Spatial simulation method for assessment of direct economic losses of typhoon flood based on remote sensing

This application claims priority to Chinese Patent Application No. 202310033916.6, filed on Jan. 10, 2023, the contents of which are hereby incorporated by reference.

The disclosure relates to a technical field of remote sensing analysis of typhoon flood disasters, and in particular to a spatial simulation method for the assessment of direct economic losses of typhoon flood based on a remote sensing.

Assessing direct economic losses is an important aspect of disaster emergency management. Scientifically, accurately, and objectively understanding the spatial distribution of economic losses related to hazard-bearing bodies provides vital guidance for disaster response decision-making and post-disaster reconstruction. Typhoon flood has the characteristics of wide influence range, strong destruction intensity and long duration, and primarily affect mostly economically developed areas, and resulting in substantial property losses annually.

The direct economic losses of typhoon flood hazard-bearing bodies encompasses the expenses associated with restoring or repairing various types of hazard-bearing bodies to their pre-disaster functionality and condition after being damaged by a typhoon flood. At present, the direct economic losses of typhoon flood hazard-bearing bodies is mostly reported at the administrative boundary level providing valuable insights into the extent of disaster-related losses for countries and regions. However, this reporting method is susceptible to significant errors due to differences in statistical criteria and human subjectivity. It also fails to account for the variations in the distribution of disaster losses intensity within administrative boundaries. Therefore, the statistical method restricts the scientific accuracy of assessing typhoon flood losses to some degree and falls short of meeting the precision requirements of flood catastrophe insurance assessment and claims settlement. At present, there are two main types of models used for simulating the direct economic losses of typhoon flood hazard-bearing bodies: statistical models and multi-index evaluation methods. The simulation of direct economic losses in typhoon flood hazard-bearing bodies involves numerous indicators that exhibit strong spatial correlations. Additionally, the determination of weights for these multiple indicators is often influenced by human subjectivity. The intelligent machine learning method represented by deep network may explore a complex nonlinear correlation mechanism between multi-feature indicators and the direct economic losses of typhoon flood hazard-bearing bodies by learning through training samples. In view of the above drawbacks, there is a need for improved systems and methods for spatial simulation of direct economic losses in typhoon flood hazards.

To address the significant estimation errors associated with existing methods for simulating the direct economic losses of typhoon flood hazard-bearing bodies, this invention presents a spatial simulation method reliant on remote sensing for assessing direct economic losses resulting from typhoon floods.

The field of disaster assessment and forecasting has seen various methods and technologies aimed at enhancing our understanding of catastrophic events and their economic impact. Notably, these methods have different focuses and applications compared to the spatial simulation of direct economic losses due to typhoon floods based on remote sensing, which is the core aim of the present invention.

The relevant prior art references collectively demonstrate the diversity of methods and technologies employed in the field of disaster assessment and forecasting with each reference offering distinct approaches, focuses, and applications.

Among these references, CN112819340A delves into urban flood assessments using multi-source data, focusing on dynamic assessment in urban areas. Building upon this, KR20210085592A addresses the complexities of mega flood and dam breach simulations caused by continuous rainfall, primarily emphasizing precise rainfall-runoff analysis and efficient flood preparation. In a related vein, CN113723849A introduces a method for flood event and risk identification based on spatial-temporal characteristics, focusing on assessing flood disaster situations and identifying risks. CN114266502A takes an innovative approach by concentrating on flood disaster losses assessment and simulation deduction using satellite remote sensing and sensor data, enhancing losses evaluation accuracy. Meanwhile, U.S. Ser. No. 11/521,379B1 leverages vision transformer technology for flood disaster monitoring and analysis, emphasizing deep learning and radar data. Further, US20070168155A1 pioneers a statistical-deterministic approach to predict natural disasters, providing valuable insights into disaster prediction. Lastly, US20220268963A1 contributes a system and method for environment-dependent probabilistic tropical cyclone modeling, offering a distinct perspective by focusing on modeling tropical cyclones and assessing cyclone risk.

None of the cited prior art references disclose the present invention's method for spatially simulating method for the assessment of direct economic losses resulting from typhoon floods based on remote sensing. The present invention uniquely addresses the shortcomings of existing methods and presents an innovative approach to assessing economic losses in the context of typhoon floods utilizing remote sensing and deep learning techniques for more accurate and objective results.

The present invention provides a spatial simulation method for the assessment of direct economic losses of typhoon flood based on a remote sensing implemented in a computer system using a set of computer-executable instructions includes following steps:

S 1 , collecting a multi-source data before and after a typhoon flood disaster in an affected area and preprocessing the said data to obtain a preprocessed data; S 2 , extracting reset parameters of a hazard-bearing body according to the preprocessed data, and constructing a feature set of the reset parameters of the hazard-bearing body; S 3 , collecting insurance disaster assessment data and historical investigation data in the typhoon flood affected area, and constructing a sample set of direct economic losses of the hazard-bearing body; and S 4 , carrying out a spatial simulation on the direct economic losses of the typhoon flood hazard-bearing body by adopting a deep learning neural network model according to the feature set of the reset parameters of the typhoon flood hazard-bearing body and the sample set of the direct economic losses of the hazard-bearing body.

Further, the multi-source data in the S 1 includes time series remote sensing data, typhoon meteorological index data, underlying surface risk exposure data and typhoon flood disaster losses data.

Further, a method for collecting and preprocessing the time series remote sensing data includes following steps:

Step A 1 , obtaining radar remote sensing data of Gaofen-3 and Sentinel-1 and optical remote sensing data of Gaofen-1, Gaofen-6 and Sentinel-2 before and after disaster in the typhoon flood affected area, where the optical remote sensing data selects cloud-free and snow-free data; Step A 2 , performing an orbit correction, an image clipping, a filtering processing, a radiometric calibration, a terrain correction, a geocoding and an image mosaic on the radar remote sensing data to obtain preprocessed radar remote sensing data; Step A 3 , performing a radiation calibration, an atmospheric correction, a geometric correction, a band registration, a multi-band image synthesis and an image fusion on the optical remote sensing data to obtain preprocessed optical remote sensing data; and Step A 4 , unifying the preprocessed radar remote sensing data and the preprocessed optical remote sensing data to 30-meter resolution through an image registration and a resampling, and completing a preprocessing on the time series remote sensing data.

Further, a method for collecting and preprocessing the typhoon meteorological index data includes following steps:

Step B 1 , selecting a precipitation product of GPM (Global Precipitation Mission) IMERG (Integrated Multi-satellitE Retrievals for GPM), downloading product data sets of time periods before and after the disaster, reading and generating tiff files through ArcGIS, and resampling to 30-meter spatial resolution through a bilinear interpolation method; and Step B 2 , selecting a near real-time product data set of Meteorological Administration Office, using python to batch process original files of the near real-time product data set and converting the original files of the near real-time product data set into a raster data set, and resampling images in the raster data set to 30-meter spatial resolution by the bilinear interpolation method, and completing a preprocessing on the typhoon meteorological index data.

Further, a method for collecting and preprocessing the underlying surface risk exposure data includes following steps:

Step C 1 , obtaining latest fine land use data at 10 m global spatial resolution with secondary classification; Step C 2 , obtaining hazard-bearing body cost data from the relevant Economic and Social Development Statistical Database; Step C 3 , selecting cloud-free, geometrically and radiatively corrected NPP-VIIRS or sustainable development satellite night light remote sensing data; Step C 4 , filtering the night light remote sensing data, eliminating noise and interfering light sources in original night light remote sensing data, and performing a median synthesis on filtered night light remote sensing data set to obtain stable night light remote sensing data; and Step C 5 , resampling the stable night lighting remote sensing data to 30-meter resolution, and completing a preprocessing on the underlying surface risk exposure data.

Further, the reset parameters of the typhoon flood hazard-bearing body in the S 2 include a flood inundation maximum range feature V 1 , a flood inundation depth feature V 2 , a flood inundation duration feature V 3 , hourly maximum rainfall V 4 , daily maximum rainfall V 5 , a maximum wind speed V 6 , a maximum atmospheric pressure V 7 and risk exposure feature V 8 ;

a formula for calculating the flood inundation maximum range feature V 1 is:

Further, the feature set of the reset parameters of the typhoon flood hazard-bearing body constructed in the S 2 is:

Further, the S 3 includes following sub-steps:

S 31 , collecting the insurance disaster assessment data and the historical investigation data in the typhoon flood affected area, dividing according to a 30 m×30 m grid, and selecting grids with more than five historical investigation data items as sample grids to form samples of the direct economic losses of the typhoon flood; S 32 , determining direct economic losses P with different property types per unit area or quantity by a statistical average method according to the historical investigation data in the typhoon flood affected area; S 33 , obtaining an affected area or quantity Q of each property type according to a comparison of high-resolution satellite remote sensing data before and after the disaster; and S 34 , constructing the sample set Y of the direct economic losses of the hazard-bearing body according to the direct economic losses P and the affected area or quantity Q:

Further, the S 4 includes following sub-steps:

S 41 , correspondingly extracting the reset parameters of the typhoon flood hazard-bearing body of 30 m×30 m grids from the feature set of the reset parameters of the typhoon flood hazard-bearing body according to spatial grid positions of the samples of the direct economic losses of the typhoon flood; S 42 , normalizing all reset parameters of the typhoon flood hazard-bearing body:

The disclosure has following beneficial effects:

(1) The disclosure integrates various reset parameters of typhoon flood, and by constructing an intelligent simulation model, the simulation results may spread spatial differences of direct economic losses on the grid scale, thus breaking through the bottleneck problem that the original investigation error of typhoon flood losses based on administrative boundaries is large.

(2) The disclosure adopts multi-source remote sensing and meteorological data to extract the reset parameters of the typhoon flood hazard-bearing body, and constructs the feature set of the reset parameters of the typhoon flood hazard-bearing body, the parameter types are various and the extraction accuracy is high.

(3) The intelligent simulation model of the typhoon flood hazard-bearing body is constructed by the deep learning neural network algorithm, has high spatial applicability and may deeply explore the complex correlation mechanism between multiple reset parameters and the direct economic losses of the typhoon flood hazard-bearing body.

FIG. 1 is a flowchart of a spatial simulation method for a direct economic loss of a typhoon flood based on a remote sensing provided by an embodiment of the present disclosure.

FIG. 2 is a schematic diagram of a remote sensing simulation effect of the direct economic losses of the typhoon flood hazard-bearing body provided by the embodiment of the present disclosure.

FIG. 3 illustrates a preprocessing of time series remote sensing data, encompassing radar and optical data integration and correction processes.

FIG. 4 illustrates steps involved in collecting and preprocessing typhoon meteorological index data, including GPM IMERG precipitation data and near real-time product data.

FIG. 5 illustrates procedures for obtaining and preprocessing underlying surface risk exposure data, involving fine land use data, hazard-bearing body cost data, and stable night light remote sensing data.

FIG. 6 illustrates a collection and processing of insurance disaster assessment data and historical investigation data, encompassing grid-based sample selection and direct economic losses determination.

FIG. 7 illustrates a spatial simulation process utilizing a deep learning neural network model, covering reset parameter extraction, normalization, input training sample construction, model configuration, loss function definition, optimization, evaluation, and final spatial simulation calculation.

FIG. 8 shows a schematic structural diagram of spatial simulation method.

FIG. 9 shows a schematic structural diagram of Radar image processing.

FIG. 10 shows a schematic structural diagram of Optical image processing.

The detailed description of the embodiments provides an in-depth exploration of the spatial simulation method for assessing direct economic losses resulting from typhoon floods based on remote sensing technology implemented in a computer system using a set of computer-executable instructions. It outlines the steps, procedures, and key features of the disclosure, including the collection and preprocessing of various data sources, the construction of feature sets, and the implementation of deep learning neural network models.

Throughout the specification, the word “comprise” or variations such as “comprises” or “comprising” will be understood to imply the inclusion of a stated integer or groups of integers but not the exclusion of any other integer or groups of integers.

As used herein, the following definitions shall apply unless otherwise indicated.

A “spatial simulation method” refers to a technique or process for creating a computer-based model or simulation that replicates and analyzes the behavior, interactions, or changes of objects, phenomena, or events within a defined physical or geographical space.

A “hazard-bearing body” refers to an entity, or organization that carries or is associated with a hazard, which is something that poses the potential for harm, damage, or danger to individuals, assets, or the environment.

The terms “Gaofen-1”. “Gaofen-6”, and “Sentinel-2” are Earth observation satellites used for remote sensing and data collection.

The “Global Precipitation Mission (GPM)” is a joint mission between NASA (the National Aeronautics and Space Administration) and the Japan Aerospace Exploration Agency (JAXA) that aims to provide global observations of precipitation, including rain and snow. GPM uses a constellation of satellites to collect data on precipitation from around the world, allowing for the monitoring of rainfall and snowfall patterns on a global scale.

The “Integrated Multi-satellite Retrievals for GPM (IMERG)” refers to a product developed as part of the Global Precipitation Measurement (GPM) mission. IMERG is a dataset that combines precipitation data from multiple satellites to create a global and near-real-time estimate of rainfall. It provides information about precipitation rates and accumulations on a global scale.

The term “Meteorological Office” typically refers to a government agency or organization responsible for monitoring and providing weather forecasts and other meteorological information for a specific region or country.

The term “ArcGIS” is derived from the term “geographic information system,” emphasizing its focus on the creation, analysis, and management of geographic or spatial information. The “Arc” in ArcGIS originally referred to the software's use of vector-based graphics and the use of arcs to represent geographic features in early versions.

The “Economic and Social Development Statistical Database” typically refers to a collection of data and statistics related to various aspects of a country's or region's economic and social development. These databases are often maintained by government agencies, international organizations, or research institutions and contain a wide range of information on economic indicators, social demographics, and other relevant data.

Exemplary embodiments of the present disclosure now are described in detail with reference to the accompanying drawings. It should be understood that the embodiments shown and described in the drawings are merely exemplary, are intended to illustrate the principles and spirit of the disclosure, and not to limit the scope of the disclosure.

The embodiment of the disclosure provides a spatial simulation method for a direct economic loss of a typhoon flood based on a remote sensing, as shown in FIG. 1 , including following steps of S 1 to S 4 :

S 1 , collecting multi-source data before and after a disaster in a typhoon flood affected area and preprocessing to obtain preprocessed data.

In the embodiment of the disclosure, the multi-source data are mainly used for extracting flood inundation features and meteorological disaster indexes, performing a heterogeneous spatial distribution of property direct economic losses information and sorting out direct economic losses investigation data, and includes time series remote sensing data, typhoon meteorological index data, underlying surface risk exposure data and typhoon flood disaster losses data.

As shown in FIG. 3 , a method for collecting and preprocessing the time series remote sensing data is as follows:

Step A 1 , obtaining radar remote sensing data of Gaofen-3 and Sentinel-1 and optical remote sensing data of Gaofen-1, Gaofen-6 and Sentinel-2 before and after disaster in the typhoon flood affected area, where the optical remote sensing data selects cloud-free and snow-free data; Gaofen-3 and Sentinel-1 satellites are equipped with a C-band synthetic aperture radar, have multiple imaging modes and a resolution of up to 1 m. The optical remote sensing data of Gaofen-1, Gaofen-6 and Sentinel-2 include 2 m panchromatic/8 m multispectral, 10 m, 20 m and 60 m multispectral images. Step A 2 , performing an orbit correction, an image clipping, a filtering processing, a radiometric calibration, a terrain correction, a geocoding and an image mosaic on the radar remote sensing data to obtain preprocessed radar remote sensing data. Step A 3 , performing a radiation calibration, an atmospheric correction, a geometric correction, a band registration, a multi-band image synthesis and an image fusion on the optical remote sensing data to obtain preprocessed optical remote sensing data. Step A 4 , unifying the preprocessed radar remote sensing data and the preprocessed optical remote sensing data to 30-meter resolution through an image registration and a resampling, and completing a preprocessing on the time series remote sensing data.

As shown in FIG. 4 , a method for collecting and preprocessing the typhoon meteorological index data is as follows:

Step B 1 , selecting a precipitation product of GPM (Global Precipitation Mission) IMERG (Integrated Multi-satellitE Retrievals for GPM), downloading product data sets of time periods before and after the disaster, reading and generating tiff files through ArcGIS, and resampling to 30-meter spatial resolution through a bilinear interpolation method.

The precipitation product of IMERG has high precision, the highest time resolution is once every half hour, and the spatial resolution is 0.1°×0.1° (10 km).

Step B 2 , selecting a near real-time product data set of Meteorological Administration Office, using python to batch process original files of the near real-time product data set and converting the original files of the near real-time product data set into a raster data set, and resampling images in the raster data set to 30-meter spatial resolution by the bilinear interpolation method, and completing a preprocessing on the typhoon meteorological index data.

The near real-time products of CLDAS-V2.0 include wind speed, ground pressure, temperature and other products, and the spatial resolution is 0.0625°×0.0625° (6750 m), and is updated every 3 hours.

As shown in FIG. 5 , a method for collecting and preprocessing the underlying surface risk exposure data is as follows:

Step C 1 , obtaining latest fine land use data at 10 m global spatial resolution with secondary classification. Step C 2 , obtaining hazard-bearing body cost data from the Statistical Database. Step C 3 , selecting cloud-free, geometrically and radiatively corrected NPP-VIIRS or sustainable development satellite night light remote sensing data. Step C 4 , filtering the night light remote sensing data, eliminating noise and interfering light sources in original night light remote sensing data, and performing a median synthesis on filtered night light remote sensing data set to obtain stable night light remote sensing data. Step C 5 , resampling the stable night lighting remote sensing data to 30-meter resolution, and completing a preprocessing on the underlying surface risk exposure data.

The losses data of typhoon flood disaster is obtained through the investigation of property insurance disaster assessment data and field sample data, and contains information such as the location, category, damage degree and direct economic losses of each hazard-bearing body.

S 2 , extracting reset parameters of a typhoon flood hazard-bearing body according to the preprocessed data, and constructing a feature set of the reset parameters of the typhoon flood hazard-bearing body.

In the embodiment of the disclosure, the reset parameters of the typhoon flood hazard-bearing body in the S 2 include a flood inundation maximum range feature V 1 , a flood inundation depth feature V 2 , a flood inundation duration feature V 3 , hourly maximum rainfall V 4 , daily maximum rainfall V 5 , a maximum wind speed V 6 , a maximum atmospheric pressure V 7 and risk exposure feature V 8 . The flood inundation maximum range feature V 1 , the flood inundation depth feature V 2 and the flood inundation duration feature V 3 are flood inundation features, the hourly maximum rainfall V 4 , the daily maximum rainfall V 5 , the maximum wind speed V 6 and the maximum atmospheric pressure V 7 are rainstorm meteorological disaster-causing features, and the risk exposure feature V 8 is the underlying surface risk exposure feature.

A formula for calculating the flood inundation maximum range feature V 1 is:

A formula for calculating the flood inundation depth feature V 2 is:

A formula for calculating the flood inundation duration feature V 3 is:

Formulas for calculating the hourly maximum rainfall V 4 are:

Formulas for calculating the daily maximum rainfall V 5 are:

In the embodiment of the disclosure, according to the preprocessed IMERG precipitation product data set N={N 1 , N 2 , . . . , N n }, where n is the number of collected precipitation product images, an hourly rainfall data set {RH} and a daily rainfall data set {RD} are respectively generated through aggregation analysis, where the hourly rainfall data set is an image product representing one-hour rainfall by superimposing two adjacent rainfall products.

A formula for calculating the maximum wind speed V 6 is:

A formula for calculating the maximum atmospheric pressure V 7 is:

A formula for calculating the risk exposure feature V 8 is:

In the embodiment of the disclosure, a linear regression model is adopted to simulate the risk exposure features of the research area, and the fitting model formula is as follows:

In the embodiment of the disclosure, according to the obtained land use data with 10 m resolution and the cost data of the disaster-stricken body, the samples of the direct economic losses of the typhoon flood are obtained by using a spatial aggregation method, the sample points are 30 m×30 m unit grids, and the direct economic losses of each sample point is as follows:

Finally, the constructed feature set of the reset parameters of the typhoon flood hazard-bearing body is:

S 3 , collecting insurance disaster assessment data and historical investigation data in the typhoon flood affected area, and constructing a sample set of a direct economic losses of a hazard-bearing body.

As shown in FIG. 6 , the S 3 includes following sub-steps: S 31 -S 34 :

S 31 , collecting the insurance disaster assessment data and the historical investigation data in the typhoon flood affected area, dividing according to a 30 m×30 m grid, and selecting grids with more than five historical investigation data items as sample grids to form samples of the direct economic losses of the typhoon flood. In the embodiment of the disclosure, at least 5000 samples of the direct economic losses of typhoon flood are formed. S 32 , determining direct economic losses P with different property types (farmland, industrial and mining, family, infrastructure and public facilities in the embodiment of the disclosure) per unit area or quantity by a statistical average method according to the historical investigation data in the typhoon flood affected area. S 33 , obtaining an affected area or quantity Q of each property type according to a comparison of high-resolution satellite remote sensing data before and after the disaster. S 34 , constructing the sample set Y of the direct economic losses of the hazard-bearing body according to the direct economic losses P and the affected area or quantity Q:

S 4 , carrying out a spatial simulation on a direct economic losses of the typhoon flood hazard-bearing body by adopting a deep learning neural network model according to the feature set of the reset parameters of the typhoon flood hazard-bearing body and the sample set of the direct economic losses of the hazard-bearing body.

As shown in FIG. 7 , the S 4 includes following sub-steps: S 41 -S 48 :

S 41 : correspondingly extracting the reset parameters of the typhoon flood hazard-bearing body of 30 m×30 m grids from the feature set of the reset parameters of the typhoon flood hazard-bearing body according to spatial grid positions of the 5000 samples of the direct economic losses of the typhoon flood. S 42 , in order to eliminate the dimensional effect of each index and unify the change range of each index, normalizing all reset parameters of the typhoon flood hazard-bearing body, with the range of 0-1, and the normalization formula is as follows:

In the embodiment of the disclosure, the deep learning neural network model is set with four hidden layers, and the learning rate η is set to 10 −4 . The capacity of training samples may have a certain impact on the model accuracy. In the embodiment of the disclosure, 80% of the samples (4000 samples) are selected as model training samples by a random method, and the remaining 20% (1000 samples) are used as model verification samples.

S 47 , evaluating the trained deep learning neural network model by verifying the samples, and verifying the accuracy of the model; where evaluation indexes are an average absolute error MAE and an accuracy ACC:

Those skilled in the art will realize that the embodiments described here are to help readers understand the principles of the disclosure, and it should be understood that the scope of protection of the disclosure is not limited to such special statements and embodiments. Those skilled in the art may make various other specific variations and combinations according to the technical teachings disclosed in the present disclosure without departing from the essence of the present disclosure, and these variations and combinations are still within the protection scope of the present disclosure.