4 Climate Change and Water Resource Assessment in South Asia: Addressing Uncertainties 4.1 INTRODUCTION Any human or natural system’s environment varies from day to day, month to month, year to year, decade to decade, and so on. It follows that systematic changes in the mean conditions that define those environments can actually be experienced most noticeably through changes in the nature and/or frequency of variable conditions that materialize across short time scales and that adaptation necessarily involves reaction to this sort of variability. This is the fundamental point in Hewitt and Burton (1971), Kane et al. (1992), Yohe et al. (1999), Downing (1996) and Yohe and Schlesinger (1998). Some researchers, like Smithers and Smit (1997), Smit et al. (2000), and Downing et al. (1997), use the concept of “hazard” to capture these sorts of stimuli, and claim that adaptation is warranted whenever either changes in mean conditions or changes in variability have significant consequences. For most systems, though, changes in mean conditions over short periods of time fall within a “coping range” - a range of circumstances within which, by virtue of the underlying resilience of the system, significant consequences are not observed for short-term variability (see Downing et al. (1997) or Pittock and Jones (2000)). There are limits to resilience for even the most robust of systems, of course. It is therefore as important to characterize the boundaries of a system’s coping range as it is to characterize how the short-term variability that it confronts might change over the longer term. This chapter is designed to reflect the sensitivity to short-term climate variability (expressed in terms of the changes in frequency of flooding events in Bangladesh along the Ganges, Brahmaputra and Meghna Rivers) to long-term secular change (expressed in terms of long-term trends in maximum monthly flows) along a wide range of not-implausible climate futures. It therefore explores a case for which the boundaries of a coping range are easily defined by flooding thresholds. When we ultimately turn a discussion of how to evaluate adaptation options that might expand the coping range (exposure to flooding) or reduce the cost of flooding (sensitivity to flooding in terms of multiple metrics), we will do so in a way that can accommodate enormous uncertainty. We begin by characterizing the sources of uncertainty in our perception of how the future climate might evolve and our associated expectations about the frequency of flooding. Section 4.3 reviews historical records of annual mean flows, annual peak monthly GARY YOHE KENNETH STRZEPEK Copyright © 2005 Taylor & Francis Group plc, London, UK flows and flooding events. A statistically calibrated reduced-form relationship between monthly peak flow and the likelihood of flooding in any one year will summarize these data. Section 4.4 follows with a description of a simple hydrologic model that relates precipitation and temperature to river flow on a monthly basis; calibration and scaling issues are also reviewed. Major sources of uncertainty in generating scenarios of future climate change are described in Section 4.5. Following a methodology developed in Yo h e et al. (1999), a systematic sampling across 14 general circulation models across three alternative carbon-emissions scenarios associated with two alternative sulfate scenarios, three alternative climate sensitivities, and two alternative sulfate forcing factors will produce a wide range of future flow scenarios (684 in number). Subsequent analysis will work with 8 representative scenarios for peak monthly flows selected from the full sample. The representatives will not be chosen to reflect a probabilistic portrait of what the future might hold. They will, rather, be selected to span a full-range of “not-implausibility” futures so that the associated inter-temporal trajectories of the annual likelihood of flooding events absent any additional adaptation presented in Section 4.5 offer pictures of profound uncertainty - possible futures that cannot, at this point, be dismissed as impossible. The scenarios will, in particular, reflect the possibility that maximum flows may or may not climb continuously over time; indeed, they reflect the distinct possibility that the monthly maxima may actually begin to fall after 2050. Further adaptation can be expected to guard against any increase in the frequency of flooding, so Section 4.6 describes how these representative trajectories might be employed to characterize the relative efficacy of various adaptation options overtime before a concluding section offers some thoughts about context. 4.2 DEFINING UNCERTAINTIES Figure 4.1 offers a schematic portrait of how the drivers of climate change might influence the likelihood of flooding events in Bangladesh. Various emissions trajectories of greenhouse gases and sulfate aerosols are shown there to produce a range of climate futures, determined in large measure by uncertainty about climate sensitivity and the radiative forcing of the sulfates. These climate futures produce ranges of change in monthly precipitation and temperature which, in turn, produce a set of futures expressed in terms of maximum monthly flows in any given year. Since the severity of possible flooding events in any year can be related statistically to these maximum flows, trajectories of the likelihood of small, modest, and extreme flooding are ultimately produced. The expanding size of the loci in Figure 4.1 illustrates pictorially how the uncertainty that clouds our understanding of each step in the causal chain cascades down the causal flow. If, for example, we knew the path of future emissions exactly, we could not precisely define associated climate change. If we knew how climate change would evolve over the next decades, we still could not accurately describe how associated patterns of precipitation and temperature would be altered and how those changes might be translated into river flows. And even if we knew exactly how flows might change, we could not accurately predict how the likelihood of flooding events might change. A second cascade of uncertainty, derived from the methods with which researchers try to describe each of the links depicted in Figure 4.1, must also be recognized. First of all, there may not be one accepted model of any given link in the causal structure. Instead, multiple modeling structures - abstractions of the real world - may exist, and they sometimes produce wildly different answers to the very same questions. This simple phenomenon is valuable in examining the relative value of one particular model or another, 78 WATER RESOURCE ASSESSMENT: ADDRESSING UNCERTAINTIES Copyright © 2005 Taylor & Francis Group plc, London, UK but it introduces model uncertainty for analysts who are looking across model results for a coherent view of the future. In addition, the ability of any particular model to offer credible scenarios is limited by the statistical boundaries that surround estimates of the critical parameters (call this calibration uncertainty). These limitations are well understood, of course, but they can be exacerbated when any one parameterization (with associated error bounds) is used to produce predictions of critical state variables (call this prediction uncertainty). Things get even worse when researchers take account of uncertainty about the track that the critical drivers of the model might take in the future (call this projection uncertainty). This compounding effect, really the point of Figure 4.1, can be especially troublesome when these drivers move beyond past experience and therefore out of the sample range upon which the model was calibrated. Finally, underlying social and economic structures might change overtime; and if they do, this evolution undermines the credibility of using historically-founded modeling structures as representations of future conditions to produce what might be called contextual uncertainty. Fig. 4.1 The cascade of uncertainty from emissions to a source of vulnerability. Our depiction of climate uncertainty in terms of the annual likelihood of flooding will, at least implicitly, confront each of these sources of uncertainty by the time we describe a framework within which to evaluate adaptation options. Calibration, prediction and projection uncertainties will, for example, cloud our understanding of the link between flow in the rivers and the likelihood of flooding events. Model and projection uncertainties will cascade through the scenarios with which we create representative “not-implausible” Emissions Precipitation Concentrations Temperature Maximum Flow Flooding Likelihood G. YOHE AND K. STRZEPEK 79 Copyright © 2005 Taylor & Francis Group plc, London, UK portraits of future climate change in terms of flow, but calibration and prediction uncertainties will also have an effect behind the scenes. Finally, the evaluation approach described in Section 4.6 must accommodate contextual uncertainty. 4.3 HYDRO-CLIMATIC ANALSIS OF FLOODING IN BANGLADESH Bangladesh is very vulnerable to flooding, principally due to intense monsoon precipitation that falls on the watershed of the Ganges, Brahmaputra and Meghna (GBM) Rivers. Figure 4.2 shows how these rivers converge into a single delta within Bangladesh. Mirza (2003) reports that the GBM watershed covers 1.75 million square kilometers of Bangladesh, China, Nepal, India and Bhutan. According to Ahmed and Mirza (2000), 20.5% of the area of Bangladesh is flooded each year, on average; and in extreme cases, floods about 70% of Bangladesh can be under water. Fig. 4.2 The Ganges, Brahmaputra and Meghna Rivers. The goal of this paper is to analyze the impact of not-implausible climate change scenarios on the flood frequency in Bangladesh. Mirza (2003) took a statistical approach to relate monsoon precipitation to peak flood flows. This paper will use a conceptual hydrologic rainfall-runoff model that incorporates evapo-transpiration, snowmelt, soil moisture and surface and sub-surface flows. Separate models of the Ganges and Brahmaputra Rivers are developed and described in the next section. The hydrologic model needs to be driven by a climate data, of course, but COSMIC reports only spatially averaged climate change variables at a nation scale. To cope with this problem, Nepal was selected as the representative country for three reasons. First of all, Nepal is located almost directly in the geographic center of the GBM watershed. Secondly, its monsoon precipitation characteristics, in quantity and timing, are representative of the average characteristics over much of the GBM basins. Finally, using the COSMIC data from China or India, two very large countries over which COSMIC averages climate variables are not representative of the conditions in the GBM watershed. 80 WATER RESOURCE ASSESSMENT: ADDRESSING UNCERTAINTIES Copyright © 2005 Taylor & Francis Group plc, London, UK 4.3.1 UNCERTAINTIES IN THE HISTORICAL CLIMATE RECORD The COSMIC scenario generator provides a base year of 1990, but does not provide any information on the statistics of climate record for the country. It is nonetheless necessary to have data on the moments and probability distributions of the hydro-climatic variables to perform a flood frequency analysis. To supplement the COSMIC scenario data for Nepal, we employed historical climate data gathered by the Tyndall Center for Climate Change Research and recorded in their TYN CY 1.1 dataset. Mitchell et al. (2004) reported that the TYN CY 1.1 data provide a summary of the climate of the 20 th century for 289 countries and territories including monthly time series data for seven climate variables for the 20 th century (1901-2000). Interestingly, the dataset creators provide the following warning: “This dataset is intended for use in trans-boundary research, where it is necessary to average climatic behavior over a wide area into statistics that are representative of the whole area.” This warming endorses the use of TYN CY 1.1 and COSMIC data for Nepal as appropriate for this modeling approach. 4.3.1.1 CLIMATE VARIABILITY Table 4.1 presents the statistics for the annual precipitation and mean annual temperature for Nepal from the TYN CY 1.1 monthly time series data for the 20 th century (1901-2000). The data shows that mean annual temperature varies very little with a COV of 0.04 and a lag-one correlation of 0.47. Precipitation exhibits variability at the total annual level. More importantly for predicting the likelihood of flooding events, though, maximum monthly precipitation per year is even more variable and strongly (positively) skewed with a high coefficient of variation. Table 4.1 Climate Statistics 1901-2000 Annual Precipitation (mm) Maximum Monthly Precipitation (mm) Mean Annual Temperature ºC Mean 2,097.1 556.1 8.17 Mode 2,600.2 489.1 8.22 Median 2,084.8 533.9 8.20 Standard Deviation 264.9 98.2 0.37 Skewness 0.102 0.433 0.07 Lag-One Auto Correlation 0.096 -0.100 0.47 Coefficient of Variation 0.13 0.18 0.04 Maximum 2670.4 813.4 9.29 Minimum 1396.0 360.6 7.20 4.3.1.2 FLOOD FREQUENCY Figure 4.3 shows that the flooded area in Bangladesh varies greatly from year to year. Flood risk is characterized by the probability that a certain level of flood will occur each G. YOHE AND K. STRZEPEK 81 Copyright © 2005 Taylor & Francis Group plc, London, UK year. The risk factor is generally expressed as a return period of T = 1/(probability of occurrence). The return period is determined from the cumulative density function of flood frequency. For flood frequency analyses, FAP (1992) recommends using the Gumbel Type I Distribution (EV1) for the major rivers in Bangladesh; it is defined by: S x ux xF π α α 6 expexp)( = ∞<<∞−             − −= u = X - 0.5772a where S is the standard deviation and 7 is the mean. The mean and standard deviation of the flood peak as well as the parameters of the EV1 distribution were determined using 100-year time series of climate data with the rainfall-runoff model. Using these statistics and the EV1 distribution, flood flows for the 2-year, 10-year, 50-year and 100-year return periods were calculated. They are presented in Table 4.2. Fig. 4.3 Bangladesh Flood Area from 1954 through 1999. 4.3.2 FLOODED AREA AND SEVERITY High river flows themselves are not a problem unless they overtop their banks and flood area in the adjoining floodplain. The determination of flood flows used the science of hydrology, while determining the extent of and depth of flooding was based on the science of hydraulics. Mirza et al. (2003) reported on the application of the MIKE 11-GIS hydrodynamic model for Bangladesh to determine flooded area as a function of peak flood flows in the Brahmaputra-Ganges-Meghna Rivers system. Figure 4.4 shows the data from their work and the non-linear relationship that was developed between peak flow and flooded area with results in an R 2 of 0.59. Flooded Area (million of hectares) = 4.3095* ln[Flow (cms)] – 45.906 82 WATER RESOURCE ASSESSMENT: ADDRESSING UNCERTAINTIES 0 20 40 60 80 100 120 1954 1961 1965 1969 1973 1977 1981 1985 1989 1993 1999 Year Flooded Area (000 sq.km) Copyright © 2005 Taylor & Francis Group plc, London, UK With a relationship between peak flow and flooded area, we have created a link between climate variables and the extent of flooding. Subsequent analysis of climate change will examine the impact of potential climate change on flooding in Bangladesh with full recognition of the possibility that this impact may not be symmetric with respect to all levels of flood risk. Table 4.3 shows four levels of flooding (low, modest, moderate and severe) that were mapped to correspond to the 2-year, 10-year, 50-year and 100-year return periods, respectively. Table 4.2 Flood flow frequency statistics 1901-2000 y = 4.3095Ln(x) - 45.906 R 2 = 0.5912 0 1 2 3 4 5 6 100000 110000 120000 130000 140000 150000 Peak Flood (CMS) Flooded Area Millon hectare Fig. 4.4 The relationship between flood flows and flooded areas in Bangladesh. Table 4.3 Flood flow frequency statistics 1901-2000 P - Annual Probability of Flood Exceeding Q 0.5 0.1 0.02 0.01 T - Return Period for Q (years) 2 10 50 100 Q - Peak Flood Flow (cms) 115,000 140,000 162,500 172,000 A- Flood Area (ha 10^6) 4.311256 5.158979 5.801248 6.046099 Level of Flooding Low Modest Moderate Severe P - Annual Probability of Flood Exceeding Q 0.5 0.1 0.02 0.01 T - Return Period for Q (years) 2 10 50 100 Q - Peak Flood Flow (cms) 115,000 140,000 162,500 172,000 4.4 A HYDROLOGIC MODEL FOR THE RIVERS Mirza et al. (2003) examined the potential climate change impacts for river discharges in Bangladesh using an empirical model to analyze changes in the magnitude of floods of the Ganges, Brahmaputra and Meghna Rivers. The present analysis uses a conceptual rainfall-runoff model, WATBAL, to analyze changes in the magnitude of floods for the same watershed. Yates (1997) describes the model. It has been applied in over forty G. YOHE AND K. STRZEPEK 83 Copyright © 2005 Taylor & Francis Group plc, London, UK country studies of climate change impact on runoff including the Nile River basin, a river basin of the same spatial scale as the GBM basin. More specifically, the WATBAL model predicts changes in soil moisture according to an accounting scheme based on the one-dimensional bucket conceptualization depicted schematically in Figure 4.5. Yates and Strzepek (1994) compared this relatively simple formulation to more detailed distributed hydrologic models and found them in close agreement with absolute and relative runoff. The advantage of this lumped water-balance model lies in its use of continuous functions of relative storage to represent surface outflow, sub-surface outflow, and evapo-transpiration in the form of a differential equation (see Kaczmarek (1993) or Yates (1996)). The monthly water-balance contains two parameters related to surface runoff and sub-surface runoff. A third model parameter, maximum catchment water-holding capacity (S max ), was obtained from a global dataset based on the work of Dunne and Willmott (1996). Fig. 4.5 A schematic conceptualization of the water-balance model. The precise structure of WATBAL is easily described. To begin with, the monthly soil moisture balance is written as: where P eff = effective precipitation (length/time), R s = surface runoff (length/time), R ss = sub-surface runoff (length/time), E v = evaporation (length/time), S max = maximum storage capacity (length), and z = relative storage (1 ≥ z ≥ 0). 84 WATER RESOURCE ASSESSMENT: ADDRESSING UNCERTAINTIES Copyright © 2005 Taylor & Francis Group plc, London, UK A non-linear relationship describes evapo-transpiration based on Kaczmarek (1990): Following Yates (1996), surface runoff is described in terms of the storage state and the effective precipitation according to: where ε is a calibration parameter that allows for surface runoff to vary both linearly and non-linearly with storage. Finally, sub-surface runoff is a quadratic function of the relative storage state: where a is the coefficient for sub-surface discharge. In certain regions, snowmelt represents a major portion of freshwater runoff and greatly influences the regional water availability. Ozga-Zielinska et al. (1994) provide a two parameter, temperature based snowmelt model which was used to compute effective precipitation and to keep track of snow cover extent. Two temperature thresholds define accumulation onset through the melt rate (denoted mf i ). If the average monthly temperature is below some threshold T s , then the all the precipitation in that month accumulates. If the temperature is between the two thresholds, then a fraction of the precipitation enters the soil moisture budget and the remaining fraction accumulates. Temperatures above some higher threshold T l give a mf i value of 0, so all the precipitation enters the soil moisture zone. If there is any previous monthly accumulation, then this is also added to the effective precipitation. where, and snow accumulation is written as, G. YOHE AND K. STRZEPEK 85 Copyright © 2005 Taylor & Francis Group plc, London, UK In writing equations (4.5) through (4.7), mf i = melt factor, A i = snow accumulation, Pm i = observed precipitation, Peff i = effective precipitation, T l = upper temperature threshold at which precipitation is all liquid (°C), T s = lower temperature threshold at which precipitation is all solid (°C), i = month The model was calibrated from the TYN CY 1.1 data for the Ganges and Brahmaputra separately over using data from monthly flow from the 1970 and 1980 and produced R 2 statistics of 0.89 and 0.87 for the Brahmaputra and Ganges, respectively. Since the climate change scenarios in COSMIC begin with a base year of 1990, the COSMIC base had to be correlated with the TYN CY 1.1 average data. Panels A and B of Figure 4.6 show the relationship between historical average and COSMIC base year data for temperature and precipitation, respectively. Fig. 4.6 Panel A - Correlation of COSMIC 1990 to historical monthly temperature. 4.5 FUTURE CLIMATE SCENARIOS Schlesinger and Williams (1998 and 1999) designed the COSMIC program so that researchers could produce literally thousands of “not-implausible” climate scenarios that are internally consistent. Each scenario is defined by a specific global circulation model (of the 14 included in COSMIC) driven by one of seven emissions scenarios for greenhouse gases that span virtually the entire range of published scenarios. Each scenario is also defined by one of three associated sulfate emission trajectories and by choosing a sulfate forcing parameter between 0 watts per meter and -1.2 watts per meter squared and a climate sensitivities between 1 o and 4.5 o (for a doubling of effective carbon-dioxide concentration from pre-industrial levels). It would be imprudent if not impossible to conduct integrated analyses along each one, so there is a fundamental need to limit the 86 WATER RESOURCE ASSESSMENT: ADDRESSING UNCERTAINTIES Copyright © 2005 Taylor & Francis Group plc, London, UK [...]... Riverbed 4th Mouth Bypass 1 Resources Total costs Distributiona 3 1 5 3 4 4 4 5 1 1 2 1 2 Institutions Structureb Participationc Criteriad 1 2 2 4 2 1 5 3 5 4 5 4 2 1 3 3 2 2 3 Human Capital 1 2 5 4 4 3 4 Social Capital 1 3 4 5 2 2 5 Risk Spreading 2 1 5 4 4 3 6 Information Management Credibility 1 1 3 2 5 4 4 5 2 3 2 3 3 3 5 5 3 3 1 1 3 4 1 1 Efficacy Factor (EF) 0.8 1.0 1.0 0.6 0.8 0.6 Coping Index... (1997), pp.28 9-3 10 Yates, D and Strzepek, K.: “Comparison of Models for Climate Change Assessment of River Basin Runoff ”, IIASA Working Paper 9 4- 4 6, Laxenburg, Austria, 19 94 Yohe, G., Jacobsen, M and Gapotchenko, T.: “Spanning ‘Not-Implausible’ Futures to Assess Relative Vulnerability to Climate Change and Climate Variability” Global Environmental Change 9 (1999), pp.23 3-2 49 Yohe, G and Schlesinger, M.:... Berlin, 1996, 662 pages Downing, T E., Ringius, L, Hulme, M and Waughray, D.: “Adapting to Climate Change in Africa” Mitigation and Adaptation Strategies for Global Change 2 (1997), pp.1 9 -4 4 Dunne and Willmott: “Global Distribution of Plant-Extractable Water Capacity of Soil” International Journal of Climatology 16 (1996), pp. 84 1-8 59 FAP (Flood Action Plan) 25: Flood Hydrology Study, Flood Plan Coordination... Intertemporal Climate , Computer Software, Electric Power Research Institute, Palo Alto, CA, USA, 1998 Schlesinger, M and Williams, L.: “Country Specific Model for Inter-Temporal Climate Climatic Change 41 (1999), pp.5 5-6 7 Smit, B., Burton, I., Klein, R J T and Wandel, J.: “An Anatomy of Adaptation to Climate Change and Variability” Climatic Change 45 (2000), pp.22 3-2 51 Smithers, J and Smit, B.: “Human... “Human Adaptation to Climatic Variability and Change Global Environmental Change 7 (1997), pp.12 9-1 46 Tol, R S J., van der Grijp, N M., Olsthoorn, A A., and van der Werff, P E.: “Adapting to Climate Change: A Case Study on Riverine Flood Risks in the Netherlands” In R S J Tol and A A Olsthoorn (eds.), Floods, Flood Management and Climate Change in the Netherlands, Institute for Environmental Studies, Vrije... relationship between income inequality and vulnerability; i.e., people in more Copyright © 2005 Taylor & Francis Group plc, London, UK 98 WATER RESOURCE ASSESSMENT: ADDRESSING UNCERTAINTIES Fig 4. 10 Panel B - Efficacy factor of protecting against modest and moderate flooding along the riverbed with or without protection against severe flooding inland from the river Fig 4. 10 Panel C - Reduction in the likelihood... provided in Sections 4. 3 and 4. 5, respectively We now turn to assessing adaptation options (Step IV) The risk-hazard approach to assessing how adaptation might increase a system’s long-term sustainability in the face of climate change and climate variability builds on the notion that its exposure to the impacts of climate, its baseline sensitivity to those impacts, and its adaptive capacity determine its... Scenarios (200 1-2 100) Journal of Climate, Forthcoming, 20 04 Copyright © 2005 Taylor & Francis Group plc, London, UK G YOHE AND K STRZEPEK 101 Ozga-Zielinska, M., Brzezinski, J and Feluch, W.: “Meso-Scale Hydrologic Modeling for Climate Impact Assessments: A Conceptual and a Regression Approach”, IIASA CP 9 4- 1 0, Laxenburg Austria, 19 94 Pittock, B and Jones, R N.: “Adaptation to What and Why?” Environmental... pp.23 3-2 49 Yohe, G and Schlesinger, M.: “Sea Level Change: The Expected Economic Cost of Protection or Abandonment in the United States” Climatic Change 38 (1998), pp .43 7 -4 72 Yohe, G and Tol, R.: “Indicators for Social and Economic Coping Capacity - Moving Toward a Working Definition of Adaptive Capacity” Global Environmental Change 12 (2002), pp.2 5 -4 0 Copyright © 2005 Taylor & Francis Group plc, London,... Monitoring and Assessment 61 (2000), pp. 9-3 5 Ross, N A., Wolfson, M C., Dunn, J R., Berthelot, J M., Kaplan, G A and Lynch, J.A.: “Relation Between Income Inequality and Mortality in Canada and in the United States: Cross Sectional Assessment Using Census Data and Vital Statistics” British Medical Journal 320 (2000), pp.89 8-9 02 Schlesinger, M and Williams, L.: “COSMIC - Country Specific Model for Intertemporal . determined using 100-year time series of climate data with the rainfall-runoff model. Using these statistics and the EV1 distribution, flood flows for the 2-year, 10-year, 50-year and 100-year. flooding, it is now certainly appropriate to begin thinking about interventions over the medium- or long-term (like building dikes or instituting programs of systematic and repeated dredging). 4 Climate Change and Water Resource Assessment in South Asia: Addressing Uncertainties 4. 1 INTRODUCTION Any human or natural system’s environment