CN116070735A - A Method for Making the Yellow Sea Green Tide Distribution Area and Its Drift Prediction Initial Field Based on the Rule of Side Length and Azimuth Difference - Google Patents

Abstract

A yellow sea green tide distribution area based on side length and azimuth difference rule and a drift prediction initial field manufacturing method thereof utilize manpower to select vertexes at distribution range boundaries as prediction initial fields, and establish vertex extraction rules based on side length and azimuth difference constraint by utilizing the initial field analysis.

Description

A Method for Making the Yellow Sea Green Tide Distribution Area and Its Drift Prediction Initial Field Based on the Rule of Side Length and Azimuth Difference

technical field

This application relates to the field of model design and forecasting technology. Specifically, the application relates to a method for making the Yellow Sea green tide distribution area and its drift prediction initial field based on the rule of side length and azimuth difference.

Background technique

The distribution area information and drift trend of green tide are the main references for the deployment of green tide interception nets and the dispatch of salvage ships.

CN201911249340.7 discloses a green tide biomass forecasting method, device, equipment and medium. The method includes: determining an initially constructed green tide biomass estimation model; wherein, the green tide biomass estimation model includes at least one undetermined parameter; according to the reference distribution data of the green tide biomass in the target area, determining the initial The value of at least one undetermined parameter included in the green tide biomass estimation model constructed to obtain the green tide biomass estimation model after parameter setting, wherein the reference distribution data is determined according to satellite remote sensing images.

CN202210253275.0 discloses a method for predicting the medium and long-term trend of the Yellow Sea green tide, comprising the following steps: a. Determine the research area (33-37°N, 119-123°E) of the Yellow Sea green tide, and take 35°N as the boundary, divide It is divided into two key areas for the generation and development of green tides, and the main factors affecting the growth and drift of green tides are determined to be meteorological and oceanic factors; b obtain multi-source monitoring data of green tides and observation data of meteorological and oceanic elements in key areas; c analyze Green tide satellite discovery time, green tide main body drift direction, green tide maximum distribution area of the three indicators of the previous meteorological impact factor and sea impact factor, and respectively establish a prediction model; and marine element values, according to the prediction model established in step c, the mid- and long-term trend predictions are carried out for indicators such as the discovery time of the Yellow Sea green tide satellite, the drift direction of the main body, and the maximum distribution area in that year, and the prediction results are obtained.

Satellite remote sensing has the advantage of being instantaneous and wide-ranging. It is the main data source for green tide monitoring in the Yellow Sea and the only means to obtain complete and comprehensive green tide information. In the response to the green tide disaster in the Yellow Sea, satellite images are used to monitor the green tide coverage information, and the green tide coverage information is used to make the distribution area, and then the apex of the distribution area is used as the initial field to predict the drift of the green tide distribution.

The green tide distribution is defined as the outer envelope of the coverage area of Enteromorpha. There are two existing methods for obtaining the distribution area: the first is manual delineation, and satellite remote sensing monitors draw the distribution along the outer edge of the coverage based on the green tide coverage information noodle. The advantages of manual delineation are: the number of distribution polygon vertices is moderate, it can be directly used as the initial field, and the time-consuming calculation of drift prediction is short. The disadvantages are: strong subjectivity, no uniform standard, and different monitors draw different distribution surfaces. The second is the buffer method, which is generated using the buffer tool in the ArcGIS toolbox. Usually, the coverage area is buffered for a certain distance to form an envelope surface as the distribution area. The number of vertices in the distribution area generated by this method is huge, often ranging from tens of thousands to More than 100,000, needs to be sparsely processed. ArcGIS’s existing thinning method is to extract vertices at intervals of equal points. When the number of vertices after suction is equal to the number of vertices drawn manually, the distribution polygon will be deformed and some green tides will be missed. The coverage points need to be manually corrected. Therefore, the buffer method is used to extract distribution vertices. The advantage of this method is that it can automatically form a uniform and standardized distribution area. The disadvantage is that the number of vertices in the distribution area is huge, and there is no standardized and scientific method for suction.

Considering the advantages and disadvantages of the existing two methods of distribution area and initial field creation, this application intends to combine the advantages of ArcGIS buffer analysis and manual selection of vertices to establish an automatic vertices sparse method in consideration of both standardization and automation. . The vertex thinning method achieves that the thinned points can fully maintain the shape of the distribution, and the number of points is reduced to be equivalent to the number of manually selected points, which greatly reduces the calculation amount of drift prediction and improves the operation efficiency.

Contents of the invention

Aiming at the problems in the prior art, this application provides a method for making the green tide distribution area in the Yellow Sea and its drift prediction initial field based on the sparse rule of side length and azimuth difference. This method has three advantages in obtaining the initial field: first, the number of the number is small, which reduces the prediction time; second, the small number can fully maintain the shape of the distribution surface; Manpower, save time, improve emergency response level. The method of the present application can serve business or emergency monitoring.

A method for making the initial field of the Yellow Sea green tide distribution area and its drift prediction based on the sparse rule of side length and azimuth difference, including:

The first step is to calculate the normalized difference vegetation index NDVI using satellite images, and then use the threshold method to extract green tide coverage points,

Among them, using the standard false color image B432 (R-nir G-rB-g) combined with the normalized difference vegetation index NDVI (Normalized Difference Vegetation Index) threshold segmentation to complete the green tide disaster information extraction semi-automatically;

The second step is to use the coverage points to generate the distribution area using the buffer space analysis method, and use the radius of 1-5km to make a buffer zone for the green tide coverage points, and merge the buffer areas as the green tide distribution area;

The third step is to use the vertex sparse rule based on the edge length-orientation change constraints between adjacent vertices to sparse the distribution boundary; considering that taking a small arc distance can ensure the degree of fit after simplification, there is also a small change in the azimuth direction of the vertices. To maintain the degree of fit, use samples with the shortest arc with the same azimuth change, and limit the vertex azimuth change within 80° to form a polygonal simplified model. The formula is as follows:

In the formula, y represents the distance between adjacent vertices, and x represents the change of the tangent direction of adjacent vertices;

Step 4: Add X, Y latitude and longitude coordinates to the sparse vertices to form the initial drift prediction field.

Further, in the first step, satellites are used to collect images of the green tide initial stage, development stage, outbreak stage, and decline stage.

Further, in the first step, the normalized difference vegetation index is based on the unique spectral characteristics of the red and near-infrared bands of green algae, and the formula is: NDVI=(R _nir -R _red )/(R _nir +R _red ), where R _nir and R _red are near-infrared and red band reflectance;

Further, in the first step, use the NDVI threshold method to extract green tide information for green tide areas, carry out image histogram analysis of green tide areas, and determine the NDVI threshold, which floats around 0; usually the NDVI threshold is 0, In the case of thin cloud interference, fine-tune the NDVI threshold;

Further, the degree of coincidence between the sparse distribution polygon and the original distribution polygon is evaluated by the overall error and the algebraic error. The total area error is defined as the area of the difference between the simplified polygon and the original polygon divided by the area of the original polygon. The difference part includes the sum of increasing area S _i and decreasing area S _d . The algebraic area error is defined as the area difference between the simplified polygon and the original polygon divided by the area of the original polygon. The error calculation formula is as follows:

oe=(S _i +S _d )/S ₀

ae=(SS ₀ )/S ₀

This method of obtaining the initial field has the following advantages:

First, the number of vertices is small, which reduces the prediction time.

Second, taking both aspects of normalization and automation into consideration, it is proposed to combine the respective advantages of ArcGIS buffer analysis and manual selection of vertices to establish an automatic vertices sparse method.

Third, a small number can sufficiently maintain the shape of the distribution surface.

Fourth, using this method can automatically obtain standardized and unified results, reduce manpower, save time, and improve emergency response levels.

The distribution area of the Yellow Sea green tide and the initial field preparation method for drift prediction of the present application can be used for business or emergency monitoring.

Description of drawings

The above and/or additional aspects and advantages of the present application will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, wherein:

Figure 1 is a schematic diagram of distribution boundary vertex selection (A, small curvature boundary, B large curvature boundary)

Figure 2 is a schematic diagram of sparse pentagons into triangles

Figure 3 is a simplified distribution vertex and distribution overlay for May 17, June 6, July 9, and August 5, 2021. Details are shown on August 5, and the rest of the dates are similar

Figure 4 is a schematic diagram of the vertex angle and side length of the buffer analysis distribution polygon

Figure 5 is a schematic diagram of the results of artificial selection of the top of the green tide distribution on May 17

Figure 6 is a schematic diagram of artificial selection of azimuth difference and side length of adjacent vertices

Figure 7 shows the simplified results of the application method, equidistance method, and Douglas-Peucker method for distribution range on August 5, 2021

Fig. 8 is the algebraic area error diagram of the method of the present application, the equidistance method, and the Douglas-Puker method

Fig. 9 is the overall area error diagram of the method of the present application, the equidistance method, and the Douglas-Puke method

Detailed ways

It should be pointed out that the following detailed descriptions are all exemplary and are intended to provide further explanation to the application, and the terms used are only for describing specific implementations, and are not intended to limit the exemplary implementations based on the application.

Example 1

A method for making the initial field of the Yellow Sea green tide distribution area and its drift prediction based on the sparse rule of side length and azimuth difference, including:

The first step is to calculate the normalized difference vegetation index NDVI using satellite images, and then use the threshold method to extract green tide coverage points,

For the data and method section, satellite data are used.

The HY-1D satellite was launched on June 11, 2020 to achieve double-star networking with HY-1C. It can achieve full coverage twice in 3 days in the area where the green tide occurs in the Yellow Sea.

The coastal zone imager carried by the HY-1C/D satellite has a spatial resolution of 50m and a width of 950km, including blue (0.42μm-0.50μm), green (0.52μm-0.60μm), red (0.61μm-0.69μm), near Red (0.76μm-0.89μm) four bands can effectively monitor green tide information.

Since 2021, it has become one of the main star sources for Yellow Sea green tide business/emergency monitoring.

The research area of this application is the southern area of the Yellow Sea (South Yellow Sea). The north wind prevails in this sea area in winter and the south wind prevails in summer. Affected by the wind, the surface current in this sea area has the same pattern as the wind. The surface current flows from south to north in summer and from north to south in winter. The western part of the sea area is the North Subei radiation sandbar. The unique geological conditions and tidal laws of the North Subei radiation sandbar provide an excellent environment for the growth of Porphyra zebra, making it a famous Porphyra zebra breeding area. Laver rafts provide conditions for the growth of green tide algae. Every year at the end of April and early May, after the laver is harvested, the green algae on the rafts enter the sea and become the source of green tides, drifting northward under the combined action of the wind current field. The northern shore of the study area is Shandong Province. The green tide drifts from south to north every year, and reaches the sea area under the jurisdiction of Shandong in late June and early July, affecting the breeding industry, tourism, port shipping, and marine ecological environment in this area.

This application selects 4 phases of green tide images (see the table below) in the early stage, development period, outbreak period, and decline period for research.

Table 1 Image Information

satellite/transmission Imaging date and time Monitoring elements use HY-1D/C 2021-05-1712:42 green tide Model building and evaluation HY-1C/C 2021-06-06 10:42 green tide Model Application and Evaluation HY-1C/C 2021-07-0910:41 green tide Model Application and Evaluation HY-1C/C 2021-08-0510:39 green tide Model Application and Evaluation

The green tide in the Yellow Sea is similar to the spectrum of land green vegetation. The vegetation index method used in the extraction of land vegetation information is widely used in the extraction of green tide information, and the normalized difference vegetation index NDVI (Normalized Difference Vegetation Index) is widely used in the extraction of green tide information.

For the part of green tide information extraction, since the research area is located in the coastal waters and the water body environment is complex, in order not to miss the information of Enteromorpha as much as possible and reduce the influence of human factors, so that the result is closer to the real value, the standard false color image B432 is used (R-nir G-rB-g) combined with normalized difference vegetation index NDVI (Normalized Difference VegetationIndex) threshold segmentation to complete the extraction of Enteromorpha disaster information semi-automatically.

In false-color images, large algal blooms appear red, which provides a stronger contrast to the water than in true-color images, which facilitates visual interpretation of green tides. The NDVI is based on the unique spectrum of green algae in the red and near-infrared bands Features, the formula is: NDVI=(R _nir -R _red )/(R _nir +R _red ), where R _nir and R _red are the near-infrared and red band reflectance.

The normalized difference vegetation index can effectively enhance the green tide information and improve the contrast between the green tide and the water body, so the combination of the two can effectively identify the green tide area.

Then use the NDVI threshold method to extract green tide information for the green tide area. Extracting green tide information for green tide areas can effectively remove cloud and fog interference information and improve the accuracy of green tide information. Carry out image histogram analysis of green tide areas to determine the NDVI threshold, which floats around 0. Normally, the NDVI threshold is set to 0, and the NDVI threshold will be fine-tuned when there is interference from thin clouds and fog.

In the second step, the distribution area is generated using the coverage point using the buffer space analysis method;

The green tide distribution area is made based on the green tide coverage points using the Buffer analysis tool in ArcToolboxs in the ArcGIS desktop software, and the buffer radius is set to 3km.

The number of polygon vertices in the distribution area formed is usually tens of thousands when the distribution area of Enteromorpha is small, and often hundreds of thousands of points when the distribution area of Enteromorpha is the largest.

If it is directly used as the initial field of the distribution for drift prediction calculation, the amount of calculation is very large, and the time-consuming is too long to meet the emergency needs.

Thinning processing is required. After thinning, the number of vertices is as small as possible, and the shape of the original polygon can be effectively maintained.

The third step is to use the vertex sparse rule based on the adjacent vertex edge length-orientation difference constraint to sparse the distribution boundary;

Vertex simplification selection follows two principles. First, the boundary arc between two adjacent sample vertices should be as long as possible. The length of the arc can effectively reduce the number of vertices after simplification. Second, the line segment formed by two adjacent vertices and the simplified The matching degree of the front boundary is better, so that the difference before and after the simplification of the distribution polygon is small, and the simplification accuracy is kept high. The analysis shows that these two principles are contradictory. The arc length of adjacent vertices is too large, resulting in excessive deformation of the distribution polygon after simplification, and the shape maintenance is too strict. The boundary length of adjacent vertices must be too small. The amount of calculation is correspondingly too large.

Focus on areas of high curvature and areas of small curvature. In the boundary area with small curvature, the side length between sample points should be as long as possible to increase the simplification rate; in the boundary area with large curvature, focus on the degree of fit before and after simplification to reduce the simplification error. As shown in Figure 1, Figure A is an area with small curvature, A0 is the selected sample point, A1 is the point to be selected, and the length of the side between them is longer, the boundary curvature in Figure B is large, B0 is the selected point, and B1B2B3 is the point to be selected , in this case, in order to maintain the consistency of the boundary after simplification and the boundary before simplification, only the B1 point with the closest distance can be selected.

It can also be understood as focusing on the selection of vertices on relatively straight boundaries and curved boundaries. For relatively straight boundary boundaries, the arcs between adjacent vertices are longer, as shown in Figure 1A, A0 is the selected vertex, A1 is the vertex to be selected, the arc A0A1 is longer, and the tangent direction of the two vertices is almost Similarly, the direction changes little, and the line segment A0A1 matches the arc A0A1 highly. For a boundary with large curvature (Boundary with large curvature), the arc distance between adjacent vertices should be small. If the arc between vertices is too long, the tangent azimuth angle changes greatly, resulting in large deformation. As shown in Figure 1B, the vertex B0 has been selected, and B1, B2, and B3 are the points to be selected. If B1 is selected, then the arc B0B1 is shorter, and the angle α1 between the tangent azimuth angles of the two vertices is smaller, and the coincidence degree between the line segment B0B1 and the arc B0B1 is compared Well, B1 is an optional point. If B2 is selected, the change of azimuth angle is relatively large, the coincidence degree of line segment B0B2 and arc B0B2 is relatively poor, and point B2 is not suitable. If B3 is selected, the direction of orientation changes more, resulting in the selection of vertices to form a polygon that cannot contain algae coverage points, and B3 is unacceptable in vertex selection.

For this reason, the present application manually selects vertices in the distribution polygon generated by the buffer analysis, so that the vertices can be kept as few as possible and the distribution shape can be effectively maintained. Then analyze the distance between adjacent vertices and the angle between each inflection point and two adjacent points, and finally establish the polygonal vertex pumping model of green tide distribution.

The main factors considered in the vertex selection of artificial samples are the side length between vertices and the local curvature of the boundary. For this purpose, two corresponding feature quantities are established: the first is the side length between vertices of adjacent samples, as shown in Figure 1 The length of the side between the vertices, the size of the side length determines the size of the simplification rate. The second is the azimuth difference of adjacent sample vertices. This feature can effectively characterize the local curvature of the boundary and determine the degree of fit between the front and rear boundaries after simplification. Here, the azimuth of the sample vertex is defined as the tangent of the point to the direction of the next vertex to be selected. Azimuth. As shown in Figure 1B, the azimuth of the ray B ₀ b ₀ is the azimuth of point B ₀ .

In order to evaluate the coincidence degree between the distribution polygon after vertex sparse and the original distribution polygon, this application uses the overall error (overall error) and the algebraic area error (algebraic error) to evaluate the advantages and disadvantages of simplification. If the error is small, it means that the distribution polygon has little change after simplification, and the simplification effect is good. If the error is large, it means that the vertex of the distribution polygon changes greatly after simplification, and the simplification effect is not good. The total area error is defined as the area of the difference between the simplified polygon and the original polygon divided by the area of the original polygon. The difference part includes the sum of increasing area S _i and decreasing area S _d . The algebraic area error is defined as the area difference between the simplified polygon and the original polygon divided by the area of the original polygon. The error calculation formula is as follows:

oe＝(S _i +S _d )/S ₀ (1)

ae=(SS ₀ )/S ₀ (2)

S ₀ is the area of the polygon before simplification, S is the area of the polygon after simplification, S _i is the area of the increased area after simplification, and S _d is the area of the reduced area after simplification.

Taking Figure 2 as an example, the pentagon ABCDE becomes a triangle ACD after the vertices are sparse, then the increased area S _i here is S _△ADE , the reduced area S _d is S _△ABC , and the area of the pentagon is S _ABCDE , Then the calculation formula of the total area error oe here is (S _△ABC +S _△ADE )/S _ABCDE . In extreme cases, if only 1 or 2 vertices remain after thinning, the thinning of polygon vertices will cause the polygon to be lost, and the area change rate oe is 1 at this time.

In order to evaluate whether the number of vertices can effectively improve the prediction calculation efficiency after the thinning, the calculation time of the original initial field and the thinning initial field are calculated by using the green tide drift prediction mode of the center's business operation, and a comparative analysis is carried out.

Using the NDVI threshold method combined with expert experience to extract the green tide coverage area of the scene image, using the buffer space analysis method, the buffer radius r is set to 3km, and the green tide distribution area is obtained.

The coverage and distribution of green tides on May 17, June 6, July 9, and August 5, 2021 are shown in Figure 3. The coverage and distribution areas of the four date images are: 5.8/12242km ² , 890/44954km ² , 1210/42501km ² , 57/8615km ² , and the number of vertices is 56763, 66439, 101881, and 35510.

Taking the distribution area on May 17, 2021 as an example, analyze the corner characteristics of the distribution formed by the buffer analysis, calculate the side length of the distribution area and the angle α corresponding to the vertex, if it is greater than 180°, the angle is converted to 360°-α, and the angle can be found All are greater than 175°, concentrated in the range of 178°-180°, the distance is in the range of 0-0.2km, and most of them are distributed in the range of 0-0.1, as shown in Figure 4.

Second, the distribution polygon simplifies the model

1) Vertex samples and features

The distribution range polygon on May 17, 2021 has 56763 vertices. Manually select vertices, follow the selection of as few points as possible but keep the shape of the original distribution polygon, the edge length between vertices on the boundary with large curvature should be as small as possible, and the edge length between vertices should be increased appropriately on the boundary with small curvature. The number of selected vertices is 287 points, which greatly reduces the number of border vertices. At the same time, the polygon formed by the selected vertices is highly consistent with the polygon of the distribution range. According to the vertices manually selected by experts, the azimuth difference and side length of adjacent selected vertices are calculated. Figure 5 is a schematic diagram of the artificial selection results of the green tide distribution apex on May 17.

2) Simplified polygonal model based on azimuth difference and side length

It can be seen from Figure 6 that after manually selecting vertices, the azimuth direction of adjacent vertices changes, and the distribution of arc length between two points can be summarized as follows:

1) If the azimuth changes between adjacent vertices are small, the arc is longer, and if the azimuth changes greatly, the arc is short. The arc length between adjacent vertices is almost greater than the buffer radius of 3km.

2) There is no strict correspondence between the azimuth change and the arc length between two adjacent points, but the azimuth change and the corresponding minimum original side length can be expressed by the formula y=-0.0333x+5 when it is 0-60°. x is the azimuth change angle of adjacent selected vertices, and y is the corresponding minimum original side length.

3) The distance between two adjacent vertices with azimuth direction change greater than 60° is 3km-4km.

Based on the analysis of the above three laws, taking into account that taking a small arc distance can better ensure the degree of fit after simplification, and the small change in azimuth direction of the vertex is also conducive to maintaining the degree of fit. Using the sample with the shortest arc in the same direction change, At the same time, the orientation change of the vertices is limited within 80°, forming a polygonal simplified model. The formula is as follows:

In the formula, y represents the distance between adjacent vertices, and x changes in the direction difference of the tangent direction of adjacent vertices.

For the distribution and its initial field production process

The distribution and initial field generation scheme of this application is divided into three steps: the first step is to extract green tide coverage points, use satellite images to calculate the normalized difference vegetation index NDVI, and then use the threshold method to extract green tide coverage points; the second step is to generate distribution areas , use the coverage points to generate the distribution area by buffer space analysis, the third step is to generate the initial field for drift prediction, use the simplified polygonal model (azimuth difference-distance constraint model) studied in this application to simplify the distribution polygon, and use it as the drift prediction of the distribution area initial field. In this step, there is often not one distribution area of a scene image, so each distribution area is simplified in order. For one of the distribution areas, the vertex numbered 1 is first selected, and then the subsequent points are judged in turn, and selected according to the model rules. The second vertex, then use the newly selected vertex as the current point, select the third vertex, and so on until the distribution area vertex is selected.

For application and accuracy evaluation

Apply the simplified polygonal model of this application to simplify the distribution polygons of the images of May 17th, June 6th, July 9th, and August 5th, and the result is shown in Figure 3. The black boundary in the figure is the buffer formation distribution, and the red boundary is the simplified result. It can be seen that the simplified boundary is highly consistent with the buffer analysis distribution boundary. After simplification, the vertices are 357, 391, 602, and 225, and the simplification rates are 159, 170, 169, and 158, respectively. The total errors in the area errors are 0.79%, 0.25%, 0.42%, 0.65%; the algebraic errors are -0.07%, 0.05%, 0.13%, -0.10%. Overall, the simplification rate is high and the error is small.

Compared with the existing methods, the existing simplification method for polygons is the method of taking points at intervals

(interval method), vertical distance limit method (Vertical distance limit method), ray-barrier method (ray-barrier method), Douglas-Peucker method (Douglas-Peucker). Boundary with little fluctuation of vertices, even if the given tolerance is small, it is possible to delete all intermediate vertices on the large curved boundary, resulting in serious distortion of the shape of line features, because the buffer forms a boundary with adjacent vertex fluctuations. Large, vertical distance tolerance method and aperture method are not suitable. For this reason, the present application uses the method of the present application to conduct a comparative analysis with the interval point method and the Douglas-Puke method.

According to the same simplification rate of the simplification result of the method of this application, adjust the simplification parameters of the equidistance and Douglas-Puke method. The parameters of the equidistance method for the four dates of 5.17, 6.6, 7.9, and 8.5 are: 4.1km, 4.22km, 4.16km , 4.19km, Douglas Puke method parameters are: 0.48km, 0.46km, 0.515km, 0.48km. Simplify the distribution polygons of the four scene images. The results on August 5 are shown in Figure 7. The results on other dates are similar and will not be shown here. In Figure 7, the simplified boundary of the equidistant method is shown in blue, which is not well matched with the black distribution polygon, especially in the boundary part with large curvature. The anterior fit is good.

Calculate the overall error and algebraic error of the simplified area of the three methods, as shown in Figure 8-9. In terms of algebraic error, the method of this application is generally better than the two methods. The Douglas-Pucker method is generally better than the equidistance method. In terms of overall error, the Douglas-Pucker method is better than the two methods. The difference in the method of gamma is small, and the error of the equidistance method is relatively large.

For Drift Prediction Analysis

Huang Juan et al. established a green tide drift prediction model based on Lagrangian particle tracking method for the Yellow Sea green tide disaster (Huang Juan et al., 2011), and integrated the model into the emergency remote sensing monitoring, forecasting and early warning system of the Yellow Sea green tide disaster (Cao Conghua et al., 2017) for business applications. In order to clarify the impact of the distribution simplification operation on the prediction time, the time spent by the 72-hour drift prediction system before and after the distribution simplification was counted (see Table 2). It can be seen that the time used for the distribution drift prediction after the simplification of the four periods is significantly reduced, especially in the distribution of the most serious outbreak period of the green tide disaster on July 9, the drift prediction time is saved by about 27 minutes, which is the best for the green tide disaster interception and salvage program. Formulating saves valuable time.

Table 2 Drift prediction running time before and after simplification of green tide distribution

Therefore, this application can give the application of the interval model, which shows that the interval model is fully capable of extracting the initial field.

Claims (4)

1. A yellow sea green tide distribution area based on side length and azimuth difference rules and a drift prediction initial field manufacturing method thereof comprise the following steps:

step 1, calculating a normalized vegetation index NDVI by using a satellite image, extracting green tide coverage points by using a threshold method,

the green tide disaster information extraction is completed semi-automatically by combining standard false color image B432 (R-nirG-rB-g) with normalized vegetation index NDVI (NormalizedDifferenceVegetationIndex) threshold segmentation;

step 2, generating a distribution area by using a buffer space analysis method by using the coverage points, manufacturing a buffer area for the green tide coverage points with the radius of 1-5km, and combining the buffer areas to be used as the green tide distribution area;

step 3, sparse the distributed boundaries by using a distance-angle constraint-based vertex sparse rule (DACR, distance-angle constraintrule); considering that the reduced camber line distance can be taken to ensure the reduced fitness, the small change of the azimuth direction of the vertex is also beneficial to the maintenance of the fitness, the sample with the shortest camber line is changed by the same azimuth direction, and meanwhile, the azimuth direction change of the vertex is limited within 80 degrees, so that a polygonal simplified model is formed, and the formula is as follows:

wherein y represents the distance between adjacent vertexes, and the tangential direction difference of the adjacent vertexes is changed;

and step 4, adding XY longitude and latitude coordinates to the sparse vertexes to form a drift prediction initial field.

2. The method of claim 1, wherein in step 1, satellite is used to capture images of the early green tide, the development period, the outbreak period, and the decay period.

3. The method of claim 1, wherein in step 1, the normalized vegetation index is based on unique spectral characteristics of the green algae red band and the near infrared band, and the formula is: ndvi= (R _nir -R _red )/(R _nir +R _red ) Wherein R is _nir 、R _red Is the reflectivity of near infrared and red wave bands.

4. A method according to any one of claims 1 to 3, wherein in step 1, green tide information is extracted for a green tide region by using an NDVI threshold method, and a green tide region image histogram analysis is performed to determine an NDVI threshold value, the threshold value being floating around 0; the NDVI threshold is typically taken to be 0, and is fine-tuned in the presence of cloud-to-mist interference.

Images