perfectly, the SS will be one, which is the total sum of the binned values in the given PDFs. If a model simulates the observed PDF poorly, the SS will be close to zero (Maximo et al. 2008; Anandhi et al. 2011). SS is expressed as 
formula
1
where and are the frequency of values in the given bin from the chosen GCM and observed data; nb is number of bins used to calculate the PDF for a given region.

DESCRIPTION OF TOPSIS

TOPSIS is based on the principle that the chosen alternative should have the shortest distance from the ideal solution and furthest distance from the anti-ideal solution (Chen & Hwang 1992; Opricovic & Tzeng 2004; Raju & Nagesh Kumar 2010). The methodology of TOPSIS consists of the following:

  1. Computation of separation measure of each alternative a from the ideal solution, that is, Euclidean distance of each criterion from its ideal value, and summing these for all criteria (j = 1, 2, …. J) for given alternative a, that is, 
    formula
    2
    where j = 1, 2, … J; normalized value of criterion j for alternative a; normalized ideal value of criterion j; weight assigned to the criterion j.
  2. Computation of separation measure of each alternative a from the anti-ideal solution, that is, Euclidean distance of each criterion from its anti-ideal value, and summing these for all criteria (j = 1, 2, …. J) for given alternative a, that is, 
    formula
    3
    where normalized anti-ideal value of criterion j.
  3. Computation of relative closeness of each alternative a is 
    formula
    4

The alternatives are ranked based on the values. The higher the value, the better the alternative. Figure 2 presents a flow chart of the present methodology.

Figure 2

Flow chart of the proposed methodology.

Figure 2

Flow chart of the proposed methodology.

RESULTS AND DISCUSSION

Case study of Upper Malaprabha catchment

Six hundred data sets grid wise relating to the period 1950–1999 (12 per year for 50 years) obtained from NCEP-NCAR reanalysis environment are compared with historical runs obtained from each of the 11 GCMs to assess the SS. Interactive code is developed for computation of the SS (grid wise) for the climate variable, temperature (° Kelvin) and precipitation (kg/m2/s). In the case of precipitation, 30 bins (nb = 30) each of size 0.0000157 kg/m2/s are employed whereas 58 bins each of size 5° Kelvin are used for temperature (Maximo et al. 2008). Frequencies of values within each bin are then calculated for observed data () and general circulation model output () based on which SS is computed for T500, T700, T850, and P (Equation (1)).

SS values for Upper Malaprabha catchment for Pr, T500, T700, T850 are presented in Table 2 (columns 2, 3, 4, 5) and Figure 3. It is observed from Table 2 that for precipitation the minimum SS is 0.6283 for PCMI whereas the maximum is 0.8567 for MIROC3 (column 2); for T500, minimum and maximum SSs are 0.2833 and 0.5533 for GISS/IPSL and GFDL2.0/GFDL2.1/BCCR (column 3); for T700, minimum and maximum SSs are 0.1783 and 0.4100 for PCMI and HADCM3 (column 4); and for T850, minimum and maximum SSs are 0.315 and 0.525 for GISS and CGCM2 (column 5). The above analysis indicates that ranking is different for each GCM with reference to each criterion. It is also observed that MIROC3, GFDL2.0/GFDL2.1/BCCR, HADCM3, CGCM2 occupy the first position (due to their higher SSs). SSs (columns 2, 3, 4, 5) for each GCM for Pr, T500, T700, T850 are the inputs to the MCDM method TOPSIS.

Table 2

SS values and outcome of TOPSIS for the chosen 11 GCMs for Upper Malaprabha catchment

  SS values related to climatic variables Outcome of TOPSIS 
GCM (1) Pr (2) T500 (3) T700 (4) T850 (5) Da+ (6) Da (7) Ca (8) Rank (9) 
BCCR 0.7717 0.5533 0.2583 0.3300 0.2796 0.2595 0.4814 
ECHAM 0.6833 0.3483 0.2800 0.3917 0.2972 0.2264 0.4324 
GFDL2.0 0.8150 0.5533 0.4000 0.4183 0.2414 0.2856 0.5420 
GFDL2.1 0.8350 0.5533 0.2633 0.3783 0.2688 0.2757 0.5063 
GISS 0.7783 0.2833 0.2033 0.3150 0.3228 0.2273 0.4132 
IPSL 0.6583 0.2833 0.2017 0.4167 0.3170 0.2133 0.4022 10 
MIROC3 0.8567 0.3583 0.3867 0.3800 0.2730 0.2688 0.4961 
CGCM2 0.7550 0.4517 0.2367 0.5250 0.2703 0.2629 0.4931 
PCMI 0.6283 0.3217 0.1783 0.3367 0.3272 0.2005 0.3800 11 
HADCM3 0.8100 0.3483 0.4100 0.4100 0.2689 0.2638 0.4952 
HADGEM1 0.7883 0.2850 0.2067 0.3983 0.3110 0.2377 0.4332 
  SS values related to climatic variables Outcome of TOPSIS 
GCM (1) Pr (2) T500 (3) T700 (4) T850 (5) Da+ (6) Da (7) Ca (8) Rank (9) 
BCCR 0.7717 0.5533 0.2583 0.3300 0.2796 0.2595 0.4814 
ECHAM 0.6833 0.3483 0.2800 0.3917 0.2972 0.2264 0.4324 
GFDL2.0 0.8150 0.5533 0.4000 0.4183 0.2414 0.2856 0.5420 
GFDL2.1 0.8350 0.5533 0.2633 0.3783 0.2688 0.2757 0.5063 
GISS 0.7783 0.2833 0.2033 0.3150 0.3228 0.2273 0.4132 
IPSL 0.6583 0.2833 0.2017 0.4167 0.3170 0.2133 0.4022 10 
MIROC3 0.8567 0.3583 0.3867 0.3800 0.2730 0.2688 0.4961 
CGCM2 0.7550 0.4517 0.2367 0.5250 0.2703 0.2629 0.4931 
PCMI 0.6283 0.3217 0.1783 0.3367 0.3272 0.2005 0.3800 11 
HADCM3 0.8100 0.3483 0.4100 0.4100 0.2689 0.2638 0.4952 
HADGEM1 0.7883 0.2850 0.2067 0.3983 0.3110 0.2377 0.4332 
Figure 3

SS related to various GCMs for Upper Malaprabha catchment.

Figure 3

SS related to various GCMs for Upper Malaprabha catchment.

Equal weights of 0.25 are considered for each criterion, Pr, T500, T700, T850 for TOPSIS analysis. In TOPSIS, the ideal value is assigned as one (higher SS is desirable), and the anti-ideal value is assigned as zero (lowest SS) while the optimal value is the ideal value. The outputs from TOPSIS are (Equation (2)), (Equation (3)), (Equation (4)) and corresponding ranking pattern (Raju & Nagesh Kumar 2010). (column 6), (column 7) and (column 8) for the 11 chosen GCMs are presented in Table 2 along with ranking pattern (column 9). GFDL2.0, GFDL2.1, MIROC3 occupied the first three positions (with values of 0.5420, 0.5063, and 0.4961) and IPSL, PCMI occupied last two positions with values of 0.4022 and 0.3800. It is suggested that an ensemble of GFDL2.0, GFDL2.1 and MIROC3 is suitable for the Upper Malaprabha catchment. Ensemble mean can be used for downscaling precipitation and temperature. A sample calculation for using TOPSIS is presented in the Appendix (available online at http://www.iwaponline.com/jwc/006/074.pdf).

Case study of river basins

The suitability of GCMs for the two river basins, Krishna and Mahanadi in India, is also evaluated. Table 3 presents the latitude and longitude (Integrated Hydrological Data Book 2012 ) along with number of grid points covered by each basin. The suitability of GCMs in terms of the first and second ranks is presented in columns 4 and 5 of Table 3. The GCMs that occupied the first three ranks in the average ranking perspective, which allows a consensus to be reached (Bui 1987), are also presented in column 6. GCMs that occupy (a) the first three ranks in the average ranking perspective and (b) the first and second ranks in more than two grid points among the relevant grid points are considered for evolving a suitable ensemble and are shown in column 7. Some salient features which are the outcome of the present study are presented below for each basin.

Table 3

Suitability of GCMs for the Krishna and Mahanadi river basins

River basin (1) Latitude and longitude (2) Number of grid points covered (3) Suitability of GCMs (first rank) (4) Suitability of GCMs (second rank) (5) First three ranks in average ranking perspective (6) Suggested ensemble (7) 
Krishna 13°07′ N to 19°25′N;
73°21′ E to 81°09′ E 
∼17 GFDL2.0 (9) MIROC3 (5) BCCR (3), GFDL2.0 (5), GFDL2.1 (5) GFDL2.0, MIROC3, GFDL2.1 GFDL2.0, MIROC3, GFDL2.1, BCCR 
Mahanadi 19°20′ N to 23°35′ N;
80°30′ E to 86°50′ E 
∼15 MIROC3 (4) GFDL2.1 (3), PCMI (3), HADGEM1 (3) GFDL2.1, MIROC3, CGCM2 MIROC3, GFDL2.1, CGCM2, PCMI, HADGEM1 
River basin (1) Latitude and longitude (2) Number of grid points covered (3) Suitability of GCMs (first rank) (4) Suitability of GCMs (second rank) (5) First three ranks in average ranking perspective (6) Suggested ensemble (7) 
Krishna 13°07′ N to 19°25′N;
73°21′ E to 81°09′ E 
∼17 GFDL2.0 (9) MIROC3 (5) BCCR (3), GFDL2.0 (5), GFDL2.1 (5) GFDL2.0, MIROC3, GFDL2.1 GFDL2.0, MIROC3, GFDL2.1, BCCR 
Mahanadi 19°20′ N to 23°35′ N;
80°30′ E to 86°50′ E 
∼15 MIROC3 (4) GFDL2.1 (3), PCMI (3), HADGEM1 (3) GFDL2.1, MIROC3, CGCM2 MIROC3, GFDL2.1, CGCM2, PCMI, HADGEM1 

Note: GCMs appearing in two grid points in first and second rank positions are not presented here and not considered for the ensemble; values in parentheses represent the number of grid points in which GCMs appears.

Krishna river basin

  • GFDL2.0, MIROC3 and GFDL2.1 attained the first three ranks (based on average ranking of 17 grids) as shown in column 6.

  • First rank was attained by GFDL2.0, GFDL2.1, GISS, MIROC3, CGCM2 in 9, 1, 1, 5, 1 grid points, respectively. It is also observed that GFDL2.0 and MIROC3 occupied 82.35% of 17 grids.

  • BCCR, GFDL2.0, GFDL2.1, MIROC3, CGCM2 attained second rank in 3, 5, 5, 2, 2 grid points, respectively. It is also observed that GFDL2.0 and GFDL2.1 occupied 58.82% of 17 grids.

  • Hence, the ensemble of GFDL2.0, MIROC3, GFDL2.1 and BCCR is suggested for Krishna river basin.

Mahanadi river basin

  • GFDL2.1, MIROC3 and CGCM2 attained the first three ranks (based on average ranking of 15 grids).

  • BCCR, ECHAM, GFDL2.0, GFDL2.1, IPSL, MIROC3, CGCM2, PCMI, HADGEM1 attained the first rank in 1, 1, 1, 1, 1, 4, 2, 2, 2 grid points, respectively. It is also observed that MIROC3, CGCM2, PCMI and HADGEM1 occupied 66.67% (out of 15).

  • BCCR, GFDL2.0, GFDL2.1, MIROC3, CGCM2, PCMI, HADCM3, HADGEM1 attained the second rank in 1, 1, 3, 2, 1, 3, 1, 3 grid points, respectively. It is also observed that GFDL2.1, MIROC3, PCMI and HADGEM1 occupied 73.33% out of 15.

  • Hence, the ensemble of MIROC3, GFDL2.1, CGCM2, PCMI and HADGEM1 is suggested for the Mahanadi river basin.

Case study of India

For this purpose, 73 grid points (2.5° × 2.5°) covering India are identified and the corresponding 73 payoff matrices are evolved. A similar process of application of TOPSIS as performed for the Upper Malaprabha catchment is repeated for 73 grid points for India and an analysis of the results is presented in Table 4. It is observed from Table 4 that GFDL2.0, ECHAM, HADCM3 and MIROC3 attained first rank in 19, 11, 11, 10 grid points, respectively (69.9%), whereas GFDL2.0, GFDL2.1, BCCR and ECHAM attained second rank in 13, 13, 13, 11 grid points, respectively (68.5%). In the first rank scenarios, GFDL2.0 and (ECHAM, HADCM3) occupied 19 and 11 grid points whereas in second rank scenarios (GFDL2.0, GFDL2.1, BCCR) and ECHAM occupied 13 and 11 grid points. Third rank is attained by BCCR, GFDL2.1, in 19 and 13 grid points. It is interesting to note that MIROC3 never attained 10th rank whereas BCCR, GFDL2.0, CGCM2 never attained 11th rank. Table 5 presents spatial distribution of GCMs for the first rank. It is observed from Table 4 that GFDL2.0 occupied the first position in 19 grid points whereas GFDL2.1 occupied first position in four grid points contributing 23 out of 73 grid points; for second position, these two together contribute 26 grid points out of 73. Figures 4(a) and 4(b) present the number of grid points in which GCMs attained particular ranks. Numbers 1 to 11 shown in the legend represent the ranking pattern. It is also observed that for 17.5° latitude and 75°, 77.5°, 80° longitude, GFDL2.0 is preferred whereas for 20° latitude and 75°, 77.5°, 80° longitude, MIROC3 is preferred. Similar types of inference were drawn for various combinations of latitude and longitudes.

Table 4

Number of grid points among 73 in India in which GCMs have attained a particular rank using TOPSIS

  Ranking positions 
GCM 10 11 
BCCR 13 19 
ECHAM 11 11 17 
GFDL2.0 19 13 10 
GFDL2.1 13 13 17 11 
GISS 11 16 
IPSL 17 18 10 
MIROC3 10 10 
CGCM2 13 15 10 
PCMI 14 15 
HADCM3 11 11 12 
HADGEM1 10 10 
  Ranking positions 
GCM 10 11 
BCCR 13 19 
ECHAM 11 11 17 
GFDL2.0 19 13 10 
GFDL2.1 13 13 17 11 
GISS 11 16 
IPSL 17 18 10 
MIROC3 10 10 
CGCM2 13 15 10 
PCMI 14 15 
HADCM3 11 11 12 
HADGEM1 10 10 
Table 5

TOPSIS selected GCMs (first rank) for different grid points in India

S. No  Lat Long GCM S. No  Lat Long GCM 
7.5 77.5 GFDL 2.0 37 22.5 92.5 HADGEM1 
7.5 95 BCCR 38 25 70 GFDL 2.0 
10 77.5 GFDL 2.0 39 25 72.5 GFDL 2.1 
10 80 BCCR 40 25 75 GFDL 2.0 
10 92.5 GFDL 2.1 41 25 77.5 MIROC3 
12.5 72.5 MIROC3 42 25 80 MIROC3 
12.5 75 GFDL 2.0 43 25 82.5 BCCR 
12.5 77.5 GFDL 2.0 44 25 85 CGCM2 
12.5 80 GFDL 2.0 45 25 87.5 ECHAM 
10 12.5 92.5 CGCM2 46 25 90 ECHAM 
11 15 75 GFDL 2.0 47 25 92.5 GFDL 2.0 
12 15 77.5 GFDL 2.0 48 25 95 BCCR 
13 15 80 GFDL 2.0 49 27.5 70 GFDL 2.0 
14 15 82.5 GFDL 2.1 50 27.5 72.5 GFDL 2.0 
15 17.5 72.5 MIROC3 51 27.5 75 HADCM3 
16 17.5 75 GFDL 2.0 52 27.5 77.5 HADCM3 
17 17.5 77.5 GFDL 2.0 53 27.5 80 ECHAM 
18 17.5 80 GFDL 2.0 54 27.5 82.5 ECHAM 
19 17.5 82.5 CGCM2 55 27.5 85 ECHAM 
20 17.5 85 IPSL 56 27.5 87.5 ECHAM 
21 20 72.5 MIROC3 57 27.5 90 ECHAM 
22 20 75 GISS 58 27.5 92.5 BCCR 
23 20 77.5 MIROC3 59 27.5 95 BCCR 
24 20 80 MIROC3 60 30 72.5 GFDL 2.0 
25 20 82.5 MIROC3 61 30 75 HADCM3 
26 20 85 PCMI 62 30 77.5 HADCM3 
27 20 87.5 HADGEM1 63 30 80 ECHAM 
28 22.5 70 GFDL 2.0 64 30 95 BCCR 
29 22.5 72.5 GFDL 2.0 65 32.5 75 HADCM3 
30 22.5 75 HADCM3 66 32.5 77.5 HADCM3 
31 22.5 77.5 MIROC3 67 32.5 80 ECHAM 
32 22.5 80 MIROC3 68 35 75 HADCM3 
33 22.5 82.5 GFDL 2.1 69 35 77.5 HADCM3 
34 22.5 85 PCMI 70 35 80 ECHAM 
35 22.5 87.5 HADGEM1 71 37.5 72.5 ECHAM 
36 22.5 90 HADGEM1 72 37.5 75 HADCM3 
    73 37.5 77.5 HADCM3 
S. No  Lat Long GCM S. No  Lat Long GCM 
7.5 77.5 GFDL 2.0 37 22.5 92.5 HADGEM1 
7.5 95 BCCR 38 25 70 GFDL 2.0 
10 77.5 GFDL 2.0 39 25 72.5 GFDL 2.1 
10 80 BCCR 40 25 75 GFDL 2.0 
10 92.5 GFDL 2.1 41 25 77.5 MIROC3 
12.5 72.5 MIROC3 42 25 80 MIROC3 
12.5 75 GFDL 2.0 43 25 82.5 BCCR 
12.5 77.5 GFDL 2.0 44 25 85 CGCM2 
12.5 80 GFDL 2.0 45 25 87.5 ECHAM 
10 12.5 92.5 CGCM2 46 25 90 ECHAM 
11 15 75 GFDL 2.0 47 25 92.5 GFDL 2.0 
12 15 77.5 GFDL 2.0 48 25 95 BCCR 
13 15 80 GFDL 2.0 49 27.5 70 GFDL 2.0 
14 15 82.5 GFDL 2.1 50 27.5 72.5 GFDL 2.0 
15 17.5 72.5 MIROC3 51 27.5 75 HADCM3 
16 17.5 75 GFDL 2.0 52 27.5 77.5 HADCM3 
17 17.5 77.5 GFDL 2.0 53 27.5 80 ECHAM 
18 17.5 80 GFDL 2.0 54 27.5 82.5 ECHAM 
19 17.5 82.5 CGCM2 55 27.5 85 ECHAM 
20 17.5 85 IPSL 56 27.5 87.5 ECHAM 
21 20 72.5 MIROC3 57 27.5 90 ECHAM 
22 20 75 GISS 58 27.5 92.5 BCCR 
23 20 77.5 MIROC3 59 27.5 95 BCCR 
24 20 80 MIROC3 60 30 72.5 GFDL 2.0 
25 20 82.5 MIROC3 61 30 75 HADCM3 
26 20 85 PCMI 62 30 77.5 HADCM3 
27 20 87.5 HADGEM1 63 30 80 ECHAM 
28 22.5 70 GFDL 2.0 64 30 95 BCCR 
29 22.5 72.5 GFDL 2.0 65 32.5 75 HADCM3 
30 22.5 75 HADCM3 66 32.5 77.5 HADCM3 
31 22.5 77.5 MIROC3 67 32.5 80 ECHAM 
32 22.5 80 MIROC3 68 35 75 HADCM3 
33 22.5 82.5 GFDL 2.1 69 35 77.5 HADCM3 
34 22.5 85 PCMI 70 35 80 ECHAM 
35 22.5 87.5 HADGEM1 71 37.5 72.5 ECHAM 
36 22.5 90 HADGEM1 72 37.5 75 HADCM3 
    73 37.5 77.5 HADCM3 
Figure 4

Number of grid points in which GCMs attained a particular rank in case of: (a) 6 GCMs and (b) 5 GCMs.

Figure 4

Number of grid points in which GCMs attained a particular rank in case of: (a) 6 GCMs and (b) 5 GCMs.

The average ranking method (Bui 1987) is employed to aggregate the ranking related to 73 grid points of India for each GCM employed to arrive at a consensus. It is noted that BCCR, GFDL2.0 and GFDL2.1 attained first three ranks whereas GISS and IPSL attained the last two ranks. It can be observed from Tables 4 and 5 that an ensemble of GCMs is to be evolved as no single GCM can be recommended for India. The GCMs which: (a) attained the first three ranks in average ranking perspective; and (b) attained the first and second ranks (in more than 10 grid points in 73 grid points covering India) are taken into consideration in formulating the ensemble. Accordingly, the ensemble of GFDL2.0, ECHAM, HADCM3, MIROC3, BCCR and GFDL2.1 is suggested for India both for precipitation rate and temperature.

To the authors' knowledge, this is the first application of SS methodology for ranking GCMs using the MCDM method TOPSIS in Indian conditions; that is, for one catchment, two river basins and the whole of India. In the future, CMIP5 data sets will be used for further studies with more decision making methods and more indicators in Indian conditions.

SUMMARY AND CONCLUSIONS

The SS measure is used to rank 11 GCMs – BCCR, ECHAM, GFDL2.0, GFDL2.1, GISS, IPSL, MIROC3, CGCM2, PCMI, HADCM3 and HADGEM1 – for Upper Malaprabha catchment, two river basins, Krishna and Mahanadi, and India, for precipitation rate and temperature. TOPSIS is employed to rank the 11 GCMs. Specific conclusions emanating from the present study are as follows:

  1. It is observed that GFDL2.0, GFDL2.1 and MIROC3 attained first three ranks and IPSL and PCMI occupied last two positions for Upper Malaprabha catchment.

  2. The ensemble of GFDL2.0, ECHAM, HADCM3, MIROC3, BCCR and GFDL2.1 is suggested for India as no single GCM is recommended.

  3. The ensemble of GFDL2.0, MIROC3, GFDL2.1 and BCCR is suggested for Krishna river basin whereas MIROC3, GFDL2.1, CGCM2, PCMI and HADGEM1 is suggested for Mahanadi river basin.

ACKNOWLEDGEMENTS

This work was supported by Council of Scientific and Industrial Research, New Delhi, through project no. 23(0023)/12/EMR-II dated 15 October 2012. Authors are grateful to Richa Ojha, Research Scholar, School of Civil Engineering, Purdue University, USA and H. R. Shwetha, Research Scholar, Department of Civil Engineering, Indian Institute of Science, Bangalore, India, for their valuable inputs. Authors acknowledge the modeling groups for making their simulations available for analysis, the Program for Climate Model Diagnosis and Intercomparison (PCMDI) for collecting and archiving the CMIP3 model output, and the WCRP's Working Group on Coupled Modelling (WGCM) for organizing the model data analysis activity.

REFERENCES

REFERENCES
Anandhi
A.
Frei
A.
Pradhanang
S. M.
Zion
M. S.
Pierson
D. C.
Schneiderman
E. M.
2011
AR4 climate model performance in simulating snow water equivalent over Catskill Mountain watersheds, New York, USA
.
Hydrol. Proc.
25
,
3302
3311
.
Bui
T. X.
1987
Coop: A Group Decision Support System for Cooperative Multiple Criteria Group Decision Making
.
Springer Verlag
,
Berlin
.
Chen
S. J.
Hwang
C. L.
1992
Fuzzy Multi Attribute Decision Making: Methods and Applications
.
Springer-Verlag
,
Berlin
.
Fordham
D. A.
Wigley
T. L.
Brook
B. W.
2011
Multi-model climate projections for biodiversity risk assessments
.
Ecol. Appl.
21
,
3316
3330
.
Fu
G.
Zhaofei
L.
Charles
S. P.
Xu
Z.
Zhijun
Y.
2013
Score-based method for assessing the performance of GCMs: a case study of southeastern Australia
.
J. Geophys. Res. Atmos.
118
,
4154
4167
.
Gleckler
P. J.
Taylor
K. E.
Doutriaux
C.
2008
Performance metrics for climate models
.
J. Geophys. Res.
113
,
D06104
.
Grose
M. R.
Brown
J. N.
Narsey
A.
Brown
J. R.
Murphy
B. F.
Langlais
C.
Gupta
A. S.
Moise
A. F.
Irving
D. B.
2014
Assessment of the CMIP5 global climate model simulations of the western tropical Pacific climate system and comparison to CMIP3
.
Int. J. Climatol.
34
(12),
3382
3399
.
Guilyardi
E.
Wittenberg
A.
Fedorov
A.
Collins
M.
Wang
C.
Capotondi
A.
Oldenborgh
G. J. V.
Stockdale
T.
2009
Understanding El Nino in ocean–atmosphere general circulation models
.
Bull. Am. Meteorol. Soc.
90
,
325
340
.
Integrated Hydrological Data Book
2012
Central Water Commission
.
New Delhi, India
.
Johnson
F.
Westra
S.
Sharma
A.
Pitman
A. J.
2011
An assessment of GCM skill in simulating persistence across multiple time scales
.
J. Clim.
24
,
3609
3623
.
Knutti
R.
Abramowitz
G.
Collins
M.
Eyring
V.
Gleckler
P. J.
Hewitson
B.
Mearns
L.
2010
Good practice guidance paper on assessing and combining multi model climate projections
. In:
Meeting Report of the Intergovernmental Panel on Climate Change Expert Meeting on Assessing and Combining Multi Model Climate Projections (
Stocker
T. F.
Qin
D.
Plattner
G. K.
Tignor
M.
Midgley
P. M.
, eds).
IPCC Working Group I Technical Support Unit, University of Bern
,
Bern, Switzerland
.
Ojha
R.
Kumar
D. N.
Sharma
A.
Mehrotra
R.
2014
Assessing GCM Convergence for the Indian region using the variable convergence score
.
J. Hydrol. Eng.
19
(
6
),
1237
1246
.
Pierce
D. W.
Barnett
T. P.
Santer
B. D.
Gleckler
P. J.
2009
Selecting global climate models for regional climate change studies
.
Proc. Natl. Acad. Sci. USA
106
,
8441
8446
.
Raju
K. S.
Nagesh Kumar
D.
2010
Multicriterion Analysis in Engineering and Management
.
Prentice Hall of India
,
New Delhi
.
Reshmi Devi
T. V.
Nagesh Kumar
D.
2010
Intercomparison of general circulation models for hydrologic predictability
. In:
Sustainable Water Resources Management and Impact of Climate Change
(
Raju
K. S.
Vasan
A.
, eds).
B.S. Publications
,
Hyderabad
, pp.
33
43
.
Smith
I.
Chiew
F.
2010
Document and Assess Methods for Generating Inputs to Hydrological Models and Extend Delivery of Projections across Victoria. Final report for Project 2.2.5P
,
South Eastern Australian Climate Initiative: CSIRO Land and Australia
.
Suppiah
R.
Hennessy
K. J.
Whetton
P. H.
McInnes
K.
Macadam
I.
Bathols
J.
Ricketts
J.
Page
C. M.
2007
Australian climate change projections derived from simulations performed for the IPCC 4th assessment report
.
Aust. Meteorol. Mag.
56
,
131
152
.

Supplementary data