Abstract
Extreme rainfall and flooding are common during the summer monsoon season in Thailand. In this study, we utilized Robust Empirical Quantile Mapping (RQUANT) to correct the bias in precipitation, and total runoff data obtained from the latest Couple Model Intercomparison Project phase 6 (CMIP6) for the upper Lam Takong river basin. Five different methods were employed to estimate the river discharge and four estimations based on Budyko functions. Our analysis revealed that the ‘Total runoff’ method yielded the most accurate representation of the observed discharge. Impacts of change in land use are examined in terms of compound roughness. The Multi-Model Ensemble (MME) precipitation under medium-emission (SSP2-4.5) and high-emission (SSP5-8.5) scenarios is projected to increase by 5.74 and 10.91%, respectively. Correspondingly, the discharges are expected to increase by 4.57 and 11.05% for the far-future periods. In general, the Flo-2D model satisfactorily simulated the water level in the main channel but it underestimated small inundation depth (<0.5 m) across the floodplain. Comparing inundation maps among different scenarios and timelines, changes in the inundation area were relatively small (0.05%), especially when compared to changes in floodplain storage (6.85%) due to the mountainous nature of the river basin.
HIGHLIGHTS
Utilizing the internal variable from the CMIP6 database to create inputs to flood simulation model.
Investigation of the functional form (aridity index) for the small mountainous river basin.
Selection of suitable observed data from several world databases.
Investigating the well-known bias correction method (RQUANT).
Examining the flood mitigation measures for this particular small mountainous river basin.
INTRODUCTION
Climate change represents a set of mechanisms that have resulted in significant impacts toward hydrological processes both at global and regional scales. Several studies have suggested that the frequency of extreme weather events is increasing as the climate continues to change. These changes have likely led to a higher occurrence of heavy precipitation over the past two decades in many areas around the globe (IPCC-SREX 2012). Floods are considered a result of extreme weather events that occur regularly and cause severe damage (Doocy et al. 2013). They have become more intense and frequent worldwide in recent decades and projected to continue in the future (Higashino & Stefan 2019).
General circulation models are widely used tools to project future climate based on established physical principles (Randall et al. 2007). It is projected that extreme precipitation will increase under a warming climate (Ali & Mishra 2017). To gain better understanding of past, present, and future climate changes, the Working Group on Coupled Modeling under the World Climate Research Program (WCRP) has established the Coupled Model Intercomparison Project (CMIP3, CMIP5, and now CMIP6). These projects aim to investigate projected changes in climate extremes using global climate models in various regions. However, it should be noted that in the IPCC AR4 (CMIP3), while changes in temperature indices tend to agree for all seasons, changes in precipitation remain uncertain (Orlowsky & Seneviratne 2012). Comparisons between model results and observations have revealed uncertainties in the projections (Alexander & Arblaster 2017). Additionally, most GCMs represent climatic variations at relatively gross spatial resolutions, typically around 100–300 km, which may not be suitable for impact assessments requiring fine spatial resolutions of just a few kilometers.
The latest advancements in climate modeling can be seen in Phase 6 of the Coupled Model Intercomparison Project (CMIP6), which has recently become available (Eyring et al. 2015). However, it is still necessary to gain better understanding of how these CMIP6 model simulations, incorporating both socioeconomic and climate change factors, can be considered as reliable sources for future climate predictions. Currently, there are only a few studies that have analyzed the CMIP6 datasets to explore future climate trends. Grose et al. (2020) conducted an evaluation of CMIP6 models and their projections for future climate changes specifically focused on Australia, and they also compared their findings with the results obtained using CMIP5 models. Another study by Almazroui et al. (2020) examined the projected changes in temperature and precipitation for six South Asian countries during the 21st century using the latest CMIP6 dataset. These studies contribute to the growing body of research on CMIP6 and its potential for providing insights into future climate conditions. However, further analysis and exploration of the CMIP6 datasets are necessary to enhance our understanding of future climate projections and their implications.
Global flood studies examining the impacts of climate change have been conducted on multiple river basins (Pechlivanidis et al. 2016; Tanoue et al. 2016; Hirabayashi et al. 2021). The outcomes of these studies have revealed uncertainties regarding the projected changes in floods, with some models suggesting increases and others indicating decreases. These variations primarily stem from difference among global climate models (GCMs) and their bias corrections, the climate scenario projections, and the specific modeling approaches employed. In addition to global-scale studies, regional-scale flood analyses have been conducted in recent years. For instance, studies have focused on rainfall–runoff floods in various basins ranging from 2.6 to 26,000 km2, such as in the United States, Puerto Rico (O'Connor & Costa 2004) and the 2011 Thailand flood (Sayama et al. 2015). Hydrological modeling techniques, such as the semi-distributed hydrologic model (VIC: Variable Infiltration Capacity) and LISFLOOD-FP hydrodynamic model (Schumann et al. 2013). Krysanova et al. (2017) performed an intercomparison of climate change impacts using nine regional-scale hydrological models across 12 large river basins. These models were forced by five GCMs under four representative concentration pathways (RCPs). The study revealed that the overall average fractions of uncertainty for annual mean flow projections by the multi-model ensemble (MME) were 57% for GCMs, 27% for RCPs, and 16% for hydrological models. These findings underscore the complex nature of flood projections and highlight the need for further research and collaboration among different modeling approaches to reduce uncertainties in flood forecasting and management.
Local-scale flood research has examined flood events in various urban areas around the world. For instance, Cherqui et al. (2015) conducted an analysis of floods in the urban areas of Bordeaux, France. Luino et al. (2012) focused on the 1994 flood in Alba, Piedmont, Italy. Villarini et al. (2009) investigated floods in metropolitan areas of Charlotte, North Carolina, USA. Flood studies have also been conducted in Guangzhou, China (Huang et al. 2018), and the Dutch polder area (Bouwer et al. 2010). Dottori et al. (2018) examined the potential impact of a 3 °C global warming on river flood exposure in Ecuador. Their findings indicated that the relative increase in the population exposed to river floods could be more than tripled compared to the reference period of 1976–2005. It is worth noting that quantifying flood projections for the past-to-present time is challenging, and obtaining reliable and detailed data for future projections is virtually impossible. Kundzewicz et al. (2019) highlighted the difficulties in obtaining accurate and long-term records of flood-related data for flood risk reduction analysis. Furthermore, current climate models still face limitations in accurately representing floods at a local scale. Nevertheless, to bridge this knowledge gap, various modeling tools and techniques have been developed to analyze flood inundation and address these challenges. These tools aim to enhance our understanding of local-scale dynamics and improve flood risk assessments.
The Flo-2D model has been widely utilized for flood simulation in various events. For instance, Luo et al. (2018) conducted a flood risk assessment for Hanoi Central Area in Vietnam using the FLO-2D model. In another study, Mtamba et al. (2015) employed remotely sensed images to achieve spatial parameterization of Manning's roughness coefficient for the Flo-2D model. The evaluation of flood mapping assessments and the associated uncertainties with the Flo-2D model have been investigated using HEC-RAS and LISFLOOD-FP by Dimitriadis et al. (2016). Mishra et al. (2017) utilized outputs from global climate models (GCMs) including changes in land use as inputs for the Flo-2D model to predict future flood events. Furthermore, the analysis of low impact development effects on flood inundation in urban areas has been conducted to develop rainfall harvesting systems and permeable pavements (Hu et al. 2017). Erena et al. (2018) performed flood hazard mapping for Dire Dawa city in Ethiopia. Moreover, Kang et al. (2021) analyzed flood damage in the Wonjucheon basin using rainfall scenarios derived from CMIP5-RCP8.5. These studies demonstrate the versatility of the Flo-2D model in flood research, encompassing various applications, including flood risk assessment, parameterization, uncertainty evaluation, future flood prediction, low impact development analysis, and flood hazard mapping.
The severity of the problem escalates in mountainous areas due to higher precipitation levels and the topography that facilitates rapid surface runoff and the formation of destructive peak flows. Precipitation in mountainous regions exhibits a discontinuous spatial distribution and is influenced by orographic factors. Consequently, precipitation is highly variable from a spatiotemporal perspective (Rotunno & Houze 2007). This variability makes mountainous basins particularly susceptible to extreme precipitation events in terms of both total volume and intensity. Given these circumstances, it is crucial to conduct comprehensive investigations into the flood behavior within small mountainous catchments. Furthermore, it is essential to understand the impact of increased heavy precipitation resulting from a changing climate. Understanding these dynamics will enable better preparation and response to the potential consequences of extreme precipitation events in mountainous regions.
The projected inundation map for Khao Yai National Park holds significant value in establishing zoning regulations, land use planning, and building standards, especially considering its status as a popular tourist attraction. This map serves as a vital tool for supporting various aspects of land use, infrastructure development, transportation planning, flood warning systems, evacuation strategies, emergency management planning, and overall flood preparedness and response efforts. The present study consists of two main components. Firstly, statistical bias correction techniques are applied to analyze the changes in precipitation and total runoff using the latest model simulations from the CMIP6 over the upper Lam Takong river basin in Thailand. Additionally, five different runoff estimation methods, including total runoff and four estimations based on Budyko functions, are utilized to calculate the discharge. The second part of the study focuses on utilizing the Flo-2D model to project regional-scale flood hazards by incorporating outputs from the CMIP6 MME under SSP2-4.5 and SSP5-8.5 scenarios. The impacts of changes in land use are also considered. The study aims to achieve five primary objectives:
- 1.
Assess the performance of the bias correction technique for hydrological variables, such as precipitation and discharge.
- 2.
Project changes in precipitation and discharge under various climate change scenarios.
- 3.
Investigate the impact of climate change and land use change on regional-scale flood hazards in the near-future, mid-future, and far-future periods.
- 4.
Investigate the well-known bias correction (RQUANT) method
- 5.
Provide scientific insights and guidance for designing effective flood prevention measures.
By addressing these objectives, the study aims to contribute valuable scientific knowledge that can inform decision-making processes related to flood management, protection, and mitigation in the study area.
STUDY REGION AND DATA METHODOLOGY
Study region
The Lam Takong river basin, extending upstream to Lam Takong dam, is influenced by both the southwest and northeast monsoons. The region experiences three distinct seasons: rainy, winter, and summer. During the rainy season, from May to October, the southwest monsoon brings moisture from the Indian Ocean, resulting in significant rainfall, with peak levels observed in August and September. The average annual rainfall in the basin is 1,454.3 mm. From October to February, the wind direction reverses, and the cooler and drier northeast monsoon wind blows from the Asian landmass, ushering in the winter season. The temperature gradually rises afterward, with a slight drop in a short transitional period between the monsoons during March and April. The Lam Takong river basin has an average annual temperature of 25.4 °C, with a maximum relative humidity of 83.4% and a minimum relative humidity of 47.8%.
The latest land use map of Pak Chong district has been enacted in 2017 (Figure 1(b)) and will be regulated for the next 20 years (Department of Public Works and Town & Country Planning). However, change in land use from 2000 to 2017 in the river basin is shown in Table 1 (Land Development Department). The forest and river-lake areas do not display significant change compared to changes in community and agricultural areas. This has been due to the first National Park Act enacted for Khao Yai since 1972. The compound Manning's roughness of four land use class (0.02, 0.06, 0.1, and 0.01 for community, agriculture, forest, and river-lake in the floodplain) is calculated. The roughness displays reducing from 2000 to 2017 due to increasing community and reducing agricultural area, respectively.
Changes in land use (%) and compound Manning's roughness in the Lam Takong river basin (Land Development Department)
Class/Year . | 2000 . | 2007 . | 2011 . | 2017 . |
---|---|---|---|---|
Community | 6.02 | 9.96 | 10.34 | 11.29 |
Agriculture | 41.39 | 37.86 | 37.55 | 36.55 |
Forest | 52.54 | 52.25 | 51.88 | 51.84 |
River-lake | 0.05 | 0.2 | 0.21 | 0.22 |
Compound Manning's roughness | 0.1477 | 0.1444 | 0.1434 | 0.1425 |
Class/Year . | 2000 . | 2007 . | 2011 . | 2017 . |
---|---|---|---|---|
Community | 6.02 | 9.96 | 10.34 | 11.29 |
Agriculture | 41.39 | 37.86 | 37.55 | 36.55 |
Forest | 52.54 | 52.25 | 51.88 | 51.84 |
River-lake | 0.05 | 0.2 | 0.21 | 0.22 |
Compound Manning's roughness | 0.1477 | 0.1444 | 0.1434 | 0.1425 |
Datasets
Observation datasets
Given the limited availability of observed rainfall records in the Pak Chong district, our study was initiated by conducting an examination of various grid precipitation datasets, namely SA-OBS, APHRODITE, CPC-UNI, CRU, GPCP1DD, TRMM, ERA-Interim, JRA55, GPCC, CHIRPS, CMORPH, GsMap, and PERSIANN (Supharatid et al. 2020). These datasets differ in terms of the original observations, resolutions, and methodologies employed. Through comparisons among these datasets, it was revealed that SA-OBS, a station-based dataset representing median values, is particularly suitable to serve as a reference. Notably, van den Besselaar et al. (2017) concluded that SA-OBS currently stands as the most reliable daily gridded observational dataset available for the Southeast Asian region (SEA). Therefore, in our study, we have chosen to utilize SA-OBS as the observation data and opted to regrid it to a resolution of 0.005° × 0.005° using the bilinear interpolation technique. This allows for a consistent and standardized representation of the precipitation data for further analysis and comparison with other datasets in the study area.
Model datasets
In our study, we examined a total of 18 CMIP6 models from the CMIP6 database website (https://esgf-node.llnl.gov/search/cmip6) (see Table 2). Ten GCMs (ACCESS-CM2, EC-Earth3, INM-CM4-8, INM-CM5-0, IPSL-CM6A-LR, MIROC-ES2L, MIROC6, MPI-ESM1.2-LR, MRI-ESM2-0, and NORESM2-LM) provided the total runoff (mrro) as an internal variable. The new generation of CMIP6 models differs from the CMIP5 in having a new set of specifications for concentration, emission, and land use scenarios (Gidden et al. 2019) as well as a new start year for future scenarios. In this phase, Shared Socioeconomic Pathways (SSPs) are combined with the RCPs of CMIP5. The SSPs are based on five narratives that describe different levels of socioeconomic development (Riahi et al. 2017): sustainable development (SSP1), middle-of-the-road development (SSP2), regional rivalry (SSP3), inequality (SSP4), and fossil fuel-driven development (SSP5). Detailed descriptions of the SSPs are available in O'Neill et al. (2016).
List of CMIP6 models used in this study
GCM . | Research center . | Resolution . |
---|---|---|
ACCESS-CM2 | Australian Community Climate and Earth System Simulator | 1.88 × 1.25 |
ACCESS-ESM1-5 | Australian Community Climate and Earth System Simulator | 1.88 × 1.25 |
BCC-CSM2-MR | Beijing Climate Center, China Meteorological Administration, Beijing, China | 1.12 × 1.11 |
CanESM5 | Canadian Center for Climate Modeling and Analysis, Environment and Climate Change Canada, Canada | 2.81 × 2.77 |
CNRM-CM6-1 | National Center for Meteorological Research, France | 1.41 × 1.39 |
CNRM-ESM2-1 | National Center for Meteorological Research, France | 1.41 × 1.39 |
EC-Earth3 | EC-Earth Consortium (EC-Earth) | 0.70 × 0.70 |
FGOALS-g3 | LASG, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, 100029, China | 0.70 × 0.70 |
GFDL-ESM4 | NOAA Geophysical Fluid Dynamics Laboratory, USA | 1.25 × 1.00 |
INM-CM4-8 | Institute for Numerical Mathematics, Russia | 2.00 × 1.50 |
INM-CM5-0 | Institute for Numerical Mathematics, Russia | 2.00 × 1.50 |
IPSL-CM6A-LR | The Institut Pierre Simon Laplace, France | 2.50 × 1.27 |
KIOST-ESM | Korea Institute of Ocean Science and Technology | 1.89 × 1.88 |
MIROC6 | JAMSTEC (Japan Agency for Marine-Earth Science and Technology, Japan), AORI (Atmosphere and Ocean Research Institute, The University of Tokyo), NIES (National Institute for Environmental Studies), and R-CCS (RIKEN Center for Computational Science), Japan | 1.41 × 1.39 |
MIROC-ES2L | JAMSTEC, AORI, NIES, and R-CCS, Japan | 2.81 × 2.77 |
MPI-ESM1-2-LR | Max Planck Institute for Meteorology, Germany | 1.88 × 1.85 |
MRI-ESM2-0 | Meteorological Research Institute, Japan | 1.12 × 1.11 |
NESM3 | Nanjing University of Information Science and Technology, China | 1.88 × 1.85 |
NorESM2-LM | NorESM Climate modeling Consortium consisting of CICERO (Center for International Climate and Environmental Research), MET-Norway (Norwegian Meteorological Institute), NERSC (Nansen Environmental and Remote Sensing Center, Bergen), NILU (Norwegian Institute for Air Research), UiB (University of Bergen, Bergen), UiO (University of Oslo) and UNI (Uni Research), Norway | 2.50 × 1.89 |
GCM . | Research center . | Resolution . |
---|---|---|
ACCESS-CM2 | Australian Community Climate and Earth System Simulator | 1.88 × 1.25 |
ACCESS-ESM1-5 | Australian Community Climate and Earth System Simulator | 1.88 × 1.25 |
BCC-CSM2-MR | Beijing Climate Center, China Meteorological Administration, Beijing, China | 1.12 × 1.11 |
CanESM5 | Canadian Center for Climate Modeling and Analysis, Environment and Climate Change Canada, Canada | 2.81 × 2.77 |
CNRM-CM6-1 | National Center for Meteorological Research, France | 1.41 × 1.39 |
CNRM-ESM2-1 | National Center for Meteorological Research, France | 1.41 × 1.39 |
EC-Earth3 | EC-Earth Consortium (EC-Earth) | 0.70 × 0.70 |
FGOALS-g3 | LASG, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, 100029, China | 0.70 × 0.70 |
GFDL-ESM4 | NOAA Geophysical Fluid Dynamics Laboratory, USA | 1.25 × 1.00 |
INM-CM4-8 | Institute for Numerical Mathematics, Russia | 2.00 × 1.50 |
INM-CM5-0 | Institute for Numerical Mathematics, Russia | 2.00 × 1.50 |
IPSL-CM6A-LR | The Institut Pierre Simon Laplace, France | 2.50 × 1.27 |
KIOST-ESM | Korea Institute of Ocean Science and Technology | 1.89 × 1.88 |
MIROC6 | JAMSTEC (Japan Agency for Marine-Earth Science and Technology, Japan), AORI (Atmosphere and Ocean Research Institute, The University of Tokyo), NIES (National Institute for Environmental Studies), and R-CCS (RIKEN Center for Computational Science), Japan | 1.41 × 1.39 |
MIROC-ES2L | JAMSTEC, AORI, NIES, and R-CCS, Japan | 2.81 × 2.77 |
MPI-ESM1-2-LR | Max Planck Institute for Meteorology, Germany | 1.88 × 1.85 |
MRI-ESM2-0 | Meteorological Research Institute, Japan | 1.12 × 1.11 |
NESM3 | Nanjing University of Information Science and Technology, China | 1.88 × 1.85 |
NorESM2-LM | NorESM Climate modeling Consortium consisting of CICERO (Center for International Climate and Environmental Research), MET-Norway (Norwegian Meteorological Institute), NERSC (Nansen Environmental and Remote Sensing Center, Bergen), NILU (Norwegian Institute for Air Research), UiB (University of Bergen, Bergen), UiO (University of Oslo) and UNI (Uni Research), Norway | 2.50 × 1.89 |
In CMIP6, the RCPs from CMIP5 are combined with the SSPs. The SSPs consist of five narratives that describe distinct levels of socioeconomic development, namely sustainable development (SSP1), middle-of-the-road development (SSP2), regional rivalry (SSP3), inequality (SSP4), and fossil fuel-driven development (SSP5) (Riahi et al. 2017). Further information regarding the SSPs can be found in O'Neill et al. (2016).
To assess model performance and establish a baseline for evaluating future changes, we utilized historical runs covering the period from 1998 to 2014, enabling us to characterize the present-day climate. For the analysis of future climate change, we examined data for three distinct periods: near-future (2015–2039), mid-future (2040–2069), and far-future (2070–2099). These projections were derived from climate simulations conducted under two SSPs, specifically the medium-emission scenario (SSP2-4.5) and the high-emission scenario (SSP5-8.5). To ensure a fair comparison, all model outputs were regridded to a consistent resolution of 0.005° × 0.005° using the bilinear interpolation technique, similar to the observation datasets used in the study.
Bias correction
Statistical transformations can be achieved by using parametric or nonparametric transformations to solve Equation (1). To address the systematic biases in the GCM simulations and perform statistical transformations, we utilized the ‘Robust Empirical Quantile (RQUANT)’ method in this study. The RQUANT method, implemented using the R software package qmap (Gudmundsson et al. 2012; Gudmundsson 2016), is a nonparametric transformation approach. The RQUANT method employs local linear least squares regression to estimate the quantile-quantile relationship between the observed and modeled time series at regularly spaced quantiles. By fitting this relationship, the method provides a means to remove systematic biases in the GCM simulations. For quantiles that are not directly available in the data, the RQUANT method employs interpolation techniques to estimate their values based on the fitted values obtained from the regression. By applying the RQUANT method, we aimed to improve the accuracy of the GCM simulations by aligning them with the observed data and reducing the systematic biases. This approach allowed us to achieve reliable statistical transformations and enhance the overall quality of our analysis.
River discharge generation for flood simulation
There are several methods available for projecting discharge using climate models, including running hydrological models, using direct climate model runoff output, utilizing the aridity index, and employing statistical downscaling methods. In the case of running hydrological models, different levels of complexity can be used, with some models considering only temperature and precipitation inputs, while others incorporate additional variables such as solar radiation, relative humidity, and wind speed. However, the calibration of hydrological models typically requires long-term data spanning 10–30 years, which may not be available for the upper Lam Takong river basin in this study.
Another commonly used approach is based on the aridity index, which estimates annual surface runoff and evapotranspiration using available energy and precipitation. Various functional forms of the aridity index have been proposed, including those by Budyko (1948), Ol'Dekop (1911), Pike (1964), Schreiber (1904), and Turc (1954). The Schreiber (1904) functional relationship has been recommended as the most accurate by McMahon et al. (2011). This method has been applied in data-restricted regions where calibrated hydrological models are not feasible, as noted by Gonzalez-Zeas et al. (2012).


Budyko function F()
Author and name . | Budyko function F(![]() |
---|---|
Schreiber (1904) | ![]() |
Ol'dekop (1911) | ![]() |
Budyko (1948) | ![]() |
Turc (1954), Pike (1964) | ![]() |
Author and name . | Budyko function F(![]() |
---|---|
Schreiber (1904) | ![]() |
Ol'dekop (1911) | ![]() |
Budyko (1948) | ![]() |
Turc (1954), Pike (1964) | ![]() |
In this study, potential evapotranspiration (PET) was calculated using the Penman–Monteith equation (PET-PM), as described by Allen et al. (1998). The PET-PM equation combines mass transfer and energy balance principles, taking into account factors such as temperature, vegetation conductance, wind speed, relative humidity, and solar radiation (or solar duration). The required input variables for PET-PM include observed maximum temperature, minimum temperature, air temperature, wind speed, relative humidity, and solar radiation. The PET-PM method is recommended by the Food and Agriculture Organization (FAO) and widely recognized as the most accurate approach for calculating PET, as noted by Chen & Sun (2015) and Gao et al. (2017).


Flood simulation
The simulation of floods plays a crucial role in effective flood mitigation strategies and assessing flood risks for both current and future scenarios. To achieve this, the Flo-2D model is utilized, which is a two-dimensional flood routing model capable of simulating runoff over a system of square grid elements of any size. This model ensures the conservation of flow volume while numerically routing the flood hydrograph. The Flo-2D model requires input data such as the flood hydrograph at the upstream section, spatial information on land use, elevation, soil conditions, and rainfall characteristics (duration and amount). By utilizing this input, the model calculates the flood depth and inundation area. The numerical computation time step is determined by the wave celerity, and smaller grid elements necessitate smaller time steps to ensure accurate results.
The observed data, including rainfall, runoff, and discharge, were obtained from the bias-corrected data described in section 2.3, as well as from the Royal Irrigation Department. Furthermore, the flood inundation depths and areas were determined through surveys of flood marks during the severe flood events that occurred on October 8–9, 2020. To calibrate the Flo-2D model, the simulation of two past flood events was conducted: one in 2019 for calibration and another in 2015 for validation, as shown in Figure 2(b). Additionally, one of the most severe flood events from 2020 was used to test the model's performance. Moreover, six future flood scenarios were simulated to project flood outcomes during the near-future, mid-future, and far-future periods under the SSP2-4.5 and SSP5-8.5 scenarios.
RESULTS AND DISCUSSIONS
GCM performance for projected changes in precipitation and discharge
Generally, most CMIP6 models (represented by red and blue symbols before bias correction) exhibit R values in the range of 0.6–0.8, which are similar to those obtained after bias correction (represented by green and pink symbols). This suggests that the individual CMIP6 models do not show significant improvement in terms of R. However, there is a notable reduction in the center RMSD and SD after bias correction compared to their values before bias correction. Among the individual CMIP6 GCMs, the MME model yields the best results, characterized by the highest R value and the lowest RMSD and SD values.
Goodness-of-fit indicators (bias statistic and cdf distribution) of the mean monthly discharge during the reference period. The KS statistic at 0.05 significance level = 0.137.
Goodness-of-fit indicators (bias statistic and cdf distribution) of the mean monthly discharge during the reference period. The KS statistic at 0.05 significance level = 0.137.
Among the discharge estimation methods based on Budyko functions, the Schreiber (1904) method, suggested by McMahon et al. (2011), provided the largest values, resulting in the largest bias statistics. On the other hand, the Ol'dekop (1911) method yielded the smallest values and correspondingly the smallest bias statistics. However, the Schreiber (1904) method did not perform well in this study.
In contrast, the discharge estimated using the total runoff flux (mrro) showed the smallest bias across all months (see Figure 4(b)). Moreover, the cdf of mrro closely matched the observations (Figure 4(c)). It should be noted that all methods struggled to agree for smaller discharge values. Nevertheless, the mrro cdf yielded the smallest KS statistics compared to the other methods and was smaller than one at the 0.05 significance level. Based on these findings, the mrro method was selected to calculate the projected changes in discharge for flood simulations in response to a changing climate.
Temporal changes in projected precipitation and discharge
The projected precipitation and discharge both exhibit a continuous increase over the upper Lam Takong river basin. The magnitude of these increases is larger under the SSP5-8.5 scenario compared to the SSP2-4.5 scenario, although the difference is not statistically significant. It is important to note that while the MME helps reduce uncertainties, individual GCMs may still exhibit large biases or uncertainties, particularly for extreme values. This is evident from the noticeable differences between the lowest and highest values among the individual GCMs.
Several sources of uncertainty contribute to the overall uncertainties in the projections. These uncertainties can arise from the internal variability of the climate system (1), inter-model variability (2), and variability among different emission scenarios (3). In this study, it is found that uncertainties stemming from inter-model variability (2) and different emission scenarios (3) increase over time, as indicated by the widening inter-modal spreads under both scenarios. Consequently, the use of MME outputs for regional-scale flood simulation may introduce additional uncertainties.
The changes in projected precipitation and discharge are determined using scaling factors, as presented in Table 4. These scaling factors were calculated by taking the mean ratio of future results (during the corresponding periods) to the baseline simulations. Under the SSP2-4.5 scenario, the projected precipitation increases by 0.64, 2.18, and 5.74% for the near-future, mid-future, and far-future periods, respectively. For the SSP5-8.5 scenario, the projected precipitation increases by 0.77, 5.40, and 10.91% for the same periods. Similarly, the projected discharge also shows an increase. Under the SSP2-4.5 scenario, the discharge increases by 1.00, 1.26, and 4.57% for the near-future, mid-future, and far-future periods, respectively. For the SSP5-8.5 scenario, the discharge increases by 1.06, 3.92, and 11.05% for the same periods.
Scaling factors in projected precipitation and discharge changes
Variables . | SSP2-4.5 . | SSP5-8.5 . | ||||
---|---|---|---|---|---|---|
Near . | Mid . | Far . | Near . | Mid . | Far . | |
Precipitation | 1.0064 | 1.0218 | 1.0574 | 1.0077 | 1.054 | 1.1091 |
Discharge | 1.0100 | 1.0126 | 1.0457 | 1.0106 | 1.0392 | 1.1105 |
Variables . | SSP2-4.5 . | SSP5-8.5 . | ||||
---|---|---|---|---|---|---|
Near . | Mid . | Far . | Near . | Mid . | Far . | |
Precipitation | 1.0064 | 1.0218 | 1.0574 | 1.0077 | 1.054 | 1.1091 |
Discharge | 1.0100 | 1.0126 | 1.0457 | 1.0106 | 1.0392 | 1.1105 |
These scaling factors serve as inputs for the flood simulations conducted using the Flo-2D model in Section 3.3. It should be noted that some individual GCMs exhibit large biases in both precipitation and discharge, ranging from approximately 30–100% for extreme values, particularly from the mid-future to far-future periods (as shown in Figure 5).
For the mean monthly discharge, the MME does not accurately capture the peak discharge. However, similar to the pattern observed in mean monthly precipitation, there is an increasing trend in discharge from May to October as we progress from the near-future to the mid-future and far-future periods. Similar to precipitation, there is a greater increase in discharge under the SSP5-8.5 scenario compared to the SSP2-4.5 scenario.
Assessment of flood depth and flood inundation area of the past and present floods
Time series of calibrated and validated water level and discharges at sta. M89.
The model generally tends to underestimate small inundation depths (<0.5 m), but it demonstrates satisfactory performance in simulating higher inundation depths (>1.0 m). It should be noted that due to the resolution of the model used in this study, comparing flood depths across the entire floodplain area can be challenging. Nonetheless, the model provides reasonable simulations of water levels and discharge in the main channel (station M89).
Flood peak at sta. M89 under base case and projected CMIP6 (SSP-RCP) scenarios.
Various structural measures such as reservoirs, diversion channels, flood walls, retention basins, and dredging have been considered for this hotel. However, none of these measures are deemed feasible due to the following reasons:
- 1.
The high land prices in the area and the location within the national park limit the availability of suitable land for implementing structural measures.
- 2.
Constructing flood walls would impede the lateral flow of water, potentially exacerbating flood risks in other areas.
- 3.
Dredging activities could lead to more severe landslides, posing additional hazards.
Therefore, non-structural measures are recommended instead. These may include implementing a flood warning system, obtaining flood insurance coverage, and utilizing flood-proof techniques for constructing high-value buildings.
Projected maximum flood peak, inundation depth, and inundation area
The Flo-2D model is utilized under projected SSP2-4.5 and SSP5-8.5 scenarios, with input data obtained from the MME. The projected changes in precipitation and discharge are incorporated into each grid and at station M43A, respectively, based on the scaling factors presented in Table 3. The compound Manning's roughness under base case and future scenarios was used according to Table 1. We use similar roughness for all future scenarios according to the 20-year land use map (see Figure 1). Figure 10 displays the flood peak at sta. M89 under the base case and future scenarios. Due to the Act of the first national park for Khao Yai, changes in roughness increase peak discharge of 0.76 and 0.83% under near-future of SSP2-4.5 and SSP5-8.5 scenarios, respectively. In addition, the impact from climate change can cause significant increases in the peak discharge of 4.31 and 19.09% under far-future of SSP2-4.5 and SSP5-8.5 scenarios, respectively.
Projected maximum inundation depth and area under base case and projected CMIP6 (SSP-RCP) scenarios. The 1st and 2nd numbers in each figure denote inundation (km2) and floodplain storage (mil. m3), respectively.
Projected maximum inundation depth and area under base case and projected CMIP6 (SSP-RCP) scenarios. The 1st and 2nd numbers in each figure denote inundation (km2) and floodplain storage (mil. m3), respectively.
Given the popularity of the Khao Yai national park among tourists, these inundation maps serve as valuable tools for supporting various decision-making processes related to land use, infrastructure development, transportation planning, flood warning systems, evacuation procedures, emergency management, and flood preparedness and response measures.
CONCLUSIONS
Extreme rainfall and the resulting floods are common occurrences during Thailand's summer monsoon season. However, hydrological measurements are lacking in many of the flooded basins, making it challenging to understand the extent of the problem. In this study, we employed statistical bias correction techniques to analyze changes in precipitation and discharge using the latest CMIP6 model simulations over the upper Lam Takong river basin in Thailand.
We utilized 18 CMIP6 models under two SSP-RCP scenarios (SSP2-4.5 and SSP5-8.5) to assess future changes for three periods: near-future (2015–2039), mid-future (2040–2069), and far-future (2070–2099). Among these models, ten provided total runoff (mrro) as an internal variable. We performed an intercomparison of several observation datasets and identified SA-OBS as a suitable reference dataset based on its representation of median values. The RQUANT mapping method was applied for bias correction. Model performance was assessed using correlation coefficient (R), center RMSD, and SD. The main findings can be summarized as follows:
- 1.
Before bias correction, most CMIP6 models showed correlation coefficients (R) in the range of 0.6–0.8, which did not exhibit distinct improvement after bias correction. However, there were significant reductions in center RMSD and standard deviations (SD) after bias correction. The MME model yielded the best results, with the highest R and lowest RMSD and SD among the individual CMIP6 GCMs.
- 2.
Discharge at the upstream station (M43A) was estimated on a monthly time scale using five different runoff estimation methods. The estimation based on total runoff (mrro) showed the smallest bias for all months. Projected precipitation and discharge exhibited a continuous increase over the upper Lam Takong river basin, with larger magnitudes under the SSP5-8.5 scenario compared to SSP2-4.5 (though the difference was insignificant). Scaling factors were employed to determine the changes in projected precipitation and discharge. The projected precipitation under SSP2-4.5 (SSP5-8.5) increased by 0.64, 2.18, 5.74% (0.77, 5.40, 10.91%) while the discharge increased by 1, 1.26, 4.57% (1.06, 3.92, 11.05%) for the near-future, mid-future, and far-future periods, respectively. Some individual GCMs exhibited significant biases (around 30–100% from the mid-future to the far-future periods) for extreme values, leading to increased uncertainties when using MME outputs for regional-scale flood simulations.
- 3.
The Flo-2D model was utilized to project regional-scale floods in the Pak Chong district, Thailand. The model incorporated outputs from the CMIP6 MME under the SSP2-4.5 and SSP5-8.5 scenarios. Changes in land use were estimated in term of changes in compound Manning's roughness for flood simulation. To calibrate and validate the model, two past flood events in 2019 and 2015 were simulated. For model testing and projection, the most severe flood event in 2020 and six future flood scenarios were simulated. It should be noted that the Flo-2D model tended to overestimate peak discharges in both calibration and validation years. However, it reasonably simulated maximum water levels. Subsequently, the model was employed to simulate the most severe flood in 2020, which was the highest recorded since 1992. In general, the model exhibited a tendency to underestimate small inundation depths (<0.5 m) but satisfactorily simulated high inundation depths (>1.0 m). It is important to acknowledge that due to the model resolution in this study, it was challenging to compare detailed flood depths over the floodplain areas. Nevertheless, the model provided satisfactory simulations of water levels in the main channel.
- 4.
During the flood event in 2020, the U Khao Yai hotel and several resorts experienced flooding with depths smaller than 0.5 m. However, most outdoor facilities such as gardens, roads, and swimming pools were severely flooded with depths ranging from 1 to 4 m. The duration of the flood event lasted only 1–2 days. Based on our detailed survey, we investigated potential structural measures including the construction of a reservoir, diversion channels, flood walls, retention basins, and dredging. However, it was determined that implementing these structural measures is not feasible. Several factors contributed to this decision. Firstly, the high land prices and the location of the area within the national park made it challenging to find suitable land for structural interventions. Additionally, constructing flood walls would hinder lateral flow, and dredging could potentially trigger more severe landslides. As an alternative, we recommend the implementation of non-structural measures such as a comprehensive flood warning system and flood insurance. These measures can enhance preparedness and response to flood events. Furthermore, incorporating flood-proof techniques in the design and construction of high-value assets, such as the U Khao Yai hotel, can help mitigate potential damage caused by future floods.
- 5.
The Flo-2D model was utilized with changes in roughness and input from the MME under projected SSP2-4.5 and SSP5-8.5 scenarios. Changes in roughness slightly increase peak discharge of 0.76 and 0.83%.The resulting inundation maps demonstrated minimal variations among scenarios and timelines. Although there were slight changes in the inundation area, with a difference of only 0.05%, the floodplain storage experienced a more significant alteration of 6.85% due to the mountainous nature of the river basin.
Given that the area, particularly the Khao Yai national park, is a well-known tourist attraction, the projected inundation map holds significant value in establishing various measures. It can be instrumental in the development of zoning regulations, land use planning, and setting building standards to ensure the safety and resilience of infrastructure and transportation systems. Furthermore, the map serves as a vital tool for flood warning systems, evacuation planning, and emergency management, enabling authorities to proactively prepare for and effectively respond to flood events. By leveraging the projected inundation map, local policymakers and stakeholders can make informed decisions regarding flood mitigation strategies, resource allocation, and public safety measures. This comprehensive approach aids in minimizing the potential impacts of floods and enhances the overall resilience of the area in the face of future flood events.
Due to the current temporary limitation of available CMIP6 models, particularly those that provide internal variables such as runoff, it is crucial to conduct further evaluation as more models are gradually released through the Scenario Model Intercomparison Project (Scenario MIP) (O'Neill et al. 2016). This ongoing evaluation will contribute to enhancing the understanding and accuracy of future projections in the field of flood research. However, based on the findings obtained from the analysis of 18 CMIP6 models and the regional-scale flood simulations conducted in this study, significant improvements have been made in projecting precipitation patterns, discharge rates, flood depths, and inundation areas.
These findings hold immense value for local policymakers, as they serve as critical decision-making tools in flood mitigation strategies, land use planning, emergency management protocols, and raising general public awareness about flood risks. While this research may not have incorporated a high level of detail, its preliminary results offer valuable insights. They can assist local policymakers and water planners in adopting effective measures to achieve better flood management by 2100. By utilizing the projections and insights gained from this study, policymakers can develop proactive strategies, allocate appropriate resources, and implement targeted interventions to minimize the potential impacts of future floods on the region.
ACKNOWLEDGEMENTS
We express our sincere appreciation to the anonymous reviewers for their invaluable comments and feedback, which greatly contributed to the improvement of this manuscript. The author would like to extend gratitude to the Thailand Research Fund, specifically the Royal Golden Jubilee Ph.D. Program, for their generous support in the form of a full scholarship (grant number PHD/0203/2560). This support has been instrumental in facilitating the completion of this research. Additional financial support for field investigation in research sandbox project from the FutureTales LAB by MQDC is acknowledged. We also would like to acknowledge the Earth System Grid Federation (ESGF) for their efforts in archiving and providing access to the CMIP6 dataset. The availability of this dataset has been instrumental in conducting the analyses and drawing meaningful conclusions. We are also grateful to the observation dataset provider mentioned in the manuscript for their contributions.
DATA AVAILABILITY STATEMENT
Data cannot be made publicly available; readers should contact the corresponding author for details.
CONFLICT OF INTEREST
The authors declare there is no conflict.