DESC:METHOD AND SYSTEM FOR GENERATING FLOOD INUNDATION FORECASTS IN NEAR-REAL TIME FOR LARGE GEOGRAPHICAL REGIONS

TECHNICAL FIELD
[0001] The present disclosure generally relates to a postprocessing framework for rainfall forecasts, and more particularly, relates to a method and system for generating copula corrected flood inundation forecast in near-real time at a large scale.

BACKGROUND
[0002] Climate change has become a universal issue and is likely to pose enormous challenges for substantial sectors like agriculture, water resources, infrastructure, and the livelihood of millions of people living at a global level. Recently, the consequences of climate change can be evident in the form of untimely and uncertain rainfalls with hazards of large-scale flood inundation with increased frequency. Therefore, the flood mitigation and forecasting have become a valuable task.

[0003] Today’s modern society’s demand for more meticulous weather forecast is expanding. This ever-increasing demand have broadened the spectrum of needs for weather forecasting with a diverse set of industries dependable on weather conditions, that largely rely on weather forecasting. Moreover, growing cities need thorough knowledge of the current weather and its consequences to design a building or to provide a safe civil infrastructure to further reduce the haphazard caused by floods.

[0004] Flood inundation forecasting is an assessment of the rise in the riverine water levels in the near future done for the region of interest. Flood inundation forecasting is an important disaster mitigation practice that allows one to predict when local flooding is likely to take place with a high degree of accuracy. Forecasts typically uses runoff data, reservoir levels and releases to predict the rise in river levels.

[0005] Globally, the flood inundation hazard models largely rely upon the corrected forecast rainfall values and the topography of the region of interest by operating globally available Digital Elevation Models (DEM’s). DEM’s are the mathematical representations of the earth’s topography used for determination of time-area histogram in the Geographic Information System (GIS) platform. Conventionally, the flood hazard measurement techniques utilized the past observational ground surveyed data. Moreover, the analysis of large-scale inundation models was also carried out utilizing the identical method with the available Digital Elevation Models (DEM’s). However, the analysis caused computational overload and increased the degree of imperfection in the corrected forecasting value when taken at a large scale. Such analysis carried out by implementing DEM’s at a large scale made the forecast an arduous task.

[0006] One of the conventional methods defines flood inundation hazard in terms of rainfall forecast value directly and uses the same in subsequent flood risk estimation. The rainfall forecast value can be used for identifying large regions that might be exposed to floods without detailing of specific areas in need of flood mitigation and relief efforts.

[0007] Another conventional method discloses a coupling process of real-time flood probability forecast with precipitation forecast, which fails to disclose the system that use of the forecast and observed rainfall values and represents only the information of the probability distribution of the observed rainfall.

[0008] Thus, there is a need for a system to effectively use forecast and observed rainfall values and non-linearly correct the entire spectrum of flood inundation forecasting at the weather to medium range based on forecasted hazard.
SUMMARY
[0009] In an aspect of the present disclosure, a method for generating flood inundation forecast rainfall in near-real time for large geographical regions is disclosed. The method uses long-term hindcast rainfall data for a time period and observed rainfall values for the same time period and generates copula corrected flood inundation forecast in near-real time. The method statistically combines the forecast and observed rainfall values for a pixel using a best-fit copula within a spatiotemporal moving window process. The method is implemented by a system of hardware components, modules, and sub-modules. The method comprises the steps of receiving by at least one processor, an observed rainfall data of a region from a first remote server, receiving by at least one processor, forecast rainfall data of the said region from a second remote server, determining a best fit copula from the observed rainfall data and the forecast rainfall data for a plurality of pixels in the region, generating the non-linear bias corrected rainfall forecast ensemble and feeding the same into Zero parameter Dynamic Budyko (DB) hydrological model to generate an ensemble of runoff raster dataset between derived from corrected rainfall forecast ensembles. The method further comprises determining a flood velocity and correcting a slope factor in the globally available Digital Elevation Model (DEM) using a Slope-corrected Calibration-free Iterative Flood Routing and Inundation (SCIFRIM) sub-module and determining probability of inundation maps for given day for a given lead time using the flood velocity and the corrected slope factor.

[0010] In another aspect of the present disclosure, a system for generating flood inundation forecast rainfall in near-real time for large geographical regions is disclosed. The system comprise a first remote server connected to a network, a second remote server connected to the network, a computing device communicably connected to the first remote server and the second remote server via the network. The computing device comprises at least one processor, at least one memory unit connected with the processor, a graphical user interface connected with the processor, and a plurality of sub-modules connected with the processor. The plurality of sub-modules comprises a Slope-corrected Calibration-free Iterative Flood Routing and Inundation Model (SCIFRIM) submodule connected to the network.

BRIEF DESCRIPTION OF DRAWINGS
[0011] The detailed description is given with reference to the accompanying figures. The numbers are used throughout the drawings to refer features and modules/charts.
[0012] Figure 1 illustrates a flow chart depicting a method for generating flood inundation forecasts in near real-time for large geographical regions, in accordance with an exemplary embodiment of the present disclosure.
[0013] Figure 2 illustrates various steps involved in obtaining a best-fit copula for generating non-linearly corrected rainfall forecast ensembles in accordance with the present disclosure.
[0014] Figure 3 illustrates workflow for generation of spatial extent of inundation hazard maps using SCIFRIM submodule.
[0015] Figure 4 illustrates a schematic diagram depicting SCIFRIM submodule characterisation of catchment behaviour, represented along with underlying assumptions in accordance with the present disclosure.
[0016] Figure 5 illustrates characterizing river network as a set of Independent Channels (ICs) with pixels along a given IC in accordance with the present disclosure.
[0017] Figures 6a – 6b illustrate an iterative process of updating channel velocity across river network reaches in accordance with the present disclosure.
[0018] Figure 7 illustrates a schematic of a system for generating flood inundation forecasts for large geographical regions, in accordance with another embodiment of the present disclosure.
[0019] Figure 8 illustrates the inputs GEFS rainfall values for different lead times along with observed IMD rainfall values for comparison for the lead day in accordance with one example of the present invention.
[0020] Figure 9 illustrates the ensemble of corrected forecast values for the lead day in accordance with one example of the present invention.
[0021] Figure 10 illustrates the forecast probability of inundation maps for the ensemble of corrected rainfall values after subsequent processing with the Zero parameter Dynamic Budyko module (DB) and Conceptual Flood Routing and HAND -based Inundation module (CFRHIM).

DETAILED DESCRIPTION
[0022] In the following description, for the purpose of clarification, details of the present invention have been specified. The various embodiments of the present invention disclose a system for method for generating copula corrected flood inundation forecast in near-real time at a large scale.

[0023] The present invention discloses the system and method for correcting erroneous forecast rainfall values and mapping the probabilistic forecast extent of inundation in near real-time at a global scale on an almost near-time temporal basis. The system uses long-term hindcast rainfall data for a given time period and observed rainfall values for the same time period. The system statistically combines the forecast and observed rainfall values for a pixel using a best-fit copula within a spatiotemporal moving window process. The system represents the apriori information of probability distribution of observed rainfall given a forecast value. The system employs a moving window to determine the spatial bias in forecast rainfall values. Further, the present invention is configured to model the worst-case framework of the predicted rainfall events by preparing for Confidence Interval (CI). The model may sample the forecasted rainfall in the given CI to feed the ensembled values (inputs) to the Zero-parameter Dynamic Budyko (DB) hydrological model for gridded runoff product.

[0024] As a result of the DB model, the in-built slope correction mechanism of an inundation sub-module corrects the slope factor in the available DEM. Moreover, the present inundation module assumes the flood velocity to be constant spatially and applies the Height Above Nearest Drainage (HAND) static flood terrain indicator to obtain forecast flood inundation hazard values on a near-real-time basis.

[0025] In an embodiment of the present disclosure, a method for generating flood inundation forecast in near-real time for large geographical regions is disclosed. The large region comprises a plurality of pixels. The method comprises the steps of receiving by at least one processor, an observed rainfall data of a region from a first remote server, receiving by at least one processor, forecast rainfall data of the said region from a second remote server, determining the a best fit copula from the observed rainfall data and the forecast rainfall data for a plurality of pixels in the region, determining generating the non-linear bias corrected rainfall forecast ensemble and feeding the same into Zero parameter Dynamic Budyko (DB) hydrological model to generate an ensemble of runoff raster dataset between derived from corrected rainfall forecast ensembles generate runoff ensembles. The method further comprises determining a flood velocity and correcting a slope factor in the globally available Digital Elevation Model (DEM) using a Slope-corrected Calibration-free Iterative Flood Routing and Inundation Model (SCIFRIM) sub-module, subsequently determining probability of inundation maps for given day of for a given particular lead time using the flood velocity and the corrected slope factor.

[0026] In an embodiment of the present disclosure, the first remote server is a regional meteorological server, and the second remote server is a global meteorological server.

[0027] In an embodiment of the present disclosure, determining a best-fit copula from the observed rainfall data and the forecast rainfall data for the plurality of pixels comprises resampling the observed rainfall data of the region received from the first server, obtaining forecast global rainfall time series by applying a moving window of pixels to the forecast rainfall data for analysing a spatial variability of rainfall patterns at multiple scales, determining a degree of correlation between the resampled observed rainfall data and forecast GEFS rainfall time series for the pixel of interest as a correlation coefficient Kendall’s Tau obtained through the moving window process, determining a plurality of copulas comprising Frank, Clayton, and Gumbel copulas using a non-parametric correlation coefficient Kendall’s Tau and selecting a best-fit Copula from the plurality of copulas using the empirical copula test, upper tail dependence test and based on AIC, BIC values. The method may comprise repeating the above steps over all the pixels in the region until a best fit copula is determined for each pixel in the region.

[0028] In an embodiment of the present disclosure, the generating bias corrected rainfall forecast ensemble from the best fit copula comprises determining a conditional probability distribution of the observed rainfall values for a given rainfall forecast value for a particular day at a given time from the best fit copula, deriving a corrected range of rainfall forecast values as the 50 – 99.5 percentile of conditional probability distribution of the observed rainfall values, generating rainfall forecast ensemble for each rainfall forecast value by sampling the rainfall values within the corrected range and ordering the corrected rainfall forecast ensemble for each pixel by considering a best case and worst case of corrected range of rainfall forecast values.

[0029] In an embodiment of the present disclosure, determining a flood velocity for each time step and correcting the slope factor in the globally available DEM by SCI-FRIM sub-module comprises determining flood velocity through iterative calibration-free routing procedure; wherein initial bankfull velocity of highest ordered reach is calculated by applying Manning’s equation using a global empirically derived river geometry data and is then assigned to all reaches within the river system for the same time step, and subsequently the discharge estimated at the mouth of the river is used to recalculate flow velocity for the highest-ordered reach, that is iteratively assigned to upstream reaches until the channel flow velocity stabilizes. This determination step further comprises correcting the slope factor in the globally available digital elevation (DEM) model by spatially sub-setting a river network into a set of independent channels and determining a derivative of the exponential distribution relationship between the distance of each channel pixel to the pourpoint and corresponding elevation values.

[0030] In an embodiment of the present disclosure, determining probability of inundation maps by the SCI-FRIM sub-module for a given day of a given lead time using the flood velocity and the corrected slope factor comprises generating the flood extents for different inputs within each Reach Contributing Area (RCA) corresponding to the discharge-height (q-h) relationship for each reach, selecting Height Above Nearest Drainage (HAND) rasters from pre-processed HAND raster library based on the flood depth obtained for RCA and mosaicing the selected HAND rasters to obtain the flood inundation map across the entire region of interest.

[0031] In an embodiment of the present disclosure, the method further comprises converting the resampled observed rainfall data and forecast rainfall time series into probability Cumulative Distribution Function (CDF) values by fitting a Gamma distribution to the non-zero rainfall values.

[0032] In an embodiment of the present disclosure, the method further comprises identifying an independent channel in the river geometry data as a set of reaches between the nearest source node for a particular junction node by the SCI-FRIM sub-module.

[0033] In an embodiment of the present disclosure, the method further comprises storing a raster library of predefined thresholded HAND rasters for each RCA and corresponding look-up-table of discharge-height-velocity (q-h-v) relationships for each reach.

[0034] In an embodiment of the present disclosure, the method further comprises determining an uncertainty associated with the empirically derived global channel bank full-depth values and the shape of cross-section of the river channel.

[0035] In an embodiment of the present disclosure, the method further comprises pre-processing and storage of the globally available DEM for a given spatial region with catchment and channel network delineation along with RCA characterisation.

[0036] Referring to Figure 1, illustrated is a flow chart depicting a method for generating flood inundation forecasts for large geographical regions, in accordance with an exemplary embodiment of the present disclosure. The method uses the Global Ensemble Forecast System (GEFS) forecast precipitation values comprising hindcast values over the past minimum of 30 years with daily forecast data having a lead time of up to 8 days. The method uses the system for automatically collating and resampling the observed rainfall data over a particular large scale region, from a respective server, or a meteorological server with GEFS rainfall data. In an example as shown in Figure 1, gridded observed rainfall data of India from the Indian Meteorological Department (IMD) is used to couple with the GEFS rainfall data of a particular lead times (from 1-day lead time until 7-days) over co-incident pixel areas.

[0037] The method comprises the steps of receiving by at least one processor, an observed rainfall data of a region from a first server. The large region comprises a plurality of pixels of small regions. The method comprises receiving by at least one processor, a forecast rainfall data of the region from a second server, determining a best fit copula from the observed rainfall data and the forecast rainfall data for the plurality of pixels in the region by the at least one processor, determining bias corrected rainfall forecast ensemble and feeding the same into Zero parameter DB hydrological model by the at least one processor, determining a flood velocity and correcting the slope factor in the globally available Digital Elevation Model using a Slope-corrected Calibration-free Iterative Flood Routing and Inundation Model (SCIFRIM) sub-module using an ensemble runoff raster dataset and determining probability of inundation maps for a given day of a given lead time using the flood velocity and the corrected slope factor.

[0038] The various steps of the method in the present invention are implemented by a system comprising a plurality of modules, a plurality of sub-modules, a plurality of servers comprising a first remote server connected to a network and a second remote server connected to a network and a plurality of computing devices connected via a communication network of the present invention. In some examples of the present invention, the computing device may be not only limited to a computer, a laptop or a mobile/portable device or a handheld device. Each computing device may comprise one or more processors. Each processor comprises a control unit, registers, an arithmetic logic unit and a processor memory unit capable of storing a plurality of instructions to perform one or more methods or tasks. As similar to the above, each server may be a computing device comprising one or more processors. In general, modules and hardware components of the system are computer alike devices. The processors are configured to perform the said steps of the method of the above embodiments.

[0039] The computing device further may comprise a graphical user interface connected with the processor and a plurality of sub-modules connected with the processor, and the plurality of sub-modules comprises a Slope-corrected Calibration-free Iterative Flood Routing and Inundation Model (SCIFRIM) submodule connected to the network and other sub-modules connected to the network.

[0040] Figure 2 illustrates various steps involved in obtaining best-fit copula for generating non-linearly corrected rainfall forecast ensemble members in accordance with the present disclosure. The system applies a moving window process for analysing a spatial variability of rainfall patterns at multiple scales. The moving window of 3*3 pixels is taken and used by the system for the analysis. The moving window uses the centre IMD rainfall pixel of interest with maximum forecast values within a temporal neighbour of 3 days provided within the window size. Further, the system determines a non-parametric correlation coefficient Kendall’s Tau between the observed IMD data and forecast GEFS rainfall time series obtained through the moving window process. The Kendall’s Tau is a degree of correlation between the observed and the forecast rainfall values for the pixel of interest. Kendall’s Tau is further used to determine the Frank, Clayton, and Gumbel copulas.

[0041] The rainfall time series are procured from the moving window of 3*3 pixel. The observed and forecast rainfall time series are converted into scale-free measures by considering probability Cumulative Distribution Function (CDF) values. In the present invention, the CDF is the probability of obtaining a zero-rainfall value followed by the probability of obtaining non-zero rainfall values by fitting a Gamma distribution to the non-zero rainfall values.

[0042] The values obtained from an empirical copula test, and upper tail dependence test are determined for Frank, Clayton, and Gumbel copulas. The system further selects a best-fit Copula selected from Frank, Clayton, and Gumbel copulas based on AIC, BIC values using the afore-mentioned tests. The above process is iterated over all the pixels in the region of study until a best fit copula is identified for each grid of interest/ each pixel in the region. Once the best fit copula for every pixel is identified, the system determines the conditional probability distribution of the observed values given the forecast value for a particular day at a given lead time from the best fit copula. In order to determine the worst flood scenarios, the system uses the higher percentile values of conditional probability distribution in the range 70 - 99.9 and derives a corrected range of rainfall forecast values. In another embodiment, the system derives a corrected range of rainfall forecast values within as the 50 – 99.5 percentile of conditional probability distribution of the observed rainfall values.

[0043] In a further embodiment of the present disclosure, the rainfall values are sampled by system within the corrected forecast intervals thereby generating rainfall forecast ensemble for each rainfall forecast value and corrected rainfall forecast ensemble is ordered for every pixel to consider the best case and the worst case of corrected rainfall forecast values. Then, the ensemble of corrected rainfall forecast values is fed to a calibration-free, continuous rainfall-runoff DB hydrological model.

[0044] The DB model of the present invention uses a calibration-free instantaneous dryness index as a function of time {f(t)} to perform continuous hydrologic partition at a fine temporal scale. The DB hydrological model uses a modified Budyko function that imitates the effect of antecedent rainfall and solar radiation on the ER at a given time instant. The modified Budyko function makes use of the empirically derived universal decay function to the past available water and potential evapotranspiration (PET) time series values to replace the soil moisture content. The decay function is assumed to be the same for the catchments located across all geographical locations. The ER thereby decays into runoff with time as a result of the decay function followed by feeding the values to a conceptual flood routing and inundation sub-module. The system further comprises an inundation sub-module, named, a Slope-Corrected Calibration-free, Iterative Flood Routing and Inundation Model (SCIFRIM), which determines flood velocity for each time step by applying Manning’s equation and uses the global empirically derived river geometry data based on Leopold’s observations. Figure 3 illustrates a flow chart depicting the flow of the process in the inundation sub-module in accordance with the present disclosure.

[0045] Further, the SCIFRIM corrects the slope factor in the globally available DEM by spatially sub-setting the river network into a set of independent channels and determining the derivative of the exponential distribution relationship between the distance of each channel pixel to the pourpoint and corresponding elevation values. Figure 4 illustrates a schematic diagram depicting SCIFRIM submodule characterisation of catchment behaviour, represented along with underlying assumptions in accordance with the present disclosure. Figure 5 illustrates characterizing river network as a set of Independent Channels (ICs) with pixels along a given IC in accordance with the present disclosure.

[0046] An independent channel is defined by a junction and source characterised as end nodes of a river network. In a channel network, a junction is defined as where two nodes/pixels meet, while an upstream reach head pixel is called the source node. The river channel can then be visualised with n source nodes and n-1 junction nodes, with the river mouth node designated as a pseudo nth junction node.

[0047] Further, the inundation sub-module of the present invention is configured to identify an independent channel as a set of reaches between the nearest source node for a particular junction node. For each reach, a set of simulated discharge (q) - velocity (v) pairs are developed for different flow heights (h) considering both over-and-under-bank full height conditions. In the case of over-bank full flow, the sub-module assumes that the entire q is routed only across the channel. Additionally, v is simulated considering the wetted channel perimeter as constant. Correspondingly, q - h values are also stored for each reach based on the pre-processing technique.

[0048] Figures 6a – 6b illustrate an iterative process of updating channel velocity across river network reaches in accordance with the present disclosure. The system further dynamically updates the flow velocity. The velocity for a given time interval, vi, across all reaches in a channel is defined to be the bank full velocity of the highest ordered (river-mouth) reach. The time interval is defined by the frequency of the ER time series. Using the bank full velocity, discharge is routed from the upstream reach to the river mouth and the velocity for the discharge obtained is re-evaluated from the set of simulated q - v values for the river mouth reach. The system iteratively updates the flow velocity until vi converges to a stable value. The obtained stable velocity is assigned to all the reaches in the river system to calculate the flow height (h) from the pre-computed values of q - h for each reach.

[0049] The respective inundation depth (DT) within a Reach Contributing Area [RCA (DT)] is obtained by reducing the bank full depth from the h obtained for a reach. The Height Above Nearest Drainage (HAND) technique is generally implemented to detect inundation areas over large spatial domain and to obtain the rasterized flood inundation map across the entire region of interest. Based on the respective inundation depth obtained, respective HAND rasters for each RCA are then selected and mosaiced together to obtain the flood inundation map across the entire region of interest.

[0050] HAND-based inundation mapping approaches generally overestimate the extent of inundation within the region of interest. Considering modularised RCA-based units for mapping inundation greatly controlled over-estimations, resulting in optimum model performance. The proposed time-varying non-linear flow velocity estimation procedure significantly improves over the constant linear velocity assumption with hydrograph-based approaches and calibration-intensive Muskingham-Cunge-based approaches, especially under sub-daily routing considerations. The model can be applied using high-resolution LiDAR DEM and even using globally available open-access but error-prone SRTM/ASTER DEMs by correcting for channel slope affecting flow during flood events. By pre-computing the rating curve relationships and corresponding HAND extents within each RCA, the model saves critical computational time crucial for flood management procedures.

[0051] The HAND flood terrain model captures the flood extents that can be pre-computed and fetched to the ensemble of flood velocities. The inputs are fed to the present invention thereby receiving an ensemble of flood maps corresponding to each runoff raster dataset which were generated between the best and worst case corrected forecast rainfall values, consequently resulting into a flood hazard map corresponding to forecast rainfall values for a particular lead time.

[0052] Referring to Figure 7, illustrated is a schematic of a system for generating flood inundation forecasts for large geographical regions, in accordance with another embodiment of the present disclosure. The system (100) comprises a first remote server (110) connected to a network (130), a second remote server (120) connected to the network (130), a computing device (150) communicably connected to the first remote server (110) and the second remote server (120) via the network (130). The computing device comprises at least one processor (140), at least one memory unit (145) connected with the processor (140), a graphical user interface (147) connected with the processor (140), a plurality of sub-modules (160) connected with the processor, and the plurality of sub-modules comprises a Slope-corrected Calibration-free Iterative Flood Routing and Inundation Model (SCIFRIM) submodule (165) connected to the network (130). The system performs the method steps of the previous embodiments for generating flood inundation forecast in near-real time for large geographical regions.

[0053] To demonstrate the forecast flood inundation probability of the present invention, non-linear correction of forecast rainfall values for a given lead day during a lead month was simulated and compared with observed data. In an example, each observational day during august 2018, with observed rainfall values recording the peak flood day as 16th August 2018, for the catastrophic flood event that Kerala witnessed on 16th August 2018 was prepared. Figure 8 illustrates the inputs GEFS rainfall values for different lead times along with observed IMD rainfall values for comparison for the lead day in accordance with one example of the present invention. Figure 8 further illustrates that the observed rainfall values are reported with higher intensities (> 100 mm) with a peak magnitude of around approx. 250 mm. In the case of GEFS, with the advancement in lead time, there is an expected progressive decrease in the magnitude of peak rainfall intensity.

[0054] Figure 9 illustrates the ensemble of corrected forecast values for the lead day in accordance with one example of the present invention. Figure 9 shows the ensemble of corrected forecast values for 16th August 2018, which has accounted for the spatial displacement in forecast rainfall patterns. Figure 9 illustrates that there is a significant displacement in forecast rainfall witnessed over ocean pixels rather than over observed inland regions, which has been corrected in the case of the spatial window approach considered above, resulting in larger ensemble values being mapped over inland regions.

[0055] Figure 10 illustrates the forecast probability of inundation maps for the ensemble of corrected rainfall values after subsequent processing with the Zero parameter Dynamic Budyko module (DB) and Conceptual Flood Routing and HAND -based Inundation module (CFRHIM). Figure 10 illustrates that the inundation probability is captured even for a lead time of 7 days, which is attributed to the larger quantile values (up to 99.5%) that the present invention has accommodated into the forecast-correction methodology. The output probability of inundation is binarised, considering pixels to be flooded even if they exhibited a 1% chance of inundation. Thereafter, Flooding Accuracy (FA) is processed upon comparing the binarised hazard map to the observed inundation point dataset obtained from Kerala State Disaster Management Agency (KSDMA). The FA showed a decreasing pattern with an increase in lead time (FA - 61.7%, 57.2%, 55.7%, 51.4%, 49.8%, 49.1%, and 48.3% for lead days 1,2,3,4,5,6,7, respectively); the same characteristic followed by the GEFS forecast rainfall intensity values.

[0056] The present invention effectively corrects the erroneous forecast rainfall values and also redefines the hazard measurement by using weather to medium range forecast values on a pixel scale and conventional flood hazard measurement techniques utilising the past observational data. Existing methods that rely on predicting inundation directly from forecast rainfall values are highly error-prone. Even those methods depend on randomly corrected ensemble forecast values that do not capture the entire flood inundation extent probability. The system estimates inundation hazard on a pixel scale (of the order of metres) rather than the usual grid scale (of the order of hundreds of kilometres).

[0057] The system is computationally time efficient and used on near-real-time basis provided forecast data is readily accessible when compared to other data-intensive and physically complex routing procedures including LIS-FLOOD, Cama-FLOOD, HEC-RAS. Moreover, the presented method is flexible to be used with any forecast/observed rainfall data and be coupled with any hydrological run-off model output at any given spatial and temporal resolution. The present invention is of particular importance in in-situ data-scarce regions including the Indian sub-continent, since the system relies on satellite-based forecast datasets that are available globally. Additionally, the system has the potential to assist in flood mitigation and relief efforts, as well as in evaluating flood insurance and assessing disaster risk.

[0058] While the present invention has been described above in terms of specific embodiments, it is to be understood that the invention is not intended to be confined or limited to the embodiment disclosed herein.
,CLAIMS:We Claim:

1. A method for generating flood inundation forecast in near-real time for large geographical regions, the method is implemented by a system comprising:
receiving by at least one processor, an observed rainfall data of a region from a first remote server, wherein the region comprises a plurality of pixels;
receiving by at least one processor, a forecast rainfall data of the region from a second remote server;
determining by the at least one processor, a best fit copula from the observed rainfall data and the forecast rainfall data for the plurality of pixels in the region;
generating bias corrected rainfall forecast ensembles for each pixel using the best fit copula by the at least one processor;
feeding the bias corrected rainfall forecast ensembles into a Dynamic Budyko (DB) hydrological model to generate an ensemble of runoff raster dataset derived from corrected rainfall forecast ensembles by the at least one processor;
determining a flood velocity for each time and correcting a slope factor in the globally available Digital Elevation Model (DEM) using a Slope-corrected Calibration-free Iterative Flood Routing and Inundation Model (SCIFRIM) sub-module using an ensemble runoff raster dataset; and
determining probability of inundation maps by the SCI-FRIM sub-module for a given day of a given lead time using the flood velocity and the corrected slope factor.

2. The method as claimed in claim 1, wherein determining the best-fit copula from the observed rainfall data and the forecast rainfall data for the plurality of pixels comprises:
resampling the observed rainfall data of the region received from the first server;
obtaining forecast global rainfall time series by applying a moving window of pixels to the forecast rainfall data for analysing a spatial variability of rainfall patterns at multiple scales;
determining a degree of correlation between the resampled observed rainfall data and forecast rainfall time series for the pixel of interest as a correlation coefficient Kendall’s Tau obtained through the moving window process;
determining a plurality of copulas comprising Frank, Clayton, and Gumbel copulas using a non-parametric correlation coefficient Kendall’s Tau; and
selecting a best-fit Copula from the plurality of copulas using the empirical copula test, upper tail dependence test and based on AIC, BIC values;
repeating the above steps over all the pixels in the region until a best fit copula is determined for each pixel in the region.

3. The method as claimed in claim 1, wherein the method further comprises converting the resampled observed rainfall data and forecast rainfall time series into probability Cumulative Distribution Function (CDF) values by fitting a Gamma distribution to the non-zero rainfall values.

4. The method as claimed in claim 1, wherein the generating bias corrected rainfall forecast ensemble from the best fit copula comprises:
determining a conditional probability distribution of the observed rainfall values for a given rainfall forecast value for a particular day at a given time from the best fit copula;
deriving a corrected range of rainfall forecast values as the 50 – 99.5 percentile of conditional probability distribution of the observed rainfall values;
generating rainfall forecast ensemble for each rainfall forecast value by sampling the rainfall values within the corrected range;
ordering the corrected rainfall forecast ensemble for each pixel by considering a best case and worst case of corrected range of rainfall forecast values.

5. The method as claimed in claim 1, wherein the moving window comprises a 3x3 pixels.

6. The method as claimed in claim 1, wherein determining a flood velocity for each time step and correcting the slope factor in the globally available DEM by SCI-FRIM sub-module comprises:
determining flood velocity through iterative calibration-free routing procedure; wherein initial bankfull velocity of highest ordered reach is calculated by applying Manning’s equation using a global empirically derived river geometry data and is then assigned to all reaches within the river system for the same time step; subsequently the discharge estimated at the mouth of the river is used to recalculate flow velocity for the highest-ordered reach, that is iteratively assigned to upstream reaches until the channel flow velocity stabilizes.
correcting the slope factor in the globally available digital elevation (DEM) model by spatially sub-setting a river network into a set of independent channels and determining a derivative of the exponential distribution relationship between the distance of each channel pixel to the pourpoint and corresponding elevation values.

7. The method as claimed in claim 1, wherein determining probability of inundation maps by the SCI-FRIM sub-module for a given day of a given lead time using the flood velocity and the corrected slope factor comprises:
generating the flood extents for different inputs within each Reach Contributing Area (RCA) corresponding to the discharge-height (q-h) relationship for each reach;
selecting Height Above Nearest Drainage (HAND) rasters from pre-processed HAND raster library based on the flood depth obtained for RCA; and
mosaicing the selected HAND rasters to obtain the flood inundation map across the entire region of interest.

8. The method as claimed in claim 1, wherein the method further comprises identifying an independent channel in the river geometry data as a set of reaches between the nearest source node for a particular junction node by the SCI-FRIM sub-module.

9. The method as claimed in claim 1, wherein the method further comprises storing a raster library of predefined thresholded HAND rasters for each RCA and corresponding look-up-table of discharge-height-velocity (q-h-v) relationships for each reach.

10. The method as claimed in claim 1, wherein the method further comprises determining an uncertainty associated with the empirically derived global channel bank full-depth values and the shape of cross-section of the river channel.

11. The method as claimed in claim 1, wherein the method further comprises pre-processing and storage of the globally available DEM for a given spatial region with catchment and channel network delineation along with RCA characterisation.

12. The method as claimed in claim 1, wherein the first remote server is a regional meteorological server, and the second remote server is a global meteorological server.

13. A system (100) for generating flood inundation forecast in near-real time for large geographical regions, comprising:
a first remote server (110) connected to a network (130);
a second remote server (120) connected to the network (130);
a computing device (150) communicably connected to the first remote server (110) and the second remote server (120) via the network (130), wherein the computing device comprises:
at least one processor (140);
at least one memory unit (145) connected with the processor (140);
a graphical user interface (147) connected with the processor (140);
a plurality of sub-modules (160) connected with the processor, and the plurality of sub-modules comprises a Slope-corrected Calibration-free Iterative Flood Routing and Inundation Model (SCIFRIM) submodule (165) connected to the network (130).