Assessing potential flood vulnerability to climate change by CMIP3 and CMIP5 models: case study of the 2011 Thailand great flood

The impact of global warming is likely to increase vulnerability to the potentially damaging impacts of climate change, especially increases in the frequency and harshness of weather events such as heavy rainfall, shifts in the rainy season, and an increase in the number of wet days. Therefore it would need rather long-term future climate projection to be able to clearly detect the change in future climate patterns. Several communities in the coastal areas are vulnerable to a range of hazards, especially coastal flooding. An increase in mean sea level and sea level extremes will mainly affect the terrestrial landscape, increasing the risk of inundation of low-lying coastal areas. Coastal areas in both developing and more industrialized economies face a range of risks related to climate change and variability. Potential risks include accelerated sea level rise, an increase in sea surface temperatures, intensification of tropical and extra tropical cyclones, extreme waves and storm surges, altered precipitation and runoff, and ocean acidification. The Intergovernmental Panel for Climate Change Fourth Assessment Report (IPCC 2007) points to a range of outcomes under different scenarios. It identifies a number of hotspots, including heavily urbanized areas situated in the low-lying deltas of Asia and Africa, as especially vulnerable to climate-related impacts. The number of major cities located near coastlines, rivers, and deltas provides an indication of the population and assets at risk. Thirteen of the world's 20 largest cities are located on the coast, and more than a third of the world's people live within 100 miles of a shoreline. Low-lying coastal areas represent 2% of the world's land area, but contain 13% of the urban population (McGranahan et al. 2007). Bangkok is one of the cities vulnerable to climate change.

Floods have had an impact on human society from time immemorial. Damage from flooding has been increasing, and floods are occurring more frequently. Flooding is a byproduct of persistent heavy rain, which is associated with the passage of a tropical cyclone or a monsoonal depression. However, flood disasters are increasing because of deforestation, poor land drainage, expanding agriculture and urbanization. All of these activities promote an increased rate of runoff and a greater flood peak. Hydrologic models provide a framework in which to conceptualize and investigate the relationship between climate change and floods. The use of hydrological models (hard computing) in climate change studies can range from the evaluation of annual and seasonal streamflow forecasting using simple water-balance models (Arnell 1992) to the evaluation of flood impact processes (Singh & Kumar 1997; Bronstert 2003). Recently, hydrologic modeling uncertainty has been examined for climate change impact study (Poulin et al. 2011; Steinschneider et al. 2015). In addition, the use of soft computing in the field of hydrological forecasting is a relatively new area of research. Soft computing technologies can produce a single, hybrid solution for the enhancement of operational river levels and flood forecasting systems. Moreover, it offers a more flexible, potentially self-adaptive approach to modeling flood processes, which by their nature are inherently complex, nonlinear and dynamic. Recent application of flood forecasting using the soft computing approach can be found (Wu et al. 2008; Chidthong et al. 2009; Taorumina & Chau 2015).

Since the Coupled Model Intercomparison Project (CMIP) was launched in 1995, coupled ocean–atmosphere general circulation models developed in dozens of research centers around the world have been compared and analyzed extensively. The program has improved our scientific understanding of the processes of Earth's climate system and of our simulation capabilities in this field. CMIP also plays an important social role by contributing to the Intergovernmental Panel on Climate Change (IPCC). CMIP phase three (CMIP3) provided the scientific base for the Fourth Assessment Report (AR4) of IPCC published in 2007. CMIP phase 5 (CMIP5) was initiated in 2008, and the CMIP5 data are now available for analysis and are expected to provide new insights on our climate for the Fifth Assessment Report (IPCC 2013). The latest generation of global climate models (GCMs), the framework of the fifth phase of the CMIP5, reflects 5–6 years of effort by multiple climate modeling groups around the world. Compared to CMIP3, CMIP5 models typically have finer resolution processes, incorporation of additional physics, and better-developed or well-integrated earth system components (Taylor et al. 2012). Emerging literature on CMIP5 (Meehl et al. 2009; Taylor et al. 2011a, 2011b, 2012) has reported improvements in simulating certain key processes. Kug et al. (2012) reported that the CMIP5 suite of models performs slightly better than the CMIP3 models in simulating two types of El-Nino events: Warm Pool El-Nino (a new type) and Cold Tongue El-Nino (the conventional El-Nino). A few studies have been performed on the intercomparison of the performance of CMIP3 and CMIP5 (Brands et al. 2013).

Previously, projected changes of the climate by GCMs were derived mainly from the simulation of CMIP3 and more recently, by the latest generation of coupled models in CMIP5. However, most research concentrates on large domains such as the global monsoon region (Chen & Sun 2013), the western Pacific region (Grose et al. 2014), or the Asian monsoon region (Sperber et al. 2013). Regional assessments of models have been attempted for a limited number of regions: South America (Blázquez & Nuñez 2013), Europe (Cattiaux et al. 2013), and Australia (Mehrotra et al. 2014). Hydrologic projections require climate projections that are derived from GCMs, the only tools we have available to us to explore the response of the global climate system to scenarios of future greenhouse gas emissions. GCMs are run at relatively coarse resolutions with a grid spacing of a 100 km or more. The regional climate response will be affected by large topographic relief, strong precipitation gradients and continentality. Statistical downscaling is an approach often adopted to translate large-scale information to a local scale, where relationships between large-scale variables in the GCM and local-scale climate variables in the observed record are used to adjust GCM projections to better represent local conditions.

In this paper, we assess the CMIP3 and CMIP5 simulations on the local scale of Thailand and look into future change in precipitation for Bangkok due to its vulnerability to climate change. First, the flood history and flood behavior of the 2011 great flood are examined. Then, the future vulnerability and statistical downscaling from nine model pairs of CMIP3 and CMIP5 are explained. Subsequently, hydrological simulations (Mike11 and Mike21 models) with input climate change variables from the climate downscaling are described, the results of several adaptation measures are given, followed by a summary and short discussion.

Bangkok is still under threat of flooding, especially the increasing flood risk due to climate change and the rapid urbanization in the floodplain. The frequency of devastating floods tends to be higher and there are increasing signs of loss of human lives and property. Flood devastation such as in Thailand in 2011 is not simply the result of extreme rainfall and poor reservoir management. It results from failure to prepare for recurrent floods (Ziegler et al. 2012).

One plausible consequence of global warming is acceleration of the hydrological cycle, which is simply the balance among global evapotranspiration, rainfall, surface runoff, and storage (Ziegler et al. 2003). Acceleration may increase the frequency and/or intensity of extreme events, which occur annually throughout monsoon Asia. Thailand is affected by two tropical weather patterns during its ‘rainy’ season: the southwestern monsoon that originates in the Indian Ocean and tropical storms that originate in the Pacific. The monsoons are the primary driver of rainfall during the rainy season, which lasts from June to December, and flooding often occurs during the regular heavy monsoon rains. Tropical storms are less common in Thailand, but combined with monsoon rains, they helped to cause the two largest floods in recent years.

In the present study, we investigated the capability of CMIP3 and CMIP5 models in simulating precipitation over land through historical 20th century skills and future projections. Because the GCM selection was not the primary objective of this study, therefore, it was done by: (1) the availability of the variables (precipitation), scenarios (CMIP3:B1 and A2; CMIP5: RCP4.5 and RCP8.5), and based line and target projection years (1980–1999 and 2010–2099); and (2) guidelines from previous studies for the southeast Asia region (Kumar et al. 2014; McSweeney et al. 2015). Finally, we used nine climate model pairs from CMIP3 and CMIP5, downscaling for Bangkok (see Table 1). The horizontal resolution of the CMIP5 models is generally improved for the majority of the models.

We downscaled precipitation for each of the nine model pairs from both generations of models (CMIP3 and CMIP5) and computed with respect to Bangkok station observed data. Then, we also evaluated the models' agreement on projecting mean climatology over the last two decades of the 20th century (1980–1999). Specifically, we generated changes in projection precipitation for all model pairs from the CMIP3 and CMIP5 models with three target periods, the near-future (2010–2039), the mid-future (2040–2069) and the far-future (2070–2099).

The similarity between the observed and model-simulated field can be described by the Taylor diagram. The reference data set is plotted along the abscissa. The model data set is plotted in the first or second quadrant depending upon whether the correlation coefficient is positive or negative, respectively. The azimuthal position of model data is given by the arccosine of the correlation coefficient between the reference and model data set. The radial distances of reference and model data points from the origin are proportional to their standard deviations. The centered RMS (Root Mean Square) error is proportional to the distance between the points representing reference and model data sets. The closer a model point is to the reference data point, the lower its centered RMS error; it implies that the model is performing relatively well. High correlation between reference and model data signifies model-simulated seasonal cycles are reasonably phased. In this study, we use observations (TMD data) as a baseline for measuring historical performance, while the ensemble medians are used as a baseline for multi-model performance measuring in the future. As a measure of central tendency, Wilks (2011) suggested the use of median rather than the mean to be robust to outliers.

The future rainfall over Bangkok was analyzed for the period 2010–2099 for all scenarios in this study (B1, A2, RCP4.5, RCP8.5). The delta change factor approach was used to indicate the difference between future and reference day climate. The delta change is compiled for the seasonal (MJJASO) scale in Figure 11. The mean precipitation during the wet season (MJJASO) of CMIP3 and CMIP5 models and observed data (1980–1999) are shown for all target future periods. The model color is similar to Figure 9. Most GCMs show higher mean precipitation than the observed data except for CSIRO-Mk3.0, IPSL-CM4 (CMIP3) and CSIRO-Mk3.6, and IPSL-CM5A-LR(CMIP5). Results show higher variability for a higher emission scenario similar to Figure 12. We observed that all CMIP3 and CMIP5 models give a significant increase in precipitation for the far-future period, but not so much difference between the near-future and mid-future periods. We also cannot see any improvement in CMIP5 models over CMIP3 models.

Regarding the concerns about the future flooding in Bangkok and its vicinity, several studies, for example, IPCC (2007), Nicholls et al. (2008), World Bank (2010) and Kundzewicz et al. (2014) indicate that Bangkok may have to face major flooding in the future. Significant drivers are increasing rainfall and upstream discharge, land use change, land subsidence and an increase in sea level. Any or all of these significant factors would result in Bangkok and its vicinity experiencing greater flood risk and vulnerability.

The ground surface elevation in the study area was derived from the benchmarks surveyed by the Department of Mineral Resources and the Royal Thai Survey Department. The ground survey elevation outside the study area is derived from the topographic map of 1:50,000 scale. The bathymetry was derived from an eco-sounding survey by the Hydrographic Department. The channel cross-section was surveyed by the Royal Irrigation Department from the upstream (Nakhon Sawan province) to the downstream at the river mouth, a distance of 375 km. The Tha Chin river cross-section was input from the Wat Sing District, Chainat Province, to the downstream station at the Tha Chin River mouth, a distance of 319 km. Several tributary channels in the Chao Phraya-Tha Chin basin were also included in the river network.

The 2011 monsoon season was exceptionally heavy and led to extensive and long-lasting flooding in the Chao Phraya river basin. Flooding was exacerbated by the rapid expansion of urban areas into flood plains, and was the costliest natural disaster in the country's history, with direct damages estimated at US$45 billion.

In this study, we examined the flood behavior and the lessons learned, then we looked into future flood risk and flood vulnerability from extreme precipitation. The comparison across two generations of the GCM ensembles CMIP3 and CMIP5 were made through historical 20th century and future projections. The majority of CMIP5 models show double peaks of precipitation (in May and September) similar to the observed data. However, both models overestimate the dry spell and underestimate the peak precipitation. The precipitation downscaling was performed using two statistical downscaling methods, LS and the DM. The DM downscaling is found to be better than the LS. Overall, our results suggest that the performance of CMIP5 models cannot be readily distinguished from CMIP3 models, although there are clear signals of improvements over Bangkok. Both model generations perform reasonably well in capturing the amplitude and phasing of past mean annual precipitation. The correlation coefficient for all GCMs lies between 0.6 and 0.8, implying most of the models simulate the mean rainfall reasonably well. Even though both model generations display approximately the same standard deviation as the observed, more spatial variability and more RMS errors are found for all future periods. Therefore, past model performance does not guarantee future results. The forecasted precipitation changes, in the rainy season were examined through the Multi Model Mean and Multi Model Median of nine GCMs. The use of the Multi Model Mean shows continuously increased rainfall from the near-future to the far-future, in the order of 2–12%. The Multi Model Median shows increased rainfall only for the far future. Both show a potential increase in future precipitation, implying more flood vulnerability. The complex terrain and land-sea contrast at Bangkok may contribute to those findings. Further spatial analysis for the whole of Thailand will be done in the near future. In addition, more GCMs in CMIP3 and CMIP5 are needed to confirm our present results.

We then applied flood simulation models (Mike 11 and Mike 21) to examine the 2011 flood behavior, including adaptation measures for a changing climate. It is found that several areas in the lower Chao Phraya river basin may be inundated for 1–2 months. The initial cost of damages was estimated at about THB 150,000 million. Four adaptation measures were investigated. The first case study is the base case of the 2011 flood (do nothing). The inundation area is found to be approximately 14,080 km². The use of 3,200 km² as a retention basin in the second case study can reduce the inundation area by 18% or 11,520 km². The third case study uses the east floodway, of 1,000 cms capacity. This can reduce the inundation area by 35%, or 9,120 km². The fourth case study uses both floodways (east and west, of 1,000 cms capacity each). This can reduce the inundation area by 42% from the 2011 flood event. However, even with this, Bangkok is still under threat of flooding. The frequency of devastating floods tends to be higher and there are increasing signs of loss of human lives and property. In summary, flooding in the Chao Phraya river basin cannot be completely avoided, Therefore, the best practice for high flood risk communities is to raise the houses and have an open space in the first floor, and this should be used as one flood resilient approach.

The author is grateful for the valuable comments of anonymous reviewers, who contributed to an improvement of this paper. Thanks to the World Bank for sponsoring the project in 2009. Thanks also to the Thai Meteorological Department and the Royal Irrigation Department for providing the observed rainfall and discharge data. The climate model data sets were obtained from the PCMDI archive.