Open Data Flood Mapping of Chao Phraya River Basin and Bangkok Metropolitan Region

Aims: To examine the utility of open data for flood mapping of the Bangkok Metropolitan Region and Chao Phraya River basin. The region is particularly vulnerable to flooding, having experienced recurrent major flooding events, including the some of the most extensive and prolonged in 2011. Study Design: Novel methodologies were innovated utilising open spatial data and open source geographical software to generate flood extent/hazard maps of the Bangkok Metropolitan Region and Chao Phraya River basin. Key geospatial data were sourced from the Thai Geo-Informatics and Space Technology Development Agency and NASA’s Shuttle Radar Topography Mission. Methodology: Given limited resources for conducting detailed hydrological-hydraulic analyses, two alternative approaches were examined for flood extent/hazard mapping of the basin and city. The first method made use of publicly available historical flood data to produce an up-to-date composite flood extent/hazard map. The second approach, using the latter output as a reference source, examined the utility of a modified topographic index for delineating flood-prone areas, as integrated into the r.hazard.flood module of the open source GRASS GIS application. Results: Compilation of multi-year historical data enabled generation of a relatively finescale (~100m spatial resolution) flood extent/hazard map for the basin and city. The optimal tau threshold for delineating flood exposed cells from the modified topographic index was linearly related to the sub-basin mean slope. The four most northerly subbasins of the Chao Phraya basin, those with higher mean slopes, gave lowest total errors, ranging from 17.5 to 35.9 percent. Conclusions: Open data in the form of multi-year spatial flood layers were effectively Original Research Article British Journal of Environment & Climate Change, 4(2): 186-216, 2014 187 combined to generate a relatively fine-scale flood extent/hazard map for the Chao Phraya River basin and Bangkok Metropolitan Region, and the modified topographic index showed promise as an alternative means for identifying flood exposed areas.


INTRODUCTION
Flooding is an increasing hazard for low-lying coastal megacities of Southeast Asia. Climate change is projected to bring increased precipitation and frequency of intense rainfall events, increased extreme rainfall events linked to tropical cyclones and sea level rise [1]. The densely inhabited mega delta regions in the Asia-Pacific region are described as being at most risk from increased flooding from the sea and in some areas by rivers [2].

Increasing Flood Potential of Chao Phraya River Basin
The Bangkok Metropolitan Region (BMR), situated at the mouth of the Chao Phraya River, is vulnerable to both riverine and coastal flooding. The area is low-lying, with Bangkok City mostly less than 2.5 meters above mean sea level and below mean sea level in some areas [3]. In a recent OECD study, Bangkok was ranked as one of the most vulnerable to future coastal flooding with regard to population and asset exposure in the context of expected urbanisation and economic growth [4]. Coastal erosion in the upper Gulf of Thailand was described as already critical and impeding sustainable development of the BMR, with sea level rise and land subsidence due to excessive water withdrawal likely exacerbating hydrological impacts in the future [5]. Factors that potentially increase the risk of flooding in the BMR, including both human-related development activities within the Chao Phraya River catchment and climate change-related impacts [1,[5][6][7][8][9], are shown in Fig. 1.

Bangkok Flood History
Bangkok has experienced recurrent major flooding, including from the mid-twentieth century : 1942, 1978, 1980, 1983, 1995, 1996, 2002, 2006 and 2011 [6,[10][11]. The flood of 2011 was unprecedented in recent times, likely representative of more than a 100-year return period event, resulting from heavy rainfall over the entire Chao Phraya catchment [12][13]. According to DHI [12], the return period is estimated above 100 years, based on modelling conducted of the Chao Phraya River basin since the 1990s, hydrological studies and measured peak water levels, with the latter likely underestimated in 2011 given diversions of upstream flows; observed and estimated peak water levels reproduced as Table A1 in Appendix. Precipitation during the rainy season of 2011 was reported as 143 percent of the average over the 20 year period from 1982 to 2002 [14]. The flood resulted in over 800 mortalities, the displacement of thousands of city residents, an estimated 45 billion USD in damage, and economic repercussions internationally [14][15][16][17].

Flood Modelling
In order to accurately model flooding in the Bangkok Metropolitan Region, requires a detailed understanding of rainfall and run-off processes in the Chao Phraya River basin and its constituent sub-basins. The Chao Phraya River basin covers an area of approximately 160,000 km 2 (see Table 3, section 3.1) with four northern tributaries (Ping, Wang, Yom and Nan Rivers) that converge at Nakhon Sawan to form the Chao Phraya River, which subsequently branches and flows through the BMR to the Gulf of Thailand (Fig. 2) [7]. Flooding can result from a number of causes including upstream storm run-off, direct precipitation over the city, coastal storm surge and sea level rise, and the extent of flooding influenced by a range of factors such as flood management infrastructure, drainage and sewerage networks, land subsidence, sedimentation, and soil properties [18], as well as measures adopted during flood events such as water diversion (e.g., [12]). Since the midtwentieth century, in the Chao Phraya basin, 3000 dams have been constructed, the two largest being the Bhumiphol and Sirikit dams which control 22 percent of the run-off of the whole Chao Phraya basin [7]; the latter two dams constructed in 1964 and 1974 respectively [14] (see Fig. 2).

Flood Modelling
In order to accurately model flooding in the Bangkok Metropolitan Region, requires a detailed understanding of rainfall and run-off processes in the Chao Phraya River basin and its constituent sub-basins. The Chao Phraya River basin covers an area of approximately 160,000 km 2 (see Table 3, section 3.1) with four northern tributaries (Ping, Wang, Yom and Nan Rivers) that converge at Nakhon Sawan to form the Chao Phraya River, which subsequently branches and flows through the BMR to the Gulf of Thailand (Fig. 2) [7]. Flooding can result from a number of causes including upstream storm run-off, direct precipitation over the city, coastal storm surge and sea level rise, and the extent of flooding influenced by a range of factors such as flood management infrastructure, drainage and sewerage networks, land subsidence, sedimentation, and soil properties [18], as well as measures adopted during flood events such as water diversion (e.g., [12]). Since the midtwentieth century, in the Chao Phraya basin, 3000 dams have been constructed, the two largest being the Bhumiphol and Sirikit dams which control 22 percent of the run-off of the whole Chao Phraya basin [7]; the latter two dams constructed in 1964 and 1974 respectively [14] (see Fig. 2).

Flood Modelling
In order to accurately model flooding in the Bangkok Metropolitan Region, requires a detailed understanding of rainfall and run-off processes in the Chao Phraya River basin and its constituent sub-basins. The Chao Phraya River basin covers an area of approximately 160,000 km 2 (see Table 3, section 3.1) with four northern tributaries (Ping, Wang, Yom and Nan Rivers) that converge at Nakhon Sawan to form the Chao Phraya River, which subsequently branches and flows through the BMR to the Gulf of Thailand (Fig. 2) [7]. Flooding can result from a number of causes including upstream storm run-off, direct precipitation over the city, coastal storm surge and sea level rise, and the extent of flooding influenced by a range of factors such as flood management infrastructure, drainage and sewerage networks, land subsidence, sedimentation, and soil properties [18], as well as measures adopted during flood events such as water diversion (e.g., [12]). Since the midtwentieth century, in the Chao Phraya basin, 3000 dams have been constructed, the two largest being the Bhumiphol and Sirikit dams which control 22 percent of the run-off of the whole Chao Phraya basin [7]; the latter two dams constructed in 1964 and 1974 respectively [14] (see Fig. 2).

Fig. 2. Three-dimensional view of Chao Phraya River, adjacent Sa Keo sub-basin, and BMR with labelled administrative areas
Note: vertical elevation of image is exaggerated. Open data sources used in figure: digital elevation model [37]; basin/sub-basin boundaries [40]; dams and reservoirs [51]; urban areas [49]; waterways [50]; and BMR administrative areas [46] Detailed flood inundation modelling efforts in the recent past by Panya Consultants [5] and the World Bank [18] examined the potential impact of climate change on flooding in Bangkok. However, modelling systems used in the latter studies, based on the commercial MIKE FLOOD software package developed by the Danish Hydraulic Institute (DHI), were not accessible, nor the data and financial resources available for the author to conduct an equivalent detailed investigation. Accurate and reliable flood hazard/risk maps were not available for the Chao Phraya River basin nor Bangkok Metropolitan Region. Data management concerns have previously been ascribed to multiple agencies responsible for maintaining water resources data and information [7].

Flood Extent/Hazard Mapping
This study specifically addresses creation of flood extent/hazard maps. In general, the term 'flood hazard map' refers to a spatial information product that illustrates geographical areas that are flood-prone or susceptible to flooding, which differs from a 'flood risk map' which indicates potential negative impacts of a flood, such as the number of people and kind of economic activities affected [19]. More specific requirements of hazard maps, as defined by the European Flood Directive (EFD) definition [19], includes identification of those areas at low, medium (return period ≥ 100 years) and high probabilities of flood, including components of extent, depth, and flow velocity where relevant. The emphasis here is on mapping the geographical extent of flooding, plus indication of flooding frequency based on historical data, with outputs to be integrated subsequently with socio-economic and health data to better understand risk and resilience as part of the Coastal Cities at Risk (CCaR) project. CCaR supports research in the megacities of Bangkok, Manila, Lagos and Vancouver, of which a central output is development of a spatial system dynamics simulator to model city resilience (Coastal Megacity Resilience Simulator) [22], of which developing a measure of flood risk is a key component, together with other economic, social and health inputs. However, the unavailability of flood hazard data became apparent during implementation of project in Bangkok, and thus critical that alternative approaches identified for flood mapping.

International Coastal Cities at Risk Initiative
Hence the focus of this paper to describe approaches taken, challenges encountered, and to share findings with wider research and practitioner communities. Moreover, with two major floods each decade since the 1980s, flooding poses a substantial climate-related risk to the livelihood of Bangkok residents. The need for accurate, accessible and up-to-date flood maps is clearly evident to facilitate decision making, whether for government disaster preparedness planning, investment by the business community, or residents wanting to better understand their potential vulnerability.

Study Objectives
The chief objective of the study was to examine the utility of open data for flood mapping of the BMR and its wider basin. While there is no single definition of 'open data' [23][24], for the purpose of this study, the broad opendefinition.org was adopted, where: 'a piece of data or content is open if anyone is free to use, reuse, and redistribute it -subject only, at most, to the requirement to attribute and/or share-alike' [25]. In this study all of the data used in the analyses were freely accessible online, apart from one set of administrative boundaries supplied by the BMA ([26-30], Table A2). However, only one dataset considered for use, NASA's Shuttle Radar Topography Mission (SRTM) data, was released and distributed without restriction; its derivative created by Consultative Group for International Agricultural Research -Consortium for Spatial Information (CGIAR-CSI) was actually used in this study, but is more restrictive with regard to data distribution (see section 2.1 and Table A2 for more  details).
Key open datasets used in the study included eight years of historical inundation data, and application of a modified topographic index (TI m ) based on SRTM digital elevation model data using methodology developed by Manfreda and colleagues [31][32]. According to the Manfreda et al. [31], the TI m -based method 'may represent a useful and rapid tool for a preliminary delineation of flooding areas in ungauged basins and in areas where expensive and time-consuming hydrological-hydraulic simulations are not affordable or economically convenient'.
Historical data and outputs from Tl m will indicate at a relatively fine scale the spatial extent of flood-prone areas. Multi-year historical data can also provide a spatio-temporal indication of flooding occurrence for hazard mapping (section 3.1). Though eight consecutive years of historical data is a relatively short time-span, the period includes two major flood episodes, including the most recent of 2011, estimated as having a return period of more than 100 years (see section 1.2); and based on discharge data from the Chao Phraya River basin, the 2006 flood had an estimated return period of 10 years [12]. Moreover, the pattern of recent historical inundation should reflect current flood management infrastructure and efforts, potentially useful given the current lack of available modelling capacity. Flood-prone areas derived from the TI m are based on the compiled eight-year historical flood map; given that 2011 likely reflected an above 100-year flood event, outputs from the TI m analysis were considered roughly correspondent with this return period. Data from all eight years were used in the TI m analysis, and not 2011 coverage alone, as collectively this should capture more areas susceptible to flooding, thereby addressing spatial variability of precipitation across the basin. Findings will be directly used for examining city resilience though the Coastal Megacity Resilience Simulator (section 1.5), and should also prove beneficial for other efforts investigating and developing risk maps for the basin and city.
The study's secondary aim is to provide a quantitative historical assessment of the impact of recent flooding events in terms of area inundated, including the above 100-year event of 2011. Additionally, open source software is used in the analyses, with the aim of presenting methods that could be adopted by researchers or practitioners working outside an institutional setting with limited access to commercial GIS software, key issues which in the opinion of this author pose barriers to the uptake and application of geospatial technologies. The r.hazard.flood module is a python-based script that can be readily integrated into the main GRASS software suite, accessed through a graphical user interface window, and enables creation of flood map rasters based on a modified topographic Index (TI m ) derived from DEM physical properties. Research on applying TI m -based flood mapping is well documented on river systems in Italy [31], and in this study, its utility further evaluated as a practical open source mapping tool by applying to the Chao Phraya River basin in Thailand.

MATERIALS AND METHODS
Two approaches were examined for creating flood extent/hazard maps based on the availability of open data. Firstly, multi-year spatial historical inundation data (see section 2.1) were used to create a GIS composite layer of the entire Chao Phraya basin and BMR, and the latter then used as a reference layer to assess the utility of the modified topographic index, as integrated in the GRASS GIS module r.hazard.flood, for delineating flood-prone areas. Open source desktop GIS and command line utilities were used, including GRASS GIS and associated add-on modules [33], and GDAL/OGR command line utilities [34]. Before examining these methods, open data sources used are highlighted in the next section.

Open Data Sources and Quality Assessment
A variety of data sources were utilised and are detailed below, together with any data quality issues encountered.

Historical flood inundation layers
Historical spatial flood data were obtained from the Thailand Flood Monitoring System of the Geo-Informatics and Space Technology Development Agency (GISTDA) [35]. Flood inundation area maps were downloaded as ESRI shapefiles from the latter site, with annual flood data accessible from 2005 to 2012 inclusive. The flood maps were derived from synthetic aperture radar (SAR) imagery acquired during the rainy season [GISTDA, personal communication].

Digital elevation data
Post-processed digital elevation data from the Shuttle Radar Topography Mission (SRTM) [36] were used in the study to create the modified topographic index (TI m ). SRTM data used were of 3 arc second spatial resolution and processed to fill no-data regions by the Consortium for Spatial Information (CSI) of the Consultative Group for International Agricultural Research (version 4.1) (CGIAR) [37][38]. ASTER Global Digital Elevation Model (ASTER GDEM) [39] data were also examined for use, but quality appeared variable in the lower Chao Phraya basin which precluded theiruse on this occasion; a linear artefact was apparent in two tiles: ASTGTM2_N14E099_dem_tif and ASTGTM2_N14E100_dem.tif.

Chao Phraya river basin and sub-basin boundaries
The Chao Phraya River basin and sub-basin boundaries for the study were defined using the ESRI shapefile -'Hydrological basins in Southeast Asia (Derived from HydroSHEDS)'accessible from the UN FAO's GeoNetwork portal [40]. The data set was published under FAO's AQUASTAT programme, and derived from the HydroSHEDS initiative of the World Wildlife Fund (in partnership with the U.S. Geological Survey (USGS) and other organisations) [41]. While basin boundaries have been derived elsewhere (as briefly mentioned below), boundaries for both basin and sub-basins are provided, all of which are named; however, some sub-basins were incorrectly labelled in the latter resource and have been re-labelled 1 to reflect main rivers/waterways as shown in Fig. 3.

Fig. 3. Chao Phraya River basin and sub-basins and adjacent Sa Keo subbasin
Coordinate reference system: EPSG 32647 (UTM47N WGS84). Basin boundaries (white lines) as per FAO AQUASTAT hydrological basins data set [40]. Other open data used in figure: CGIAR-CSI SRTM digital elevation model [37] HydroSHEDS data are primarily based on elevation data acquired from NASA's Shuttle Radar Topography Mission [36] as well as the Digital Chart of the World (DCW), ArcWorld data set, and the Global Lakes and Wetlands Database (GLWD) [42]. Unlike HydroSHEDS, the basins/sub-basins are named in the FAO-amended dataset, and the boundary of the Chao Phraya basin closely matches that in HydroSHEDS, apart from some differences in the southern part of the basin, such as inclusion of the Chai Si sub-basin and small differences near the coast. The technical literature for HydroSHEDS describes the drainage basins as not yet finalised, though generally they have better accuracy than the earlier HYDRO1k elevation derivative database, ArcWorld and DCW, and there may be issues concerning the accuracy of river networks in low relief areas [42].
HYDRO1k Elevation Derivative Database is derived from the Global 30 Arc-Second Elevation (GTOPO30) data set, a 30 arc second (~1 km) digital elevation model of the world created by USGS in 1996 [43][44]. Layers created as part of the data set include derivatives for hydrological analysis including basin boundaries, streams, slope, aspect, flow direction and accumulation.
Another data set delineating the Chao Phraya basin is the Global Runoff Data Centre (GRDC) 'Major River Basins of the World' [45], derived from HYDRO1k. The chief differences between the latter and FAO's data set concerns the inclusion of the Mae Nam Moei (Salween basin) and the Sa Keo (Gulf of Thailand Coast) sub-basins into the Chao Phraya River basin in the GRDC layer.
For the purpose of this study, FAO's delineation of the Chao Phraya basin was adopted [40], where the Bangkok Metropolitan Region (BMR) crosses two major basins -the Chao Phraya and Gulf of Thailand Coast basins. The BMR extends from the Chao Phraya basin into the Sa Keo sub-basin of the latter. The BMR and entire Chao Phraya basin are located in the UTM47N zone and all results are presented based on this projection.

Administrative boundaries and urban areas
The BMR, or Greater Bangkok, comprises the city of Bangkok (special administrative area) and conterminous provinces of Samut Prakan, Samut Sakhon, Nakhon Pathom, Nonthaburi, and Pathum Thani. The BMR boundary can be freely accessed from the GADM database of Global Administrative Areas [46]. The BMR covers an area of approximately 7652 km 2 (calculated from GADM) with a population of 10.5 million in 2012 [47], though the population varies by several million more in consideration of unregistered residents and commuters [48]. A BMR boundary layer supplied by the Bangkok Metropolitan Administration proved to be mostly congruent with the GADM-derived data with a total BMR area of 7689 km 2 ; some minor boundary differences are apparent generally with the largest differences in Samut Sakhon and Samut Prakan provinces. The BMA-supplied layer was used for determining flooded areas for each BMR administrative area, as reported in Tables 1 and 2 (Results and Discussion). Urban areas flooded in the BMR were based on open data from the Global Rural-Urban Mapping Project (GRUMP) by CIESIN et al. [49].
Other open data sets accessed include those used for displaying rivers and water bodies in Fig. 2 above [50][51].

Creation of Composite Historical Flood Map
The method adopted for creating the composite flood map is summarised diagrammatically in Fig. 4. Shapefiles were clipped and transformed from EPSG 4326 to EPSG 32647 (UTM 47N/WGS84) and rasterized to a spatial resolution of 100m using GDAL command line utilities -ogr2ogr and gdal_rasterize; the appropriate spatial resolution was determined following consultation with GISDTA. After rasterizing, a single compiled raster image was created from the series of eight flood inundation images (2005 -2012 inclusive) by summing cell values using the GRASS r.series module [52]. The output layer contained cell values ranging from 1 to 8 representing areas with the lowest to highest occurrence of flood; areas unflooded during the 8 year period were assigned a value of zero. Based on the accuracy of the input inundation maps, identification of flooded areas should be approximately 80 to 90% correct, though the error might be higher in urban areas [53].

Development of Flood Map Using Modified Topographic Index and Error Analysis
The second approach examined the utility of the modified topographic index (TI m ), a procedure developed by Manfreda et al. and incorporated into the GRASS r.hazard.floodadd-on extension module [31][32][54][55]. The module delineates flood-prone areas by calculating a modified topographic index from a digital elevation model (see formula below), and categorises those cells above a given threshold value as exposed to flooding [31]. In other words, flood-prone areas are derived from topographic parameters, the supposition being that physical surface features can reveal areas exposed to flood inundation given that the land surface is shaped by hydrological processes [31]. Areas susceptible to inundation are determined by a value above a calculated threshold (tau, τ).
The TI m is based on the Topographic Index (TI), also called the Topographic Wetness Index (TWI), which reflects the quantity of water flowing to a particular location and accumulation in low slope areas (e.g., [56]), but differs only by the inclusion of an exponent, n [57]. The TI m is a function of the specific upslope contributing area (a d ) (determined by multiplying accumulation x cell size), the local slope (β), and the exponent n is a function of DEM cell size.
The threshold tau and exponent n values incorporated into the r.hazard.flood module were derived from research conducted in the Arno River basin in Italy [31], employing a calibration procedure based on error analysis of TI m outputs compared to reference flood maps generated from an ensemble of historical flood maps, geomorphological and hydraulic studies [31,57]. The TI m threshold value used in r.hazard.flood corresponds to the minimum total error with reference to the higher 500-year return period of the reference maps, where τ = 10.89n + 2.282 [31]. The exponent n was found to be strongly correlated to DEM cell size, where a plot of n against DEM cell size could be expressed as n = 0.016x 0.46 [31]. Additional findings in the Italian study showed that good results were possible using DEM's with resolutions of up to approximately 100m, including reliable results from NASA's SRTM 90m DEM. For example, using this expression, a DEM of 90m cell size will give a threshold of 3.6627, with cells above this value defined as being exposed to flooding.
The method adopted in this study was to use the CGIAR-CSI SRTM 90m DEM of the Chao Phraya River basin and the composite historical flood map developed previously as a reference for assessing the utility of the modified topographic index for defining areas as exposed to flooding (see Fig. 5). Error estimation is based on the approach by Manfreda et al. [31] (see Fig. 6). While the analysis depends on the accuracy of the multi-year flood maps (~80-90% as noted previously), these currently represent the best source of information available for the Chao Phraya River basin; detailed and accurate flood maps based on hydrological-hydraulic mapping (as used in the Arno River basin study by Manfreda et al. [31]) were not available.
Additional potential constraints to error analysis of the TI m layer include land use/land cover changes in the Chao Phraya basin, including urbanisation and flood infrastructure (e.g., construction of dikes), and water diversion during a flooding event (e.g., [12]), which could modify the extent of flooding in the reference layer. For instance, greater flood protection given to key commercial and business districts of the BMR (located in the Delta sub-basin). Furthermore, an appropriate tau threshold needs to be determined to identify flood-prone areas based on the reference historical flood map. The reference map likely represents more than a 100-year flood (see comments above concerning the 2011 flood) and thus if the default threshold in r.hazard.flood were applied, it would give an overestimation of area delineated by the module as flood-prone. The optimal tau value, corresponding to minimum total error, was interpolated from error analysis of each sub-basin, as summarised in Fig. 6.

. Error analysis of TI m -based flood rasters to derive optimal tau threshold values
Error estimation based on the approach by Manfreda et al. [31] A DEM of the entire Chao Phraya River basin was created by patching together SRTM tiles downloaded from CGIAR-CSI [58] and this input to the r.hazard.flood add-on module to generate the TI m layer, with flood output layers subsequently manually generated using selected tau thresholds between 2.0 and 7.0 (including the r.hazard.flood default of 3.6627 which corresponds to a 90m resolution DEM based on a 500-year flood return period) (Fig. 5). Selected tau thresholds and total errors were plotted for each sub-basin and the minimum threshold interpolated using a cubic spline (using GRACE 2D plotting software 2 ). The mean slope of each sub-basin was also determined (output from r.slope.aspect [59]) and plotted against minimum total errors and tau thresholds to assess potential correlation.

Composite Historical Flood Map
Based on historical data, flood extent/hazard maps were generated for the entire Chao Phraya River basin (Fig. 7) and Bangkok Metropolitan Region (Fig. 8). Areas of highest flood occurrence are displayed in red and least risk coloured yellow, and those areas not flooded in the eight year study period remain uncoloured. Cumulative historical inundation (i.e.,

Fig. 6. Error analysis of TI m -based flood rasters to derive optimal tau threshold values
Error estimation based on the approach by Manfreda et al. [31] A DEM of the entire Chao Phraya River basin was created by patching together SRTM tiles downloaded from CGIAR-CSI [58] and this input to the r.hazard.flood add-on module to generate the TI m layer, with flood output layers subsequently manually generated using selected tau thresholds between 2.0 and 7.0 (including the r.hazard.flood default of 3.6627 which corresponds to a 90m resolution DEM based on a 500-year flood return period) (Fig. 5). Selected tau thresholds and total errors were plotted for each sub-basin and the minimum threshold interpolated using a cubic spline (using GRACE 2D plotting software 2 ). The mean slope of each sub-basin was also determined (output from r.slope.aspect [59]) and plotted against minimum total errors and tau thresholds to assess potential correlation.

Composite Historical Flood Map
Based on historical data, flood extent/hazard maps were generated for the entire Chao Phraya River basin (Fig. 7) and Bangkok Metropolitan Region (Fig. 8). Areas of highest flood occurrence are displayed in red and least risk coloured yellow, and those areas not flooded in the eight year study period remain uncoloured. Cumulative historical inundation (i.e.,

Fig. 6. Error analysis of TI m -based flood rasters to derive optimal tau threshold values
Error estimation based on the approach by Manfreda et al. [31] A DEM of the entire Chao Phraya River basin was created by patching together SRTM tiles downloaded from CGIAR-CSI [58] and this input to the r.hazard.flood add-on module to generate the TI m layer, with flood output layers subsequently manually generated using selected tau thresholds between 2.0 and 7.0 (including the r.hazard.flood default of 3.6627 which corresponds to a 90m resolution DEM based on a 500-year flood return period) (Fig. 5). Selected tau thresholds and total errors were plotted for each sub-basin and the minimum threshold interpolated using a cubic spline (using GRACE 2D plotting software 2 ). The mean slope of each sub-basin was also determined (output from r.slope.aspect [59]) and plotted against minimum total errors and tau thresholds to assess potential correlation.

Composite Historical Flood Map
Based on historical data, flood extent/hazard maps were generated for the entire Chao Phraya River basin (Fig. 7) and Bangkok Metropolitan Region (Fig. 8). Areas of highest flood occurrence are displayed in red and least risk coloured yellow, and those areas not flooded in the eight year study period remain uncoloured. Cumulative historical inundation (i.e., maximum extent of flooding derived by overlaying all annual inundation layers) are presented for: the entire BMR, Bangkok City and constituent provinces (Table 1); urban areas (Table 2); and entire Chao Phraya River basin and sub-basins (Table 3). Stacked bar charts showing the annual extent of flooding for the BMR (Bangkok City and its contiguous provinces) and the entire Chao Phraya basin (all 10 sub-basins) are displayed in Figs. A1 and A2 respectively (see Appendix). Key statistics of the 2011 flood disaster indicate that approximately 4200 km 2 (56 percent) of the BMR was inundated, and at the basin level over 26,000 km 2 (17 percent) was flooded.  3

flood pixels overlaying river/waterways/water bodies omitted from flood area calculations
Overall, at some time during the eight year period from 2005 to 2012, nearly 60 percent of the BMR's land area and 50 percent of its urban area was flooded. The province of Pathum Thani was the most flood-prone province with 92 percent of its land area inundated and 95 percent of its urban area. The provinces of Nonthaburi and Nakhon Pathom were also especially vulnerable with 78 and 66 percent of land inundated respectively. Bangkok City (the special administrative area), experienced flooding mainly in its eastern and western regions, while the coastal provinces of Samut Sakhon and Samut Prakan experienced flooding to a lesser extent, though not insubstantial; however, coastal flooding caused by tidal and storm surge may not be captured in the GISTDA inundation layers as they were derived from SAR data during the rainy season and focused on flooding caused by precipitation and downstream run-off (GISTDA, personal communication).
The results clearly indicate that millions of residents are vulnerable to flooding in the BMR: with a total population of over 10 million and some 50 percent of its urban area inundated on one or more occasions between 2005 and 2012 (see Table 2). It should also be noted that past historical inundations may not indicate the frequency and extent of present and future flooding events in the context of climate change and land cover/land use changes in the basin. However, given the lack of existing hydrological-hydraulic modelling data, these data do provide some indication of spatial vulnerability to flooding.  [49] and excluding river/waterway/waterbodies. 3 2012 population data from BMA (2012) [47]

Modified Topographic Index (TI m )
For each Chao Phraya sub-basin, the optimal threshold tau value, corresponding to minimum total error, was interpolated from a plot of calculated total errors and associated tau threshold values used for extracting the flood raster from the TI m layer. Fig. A3 (Appendix) presents total error and tau threshold plots of each Chao Phraya sub-basin, and Table 4 shows the interpolated minimum total errors and associated optimal tau thresholds, as well as area and mean slope for each sub-basin. Optimal tau threshold values varied from 3.96 (Mae Nam Ping) to 5.23 (Delta). It was anticipated that the estimated tau threshold would be higher than the 3.6627 calculated using the linear function derived by Manfreda et al. [31] (see Fig. 6); at the latter threshold, much of the error in the sub-basins is attributable to overestimation of flooded areas. The reference layer used in this study was derived from eight years of inundation data, with the 2011 flood likely reflecting more than a 1-in-100 year flood, whereas the tau threshold derived by Manfreda et al. [31] was based on a considerably higher return period of 500 years.
Total minimum error decreased with increasing sub-basin mean slope. Accuracy was substantially lower when applied to areas of lower topographic relief, with minimum total error increasing linearly from approximately 10 to 70 percent for a mean slope decreasing from 20 to 2.5 percent respectively. A linear regression of minimum total error and mean sub-basin slope shows a correlation (correlation coefficient -0.9449) between increasing mean slope and reducing total error; such a linear correlation does not appear present in Manfreda et al. [31] data. The minimum total error (y) can be estimated from the linear function: y = 78.174 -3.3377x, where x is the mean slope (Fig. 9). Thus, minimum total errors of 10, 25 and 50 percent correspond to mean slopes of approximately 20, 16 and 8% respectively. Significant error may be introduced into low relief areas given that the SRTM DEM comprises integer and not floating point data [31]. Unfortunately, it was not possible to further examine whether this error was related to using SRTM DEM given artefacts evident in the higher resolution ASTER DEM for the region. Overall, the four most northerly subbasins of the Chao Phraya basin, those with higher mean slopes, gave lowest total errors, ranging from 17.5 to 35.9 percent (see Fig. 10). All the remaining sub-basins had errors ranging from 45.8 to 87.2 percent.  [49] The optimal tau threshold representing minimum total error appears related to sub-basin mean slope. The four northernmost sub-basins, with highest mean slope values, presented optimal tau thresholds ranging from 3.96 to 4.50, with thresholds from 4.64 to 5.23 for the remaining sub-basins (Table 4). A linear relationship between optimal tau threshold and mean slope is given by the function: y = 5.1912 -0.06162x, where y is the optimal threshold (Fig. 11). It is interesting to note that Manfreda et al. also noted improved performance of the TI m method in basins of lower order [31]. Given that tau appears related to mean slope in this study, more accurate representation of flooding using TI m may be best obtained by customising the threshold to the mean slope of each sub-basin; in this study a single tau value does not appear practicable for accurately delineating flood exposed areas. Potential limitations and constraints to error analysis of the modified topographic index output are depicted in Fig. 12. Limitations include the unavailability of flood data derived from hydrological-hydraulic models of the basin for strengthening the accuracy of the reference data, and more accurate DEM data for generation of the modified topographic index layer.
Other potential constraints include accuracy of input historical inundation data, such as underestimation of flooded area in urban settings related to the satellite imagery, and the influence of land cover/land use changes such as urbanisation and flood defence management on flooding extent, e.g., greater flood protection of key commercial and business areas of the BMR (as located in the Delta sub-basin). Potential limitations and constraints to error analysis of the modified topographic index output are depicted in Fig. 12. Limitations include the unavailability of flood data derived from hydrological-hydraulic models of the basin for strengthening the accuracy of the reference data, and more accurate DEM data for generation of the modified topographic index layer.
Other potential constraints include accuracy of input historical inundation data, such as underestimation of flooded area in urban settings related to the satellite imagery, and the influence of land cover/land use changes such as urbanisation and flood defence management on flooding extent, e.g., greater flood protection of key commercial and business areas of the BMR (as located in the Delta sub-basin). Potential limitations and constraints to error analysis of the modified topographic index output are depicted in Fig. 12. Limitations include the unavailability of flood data derived from hydrological-hydraulic models of the basin for strengthening the accuracy of the reference data, and more accurate DEM data for generation of the modified topographic index layer.
Other potential constraints include accuracy of input historical inundation data, such as underestimation of flooded area in urban settings related to the satellite imagery, and the influence of land cover/land use changes such as urbanisation and flood defence management on flooding extent, e.g., greater flood protection of key commercial and business areas of the BMR (as located in the Delta sub-basin).

CONCLUSION
Open data in the form of multi-year spatial flood layers were effectively combined to generate a relatively fine-scale (~100m spatial resolution) flood extent/hazard map for Thailand's Chao Phraya River basin and Bangkok Metropolitan Region. The maximum flooding extent is likely representative of above a 100-year return period event; however it should be noted that future flooding in the basin may not reflect the past, given new development within the basin and climate change impacts. The flood extent/hazard map offers potential to be updated with more recent data so as to better reflect land cover/land use changes and flood infrastructure developments, as well as climate change impacts. As well as being used for the CCaR project city resilience modelling, the flood maps created in this study will be made available to the Bangkok Metropolitan Administration, requesting that they are added to their existing BMA Web Map Service 4 and also to GISTDA for sharing through their flood monitoring portal [35].
The utility of using a modified topographic index from SRTM DEM for delineating flood-prone areas also offers promise for flood mapping, where other means of assessment, such as detailed hydrological-hydraulic assessments and historical inundation data are unavailable. However, as also noted by Manfreda et al. [31], the index cannot be considered a replacement for more detailed analyses, but as a preliminary assessment of flood-prone areas. For instance, the procedure could be used to identify areas especially at risk from flooding. Findings presented here indicate good correspondence between reference historical inundation data and the results of the TI m method, at least for the four most northerly sub-basins of the Chao Phraya River basin characterised by higher mean slopes. However, selection of an appropriate tau threshold value is critical for generating an accurate flood map -the threshold value reflecting the return period to be presented, and as shown in this study, the mean slope of the sub-basin. No single tau threshold value appears practicable and might be best determined from a sub-basin's slope given the linear relationship observed above.
Performance of the TI m -based method is reduced in areas of low topographic relief, which may relate to the limited vertical resolution of the SRTM DEM (see section 3.2). The latter observation reflects the comment by Manfreda et al. [31] that the method may be especially useful in basins with marked topography, and greater error may relate to the SRTM DEM used to generate the TI m . Other potential errors may derive from the accuracy and nature of the reference layer compiled from historical inundation data, where flooding extent may be influenced by land cover/land use changes, such as urbanisation and flood protection measures. The two southernmost sub-basins, characterised by lowest mean slopes and highest urban extents, presented the highest total errors; however, the low mean slope and minimal urbanisation of the central Khlong Kriangkrai sub-basin suggests that mean slope might be the key factor influencing their observed error. Findings indicated that a total error of 20 percent or less required a mean slope of at least 17 percent (~10 degrees).
The modified topographic index is potentially a highly valuable tool for environmental and disaster risk management. To further validate the utility of using the modified topographic index as a tool for delineating flood-prone areas, it is recommended that future studies examine higher resolution DEMs, such as the 30m resolution ASTER GDEM, and other 4 BMA Web Map Service: http://203.155.220.230/wms/ drainage basins as study sites. Moreover, it would be informative to further examine optimal tau thresholds for delineating flood exposed areas and if a linear relationship between the tau threshold and mean slope is observable elsewhere.
The methodology presented here also offers scope to create flood maps representing multiple return periods. For example, according to the report by DHI [12], the 2006 flood was estimated as a 10 year flood and spatial flood data is already accessible from GISTDA. Furthermore, given estimated return periods of 25 and 50 years for the 1996 and 1995 floods respectively [12], there should be scope to derive flood maps for these return periods.
It should also be highlighted that given the emphasis on using open data and open source GIS software, the methods adopted here offer scope for independent non-institutional researchers to conduct further investigations. Open data made generation of Chao Phraya River basin and the BMR flood hazard maps feasible in a reasonable time and with limited financial resources, apart from expertise required to conduct the GRASS-based GIS analyses.