Resilience to unusual flooding after 2021 tropical storms in part of mainland Southeast Asia

In Southeast Asian countries, flooding is common and occurs a few weeks after tropical storms. In Thailand, the government’s approach to dealing with flooding is focused on the construction of anthropogenic megastructures. The loss of life, property, and ecosystem from flooding still occurs and will likely become increasingly more severe soon due to global climate variation. Here, in this brief report, we show a case study analyzing the cause of unusual flooding that occurred from September to November 2021 during the tropical depression “Dianmu” and the tropical cyclone “Kompasu” that hit the Khorat Plateau, northeastern Thailand. These two tropical storms left behind vast flooded urban and agricultural areas in the upstream part of the Chi River on this, the largest plateau in mainland Southeast Asia. The flood water stagnated for a month and unusually drained downslope in the upstream zone. In this work, a series of free online access satellite images taken by the Sentinel-1 C band GRD and Sentinel-2 available in the Google Earth engine platform was analyzed. As a result, we confirmed that the normal nature of water flow was problematic due to the presence of anthropogenic structures, especially roads. The installation of different design and dimensions of culverts are urgently recommended, particularly in the floodplain zone. The quick responsible idea provided in this paper highlights and simplifies the resilience to future extreme weather events of the rural community and ecosystem.


Introduction
Floods were counted as the most frequent type of disaster with 43% of all recorded events from 1998 to 2017 (United Nations Office for Disaster Risk Reduction, 2018). From early September to November 2021, the tropical depression "Dianmu, " which originated from the South China Sea, made landfall on the Vietnamese coast, then struck westwards to Laos PDR and ended at the Khorat Plateau, northeastern Thailand. Dianmu left behind a water volume (average of 27 mm/day in early September) that was slightly higher than Frontiers in Ecology and Evolution 02 frontiersin.org the average rate (about 17 mm/day) of monsoon rain in Thailand (Rain Water, 2021). The subsequent flooding devastated agricultural areas, ecosystems, transportation routes, and properties in the western part of the Khorat Plateau, especially the upstream part of the Chi River catchment basin on the dip slope of the cuesta. This massive volume of water slowly flowed downstream with a longer flooding period (4-6 weeks) than in a normal year (basically 1-2 weeks), even though the topographical slope of the western part of the basin is high (10-15 degrees). Later in November, the tropical cyclone "Kompasu, " generated in the Pacific Ocean, moved westwards to the South China Sea and made landfall in northern Vietnam ( Figure 1A). The water level in the Chi River once again increased and led to an unusual expansion of the floodplain boundary. Two provinces (Chaiyaphum and Khon Kaen) inside the Chi River's floodplain were severely flooded (Arunyanart et al., 2017;Vanichsan and Taesombat, 2022). We hypothesized that there were either natural or anthropogenic factors that caused this unusual water flow behavior. This brief report emphasized the application of satellite images on the Google Earth platform, and analysis of the free, onlineaccess Sentinel-1 and Sentinel-2 images. The optical remote sensing received image data from a spectral sensor in the range of short-wave infrared through to ultraviolet. It can be applied for flood monitoring because the infrared band reflectance on the water is low, while the reflectance of the green and red bands on the water is high. However, optical remote sensing has the limitation that it is unable to detect flooded areas in severe weather, such as a high percentage of cloud cover (Arusha and Bharathi, 2019). In contrast, the image from radar remote sensing is generated from a spectral sensor in the microwave range that can penetrate in every condition (Refice et al., 2018).
Recently, optical and radar remote sensing was found suitable for flood management and mapping (Refice et al., 2018). Synthetic aperture radar (SAR) is the most used technology for imaging radars in every weather zone and has been used for flood monitoring since the 1990s (Curlander and McDonough, 1991). This method has been employed to detect river floods in many countries of the world, including Bangladesh (Uddin et al., 2019), Canada (Brisco et al., 2019), China (Dong et al., 2021), Eastern Europe (Gan et al., 2012), Germany (Martinis and Rieke, 2015), and the United Kingdom (Martinis et al., 2009). Moreover, SAR can be assessed using many free online datasets, such as Sentinel-1, TerraSAR-X, RADARSAT-2, and COSMO-SkyMed (Pulvirenti et al., 2016).
Several methods of SAR-based analyses have been employed for flood monitoring, such as statistical active contouring (Horritt et al., 2001), the application of neural networks in a grid system (Kussul et al., 2008), pixel-based segmentation (Martinis et al., 2009), fractal dimensioning of multi-temporal images (Huang et al., 2011), texture analysis (Senthilnath et al., 2013;Pradhan et al., 2014;Sghaier et al., 2018), thresholding (Long et al., 2014), changes detection (Long et al., 2014), and clustering and classification (Twele et al., 2016;Martinis et al., 2018). All these methods use only one image to classify flooded areas. The change detection aims to compare the flood scene to an earlier dry image to classify the flooded area (Giustarini et al., 2015;Schlaffer et al., 2015). In addition, the change detection methods can be combined with other methods for classifying a flooded area that has an unusually low backscatter value (Clement et al., 2018). Thus, the aims of this work are set up to analyze the cause of unusual and long flood periods and to provide a simple way how to return the water flow to the normal rate.

Study area
In this brief report, the case study area is focused on the Chi River catchment basin, in northeastern Thailand. The Chi River is one of the two main rivers on the Khorat Plateau. The study is Maps showing the location of the Khorat Plateau in mainland Southeast Asia and the geographical names referred to in this brief report. Frontiers in Ecology and Evolution 03 frontiersin.org bound in the upstream part of the Chi River catchment. The Chi River is 765 km long with the upstream position on the back slope of a cuesta, where the slope is gentle, being about 150 km in length and situated in the Phetchabun mountain range, flowing eastwards through the center of northeastern Thailand (via Chaiyaphum, Khon Kaen, and Maha Sarakham provinces; Figure 1A). After that, the Chi River turns south in Roi Et, runs through Yasothon, and joins the Mun River in the Kanthararom district of Srisaket province. With such a long length, flooding often occurs within the Chi River floodplain. Several flood analyses of the Chi River have been reported from Chaiyaphum and Khon Kaen provinces (Kuntiyawichai et al., 2008;Arunyanart et al., 2017;Waiyasusri and Chotpantarat, 2020;Sakmongkoljit et al., 2021); however, none of the SAR data for this river catchment basin has been analyzed so far. Moreover, we also analyzed the satellite dataset from the middle and downstream parts of the river to help synthesize the behavior of the September 2021 flood throughout the entire catchment. We separated the catchment basins into five sub-areas: A1 (Chaiyaphum), A2 (Chonnabot and Ban Phai District), A3 (Khon Kaen), A4 (Downstream area of Ubol Rattana dam), and A5 (Maha Sarakham; Figure 1B). The upstream part includes sub-areas A1 to A3.

Satellite dataset
The SAR from the Sentinel-1 dataset taken between September and November 2021 that covers Chaiyaphum, Khon Kaen, and Maha Sarakham provinces were selected. The change detection and thresholding (CDAT) method was based on Long et al. (2014). This study used the Sentinel-1 data available from the European Space Agency Copernicus program, which consists of two satellites: Sentinel 1-A (Operated in April 2014) and Sentinel 1-B (Operated in April 2016). The satellites are in opposite polar sun-synchronous orbits at an altitude of 693 km, with a repeat cycle of 12 days, having 175 orbits. This results in a repeat frequency of 24 h at high latitudes and 3 days at the equator. The SAR system runs within C-band (5.407 GHz) in four modes: strip map (SM), interferometric wide swath (IW), extra-wide swath (EW), and wave (WV). For land surface, the IW swath mode is the most suitable model that acquires data with a 250 km swath at a 5 m by 20 m spatial resolution. It consists of four bands: vertical transmit and vertical receive (VV), vertical transmit and horizontal receive (VH), horizontal transmit and vertical receive (HV), and horizontal transmit and horizontal receive (HH). Each band can be used for flood detection. The cross-polarized band (VH or VV) has more noise and is less efficient for detecting floods than the co-polarized band (HH or VV; Manjusree et al., 2012). However, since the HH and HV bands were not available in our study area, we used the VV band for detecting floods.
The combination of satellite datasets used in this study included the flood extension from remote sensing and hydrology data. For the remote sensing data, we used the Sentinel-1 C band GRD collection provided by the Google Earth Engine (GEE) cloud computing platform. All scenes in this dataset have been pre-processed with the Sentinel-1 toolbox that consists of thermal noise removal, radiometric calibration, terrain correction, and converting backscatter coefficient to decibels (dB). In this study, we employed VV polarize band images that were taken on August 28, 2021;September 9, 2021;September 21, 2021;September 27, 2021;October 3, 2021;October 15, 2021;October 21, 2021;October 27, 2021;November 2, 2021;November 8, 2021;November 20, 2021;and November 26, 2021.
The analysis of the flood extent was divided into six steps. First was the selection of the VV band and then the next step was to speckle filter it using the focal method. The third step was to create a different image between the reference image (pre-flooded image taken on August 28, 2021) and the flooded image that was taken from September 9 to December 8, 2021. The fourth step was to identify any potentially flooded area from the different images. The fifth step was to perform model validation. In this study, we used the sentinel-2 image that was taken on October 25, 2021, to validate the flooded area on October 27, 2021, that was generated from the Sentinel-1 C band GRD collection. Finally, the last step was to interpret any potentially flooded area with the daily rainfall data.

Rainfall data
The daily rainfall (mm/day) was collected from the Department of Royal Rainmaking and Agricultural Aviation of Thailand from September 1, 2021, to December 1, 2021, from five stations (Rain Water, 2021). However, we mainly analyzed the data from three stations in the upstream part of the Chi catchment basin where the flow rate of water was problematic ( Figure 2). As seen in the graphs of Figure 2, the daily rainfall in the upstream station (Chaiyaphum) peaked for only 3 days with a 108 mm maximum. Whereas, the frequency of rainfall at the downstream stations (Khon Kaen and Maha Sarakham) was higher.
The upstream station (CHI003) is close to Chaiyaphum town. The daily rainfall at this station fluctuated between 1 and 24 September 2021 and then dramatically rose to 108 mm on September 25, which was caused by the Dianmu tropical depression. The tropical rainfall then decreased after September 25, 2021, but increased again on October 13, 2021, due to tropical cyclone Kompasu.
At the Khan Kaen station (CHI006), the daily rainfall (mm) averaged about 7.595 mm/day from 1 September to 16 October 2021. The highest daily rainfall was 42 mm on October 16, 2021, which was caused by tropical depression Dianmu and tropical cyclone Kompasu.
The last station was at Maha Sarakham irrigation dam (CHI007). The daily rainfall in this station averaged about 8.25 mm/day in the period of 1-23 September 2021. The highest daily rainfall in this period was 22.8 mm, which was caused by tropical depression Dianmu. It then sharply decreased but increased rapidly from 0 mm on October 12 to 32.8 mm on Frontiers in Ecology and Evolution 04 frontiersin.org October 13, 2021, due to Tropical cyclone Kompasu. Thereafter, the daily rainfall decreased.

Drainage network and existing roads
The upstream part of the Chi catchment basin is dominated by cuesta with an eastward dipping downslope. The drainage network is mainly concentrated in the western highland. Some networks originate from the mountain in the north with a longer drainage length than the western upstream origin (Pakoksung and Torsakul, 2022).

Accuracy and advantage of SAR
This case study assessed the accuracy of the model for detecting flooded areas in two periods: October 27 and November 20, 2021. As of October 27, 2021, the Sentinel-2 images taken on October 25 in sub-areas A2, A3, and A5 were analyzed for assessment accuracy. The overall accuracy, producer accuracy of the flooded area, user accuracy of the flooded area, and kappa coefficient were 99.264, 86.992, 93.274% and 0.8964, respectively. The kappa coefficient (0.8964) is extremely high (maximum of 1). Therefore, this model of classifying the flooded area in October 2021 was deemed to be dependable. Moreover, many non-flooded pixels were classified as non-flooded pixels.
As of November 20, 2021, this study used the Sentinel-2 image taken on November 22, 2021, in sub-area A3 for analysis of the accuracy. The overall accuracy, producer accuracy of the flooded area, and user accuracy of the flooded area were 90.909, 44.62, and 44.84%, respectively. The kappa coefficient was only 0.407% and many flooded pixels were classified as non-flooded pixels. The potentially flooded area and the Sentinel-2 image taken on November 22, 2021, showed many non-flooded pixels were classified as flooded pixels ( Figure 3A), which may be due to moist soil and surface water in agriculture zones. Areas close to non-flooded areas were also classified as flooded areas ( Figure 3B). In addition, non-flooded pixels that were classified as flood pixels first appeared in November 2021.
The classification of flooded pixels from this case study has some advantages. First, this method does not classify usual surface water areas, such as lakes, to be potentially flooded areas because the reference images were taken shortly before the flooding event. Second, the flooded pixels were classified using thresholding. Thus, it holds no bias from visual interpretation. Third, this SAR image-based method can classify potentially flooded areas in every weather situation because SAR can penetrate in all weathers, even under a high percentage of cloud cover.

Unusual water flow
The water mass had extended from sub-areas A1 and A2 of the upstream area to downstream of the Chi River area and joined with massive rainwater from sub-area A3. The accumulation of water from the upstream area started on September 21, 2021, while sub-areas A2 and A3 accumulated water from the tropical depression Dianmu from September 27 to October 3, 2021, respectively ( Figures 3C,D). The water level at A1 went down and the flooded area decreased since October 3, 2021. However, the flooded area in sub-areas A2 and A3 did not decrease in this period. After that, the potentially flooded area in sub-area A1 dramatically rose on October 21, 2021, which was caused by tropical cyclone Kompasu, while the potentially flooded area in sub-areas A2 and A3 in late-October 2021 did not change much from the beginning of October 2021.
As mentioned earlier, the erroneous allocation of non-flooded pixels to flooded pixels in November 2021 may be due to soil moisture and surface water in agriculture zones. However, many flooded pixels in October 2021 were classified as a non-flooded areas. Perhaps, the sum area of the potentially flooded area in September and October 2021 generated from the CDAT method was less than the flooded area.

Simplified way for reducing the flood period
We recognized that highways and local roads are constructed without proper planning. Although the purpose was to form a transportation network among communities, some roads were built across the stream in such a position as to obstruct the waterway, not only in the floodplain but also in the upstream part ( Figure 4A). Undoubtedly, the failure of the anthropogenic construction led to the road becoming a dam or barrier and resulted in a change in the direction of the water flow. This issue has become clear as being important but unfortunately has elicited little concern from responsible decision-makers in different government departments, due to a lack of government agency integration. However, the way to solve this is simple. In places where the water flow is problematic or obstructed by roads, the proper design of culverts underneath the road is recommended ( Figure 4B). This recommendation for reducing the flood period is likely simple, on the other hand, it is very significant and can be applied to other areas that have similar landforms. We also suggest that the government and local governors should be concerned about this case study and take into account the mitigation of not only the economic losses but also the reduce the cost of a non-necessary budget for constructing some structures.

Conclusion
The tropical depression and cyclones are the result of global climate variation and commonly produce a direct effect on the Khorat Plateau in mainland Southeast Asia. In the case of the 2021 flooding event at the Chi River floodplain, the water flow on the floodway was problematic. We recognized the disturbance from anthropogenic factors made it clear that the Frontiers in Ecology and Evolution 06 frontiersin.org channel planform of the Chi River has a high risk and was, more or less, modified after the November 2021 mega-flooding. After the overall terrain analysis using SAR satellites, we recommend a post-flooding investigation soon after the event, especially in places where infrastructures disturb the water floodway. We found that one of the causes of the unnatural flow of water in this upstream area of the Chi River catchment basin was anthropogenic constructions, especially the inappropriate design of local roads without culverts. Therefore, we provide the specific locations where culverts can be installed to alleviate future flooding in the coming years.

Data availability statement
The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author.

Author contributions
PW prepared GIS and conducted satellite analysis. NC edited and drafted the manuscript. SP prepared GIS materials and contributed to discussion. MC wrote and edited the manuscript, provided the figures, and contributed to discussion. All authors contributed to the article and approved the submitted version. Maps showing anthropogenic structures (roads) constructed across the water flow (A) with aerial pictures showing drowned main roads from Chaiyaphum (left; www.dailynews.co.th) and Khon Kaen (right picture; www.prachachat.net). (B) A close-up of the upstream part of the Chi River catchment basin shows the place where culverts are recommended (red triangle) to return a natural water flow during flooding.