FRAMEWORK FOR DEVELOPING ENSEMBLE OF GCMS AND ITS APPLICATION IN CLIMATE CHANGE STUDIES

Abstract

Climate change, as defined by the Intergovernmental Panel on Climate Change (IPCC), refers to “a change in the state of climate that can be identified by changes in the mean and/or the variability of its properties, and that persist for an extended period, typically decades or longer.” The IPCC Special Report (IPCC, 2018) outlines the potential risks associated with the 1.5 ºC warming over pre industrial levels. These include rise in extreme temperature in many regions; and increase in intensity and frequency of heavy precipitation in some areas while increase in droughts in other areas. Moreover, the temperature change, in terms of hot days are projected to be more severe on land especially in tropical countries like India. The IPCC Assessment Report 5 (AR5, 2014) states that areas in India and other parts of Asia will face the potential risk of increased riverine and urban flooding along with increased probability of heat related mortality. It is also highlighted that the food shortages due to climate change impacts on agriculture sector will pose more challenges to the developing nations like India. These impending risks underline the necessity for studying climate change impacts rigorously. Future planning and management of water resources requires projected data of at least two important meteorological variables, temperature and precipitation. Therefore, spatiotemporal analysis of these variables and generation of runoff under future projected climate scenarios provides the basis for climate resilient planning and management of water resources. Moreover, as reservoirs are considered the stable pillar of water resources management, the study of impact of climate change on reservoirs and reservoir operations is also necessary to formulate future plans. It may be noted that the degree of reliability of such studies and in turn the effectiveness of mitigation strategies is constrained by the degree of accuracy with which GCMs are able to represent climate of a particular location. Hence the evaluation of the characteristics of the model simulated variables, in terms of their accuracy (with respect to reference data), variability and uncertainty constitute important aspects of climate change analysis studies. Consequently, the exercise involves efforts towards enhancing the efficiency of the processes that work with GCM outputs and increases their accuracy. These efforts encompass review and analysis of existing methodologies, bridging the gaps between them, and developing new methodologies. In this background, the major thrust of this dissertation work was on the development of a new framework for performance evaluation of GCMs and generation of weighted ensemble data of v climatic variables. This data was further used to identify and quantify the climate change signals all over India. Finally, data generated from the framework was also used to check the Tehri reservoir operations under the climate change scenarios. The methodology of the dissertation is presented in the following diagram schematically. The details of the methodology are presented in the subsequent sections. 1. Framework Development The weighted average ensemble data was created for each zone (Koppen Climatic Zone) for three variables (precipitation, minimum and maximum temperature). The generated data was evaluated on a zonal basis to assess the efficiency of the data created. This evaluation was made against the reference data. To assess the comparative advantages of the proposed framework, the performance of the generated data was also compared to the individual GCMs and mathematical average of all GCMs (since mathematical average is the most frequently used method in GCM ensemble). This comparison was made using Taylor diagram and goodness of fit of CDFs. Moreover, the frequency with which a particular methodology was selected during each of the two stages of ranking of models and the pattern of distribution of weight between models were also analysed. Framework development •Development of a framework for the ensemble modelling of GCM outputs, incorporating some existing techniques and introducing non-parametric distribution in Bayesian framework. •Validation and efficiency assessment of the framework for different climatic zones in India. Projected climate analysis •Application of ensemble projected climatic data generated using the developed framework for studying the future climates across India. • Study the characteristics of projected precipitation using ETCCD Indices. Case study of reservoir operation under climate change •Application of the framework to generate future climatic data for Tehri reservoir catchment. • Simulation of future streamflows at Tehri reservoir using the ensemble data in SWAT model. •Performance evaluation of Tehri reservoir, within current reservior operation rules, under climate change scenarios. vi The analysis of the frequency of the occurrence of each methodology in ranking of GCMs showed that the Bayesian methodology occurred 21 (out of 24) times thus reaffirming its comparative advantage. The spread and variability of data is more in case of precipitation which is better captured by nonparametric distribution. Since the variation in temperature data is lower, the parametric distribution is selected more frequently in this case. The distribution of weights, for precipitation, maximum temperature and minimum temperature, resulting from this framework showed an overall uniformity in terms of number of GCMs (three to five GCMs were present in the final combinations). However, there were subtle variations in the pattern of weight distribution among them. The developed framework is able to capture the characteristics of the reference data (standard deviation) more efficiently than the mathematical average data model without any substantial compromise in error parameters. It must also be noted that the reduction in errors in case of precipitation is substantially higher in comparison to the reductions in case of temperature. The CDFs of the reference data, the weighted average data and the mathematical average data were compared visually as well as with the Root Mean Square Error (RMSE), Bayesian Information Criterion (BIC) and Akaike Information Criterion (AIC). In case of precipitation and maximum temperature, the weighted average ensemble data showed minimum values of RMSE/BIC/AIC (indicating nearest fit with reference CDF) for all the zones. In the case of minimum temperature, the values of RMSE/BIC/AIC were found to be minimum for five zones which cover 87% of the area. The visual interpretation of the CDFs showed a considerably higher degree of overlap between the CDFs of the reference data and the weighted average ensemble data, in comparison to the overlap between the observed data and the mathematical average ensemble data. These results reinforce the fact that the weighted average model has an advantage over the conventionally used mathematical average, in terms of replicating the behaviour of the observed climatic variable, thus underlining the success of the framework developed. Therefore, it is concluded that the proposed framework is an improvement over the existing methods of GCM data ensemble. 2. Analysis of Projected Precipitation and Temperature The uncertainty analysis (computed as the interquartile range of PDF of anomalies) of precipitation data, maximum temperature and minimum temperature data was performed for the end century (2071 to 2099) scenario for both RCP 4.5 and RCP 8.5. It was seen that the weighted average data had the minimum uncertainty in all the grids (for both RCP scenarios) vii for precipitation. For maximum temperature, the weighted average data had the minimum uncertainty in 73% grids (for both RCP scenarios). For minimum temperature, the weighted average data had the minimum uncertainty in 31% and 14% grids for RCP 4.5 and 8.5 scenarios respectively. Thus, it can be seen that minimum uncertainty was achieved in most of the cases using three to five models whereas the mathematical average of 21 models do not achieve minimum uncertainty in any grid. Hence, the weighted average data can be said to have a comparative advantage over the mathematical average in this regard. 2.1 Analysis of Future Precipitation: The precipitation data generated for future (2016 to 2099) was analysed for trend on seasonal and annual basis. The analysis reveals that precipitation shows an increase in the volume in all cases under RCP 8.5 scenario in all the seasons over all parts of India. However, the increase seen in central, western and north to north-western parts are conspicuous. In case of monsoon rainfall, the increasing trends under the RCP 8.5 scenario, in the low rainfall areas of the lower Deccan plateau and the semi-arid western parts are important observations. The post monsoon season shows increasing trends all over India in both the RCP scenarios and in some cases show more than 100% increase in precipitation over 2016 to 2099. This is an important observation from water resources planning perspective. In winter, the western Himalayan ranges show increase in precipitation in the RCP 8.5 scenario, indicating an increase in the snowfall. RCP 8.5 scenario also shows decrease in the precipitation in the winter over the northern plains of the country. The premonsoon precipitation in both RCP 4.5 and 8.5 scenarios do not show any trend for most parts of the country. Three ETCCD Indices (SDII, CDD and CWD) were analysed for future precipitation. Increasing trends in Simple Daily Intensity Index (SDII) are seen in the eastern parts of the country which receives moderate to high rainfall and receives the rainfall mainly during the monsoon season. These regions show increasing trends in monsoon precipitation in both scenarios, while they do not show any decreasing trends in continuous dry days (CDD) under either of the RCP scenarios. Thus it can be concluded that the rainfall intensity will increase in the eastern regions of the country with the rainfall concentrated in the monsoon period, while the western and relatively arid regions will see a more temporally equitable distribution of rainfall under climate change scenarios. This underlines the fact that the water resources management strategies required for the different regions of the country has be to be based on regional analysis. viii 2.2 Analysis of Future temperature: The analysis of trend of annual maximum, annual minimum and annual mean temperature showed that increasing trends were more prominent across all zones in the 8.5 scenario. The annual maximum temperature showed a decrease in the percentage of grids showing statistically significant increasing trends in the RCP 4.5 far century (2071 to 2099) scenario whereas for the RCP 8.5 scenario the percentage of grids showing statistically significant change increased in the far century scenario. The annual minimum temperature showed significant increasing trends in a higher percentage of grids as compared to annual maximum temperature, except for the hot desert zone (RCP 4.5 mid and far and RCP 8.5 near scenarios). In case of annual mean temperatures all the zones showed varying percentage of grids with increasing trends. In case of annual mean temperature, as in the case of maximum temperature, the percentage of grids with significant increasing trends decreased in RCP 4.5 far century timeline. The analysis of anomalies in annual maximum and annual minimum temperature was performed. The annual maximum temperature showed a mixed pattern of anomaly (ranging from negative to positive) where the positive anomalies increased from near to far century and from RCP 4.5 to RCP 8.5 scenarios. Maximum anomalies were noticed in annual minimum temperature where all the grids showed a positive anomaly for both scenarios. This indicates that the rise in minimum temperature is more pronounced. This is indicative of the fact that the ability of the earth to radiate back the heat diminishes as we move to higher radiative forcing pathways and from near to far century. This empirically proves, once again, the basic idea of greenhouse effect induced global warming. The anomalies in the annual mean temperature. The findings of temperature are in line with the findings reported many researchers (Kundu et al., (2017), Rajbhandari et al., (2018) etc.). 3. Case Study of Reservoir Operation under Climate Change: A Case Study of Tehri Dam The developed framework was used to produce the projected climatic data for Tehri catchment from 2016 to 2099. These data were used in the SWAT model to generate the future streamflow. This streamflow was finally used to simulate the reservoir operation and analyse the performance of the reservoir under changing climate. The SWAT model was setup using the historical data and run from 2005 to 2010. Out of 5 years’ data, 3 years data (2006 to 2008) was used for calibration and 2 years (2009 to 2010) for validation. The Nash Sutcliffe Efficiency (NSE) of the model during calibration at daily and ix monthly scale were found to be 0.53 and 0.60, respectively. The values of NSE during validation at daily and monthly scale were 0.79 and 0.92 respectively. The SWAT model was also run with the mathematical average of the GCMs which yielded a NSE value of 0.41 thus establishing the superiority of the ensemble average. After calibration, the SWAT model was run for generating future streamflow. The streamflow data was then used to simulate the reservoir operation under operational constraints and consequently the hydropower generation was computed. The daily final elevation of the reservoir was calculated from the above simulation. It was seen that the reservoir level stays above 740 m (MDL) and below 840 m (FRL) while supplying all the dedicated flows. It was also found that the lower precipitation under RCP 8.5 results in the reservoir not getting fully filled in certain instances. These instances result in the lack of creation of head for power generation. Finally, the operation of the reservoir was evaluated on the basis of its ability to meet the target hydropower generation. It was seen that while the target power met for all years under RCP 4.5 scenario, it misses for a few years under RCP 8.5 scenario. Hence the Reliability Resilience and Vulnerability (RRV) analysis was carried out for hydropower generated under RCP 8.5 scenario. The indices were computed to be 0.916 (reliability), 0.857 (resilience) and 0.265 (vulnerability). It may be noticed that the reservoir has high values of reliability and resilience with a comparatively low value of vulnerability under the RCP 8.5 scenario. 4. Summary It can be concluded by underlining the fact that, the hypothesis of improvement of efficiency in ensemble of GCMs using non parametric distribution within the Bayesian framework, hold true empirically. This may be construed as an addition to existing body of theoretical scientific knowledge thus highlighting the major contribution of this dissertation work. The application of the framework to climate change analysis of the future and in reservoir operation aims to aid in water resources planning and management in both national and site-specific context, thus justifying the social contribution of this thesis work. The above flow of work addresses the objectives set out at the beginning. However, there is scope of improvement, in terms of incorporating more sophisticated techniques like Monte Carlo sampling in the framework and application of the framework for other regions and other variables.