Ranking of general circulation models for Surat City by using a hybrid approach

In hydrological designs for non-structural and structural flood control measures, especially in urban settings, quantifying precipitation extremes is essential. It is crucial to have reliable estimates of extreme rainfall intensities for efficient urban infrastructure design. Numerous examples abound in gross overestimation or underestimation of designs based on established techniques that lead to devastating consequences (Rupa et al. 2015). General circulation models (GCMs) are numerical illustrations of the land surface, ocean, and atmospheric processes developed based on physical-based empirical relationships and physical laws (Hassan et al. 2020). These can easily simulate/forecast different climatic parameters’ present/future values. As a basis for evaluating impacts on hydrological systems, GCM outputs can be used. However, these outputs are significantly affected by uncertainties in development and application of GCMs, boundary condition, initial condition, emission scenarios, and model structure. This requires the selection of an appropriate GCM or set of GCMs (since sometimes no solitary model is found to be consistently supercilious) (Raju & Kumar 2015). The GCM ensemble subset selection needs a method tailored towards the efficiency of model performance or dependence in climate projection impact investigation. Usually, existing methods adopt two approaches. First is the ‘past performance approach,’ which relies on the ability of GCMs to imitate past climates but does not take future forecasts into account. Second is the ‘envelope approach,’ which selects GCMs but does not contemplate the potential of GCMs to replicate the past climate following their agreement with future climate projections. The ‘hybrid approach’ refers to combining the envelope approach with the past performance approach (Hassan et al. 2020). The hybrid approach takes past performance of GCMs and future climate projections of GCMs into account. Thus, the hybrid approach is used in the present study.

The NASA Ames Research Center and the Climate Analytics Group prepared the NEX-GDDP dataset by utilizing the NASA Earth Exchange. They distributed it through the NASA Center for Climate Simulation (NCCS) (https://developers.google.com/earth-engine/datasets/catalog/NASA_NEX-GDDP, accessed 5 May 2020; Thrasher et al. 2012). For various earth science groups, NEX-GDDP has considerable potential to become a widely used high-resolution dataset and a modern climate change standard (Bao & Wen 2017). Therefore, this dataset is used in the present study. India's coastal cities behave differently from the viewpoint of climate change, and one of them is Surat (Desai et al. 2015). Surat remains at high risk from flooding, and to address these risks, urban expansion has not been managed. Climate change is likely to worsen flood risks by incrementing precipitation in and around Surat City (Bhat et al. 2013). Thus, to study the effect of climate change on Surat City's stormwater drainage network, a selection of suitable ensemble of GCMs and generation future annual maximum rainfall series is necessary. Therefore, ranking of GCMs and generation of future annual maximum rainfall series is needed for Surat City.

Some recent studies have been carried out outside of India on the ranking of GCMs and/or the generation of future precipitation/rainfall time series by using suitable GCMs (Hassan et al. 2020; Khayyun et al. 2020; Rana et al. 2020; Salman et al. 2020). Various studies have also been carried out in India on the topic of Ranking of GCMs and/or generation of future precipitation/rainfall series by using suitable GCMs (Bal et al. 2016; Shashikanth & Sukumar 2017; Das et al. 2018; Hengade et al. 2018; Khan & Koch 2018; Thasneem et al. 2019). A few critical studies carried out in India are discussed below.

Rupa et al. (2015) considered 26 GCMs of Coupled Model Intercomparison Project Phase 5 (CMIP5) along with four Representative Concentration Pathways (RCP) scenarios for the case study of Bangalore City in India to study the climate change effects and to obain predicted IDF relationships.

Raju & Kumar (2015) used the skill score (SS) and performance indicator to rank 11 GCMs for the Upper Malaprabha catchment, two river basins, viz. Mahanadi and Krishna basins, and India corresponding to two variables, namely temperature and rate of precipitation. For the ranking of eleven GCMs, the TOPSIS technique was used.

Raju et al. (2016) assessed the minimum temperature (T_min) and maximum temperature (T_max) simulations for India, derived from 36 CMIP5 GCMs and corresponding data extending over 40 grid points. The correlation coefficient (CC), SS, and normalized root mean square error (NRMSE) were used for assessing GCMs. The entropy method was utilized to calculate the weights for the indicators mentioned above. However, equal weights were also considered as part of the sensitivity analysis tests. For the ranking of GCMs, the compromise programming (CP) technique was utilized.

From the reviewed literature the following research gaps are inferred: (i) none of the reviewed studies carried out ranking of GCMs for Surat City by using fine resolution dataset, (ii) none of the reviewed studies used a combination of ranking of GCMs (past performance) and reliability ensemble approach (REA) (envelope approach), i.e. hybrid approach for generation of future annual maximum rainfall series for Surat City to reduce the uncertainty in the projection of future annual maximum rainfall, (iii) none of the reviewed studies used the REA method for generation of future annual maximum rainfall series for Surat City.

All the aforesaid research gaps are addressed in the present study by: (i) ranking of GCMs for Surat City by using fine resolution dataset, (ii) combination of ranking of GCMs and REA method, i.e. hybrid approach for generation of future annual maximum rainfall series for Surat city to reduce uncertainty in the projection of future annual maximum rainfall and (iii) use of the REA approach for generation of future annual maximum rainfall series for Surat City (by using an ensemble of the top five ranked GCMs). Thus, in the current work, a hybrid approach is used to generate future annual maximum rainfall series for Surat City (having less uncertainty) based on RCP scenarios 4.5 and 8.5.

The methodology used in the present study consisted of firstly extracting CMIP5 GCMs data for the study area. Then, by applying performance criteria, performance indicators were evaluated for each GCM. The weight for each performance indicator was then determined by using the entropy approach. Each GCM was then ranked by applying these weights to each criterion by using TOPSIS algorithm. The top five GCMs were then used to generate future annual maximum rainfall series by using the REA method corresponding to RCP scenarios 4.5 and 8.5. Ranking of GCMs takes past performance into account while the use of REA for generation of future annual maximum rainfall series takes future climate projections of GCMs into account. Thus, a hybrid approach is used in the present study.

Surat city is situated in the Gujarat state covering an area of 326.515 km². It is located between longitudes 72°45′ to 72°54′ E and latitudes 21° 06′ to 21°15′ N. It is situated on the banks of the Tapi River with the coastline of the Arabian Sea on its west side (Joshi et al. 2012). According to the 2011 census, Surat city had a population of 4.5 million (Patel et al. 2017). The study area experiences hot summers with temperatures varying from 38 to 45 °C. The winters are mild, but the month of January is specifically cold, with temperatures ranging from 10 to 15.5 °C. The mean annual rainfall is 1143 mm (Sharma et al. 2013). The study area is shown in Figure 1.

The NASA Earth Exchange-Global Daily Downscaled Projections (NEX-GDDP) dataset contains downscaled projections for RCPs 4.5 and 8.5, obtained from 21 models and scenarios corresponding to daily scenarios developed and disseminated under CMIP5. Each climate projection included a daily temporal scale data of minimum temperature, precipitation, and maximum temperature for the period 1950–2100. The spatial resolution of the dataset is 0.25° (nearly equal to 25 × 25 km) (NEX-GDDP India). The NEX-GDDP dataset is utilized in this study. India Meteorological Department (IMD), Pune, provided rainfall data of the Surat City's observation station corresponding to the period 1969–2005, which is utilized in the present study.

The CC, NRMSD, ANMBD and NSE are employed as statistical performance metrics in the present study.

Regardless of the decision maker's point of view, the entropy approach is used to evaluate the weights of various indicators based on the specified payoff matrix. The entropy technique's analysis is based on the amount of available information (as measured by its entropy value) and its association with the importance of indicators (Raju et al. 2016). Below is an explanation of the entropy approach.

When the entropy value is large, the criteria vector contains high uncertainty, information diversification is low and criterion is less important (Raju & Kumar 2014).

TOPSIS is based on the premise that the chosen option should have the shortest distance from the ideal solution and the greatest distance from the anti-ideal solution (Raju & Kumar 2015).

TOPSIS methodology is comprised of:

• 1.
  Calculation of the separation measure of every alternative a from the ideal solution, i.e., the Euclidean distance from the ideal value of each criterion and for all criteria (j is equal to 1, 2, 3, …. J), summing these for the particular alternative a, i.e. is calculated by using Equation (8) (Raju & Kumar 2015):

  

  (8)

  where j is equal to 1, 2, …J; for criterion j, alternative a's normalized value.

• = criterion's normalized ideal value j; = assigned weight to the criterion j.

• 2.
  Calculation of separation measure every alternative a from the anti-ideal solution, i.e., Euclidean distance from the anti-ideal value of each criterion and for all criteria (j is equal to 1, 2, 3, …. J), summing these for the particular alternative a, i.e. is calculated by Equation (9) (Raju & Kumar 2015).

  

  (9)

  where f_j** = Criterion j's normalized anti-ideal value
• 3.
  Calculation of relative closeness C_a for each alternative a by using Equation (10):

  

  (10)

Based on the C_a values, alternatives are ranked. The higher the value of C_a, the superior is the alternative (Raju & Kumar 2015).

The entropy technique is used to assign weights to the performance indicators, with various weights being assigned to different performance indicators rather than equal weights. This, in turn, aids in the GCMs ranking. Table 1 shows the weights assigned by the entropy technique to four performance metrics, which are evaluated by comparing simulated and observed monthly rainfall data. Amongst the four performance indicators, NSE is found to have the highest weightage (0.79), which showed a significant impact on the ranking of GCMs. The combined contribution is found to be less than 21% for ANMBD, CC, and NRMSD.

The TOPSIS technique takes the weights assigned by the entropy technique as input. Then ranks were assigned to each GCM according to TOPSIS metric values, and these metrics were determined by comparing simulated monthly rainfall data of 21 GCMs with observed monthly data of Surat City corresponding to the period 1969–2005. Table 2 displays the rankings given to 21 GCMs. TOPSIS metric values for the top five ranks are 0.986, 0.936, 0.910, 0.863, and 0.843, respectively. As a result, simulations of the top five GCMs are employed in the further analysis.

The REA method was applied to assign weights to the top five GCMs corresponding to three time slices (2020–2049, 2050–2079, and 2070–2099) and RCPs 4.5 and 8.5 scenarios, respectively, and these weights are shown in Tables 3 and 4, respectively.

In the present study, four statistical performance indicators, namely CC, NRMSD, ANMBD, and NSE, are employed, which are estimated based on observed monthly rainfall and GCM simulated monthly rainfall data for Surat city. By applying performance criteria, performance indicators are evaluated for each GCM. The weight for each performance indicator is then determined by using the entropy approach. GCMs are then ranked by applying these weights to each criterion by using the TOPSIS algorithm. The top five GCMs are then employed to generate future annual maximum rainfall series for Surat City by using the REA method corresponding to RCP scenarios 4.5 and 8.5. Thus, a combination of ranking of GCMs (past performance approach) and generation of future annual maximum rainfall series by using REA (envelope approach), i.e. the hybrid approach is used in the present study to reduce uncertainty in the projected annual maximum rainfall for Surat City. The NSE is found to have a major influence on the GCM's ranking, which secured the highest weightage (0.79) amongst the four performance indicators. The top five ranked GCMs are NOrESM1-M, bcc-csm1–1, CNRM-CM5, MIROC-ESM, and GFDL-CM3. Future annual maximum rainfall series are then generated corresponding to two scenarios, RCP 4.5 and 8.5, by using an ensemble of the top five ranked GCMs through application of the REA technique. Based on the results of the projected annual maximum rainfall series, it can be found that for both the scenarios, i.e. RCP 4.5 and 8.5, the study area could experience more severe events in the future than in the past.

Climate change is likely to worsen flood risks by incrementing precipitation in and around Surat city. Thus, to study the effect of climate change on Surat City's stormwater drainage network, ranking of GCMs and generations of future annual maximum rainfall series are needed. A hybrid approach is used in the present study to generate future annual maximum rainfall series for Surat City corresponding to RCP 4.5 and 8.5 scenarios, which has not been performed by any reviewed study. The hybrid approach used in the present study consists of applying both a past performance approach through ranking of GCMs and an envelope approach through the REA technique for generation of future annual maximum rainfall series, which reduced the uncertainty in the projected future annual maximum rainfall of Surat City.

From the study, NSE is found to have a major impact on the ranking of GCMs. On the basis of projected annual maximum rainfall series, it can be found that, corresponding to RCP 4.5 and 8.5 scenarios, the study area could experience more severe events in the future than in the past. This can therefore increase the risk of flooding in the study area and affect the performance of the stormwater drainage network. The findings of the study can be used for flood risk analysis of Surat city, including stormwater drainage network assessment and urban flood assessment under climate change. The combined hydraulic and hydrological modelling can be used to evaluate the stormwater drainage system's capacity to handle floods under future climate change scenarios which can be generated by using annual maximum rainfall series derived in the present study for Surat City. The current study identified the best GCMs for Surat city and generated annual maximum rainfall series for the future by using an ensemble of top five ranked GCMs, which can be used in future hydrological or meteorological investigations.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.