Development of Flood Hazard Maps for India

Transcription

1 Development of Flood Hazard Maps for India Srinivas Kondapalli 1, Ramesh Jakka 2 Samyadeep Ghosh 3 Raj Kumar N 4 Ismail Mulla 5 1 Manager (Flood Research), Verisk Analytics India Pvt. Ltd, d.b.a AIR Worldwide 2 Manager (GIS), Verisk Analytics India Pvt. Ltd, d.b.a AIR Worldwide 3 Senior Research Engineer, Verisk Analytics India Pvt. Ltd, d.b.a AIR Worldwide 4 Research Engineer, Verisk Analytics India Pvt. Ltd, d.b.a AIR Worldwide 5 Lead Engineer (GIS), Verisk Analytics India Pvt. Ltd, d.b.a AIR Worldwide 2 nd Floor, Block C, Divyasree Omega, Kondapur, Hyderabad, Telangana aper should not be Words = 7/8 Pages) Abstract About the Author Flood is a pervasive global peril, and effective flood risk assessment and management is one of the top concerns in the insurance and reinsurance industry. As exposure continue to grow in flood-prone areas, the need for having comprehensive flood hazard maps increases. To enable a understanding of the threat posed by complex river systems in India, AIR Worldwide (AIR) has developed probabilistic flood hazard maps for different probability of Mr. Srinivas Kondapalli exceedances (return periods) that provide spatial extents as well as Srinivas Kondapalli is a member of AIR s Flood flood depths. The flood maps are based on the flood vulnerability of Research team. Srinivas has 8+ years of working experience in development of catastrophe risk India s river network and incorporate the physical properties of models worldwide. He contributed to the India s terrain along with a thorough analysis of the frequency of development of AIR s inland flood models for flooding using annual peak flows in the rivers. The hazard maps United Kingdom, Germany, United States and Japan include all eleven major river basins that in India, as well as areas and flood hazard maps for China, Thailand, Brazil, outside India that contribute to flooding within India. Within these Canada and India. He has worked on the project for basins, the maps cover variable density river network systems the enhancement of DTM quality for future models. including every stream draining 10, 100 and 500 Sq.km or more for He has also worked on the development of AIR s all river basins covering a total drainage area of approximately 5.1 cyclone model for India. Srinivas earned M.S. in Water Science Engineering from UNESCO-IHE, Delft, million Sq.km. A physically based hydraulic model employing 1- Netherlands and MBA specializing in finance from dimensional kinematic wave approximation is used to translate flow IGNOU, New Delhi. Besides this, he also conducted quantiles to water surface profile along river center lines. AIR s flood post-disasters surveys for Cyclone HudHud (2014) mapping tool, based on 2-dimensional diffusive and kinematic wave and Chennai Floods (2015). equations, is then used for generating the spatial flood hazard maps. These maps are validated with historical event flood inundation E mail ID: skondapalli@air-worldwide.com extent maps. The flood hazard maps are available as geospatial layers Contact: for use in AIR s patented product Touchstone, to enable understanding of the threat posed by complex river systems in India. Page 1 of 9

2 Introduction Floods are one of the most common natural disasters that could pose a significant threat to millions of people around the world. 9% of India`s population are acutely exposed to floods (Maplecroft, 2015). Floods, being a recurrent phenomenon cause huge losses to lives, livelihoods, infrastructure, public utilities and properties every year. On an average, 75 lakh hectares of land is affected and 1800 crores of public properties, houses and crops are destroyed due to floods every year (NDMA, 2008). Recently, many major cities like Delhi, Bangalore, Mumbai and Chennai have witnessed heavy economic losses due to floods. On July 2005 Mumbai, India's largest city by population, have witnessed heavy flooding, which brought the city to a standstill for several days causing huge economic losses. Chennai experienced severe floods in 2015 resulting from heavy rainfall during the annual northeast monsoon in November December which overflowed two rivers Adyar and Cooum. Massive flooding along the Krishna River in 2009 wreaked havoc in districts of Kurnool, Guntur, Nalgonda and Vijayawada city of Andhra Pradesh. Floods along the Ganges disrupted the state of Uttarakhand in 2013 which caused many casualties in terms of lives and property. Assam, Odisha, West Bengal and Meghalaya are among other areas which experience recurrent flooding. These increasing risks in the flood prone areas indicate the dire need for the development of flood hazard maps. Many studies regarding flood hazard has been carried out in India in the past, however, no such study or dataset exist which covers the entire country and can be used to mitigate the risk effectively by the insurance and re-insurance industry in general. Hence, AIR has carried out this research considering the specific requirements of the industry. This paper focuses on the development of comprehensive flood hazard maps that capture the complete risk of inland floods across all river basins of India corresponding to the 25-, 50-, 100-, 200-, 250-, and 500-year return periods. The AIR researchers collected all required datasets such as (i) Shuttle Radar Topography Mission (SRTM) Digital Elevation Model (DEM) at 90m resolution (ii) Annual Peak flow information of stream gauges throughout India and its contributing catchments from the Government of India and Nepal and (iii) Land Use Land Cover (LULC) data from the NASA Moderate Resolution Imaging Spectro radiometer. All these datasets were processed in ESRI ArcGIS and multiple geospatial analyses were used for pre and post processing of data. Data and Approach India has eleven major river basins. The largest river basin is the Ganga-Brahmaputra-Meghna river system, with a catchment area of about 11.0 lakh sq.km. The other major river basins with catchment area more than 1.0 lakh sq.km are rivers Indus, Mahanadi, Godavari and Krishna. India shares its international boundaries with the neighboring countries such as Afghanistan, Pakistan, Nepal, Bhutan, China, Myanmar and Bangladesh. Three of major rivers flowing through India also share its course with Page 2 of 9

3 these countries. In order to effectively model the actual fluvial characteristics, AIR has taken into account the upstream and downstream effects originating from these countries. The vital datasets required to develop flood maps are hydrological data in the form of historical peak flows for stream gauges and terrain data in the form of digital elevation models, physiographic data and land use land cover data. DEM Enhancements The most critical aspect of flood hazard mapping is a reliable DEM, which provides the physical geometry of the floodplain and the river channel above the surface of the water. AIR has considered SRTM 90m resolution DEM developed by NASA which effectively impacts water flow by capturing the morphology of the landscape. The 90m resolution is detailed enough to capture the flood hazard along major rivers and areas of concentrated exposure. However, to provide a sound assessment of water levels, AIR researchers ensured that the data should reflect the bare-earth terrain. The SRTM 90m elevation dataset has artifacts due to vegetation or urban structures (Baugh, et al., 2013). Therefore, the DEM has been post-processed to remove such artifacts due to vegetation or urban structures to resemble bare earth conditions. To correct the data in areas of dense vegetation, AIR employs a streamlined approach that was closely based on the procedure suggested by Baugh, et al The procedure uses a global vegetation height dataset at a 1 km resolution (Simard, et al., 2011). Analyses indicated that the urban areas always included some bare ground cells; therefore, AIR employed a smoothing process, which is computationally efficient and scientifically sound. The DEM is then hydraulically enforced to allow water to flow naturally along the surface by removing physiographical features in the DEM representing bridges, weirs, flood gates or vegetation along the river. Variable Density River Network The river networks are generated with variable density, based on the contributing catchment area. Higher density river networks are generated in the proximity of populated places; medium and low density networks are created in other areas. A variable density of a minimum drainage catchment size of 10, 100 and 500 Sq.km is maintained for high, medium and low density networks respectively. River network has been derived from the SRTM DEM raster dataset with 90m resolution. The sinks in the DEM was filled in order to ensure the proper delineation of streams. The flow direction raster (FDR) is then generated to calculate the cell to cell surface water flow direction. Using the FDR, flow accumulation grid is generated based on the cell weightage. i.e., the number of cells from which water flows to a downstream cell. Using the weightage of each cell the variable density networks are derived. Three different networks are delineated in such a way that every urban area, within a buffer of 10km has a 10 sq.km river network, similarly, 50km buffer is applied and this area has a 100 sq.km river network, and the rest of the area has a 500sq.km network. In this process the flow lines considered in Page 3 of 9

4 500sq.km, 100sq. km and 10sq.km networks will have drainage area greater than or equal to 500sq.km, 100sq.km and 10sq.km respectively. These three different networks are merged together in such a way that denser flowlines are present within 10km from the urban regions and the number of flowlines decreases while moving farther from urban area. ArcGIS hydrology tools available in Spatial Analyst extension were used heavily for the delineation of the stream networks. Figure 1 provides a preview on the variable density network in the populated places and in the other areas. All urban areas with a population of more than 1 lakh have been considered for this variable density network creation. Fig: 1 - Variable Density River Network The total river network extends over 0.32 million km and drains an area of around 5.1 million sq.km. The modeled domain contains over 14,293 distinct river catchments, with over 11,442 catchments located partially or entirely in India. The river network within India s borders extends over 0.24 million km and drains an area of approximately 3.2 million sq.km. Cross Sections To perform hydraulic calculations for a given river, the model uses river cross sections to include the details of the terrain along the river centerline and across the floodplain of each river link. To capture enough details, AIR researchers created cross sections perpendicular to the river channel spaced at 500m intervals along the length of all river segments. This interval length is small enough to capture changes in the geometry and surface roughness of the floodplain, but large enough to be efficient from a computational standpoint. As the cross sections were generated, they were placed Fig: 2 - River Cross Sections so that their length spanned an area with an elevation of 10m above the regular water surface elevation, thereby capturing both the main river channel and the surrounding floodplain. Along each cross section, the Page 4 of 9

5 shape of the main channel, including the floodplain, was extracted from the DTM and the surface roughness was extracted from the LULC data. A sample of cross sections across a few river segments is shown in Figure 2. The model incorporates over 504,847 river cross sections spaced approximately every 500m along the modeled river network. Hydrological Data Annual peak flows and its corresponding water levels data for over 750 river gauging stations was collected from the Central Water Commission, Government of India and also from the online resources of the Water Resources Information System (WRIS) and 101 stream gauge stations data collected from the Department of Hydrology and Meteorology, Government of Nepal. Comprehensive data quality checks were carried out before utilizing the data into the model. Annual peak flow data are checked for inconsistencies and potential outliers were removed. Peak flows were also checked for change in trend and homogeneity. Effects for potential regulations were removed by screening the data affected by regulation. Stream gauges with significant number of years were only Fig: 3 Spatial Distribution of Stream gauges considered for the model. The spatial distributions of stream across India gauges are shown in the Figure 3. Flood Frequency Analyses In order to estimate the desired flood flow quantiles for each river segment, AIR researchers implemented a two-step process. In the first step, flow quantiles are estimated for river segments for which sufficient observed annual peak flow information is available. This is called at-site flood frequency analysis and is typically carried out by fitting suitable statistical frequency distributions to the available annual peak flow data. In the second step, flood quantiles for all ungauged river segments are estimated by regionalizing the estimated at-site flood flow quantiles. This regionalization process involves consideration of relevant physical properties of catchments that explain variance in the estimated flood quantiles. Based on the regionalized frequency distribution parameters, the flood flow quantiles corresponding to 25-, 50-, 100-, 200-, 250-, and 500-year return periods (i.e., corresponding to an exceedance probability of 4%, 2%, 1%, 0.5%, 0.4%, and 0.2%, respectively) for all catchments in the entire river network were then estimated. AIR researchers used Jack-knife and Monte Carlo methods to cross validate the performance of the regional flood frequency analysis. In the Jack-knife resampling method, performance Page 5 of 9

6 of the model is checked by systematically leaving out one observation from the dataset. Whereas, in the Monte Carlo validation method data is randomly split into training and testing sets, with which model performance is assessed. The results of Jack-knife and Monte Carlo validation methods are displayed in the Figure 4. Finally, these estimated flood quantiles are used to prepare the corresponding return period flood hazard maps. (a) (b) (c) Fig: 4- (a) Jack-knife Cross-Validation for 1 in 100 year return period; (b) Monte Carlo Training set for 1 in 100 year return period; (c) Monte Carlo Validation (Testing set) for 1 in 100 year return period The Hydraulic Model A physically based hydraulic modeling process (that accounts for factors like terrain, soil type, and land use/land cover) then translates flows to water surface elevation at each river cross section before determining the flood extents. The AIR, in-house hydraulic model is based on a computational algorithm similar to the steady state mode of HEC-RAS, a widelyused hydraulic model developed by the United States Army Corps of Engineers (USACE). In AIR s hydraulic model, a 1-dimensional energy equation is solved iteratively to compute a steady state water surface Fig: 5 - Calibration performance of hydraulic model profile along the dendritic (branching) river network. The hydraulic parameters of a channel cross section are estimated by using subdivisions, within each cross section, in which the flow velocity is assumed to be uniformly distributed. By using these subdivisions, the hydraulic model provides reliable flow calculations for the main channel as well as for floodplain areas. The subdivision units are based on changes in elevation obtained from the DEM, and the roughness obtained from the LULC data. The model allows roughness coefficients to vary within a cross section, by taking advantage of the high resolution LULC data. Rating curves are constructed using the Page 6 of 9

7 stage-discharge data from river gauging stations, which would ideally be well-distributed throughout the river network. The hydraulic model is calibrated using these rating curves, the results of which are shown in Figure 5. In order to generate the flood extent maps, the model exports the flood elevation data into AIR s flood mapping tool, which incorporates a multi-source point expansion methodology to provide realistic flood extents. This methodology applies 2- dimensional diffusive and kinematic wave equations from given points on the water surface while incorporating the dataset from the DTM. This process delineates the flood extent twodimensionally across the entire floodplain creating a continuous water surface, which is represented at the same Fig: 6 - Flood extent map along Yamuna River near New Delhi resolution as the underlying DTM. To provide multiple views of the flood hazard, AIR generated 25-, 50, 100-, 200-, 250-, and 500-year return period flood extent maps (i.e., with exceedance probabilities of 4%, 2%, 1%, 0.5%, 0.4%, and 0.2%, respectively). Figure 6 shows the extent of flooding for 25-, 100-, and 500- year return period across the river Yamuna near New Delhi. Validation of Flood Hazard Maps The AIR flood hazard maps do not account for the flood protection structures such as levees, but instead provide a complete view of hazard behind such protection structures or levees. In order to validate the modeled flood extents, AIR collected information on several historical events, such as inundation extent maps, depths, peak discharges, reported return period etc. from the local or international agencies. AIR compared the results of several return period maps for a number of regions to validate against the analyses and results from studies for those regions. Such comparative analyses provide additional basis of validation and soundness of the AIR flood maps. Examples of validation of modeled extents for the fairly recent flood events for Mumbai, Chennai and Kurnool floods are given below Mumbai Floods (a) AIRPORT (Protected) Fig: 7 - (a) 2005 Mumbai floods event foot print (Gupta, 2007); (b) (b) AIR 100 year return period flood map overlaid with blue 2005 monsoon caused extreme floods with a return period of more than 100 years (Hallegatte, et al., 2010) in the state of Maharashtra and particularly in Mumbai. Figure 7(a) shows the inundation Page 7 of 9

8 extents in and around Mumbai and also along the Mithi River on July 27, Figure 7(b) compares the same image with AIR 100-year flood hazard map by overlaying on the flood extent map Chennai Floods During the course of northeast monsoon in the months of November and December 2015, three low pressure systems developed consecutively on 9th, 15th November and 2nd December resulting in an approximate 100 year rainfall event in parts of Tamil Nadu, Andhra Pradesh, Puducherry and Chennai (Mujumdar, et al., 2016). Due to heavy rainfall and release of excess water from Chembarambakkam dam to the river Adyar (South of Chennai), a widespread flooding occurred in the many parts of Chennai. Figure 8(a) shows the RISAT satellite image of flood extents along the Adyar and Couum rivers and Figure 8(b) shows the same image with an overlay of AIR 100 year flood hazard map extents Kurnool Floods A massive flood worse in the last 100 years Tungabhadra river, which is a tributary of river Krishna, wreaked havoc in Kurnool district of Andhra Pradesh (Unified Response Strategy, 2009). Around 478 villages were severely flooded. The Figure 9(a) shows an event foot print of the event and Figure 9(b) previews the AIR 100 year return period map overlaid on the event map. (a) Fig: 8 - (a) 2015 Chennai floods event foot print (NRSC, Bhuvan RISAT); (b) AIR 100 year return period flood map overlaid with orange color (b) (a) (b) Fig: 9 - (a) 2009 Kurnool Floods event foot print (APSRAC); (b) AIR 100 year return period map overlaid with blue color Page 8 of 9

9 Conclusions There is significant flood risk in India which has been mostly unassessed and uncovered by insurance industry until now These AIR flood maps are the industry first to provide countrywide flood maps and flood risk assessment based on comprehensive flood data collation, analyses, and modeling efforts These flood maps are highly user-friendly and in electronic form that can be easily harnessed by various stakeholders in flood risk assessment and management arena government institutions, town planners, engineering companies, and flood insurance industry among others. These flood hazard maps are available as a geospatial layers for use in AIR s patented product of Touchstone, to enable us to understand the threat posed by complex river systems. The maps can be used for effective management of flood risk accumulations, for determining whether a risk meets underwriting guidelines, and for developing effective portfolio management and risk transfer strategies. References Baugh, A., D.Paul, B., Guy, S. & Mark, A., SRTM vegetation removal and hydrodynamic modeling accuracy. Water Resources Research. Gupta, K., Urban flood resilience planning and management and lessons for the future: a case study of Mumbai, India. Urban Water Journal. Maplecroft, V., Natural Hazard Risk Atlas, : Verisk Maplecroft. Mujumdar, P. et al., Chennai Floods 2015, : Interdiscipilinary Centre for Water Research, IISC Bangalore. NDMA, Management of Floods, : National Diaster Management Authority, India. Simard, M., Pinto, N., B.Fisher, J. & Baccini, A., Mapping forest canopy height globally with spaceborne lidar. Journal of Geophysical Research, p Unified Response Strategy, Situation Report Andhra Pradesh Floods Hallegatte, S., et al. (2010), "Flood Risks, Climate Change Impacts and Adaptation Benefits in Mumbai: An Initial Assessment of Socio-Economic Consequences of Present and Climate Change Induced Flood Risks and of Possible Adaptation Options", OECD Environment Working Papers, No. 27, OECD Publishing, Paris Page 9 of 9