Investigation of Role of Retention Storage in Tanks (Small Water Bodies) on Future Urban Flooding: A Case Study of Chennai City, India

: The Adyar River ﬂowing through Chennai Metropolitan Area (CMA) in Southern India functions as a surplus course of upstream water bodies that are locally known as tanks. During northeast monsoons, the river frequently ﬂoods the adjoining city areas. In this study, the impact of dredging and disappearance of tanks on ﬂooding in CMA is analyzed under historical, urbanization, and extreme rainfall scenarios utilizing an urbanization-hydrologic-hydraulic modelling framework. The simulated scenarios highlight the importance of the tanks as a ﬂood control measure for CMA. The major conclusions are (a) dredging the tanks uniformly by 2 m can compensate the increase in ﬂooding due to urbanization by 2050 for 1 in 50-year rainfalls and, (b) for disappearance of tanks, 1 in 50-year rainfall can inundate the city akin to 1 in 100-year rainfalls. The study can be useful for making informed decisions on dredging the tanks, land use planning, and ﬂood control measures for the CMA.


Introduction
Earth is presently subjected to more frequent natural disasters than ever [1,2]. In the past few decades, a huge fraction of mortalities (nearly 95%) linked with extreme events has been observed in developing nations [1]. Developing nations not only experience higher death rates but also greater economic impacts and are more vulnerable to natural hazards than developed countries [1]. Urban areas in these countries are becoming increasingly vulnerable to frequent and intensive floods [3][4][5] for a number of reasons: (a) the rise in population due to migration from countryside and other cities [6,7], resulting in extensive unplanned developmental activities and drastic changes in land use; (b) sea-level rise, storm surge, and tropical cyclones under changing climate [5,[8][9][10]; and (c) inefficient flood control measures [11][12][13]. Heavy rainfall over the Adyar basin in the state of Tamil Nadu, India is a recurring phenomenon during the monsoon seasons due to deep depressions and cyclones originating over the Bay of Bengal. The coastal Chennai city in Adyar basin suffers from frequent inundation during

Study Area
The Adyar basin, chosen as the study site, occupies an area of about 724 km 2 ( Figure 2). As mentioned earlier, the Adyar River originates as a surplus course from water bodies in the upper part of the basin and flows through the city before falling into Bay of Bengal. The total length of the river is about 43 km. The river remains dry for most of the year with flows largely occurring during the North-East (NE) monsoon (October to December). Chennai International Airport is located by the Adyar River, and its secondary runway is across the river. Many industries, business centers, hospitals, schools, and residential colonies are located on either side of the river. The CMA being a business hub has witnessed steep growth in the manufacturing, health care, retail, and IT sectors in the past few decades [65]. The city is the fourth largest metro in India, and it has become a center for tremendous economic activities. On the downside, the city is experiencing rapid and unplanned expansion due to migration of population from various parts of the country. This has resulted in significant increase in impervious areas and encroachment of water bodies, riverbanks, drains, and marsh lands. Chennai being a coastal city, lying on the thermal equator, has a tropical hot and humid climate. The average annual rainfall of CMA is around 1400 mm. During the NE monsoon, the city receives most of its annual rainfall (approximately 800 mm). Therefore, the city is in a critical position to harvest this seasonal rainfall using its water bodies and reservoirs to meet the increasing water demand. As mentioned earlier, the deep depressions and cyclones that develop over the Bay of Bengal during the NE monsoon period result in frequent flooding of the city. A 15-60 m wide Buckingham canal and some major and minor drains aid in draining the flood waters in the city into the sea. However, the sewage outfalls, bridges, and metro stations disrupt the flow in Buckingham canal and its width is restricted at many places to 10 m [13]. The Mambalam drain that lies in the center of the city draining storm water into the river is also restricted at many places due to sewage outfalls and urbanization. The upper portion of the Mambalam drain is largely the remnants of a disappeared water body. According to [13], the places in and around the Mambalam drain, namely T. Nagar, Nandhanam, etc., were the 2005 flood hotspots. Additionally, the urbanized R.A. Puram area located near the Adyar creek is susceptible to flooding due to the low-lying terrain.

Study Area
The Adyar basin, chosen as the study site, occupies an area of about 724 km 2 ( Figure 2). As mentioned earlier, the Adyar River originates as a surplus course from water bodies in the upper part of the basin and flows through the city before falling into Bay of Bengal. The total length of the river is about 43 km. The river remains dry for most of the year with flows largely occurring during the North-East (NE) monsoon (October to December). Chennai International Airport is located by the Adyar River, and its secondary runway is across the river. Many industries, business centers, hospitals, schools, and residential colonies are located on either side of the river. The CMA being a business hub has witnessed steep growth in the manufacturing, health care, retail, and IT sectors in the past few decades [65]. The city is the fourth largest metro in India, and it has become a center for tremendous economic activities. On the downside, the city is experiencing rapid and unplanned expansion due to migration of population from various parts of the country. This has resulted in significant increase in impervious areas and encroachment of water bodies, riverbanks, drains, and marsh lands. Chennai being a coastal city, lying on the thermal equator, has a tropical hot and humid climate. The average annual rainfall of CMA is around 1400 mm. During the NE monsoon, the city receives most of its annual rainfall (approximately 800 mm). Therefore, the city is in a critical position to harvest this seasonal rainfall using its water bodies and reservoirs to meet the increasing water demand. As mentioned earlier, the deep depressions and cyclones that develop over the Bay of Bengal during the NE monsoon period result in frequent flooding of the city. A 15-60 m wide Buckingham canal and some major and minor drains aid in draining the flood waters in the city into the sea. However, the sewage outfalls, bridges, and metro stations disrupt the flow in Buckingham canal and its width is restricted at many places to 10 m [13]. The Mambalam drain that lies in the center of the city draining storm water into the river is also restricted at many places due to sewage outfalls and urbanization. The upper portion of the Mambalam drain is largely the remnants of a disappeared water body. According to [13], the places in and around the Mambalam drain, namely T. Nagar, Nandhanam, etc., were the 2005 flood hotspots. Additionally, the urbanized R.A. Puram area located near the Adyar creek is susceptible to flooding due to the low-lying terrain.

Scenarios Definition
The initial storage in the tanks at the onset of the extreme rainfall event dictates the role of tanks in flood moderation. Based on volume estimates of the tanks and simulations of hydrological model (refer to Section 2.5 for details on the model set up), it is found that a rainfall between 75 mm and 100 mm is good enough to fill the tanks at the most by 50% and 75% of its capacity, respectively, in the rural portion of the watershed. The NE monsoon rainfall (NEMR) exhibits both spatial and temporal vagaries in the basin. A study by [66] presented preliminary results that suggest considerable intra-seasonal variability of NEMR. It has to be noted that the occurrence of extreme rainfall events is limited. This in turn makes it difficult to analyze the wet spells preceding the extreme events. Consequently, with a view to address the uncertainty in assuming the initial storage of the tanks, 50% and 75% initial storages in the tanks were considered for various historical, urban sprawl, and climate change scenarios. The study considers not only the existing volume of tanks but also the increased volumes of dredged tanks for various scenarios. As mentioned earlier, based on the proposal of the WRD of Tamil Nadu [41][42][43], the dredging of the tanks by 1-m and 2-m scenarios are considered in the study. The scenarios for disappearance of tanks are included only in future simulations as the tanks might lose the storage completely with time due to poor maintenance and management. Tables 1 and 2 give a detailed description of the scenarios that are used in analyzing the effectiveness of the proposed flood management strategies. The notations used for depicting the scenarios comprise two parts: one to denote the Rainfall/LULC and the other to indicate tank-related conditions. The two parts listed in Table 1 in the last two columns can be combined to give various scenarios that are used in this study. For example, US100-RCP4.5-TP2m-F50 represents the scenario with urban sprawl corresponding to the year 2050 and 100-year 24-h duration return-period rainfall derived from CNRM-RCP 4.5 for the period (2041-2071). All 163 tanks presently available are dredged by 2 m uniformly throughout the water spread area, the dredged tanks being filled up by 50% at the beginning of the simulation.

Scenarios Definition
The initial storage in the tanks at the onset of the extreme rainfall event dictates the role of tanks in flood moderation. Based on volume estimates of the tanks and simulations of hydrological model (refer to Section 2.5 for details on the model set up), it is found that a rainfall between 75 mm and 100 mm is good enough to fill the tanks at the most by 50% and 75% of its capacity, respectively, in the rural portion of the watershed. The NE monsoon rainfall (NEMR) exhibits both spatial and temporal vagaries in the basin. A study by [66] presented preliminary results that suggest considerable intra-seasonal variability of NEMR. It has to be noted that the occurrence of extreme rainfall events is limited. This in turn makes it difficult to analyze the wet spells preceding the extreme events. Consequently, with a view to address the uncertainty in assuming the initial storage of the tanks, 50% and 75% initial storages in the tanks were considered for various historical, urban sprawl, and climate change scenarios. The study considers not only the existing volume of tanks but also the increased volumes of dredged tanks for various scenarios. As mentioned earlier, based on the proposal of the WRD of Tamil Nadu [41][42][43], the dredging of the tanks by 1-m and 2-m scenarios are considered in the study. The scenarios for disappearance of tanks are included only in future simulations as the tanks might lose the storage completely with time due to poor maintenance and management. Tables 1 and 2 give a detailed description of the scenarios that are used in analyzing the effectiveness of the proposed flood management strategies. The notations used for depicting the scenarios comprise two parts: one to denote the Rainfall/LULC and the other to indicate tank-related conditions. The two parts listed in Table 1 in the last two columns can be combined to give various scenarios that are used in this study. For example, US100-RCP4.5-TP2m-F50 represents the scenario with urban sprawl corresponding to the year 2050 and 100-year 24-h duration return-period rainfall derived from CNRM-RCP 4.5 for the period (2041-2071). All 163 tanks presently available are dredged by 2 m uniformly throughout the water spread area, the dredged tanks being filled up by 50% at the beginning of the simulation.

TP
Existing volumes of the 163 tanks that are considered for the study are included in the hydrologic model. These scenarios represent the baseline conditions or the tanks' present conditions.

TP1m
The 163 tanks are deepened by 1 m uniformly throughout the water spread area, and the increased volumes are included in the hydrologic model.

TP2m
The 163 tanks are deepened by 2 m uniformly throughout the water spread area, and the increased volumes are included in the hydrologic model. F50 Prestorage of 50% of the total storage capacity of the tanks.

F75
Prestorage of 75% of the total storage capacity of the tanks.

TD
This scenario considers the absence of tanks or tank disappearance in the study area. In other words, the depression storage is considered 0 in the study area.

Frequency Analysis of NEX-GDDP Data
The rainfall for the midcentury (2041-2071) period in this study uses daily time series of rainfall obtained from NASA Earth Exchange Global Daily Downscaled Projections (NEX-GDDP) datasets. The NEX-GDDP dataset includes downscaled climate scenarios obtained from General Circulation Model (GCM) runs that were carried out under the Coupled Model Intercomparison Project Phase 5 (CMIP5) for two greenhouse gas emission scenarios called Representative Concentration Pathways (RCPs) 4.5 and 8.5 at a spatial resolution of 0.25 • (approximately 25 km × 25 km). Studies by [67,68] pointed out that CNRM-CM5.0 (Centre National de Recherches Meteorologiques Coupled Global Climate Model, Version 5) and GFDL-CM3.0 (Geophysical Fluid Dynamics Laboratory Climate Model, Version 5) CMIP 5 models can realistically simulate the June-September summer monsoon rainfall over India. However, to limit the length of the paper, midcentury (2041-2071) rainfall data corresponding only to the CNRM-CM5.0 model is used. The study area extends across four CNRM-CM5.0 grids that are represented in this paper as grids 1, 2, 3, and 4 for convenience. Grids 13.125 • N), respectively. The annual maximum daily rainfall series is obtained for the period 1970-2071 for all CNRM-CM5.0 model grids. Gumbel (extreme value type I) distribution is then used to estimate the rainfall depths corresponding to various recurrence intervals/return periods. In an attempt to correct the bias in CNRM-CM5.0 data, the frequency analysis is repeated for the Indian Meteorological Department (IMD) grid data available at the resolution 0.25 • × 0.25 • . IMD data has a similar resolution to that of the CNRM grids for the baseline period . For each return period and grid, the ratio of the corresponding IMD to CNRM-CM5.0 rainfall depths is obtained for the baseline period. The rainfall depth ratios for various return periods are then multiplied with the corresponding CNRM-CM5.0 midcentury rainfall depths to get the bias-adjusted values. Figure 3 shows the relationship between bias-adjusted rainfall depths for different return periods corresponding to CNRM-CM5.0 datasets for the RCP 4.5 and 8.5 scenarios.

Urban Sprawl Prediction
In this study, the SLEUTH model [69][70][71] is used for urban sprawl prediction of Chennai city by 2050. "SLEUTH" is an acronym based on the inputs used in the model: slope, land use, exclusion, urban, transport, and hill shade. It is a probabilistic cellular automata model, which is widely used for urban growth prediction. The model requires urban extents corresponding to four years, while two land use maps are required: one at the beginning and the other at the end of the calibration period. In this study, these maps are prepared by performing unsupervised classification on Landsat imageries belonging to 28

Urban Sprawl Prediction
In this study, the SLEUTH model [69][70][71] is used for urban sprawl prediction of Chennai city by 2050. "SLEUTH" is an acronym based on the inputs used in the model: slope, land use, exclusion, urban, transport, and hill shade. It is a probabilistic cellular automata model, which is widely used for urban growth prediction. The model requires urban extents corresponding to four years, while two land use maps are required: one at the beginning and the other at the end of the calibration period. In this study, these maps are prepared by performing unsupervised classification on Landsat imageries belonging to 28  The basic processing units of the model are termed cells. The cells get new states for each time step based on the transition rules applied to the previous states of the cell and its neighbors. The transition rules that influence the probability of urbanization are determined based on growth parameters that are to be calibrated [72,73]. SLEUTH uses brute force calibration, and in order to mimic the random processes associated with urban growth, Monte-Carlo (MC) simulations are used [74]. The results of the calibration run can be interpreted with the help of fit statistics, such as compare metric and the Lee-Sallee index. The compare statistics is the ratio of the number of modelled populations of urban pixels to the observed populations for the final control year [75]. The Lee-Sallee shape index is the ratio of intersection and union of the simulated and observed urban areas, averaged over all control years [72,76]. Many studies have used Lee-Salle metric to narrow down the coefficient space [73,[77][78][79][80][81]. Apart from the aforementioned spatial metrics, the SLEUTH model also estimates non-spatial pattern metrics, namely clusters and edges metrics. Clusters and edges metrics are ordinary least-squares regression scores for the modelled and actual urban clusters and edge pixels, respectively [75]. For the Adyar basin, the final calibration phase resulted in values of 0.6, 0.61, 0.88, and 0.7 for the compare, Lee-Sallee, clusters, and edges metrics, respectively. It can be found from literature that the obtained values of the metrics are within the ranges as reported in similar studies [77][78][79]81]. The calibrated values of the diffusion, breed, spread, slope, and road coefficients are 68, 67, 25, 25, and 44, respectively. In the prediction mode, the urban extent is simulated for every year in the period 2000-2050. The predicted LULC maps for the years 2005, 2015, and 2050 are shown in Figure 4. Table 3 presents the distribution of predicted LULC classes for the years 2005, 2015, and 2050. It can be seen from Table 3 that % of urban pixels increased from 9.6% to 45.9% in the period 2005-2050.

Hydrological Modelling
The HEC-HMS (Hydrologic Engineering Center-Hydrologic Modeling System) model [82] developed by U.S. Army Corps of Engineers is used in this study to generate flood hydrographs at the inlet of the hydraulic domain for the scenarios in Table 1. HEC-HMS requires DEM, soil type, LULC, and rainfall data as inputs. An SRTM DEM of 30 m × 30 m resolution is used for delineating the sub-basins of the study area ( Figure 2). The Curve Number (CN) method is used to estimate the effective rainfall [83]. The CN method requires soil and LULC information as inputs. The European Soil Data Centre (ESDAC) [84] soil repository and LULC maps (refer to Section 2.4) for 2005, 2015, and 2050 are used for this purpose. The meteorological model includes four rain gauges for the 2015 flood and five rain gauges for the 2005 flood that are spread across the basin. Inverse distance method is used to interpolate the precipitation data throughout the basin. For the 2015 flood event, the hourly rainfall data is available from the rain gauges [36,37], while for the 2005 flood event, the daily accumulated rainfall that was reported in [30] is distributed into 3-hourly data using the TRMM (Tropical Rainfall Measuring Mission) data. For the future climate change scenario, the adjusted extreme rainfall data for the period 2041-2071 is obtained from CNRM-CM5.0 model (refer to Section 2.3). Similarly, the adjusted mid-century 24-h rainfall values corresponding to 50-and 100-year return periods are distributed into 3-hourly data using the TRMM data of the 2005 and 2015 flood events, respectively. Kinematic wave routing is chosen to route the channel flow through the basin.
Around 163 tanks are identified and delineated in the upper part of the Adyar basin with the aid of Google Earth image repository. The LIDAR DEM of 2 m × 2 m spatial resolution is used to calculate water spread area and volume of the tanks. The minimum, mean, and maximum water spread areas of the tanks are estimated to be 0.003, 0.41, and 5.25 km 2 , respectively. The sum of the volume of tanks falling under each sub-basin is entered as depression storage in the hydrologic model. For the "TP1m" and "TP2m" dredging scenarios, the DEM values within the water spread areas of the tanks are reduced by 1 m and 2 m, respectively; the corresponding increased volumes are summed up sub-basin-wise; and depression storages are estimated accordingly. Therefore, different dredging scenarios are defined by varying the depression storage for each sub-basin. As initial water levels in the 163 tanks are not known for the 2005 and 2015 rainfall events, the percentage of initial storage in the tanks are varied as 50% and 75% of the tanks' original volumes in the HEC-HMS setup, which are represented as "F50" and "F75" scenarios, respectively, in Table 1. Eventually, the runoff hydrographs that are upstream boundary conditions for the hydraulic model are generated for all the scenarios. While for the "US" scenarios, the CN-related parameters are updated by using the simulated urban sprawl by 2050.

Validation of Hydrological Model
The hydrological model that is set up for the historical 2005 and 2015 flood events are validated separately. As the Adyar basin is not well instrumented, the observed data is extremely limited. Hence, the observed flow data for calibration is limited to the reservoir inflow data estimated from the measured release over the spillways and change in observed storage. Moreover, there was a significant downpour before the isolated peaks in 2005 and 2015 rainfall events. Hence, to be on the conservative side and to account for the water released from the tanks, 75% initial storage in the tanks for historical

Validation of Hydrological Model
The hydrological model that is set up for the historical 2005 and 2015 flood events are validated separately. As the Adyar basin is not well instrumented, the observed data is extremely limited. Hence, the observed flow data for calibration is limited to the reservoir inflow data estimated from the measured release over the spillways and change in observed storage. Moreover, there was a significant downpour before the isolated peaks in 2005 and 2015 rainfall events. Hence, to be on the conservative side and to account for the water released from the tanks, 75% initial storage in the tanks for historical 2005

Hydraulic Modelling
The study [30] simulated different return period floods with 1976 and 2005 land use patterns using the HEC-RAS (Hydrologic Engineering Center-River Analysis System) 1D model developed by U. S. Army Corps of Engineers to highlight the negative impact of urbanization on flooding in CMA. However, it has to be noted that a 1D model cannot simulate the 2015 flood event with high accuracy due to its widespread inundation. The study [37] simulated the 2015 Chennai flood event using HEC-RAS 2D version 5.0 [85] and SRTM DEM of the resolution 30 m. Such a coarse resolution DEM cannot accurately represent the bathymetry of important topographical features such as the Adyar River in the CMA (approximately 80-200 m wide), the Buckingham canal (approximately 15-60 m wide), and the Mambalam drain (approximately 15 m wide). Therefore, a LIDAR DEM of a high spatial resolution of 2 m × 2 m that better represents the smaller topographical features of CMA is used to set up the hydraulic domain in this study. For the ocean side, the publicly available General Bathymetric Chart of the Oceans (GEBCO) gridded data is used. As the maximum number of grids that are allowed in the HEC-RAS 2D model is approximately 350,000, computation is not feasible on such a high-resolution grid. Therefore, the HEC-RAS sub-grid modelling is utilized herein with a computational grid of 30 m resolution together with 2 m resolution LIDAR DEM for topography to capture the detailed hydrodynamics. The sub-grid concept helps in achieving high accuracy by considering minute topographical features in lesser simulation time [86]. The flood hydrographs obtained at the river inlet from the hydrologic model are prescribed as inflow boundary conditions to the hydraulic model. In the absence of observed tidal gauge data at the mouth of the Adyar River during the flood events, predicted tide data from WXTide software (available at http://www.wxtide32.com/index.html) is supplied as the downstream boundary condition. The WXTide software was used by [87] to assess the vulnerability of the Chennai coast. According to [88], high sea levels are observed in the month of November, during which the city usually receives the maximum amount of rainfall. The predicted tidal data exhibited a similar trend and is in good agreement with the observed data as reported by [89]. For simplicity, a single value of the Manning's coefficient of 0.035 m −1/3 s is used for bed roughness as suggested in [37] for the same hydraulic domain.

Validation of Hydraulic Model
The simulated maximum inundation extent for 2015 flood is shown in Figure 6. There is no observed inundation extent available for fluvial flooding separately as the CMA also suffered from pluvial flooding due to its low-lying terrain, urbanization, encroachment of water bodies, and poor storm water management practices. Hence, the simulated inundation extents are qualitatively compared with those reported in [37,46]. As mentioned earlier, the simulated extent in [37] is obtained using SRTM 30 m DEM, whereas the inundation extent in [46] is obtained using the same 2 m high-resolution LIDAR DEM used in this study. Therefore, inundation along the smaller topographical features like Mambalam drain, Buckingham canal, and R.A Puram area closer to the Adyar creek is better captured in this study (Figure 6), similar to that reported in [46]. It needs to be emphasized that the use of high-resolution DEM and the concept of sub-grid helps to capture the flood dynamics more realistically. However, the authors of [37] could not capture such flow dynamics due to coarse resolution DEM. Overall, it is found that the simulated inundation extent for the 2015 flood event is as accurate as [46] and better than [37]. There is no inundation map available for the 2005 flood event. However, it can be considered that the simulated inundation extent for the 2005 flood event is reasonably accurate as the same hydraulic model setup is used with the corresponding boundary conditions. In addition to the inundation area, the hydraulic model is also validated with observed flood depths. Surveyed watermarks that were collected immediately after the 2015 flood event are available at various locations within the study area through field campaigns by different agencies such as the National Remote Sensing Centre, Institute for Remote Sensing-Anna University, National Centre for Coastal Research, and Indian Institute of Technology Madras [37]. As the study focuses on only fluvial flooding within the hydraulic domain, validation points that are closer to the river, its macro-drains, and canals are considered (Figure 7a). Upon comparison of simulated flood depths with the observed depths, R 2 = 0.81 is obtained for the 75% initial tank storage scenario (Figure 7b). For the 2005 flood event, the maximum flood depths are available only at Maraimalai In addition to the inundation area, the hydraulic model is also validated with observed flood depths. Surveyed watermarks that were collected immediately after the 2015 flood event are available at various locations within the study area through field campaigns by different agencies such as the National Remote Sensing Centre, Institute for Remote Sensing-Anna University, National Centre for Coastal Research, and Indian Institute of Technology Madras [37]. As the study focuses on only fluvial flooding within the hydraulic domain, validation points that are closer to the river, its macro-drains, and canals are considered (Figure 7a). Upon comparison of simulated flood depths with the observed depths, R 2 = 0.81 is obtained for the 75% initial tank storage scenario (Figure 7b). For the 2005 flood event, the maximum flood depths are available only at Maraimalai Adigal and Thiru vi ka bridges, as reported in [13]. The simulated depths (5.6 and 9.4 m) are similar to the observed depths (5.4 and 9 m). drain, (d) Chennai Bypass road, and (e) Pallavaram-Kundrathur road.
In addition to the inundation area, the hydraulic model is also validated with observed flood depths. Surveyed watermarks that were collected immediately after the 2015 flood event are available at various locations within the study area through field campaigns by different agencies such as the National Remote Sensing Centre, Institute for Remote Sensing-Anna University, National Centre for Coastal Research, and Indian Institute of Technology Madras [37]. As the study focuses on only fluvial flooding within the hydraulic domain, validation points that are closer to the river, its macro-drains, and canals are considered (Figure 7a). Upon comparison of simulated flood depths with the observed depths, R 2 = 0.81 is obtained for the 75% initial tank storage scenario (Figure 7b). For the 2005 flood event, the maximum flood depths are available only at Maraimalai Adigal and Thiru vi ka bridges, as reported in [13]. The simulated depths (5.6 and 9.4 m) are similar to the observed depths (5.4 and 9 m).

Hydrological Results
Various aspects of flood management regarding tanks like initial storage, dredging, and disappearance are discussed in detail.

Impact of Initial Storage in Tanks
The upstream flood hydrographs for various scenarios are compared in this section in Figures 8-14. It can be seen from Figures 8-14 that the decrease in peak flood discharge in urban sprawl-tank dredging scenarios is more pronounced only when the initial storage in the tank is less than 50% prior to the occurrence of flood. For the US05-TP2m dredging case, the peak discharge in a 2005-like flood will be closer to that of the historical flood itself if 50% of storage is available in all the tanks prior to occurrence of the event. However, such a significant reduction is not observed when 75% of the storage is already filled prior to occurrence of the event. Table 4 shows relative change in peak flows with respect to the corresponding US or US-RCP cases for different conditions of dredging and filling of tanks prior to occurrence of events. It can be seen from Table 4 that, as far as adaptation measures to counter impacts of future urbanization and climate change are concerned, the effect of initial storage on the reduction of flood peak is more pronounced when tanks are dredged by 2 m as compared to dredging them by 1 m. For example, in the US50-RCP4.5-50-year return period rainfall events, for 50% initial storage in the tanks, the percentage decrease of flood peaks are 3.7 and 4.4 times greater than that for the 75% initial storage in the tanks in the cases of the 1-m and 2-m dredging scenarios, respectively. Hence, it can be said that the degree of flood moderation will vary depending on the initial storage in tanks. In addition, the effectiveness of dredging of tanks in flood control is also dependent on the initial storage or the volume available in the tanks for harvesting the rainfall. events, for 50% initial storage in the tanks, the percentage decrease of flood peaks are 3.7 and 4.4 times greater than that for the 75% initial storage in the tanks in the cases of the 1-m and 2-m dredging scenarios, respectively. Hence, it can be said that the degree of flood moderation will vary depending on the initial storage in tanks. In addition, the effectiveness of dredging of tanks in flood control is also dependent on the initial storage or the volume available in the tanks for harvesting the rainfall.                 Table 4 shows that the 2-m dredging scenarios exhibit higher peak reduction in the case of moderate flood events like the 50-year return period or 2005-flood-like rainfall scenarios. From Figure 8a, it can be noticed that the peak discharge of a 2005-like flood for the US05-TP2m-F50 scenario is closer to that for H05-TP-F50 scenario. In other words, for moderate 2005-flood-like scenarios with lesser initial storage in the tanks, dredging them by 2 m may neutralize the impact of urbanization on flooding. However, such moderation is not observed for F75 and 2015-like flood scenarios. Thus, as mentioned earlier, dredging of the tanks will be more effective when initial storage in the tanks is moderated to less than 50% prior to a flooding event.  Figure 14 and Table 4 show that the effect of dredging and disappearance of tanks on peak flows is more significant for moderate floods such as that occurred in 2005 as compared to extreme floods that occurred in 2015. Peak flows for the 24-h duration 100-year return period rainfall BL100-TP-F50 and BL100-TP-F50 scenarios are found to be 1705 m 3 /s and 2954 m 3 /s, respectively. There is an increase in the peak flow by 295 m 3 /s and 226 m 3 /s for the US100-RCP4.5-TD and US100-RCP8.5-TD, respectively, from the corresponding US100-RCP4.5-TP-F50/US100-RCP8.5-TP-F50 scenarios. Similarly, for the 24-h duration 50-year return period flood, the peak flows are increased by 528 m 3 /s and 463 m 3 /s for the US50-RCP4.5-TD and US50-RCP8.5-TD scenarios, respectively, from the corresponding TP-F50 scenarios. As one would expect, the peak flows and flood extents of 24-h duration RCP scenarios are less than that of actual 2005 or 2015 flood events. This can be attributed to the fact that, in the hydrologic simulations with the NEX-GDDP projected rainfall (design storm rainfall) from RCP, we did not account for pre-event rainfall (antecedent conditions), which plays an important role in changing the initial storage in a basin. Also, it has to be noted that only 24-h duration rainfall events are considered herein for future design floods.

Hydraulic Results
The effect of tanks in flood moderation in CMA needs to be looked at in terms of inundation depths and extents for a holistic understanding of flood hazards.

Analysis of Inundation Depth
Important places (Figure 15) that are located on the banks and floodplains of Adyar River are chosen for comparison of the inundation depths of various scenarios (Figure 15). It has to be mentioned that only US50-RCP8.5 scenarios are presented in the paper as they represent the worst-case possibilities. Irrespective of the initial storage levels in the tanks, dredging of tanks by 2 m seems to reduce the inundation depths considerably for moderate floods such as 2005 flood and 50-year return period floods, when compared to those for extreme events such as 2015 flood and 100-year return period floods (Figures 16-19). For the US05-TP2m-F75 scenario, the inundation depths at the chosen locations are lesser when compared to that of the H05-TP-F75 scenario. In other words, it can be said, that if the 2005 and similar flood events occur in 2050 on an urbanized domain that has its tanks preserved and dredged by 2 m, then irrespective of the initial storage in the tanks, the effect of urbanization on flooding will be almost neutralized. However, as anticipated, the reduction in inundation depth is greater for 50% initial storage with 2-m dredging scenarios. In addition, the impact of dredging of tanks on flood moderation is reflected more in the case of moderate 2005-flood-like scenarios. On the other hand, for the extreme 2015-flood-like scenarios, dredging will not result in appreciable reduction of inundation depths. In all scenarios, the reduction in inundation depths is seen to exhibit a decreasing trend from upstream to downstream of the domain. This can be attributed to the fact that the areas in the downstream portion are closer to sea level and are relatively flat.
Water 2020, 12, x FOR PEER REVIEW 18 of 34 that has its tanks preserved and dredged by 2 m, then irrespective of the initial storage in the tanks, the effect of urbanization on flooding will be almost neutralized. However, as anticipated, the reduction in inundation depth is greater for 50% initial storage with 2-m dredging scenarios. In addition, the impact of dredging of tanks on flood moderation is reflected more in the case of moderate 2005-flood-like scenarios. On the other hand, for the extreme 2015-flood-like scenarios, dredging will not result in appreciable reduction of inundation depths. In all scenarios, the reduction in inundation depths is seen to exhibit a decreasing trend from upstream to downstream of the domain. This can be attributed to the fact that the areas in the downstream portion are closer to sea level and are relatively flat.         (Figures 16, 18, and 20a,b). Thus, in terms of inundation extents, dredging of the tanks by 2 m can neutralize the effect of urbanization for a moderate 2005-flood-like scenario. In the case of disappearance of tanks and urban expansion, a greater increase in inundation areas with respect to the H05-TP-F50 scenario is observed for the 2005-flood-like scenario than the H15-TP-F50 2015-flood-like scenario. Nevertheless, for moderate 50-year return period floods or 2005-flood-like scenarios, conservation and dredging of tanks results in reduction in inundation extents irrespective of the initial storage in the tanks (Figure 20a,b,d,e). Inundation extents for TB2m-F75 and tank disappearance scenarios for extreme floods did not show noticeable differences,     (Figure 20a,b,d,e). Inundation extents for TB2m-F75 and tank disappearance scenarios for extreme floods did not show noticeable differences, and hence, the figures for the same are not provided in the paper. The inundation pattern of the H05-TD scenario is similar to that of the H15-TP-F50 scenario (Figures 17a and 20a). In other words, in the future, a complete loss of storage of tanks in the upstream may cause a moderate 2005-flood-like event (50-year return period rainfall) to inundate to the scale of the extreme 2015-flood-like event (100-year return period rainfall). As a consequence, this will result in an excessively high inundation, especially in the areas (Figure 20a) closer and upstream of Chennai International Airport and those that are closer to the Adyar creek and Mambalam drain. On the other hand, different 2015-flood-like scenarios exhibit only marginal differences in the inundation extents (Figure 20c,e). It is noteworthy to mention that similar inferences are drawn in the previous section.

Analysis of Inundation Extent
in the future, a complete loss of storage of tanks in the upstream may cause a moderate 2005-flood-like event (50-year return period rainfall) to inundate to the scale of the extreme 2015-flood-like event (100-year return period rainfall). As a consequence, this will result in an excessively high inundation, especially in the areas (Figure 20a) closer and upstream of Chennai International Airport and those that are closer to the Adyar creek and Mambalam drain. On the other hand, different 2015-flood-like scenarios exhibit only marginal differences in the inundation extents (Figure 20c,e). It is noteworthy to mention that similar inferences are drawn in the previous section.  Findings on relative percentage changes in the total inundation area for various scenarios are reported in Table 5. For the 2005 flood, the total area under inundation increases from 18.6 km 2 to 27.8 km 2 as one moves from the historical existing scenario to the future (2050) urban sprawl and disappearance of tanks scenarios. Similarly, in the case of the 2015 flood, the area of inundation increases from 33 km 2 to 40.3 km 2 . From Table 5 and Figure 20, it can be seen that, in the combined urbanization and future climate F50 scenarios, the dredging and disappearance of tanks have a Findings on relative percentage changes in the total inundation area for various scenarios are reported in Table 5. For the 2005 flood, the total area under inundation increases from 18.6 km 2 to 27.8 km 2 as one moves from the historical existing scenario to the future (2050) urban sprawl and disappearance of tanks scenarios. Similarly, in the case of the 2015 flood, the area of inundation increases from 33 km 2 to 40.3 km 2 . From Table 5 and Figure 20, it can be seen that, in the combined urbanization and future climate F50 scenarios, the dredging and disappearance of tanks have a greater impact on the inundation areas for 50-year return period floods in comparison to 100-year return period floods. However, dredging-US05-TP2m-F75 scenarios show greater reduction in inundation areas (27.4%) with respect to US05-TP-F75 in comparison to the F50 scenarios, unlike the trend in peak flows (−8.5%). The disappearance of tanks has caused an increase in inundation areas by 4 km 2 and 4.8 km 2 relative to the US50-RCP4.5-F50 and US50-RCP8.5-F50 scenarios, respectively, for the 50-year return period rainfall. In the case of 100-year rainfall, the disappearance of tanks resulted in an increase in inundation areas by 2.5 km 2 and 2.4 km 2 relative to the US100-RCP4.5-F50 and US100-RCP8.5-F50 scenarios, respectively. As seen earlier, a similar trend is reported in the peak flows (Tables 4 and 5). This reiterates the fact that tanks play a crucial role in improving the damages caused by floods, especially in the case of a moderate, one in 50-year flood events.

Flood Hazard Zonation: Impact Analysis
The inundation depths 0.6 m, 1.4 m, and 3.5 m are chosen as threshold depths for defining flood hazard zones (Table 6) based on the satellite-and field-based damage for 2015 [37,64]. Higher differences in the inundation area are observed under moderate hazard zones across various scenarios for the 2015 flood. However, for the 2005 flood, higher differences are observed under low hazard zones across all scenarios. In the case of moderate rainfall events like 2005/RCP 4.5-flood-like events, the inundated area under "very high" hazard category is more when compared to the "low" category. It is the reverse for extreme 2015/RCP 8.5-flood-like events. This is due to the fact that, in the 2005 flood, the floodplains were inundated minimally, unlike the one in hundred-year 2015 flood. However, Figure 21 shows that, in all scenarios, the area under the high hazard zone category is the highest. It can also be seen from Figure 21 that the reduction in total and hazard zone-wise inundation areas is greater for lesser initial storage in the tanks across all the flood scenarios. This inference is consistent with the results reported in the previous sections.

Limitations of the Present Study
It has to be borne in mind that the future climate scenarios used in this study do not account for impact of urbanization on the urban microclimate. The tide gauge station data analysis shows a sea level rise at a mere rate of 0.7 mm/year (data available at http://www.psmsl.org/data/obtaining/stations/205.php). Therefore, sea level rise by 2050 is not accounted for in this study. Due to lack of data on storm drains and their status of functioning, the study considers only flooding in the vicinity of Adyar River and not the widespread pluvial flooding. To get a holistic view of the role of tanks in flood moderation, different return period floods of different duration and climate models are to be considered. Nevertheless, in the study area, the extreme NEMR usually lasts for 2-3 days and these events may occur in combination with a lesser or heavier pre-event spell. Therefore, the cases "what if a historical extreme event happens in the future that entails urban expansion" and "what if the climate projected 24-h duration extreme event occurs in the future that entails urban expansion" have been considered in this study. Moreover, from Figure 14, it can be seen that the negative impact of disappearance of tanks is reflected more in event-based simulations in comparison to that of the climate projected 24-h duration rainfall. Nevertheless, this study is highly valuable as it provides the necessary scientific reasons for preservation and maintenance of the existing tanks at the very least. Further studies need to be carried out to evaluate other associated benefits of the same, such as increased ground water recharge, water availability for domestic supply, and irrigation.

Limitations of the Present Study
It has to be borne in mind that the future climate scenarios used in this study do not account for impact of urbanization on the urban microclimate. The tide gauge station data analysis shows a sea level rise at a mere rate of 0.7 mm/year (data available at http://www.psmsl.org/data/obtaining/ stations/205.php). Therefore, sea level rise by 2050 is not accounted for in this study. Due to lack of data on storm drains and their status of functioning, the study considers only flooding in the vicinity of Adyar River and not the widespread pluvial flooding. To get a holistic view of the role of tanks in flood moderation, different return period floods of different duration and climate models are to be considered. Nevertheless, in the study area, the extreme NEMR usually lasts for 2-3 days and these events may occur in combination with a lesser or heavier pre-event spell. Therefore, the cases "what if a historical extreme event happens in the future that entails urban expansion" and "what if the climate projected 24-h duration extreme event occurs in the future that entails urban expansion" have been considered in this study. Moreover, from Figure 14, it can be seen that the negative impact of disappearance of tanks is reflected more in event-based simulations in comparison to that of the climate projected 24-h duration rainfall. Nevertheless, this study is highly valuable as it provides the necessary scientific reasons for preservation and maintenance of the existing tanks at the very least. Further studies need to be carried out to evaluate other associated benefits of the same, such as increased ground water recharge, water availability for domestic supply, and irrigation.

Summary and Conclusions
Chennai city is prone to frequent flooding due to deep depression and cyclonic activities during the NE monsoon. In order to alleviate fluvial flooding in the Adyar basin that comprises the southern part of the CMA, the effect of dredging of 163 existing tanks under urbanization and various rainfall scenarios are analyzed in this study. The Adyar River flowing through the CMA functions as a surplus course for the tanks. The impact of dredging of tanks by 1 m and 2 m on the flooding in CMA is investigated by simulating a moderate 1 in 50-year flood, i.e., a 2005-flood-like event, and an extreme 1 in 100-year flood, i.e., a 2015-flood-like event. The extreme rainfall events for the mid-century period obtained from the CNRM models are also simulated. The study is based on the urbanization-hydrologic-hydraulic modelling framework under various rainfall scenarios and brings out the importance of maintaining and dredging of the tanks in flood moderation of CMA. The major conclusions from this study that are specific to the Adyar basin are as follows: (i) It can be concluded that the tanks in the upstream catchments provide effective flood mitigation for a downstream city, when the initial storage in the tanks is minimum or when there is an insignificant wet spell prior to occurrence of the extreme rainfall. Hence, an accurate short-range rainfall forecast system for the Adyar basin can help the authorities in charge of the tanks release excess water in the tanks to downstream areas or store it in secondary reservoirs on time. (ii) The dredging of tanks results in considerable differences in inundation depths and extents, irrespective of the 50% and 75% initial storage for moderate 2005-flood-like events, i.e., one in 50-year return period floods. (iii) Dredging of tanks uniformly by 2 m within the tanks' water spread area does not reduce inundation significantly for 1 in 100-year flood-like scenarios. However, a similar dredging of the tanks can neutralize the adverse effect of urbanization in 2050 in the case of 1 in 50-year flood-like scenarios. (iv) In the future, if the tanks disappear due to urbanization, then even for a moderate 1 in 50-year rainfall, there will be excessive inundation due to fluvial flooding closer to that caused by a 1 in 100-year rainfall, i.e., similar to the 2015 flood event. (v) As the Adyar basin is prone to occurrence of frequent extreme events, it is important that the uncontrolled tanks are equipped with flood gates to regulate storage prior to a forecasted flooding event to minimize flood peaks. (vi) The flood hazard analysis performed in this study indicates that higher differences in the inundation area due to dredging and disappearance of tanks get reflected under the low hazard zone category for 1 in 50-year flood-like events. On the other hand, the differences are greater in moderate zones for 1 in 100-year flood-like events. Such a trend is due to the fact that the 1 in 100-year floods are greater in magnitude than the 1 in 50-year floods. Hence, even though the reduction/increase in inundation area due to dredging/disappearance of tanks is less for 1 in 100-year floods, the implications of dredging of tanks in flood management cannot be discounted. (vii) This study provides insights into dredging of tanks by 1 m and 2 m as possible flood mitigation scenarios. The observations of this study can be useful to the government for the proposed plan of dredging of tanks in the Adyar basin.
It is obvious that dredging is a very expensive process. However, only a very few options are available for flood mitigation in the CMA owing to the dense population, space constraint for constructing hydraulic structures, and the almost flat terrain of CMA. Therefore, the key takeaway from this study is that, at the very least, urban planning should consider preserving the storage and functionality (maintenance of tank bund and feeder channels) of the existing tanks so as to minimize damage from moderate fluvial floods (1 in 50-year rainfall-like events) in the Adyar basin in the future, given the reality that urban development cannot be efficiently controlled in a developing economy. The aforementioned inferences from the study can be used as guidelines for the government for better preparedness and management of floods in CMA.