A method for remote sensing determination of spartina alterniflora removal time

Abstract

The application discloses a kind of remote sensing determination methods of Spartina alterniflora removal time, characterized in that the high-resolution No.1 image in the continuous time period before and after Spartina alterniflora removal covering monitoring area is collected, and the four indexes corresponding to each image pixel are calculated;Remove invalid pixels in each image one by one and construct effective pixel index time series;Using harmonic function to establish the observation model of each effective pixel index before Spartina alterniflora removal, calculate the change amplitude of each effective pixel in the Spartina alterniflora removal time period, and determine the change detection threshold;By comparing the size of the change amplitude of each effective pixel and the change detection threshold, determine whether Spartina alterniflora is removed, if removed, determine the removal time;The advantage is that the method can realize accurate monitoring of Spartina alterniflora removal process in a long period of time, realize the automatic determination of Spartina alterniflora removal time and range.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing monitoring of plants, and particularly relates to a remote sensing determination method for Spartina alterniflora removal time. BACKGROUND

[0002] As a typical invasive alien species, Spartina alterniflora has strong reproduction and diffusion ability, and seriously occupies the living space of native vegetation, resulting in a series of problems such as estuary channel siltation, wetland ecological function degradation, biological diversity sharp reduction and hydrological dynamic environment change. At present, large-scale Spartina alterniflora removal projects are being carried out in coastal areas of China, mainly involving physical control measures such as mowing and plowing. After mowing and plowing, the ground surface is close to bare beach, which is obviously different from the ground surface covered by Spartina alterniflora. Under this background, accurate and rapid acquisition of information such as Spartina alterniflora removal time and range through mowing and plowing can provide important decision support for evaluating the treatment effect and optimizing the later management and control.

[0003] However, the traditional method of obtaining the removal time and range of Spartina alterniflora through field investigation has low work efficiency and high cost, and it is difficult to carry out application in a large area and at a high frequency. Remote sensing technology provides an important means for realizing large-scale, long-term and continuous earth observation. However, the remote sensing monitoring of the Spartina alterniflora removal process essentially belongs to the category of change detection. Traditional remote sensing change detection is mainly based on image difference and ratio operation methods, and the difference is detected through multi-temporal comparison. However, due to the influence of vegetation phenology, light conditions and atmospheric conditions, the false detection rate of the above methods is high. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a remote sensing determination method for Spartina alterniflora removal time, which can accurately monitor the removal process of Spartina alterniflora and realize automatic determination of the removal time and range of Spartina alterniflora.

[0005] The technical solution adopted by the present application to solve the above technical problem is as follows: a remote sensing determination method for Spartina alterniflora removal time, comprising the following specific steps:

[0006] (1) Collect high-resolution 1 images covering the monitoring area and in the continuous time period before and after the removal of Spartina alterniflora, and pre-process all high-resolution 1 images respectively;

[0007] (2) Calculate the normalized difference vegetation index NDVI, enhanced vegetation index EVI, soil-adjusted vegetation index SAVI and normalized difference water index NDWI corresponding to the pixels of each image based on the pre-processed image;

[0008] (3) Remove invalid pixels in each image one by one, and sort them in the order of observation time to construct an effective pixel time series;

[0009] (4) Using the harmonic function to fit the effective pixel index time series of Spartina alterniflora before removal, respectively establishing the observation model of each effective pixel index NDVI, EVI and SAVI, and solving the undetermined parameters in the observation model;

[0010] (5) Calculating the root mean square error of the observation model and the change amplitude of each effective pixel in the effective pixel index time series in the Spartina alterniflora removal period, and determining the change detection threshold by using the chi-square distribution;

[0011] (6) By comparing the change amplitude of the effective pixel in the Spartina alterniflora removal period with the change detection threshold, it is determined whether the Spartina alterniflora in the region corresponding to the effective pixel is removed or not, if removed, the removal time is determined, otherwise, the observation model is updated and steps (4)-(6) are repeated.

[0012] Further, the pre-processing of the GF-1 image in step (1) includes geometric correction, radiation correction, splicing and cutting, and the cloud mask file in the GF-1 image is geometrically corrected, specifically: using the same named ground feature points of the GF-1 image corresponding to the cloud mask file for geometric correction of the cloud mask file to ensure that the pre-processed cloud mask file is accurately matched with the spatial position of the GF-1 image.

[0013] Further, in step (3), the invalid pixels include cloud and cloud shadow pixels, index abnormal pixels and tide inundation pixels, and the removal method is:

[0014] Removal of cloud and cloud shadow pixels: using the geometrically corrected cloud mask file to remove the pixels covered by cloud and cloud shadow in the image;

[0015] Removal of index abnormal pixels: the effective values of index NDVI, EVI, SAVI and NDWI are located in [-1, 1], the pixels corresponding to the index abnormal values less than -1 or greater than 1 are removed;

[0016] Removal of tide inundation pixels: visually find out the pixels in the image whose Spartina alterniflora boundary is inundated by tide and not inundated, and obtain the NDWI values of these pixels, then obtain the best threshold of NDWI for segmenting the two states of inundation and non-inundation through the ROC curve, remove the pixels corresponding to the NDWI values greater than or equal to the best threshold, so as to suppress the influence of tide fluctuation change on the determination of Spartina alterniflora removal.

[0017] Further, in step (4), the observation model is:

[0018] ,

[0019] in: This indicates that a certain effective cell is in the th... t Its index at the time of the second observation i The model's predicted values, i These are NDVI, EVI, and SAVI, respectively. w k For the first k The frequency of the first harmonic function is set as a constant. ; k The value is either 1 or 2, based on the number of observations of that effective pixel in the effective pixel index time series before Spartina alterniflora removal. m Determine if the number of observations m ≥10, then k =2, otherwise k =1; , For the first k The first harmonic function coefficients represent the exponent. i The cyclical trend of change; c i The intercept represents the exponent. i The overall level of change.

[0020] Furthermore, in step (4), the undetermined parameters in the observation model are: , and c i The solution is obtained using the lasso regression algorithm.

[0021] Furthermore, in step (5), the formula for calculating the root mean square error of the observation model is:

[0022] ,

[0023] ,

[0024] in: RMSE i The index of a certain effective cell i The root mean square error of the observation model, m The number of observations for that effective pixel in the effective pixel index time series before Spartina alterniflora removal. p The index of the effective pixel i The number of undetermined parameters in the observation model; residual ( i,t ) is the index of the effective pixel. i In the t The residual at the time of the second observation is determined by the index of that effective pixel. i In the t The actual observed value at the time of the observation index( i,t ) and model predictions The difference is obtained, t =1, 2, ..., m Actual observed values index ( i,t ) is the index of the effective pixel. i In the t The value at the time of the observation.

[0025] Furthermore, in step (5), the formula for calculating the magnitude of change is:

[0026] ,

[0027] in: magnitude ( T The index of effective pixels in the Spartina alterniflora eradication period represents the change in the effective pixel index of a specific effective pixel in the time series observed on the Tth observation, where T = 1, 2, ... n ; residual ( i,T ) is the index of the effective pixel. i The residual at the Tth observation is determined by the exponent of that effective pixel. i Actual observed value at the Tth observation index ( i,T ) and model predictions The difference is obtained from the actual observed value. index ( i,T ) is the index of the effective pixel. i The value at the Tth observation, the model prediction value The observation model in step (4) is obtained by substituting the time of the Tth observation into the observation model.

[0028] Furthermore, in step (5), the change detection threshold is set to a chi-square distribution value with 3 degrees of freedom at a 95% confidence level, i.e.:

[0029] ,

[0030] in: threshold change Indicates the change detection threshold. This represents a chi-square distribution with 3 degrees of freedom, corresponding to 3 types of indices: NDVI, EVI, and SAVI.

[0031] Furthermore, in step (6), the method for determining whether Spartina alterniflora has been completely cleared is as follows:

[0032] If the variation of the effective pixel in the effective pixel index time series in the Spartina alterniflora removal period is greater than the variation detection threshold in the first n observations, it is determined that the Spartina alterniflora in the region corresponding to the effective pixel has been removed, and the first observation of the effective pixel in the Spartina alterniflora removal period is the start time of the Spartina alterniflora removal in the region.

[0033] Otherwise, the index of the first observation of the effective pixel in the Spartina alterniflora removal period is included in the observation model of step (4) to update the model, and steps (4)-(6) are repeated until the variation of the effective pixel in the Spartina alterniflora removal period is greater than the variation detection threshold in the first n observations, or all the indexes of the effective pixel in the observation model are included in the effective pixel index time series before and after the Spartina alterniflora removal.

[0034] Compared with the prior art, the present application has the following advantages:

[0035] (1) The present method can accurately monitor the Spartina alterniflora removal process over a long period of time, and automatically determine the Spartina alterniflora removal time and range;

[0036] (2) The present method takes advantage of the high spatial and temporal resolution of high-resolution satellites, and combines multiple remote sensing indexes to enhance the monitoring ability of coastal vegetation changes and reduce the misjudgment caused by vegetation phenology changes and tidal fluctuation disturbance;

[0037] (3) The present method introduces harmonic function and lasso regression method, which can accurately detect the mowing and plowing removal events of Spartina alterniflora by fitting and parameter optimization of time series pixel by pixel. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 The flowchart of the present application;

[0039] Figure 2 The process chart of the present application for removing the tidal inundation pixels and constructing the effective pixel index time series;

[0040] Figure 3 The time and range distribution chart of mowing and plowing removal of Spartina alterniflora in Ningbo City from 2022 to 2024 obtained by the present application. DETAILED DESCRIPTION

[0041] The present application will be further described in detail below with reference to the embodiments of the drawings.

[0042] As shown in FIG. 1, the monitoring area takes Ningbo City as an example, a remote sensing method for determining the Spartina alterniflora removal time, comprising the following specific steps:

[0043] (1) Collect GF-1 images covering the area where the cloud cover is less than 40% in Ningbo City from 2020 to 2024, among which 2020-2021 is before the removal of Spartina alterniflora, and 2022-2024 is during the removal of Spartina alterniflora. All GF-1 images are preprocessed by geometric correction, radiometric correction, mosaicking and cutting, and the cloud mask file in GF-1 images is geometrically corrected, specifically:

[0044] Firstly, based on the Sentinel-2 composite image of Ningbo City, the collected GF-1 images are spatially registered with it, and the GF-1 images are projected into the coordinate WGS_1984_UTM_Zone_51N and resampled to 8 m. In this process, the vector data of the cloud mask file is converted to raster data and added as an additional band to the GF-1 image, ensuring that the cloud mask file uses the same ground feature points for geometric correction. Secondly, the FLAASH module of ENVI software is used to perform radiometric correction on the geometrically corrected GF-1 images. Then, the GF-1 images of the same time are mosaicked and cut to obtain multiple images of Ningbo City from 2020 to 2024, one image representing one observation. Among them, there are 14 images in 2020, 13 images in 2021, 24 images in 2022, 25 images in 2023, and 15 images in 2024;

[0045] (2) Based on each preprocessed image, the normalized difference vegetation index NDVI, the enhanced vegetation index EVI, the soil-adjusted vegetation index SAVI and the normalized difference water index NDWI corresponding to the pixels of the image are calculated;

[0046] (3) Remove cloud and cloud shadow pixels, index abnormal pixels and tidal inundation pixels from each image, and sort them in chronological order to construct the effective pixel index time series. The specific removal method of invalid pixels is as follows:

[0047] Removal of cloud and cloud shadow pixels: In the rasterized cloud mask band, cloud and cloud shadow covered pixels are marked as 1, and clear pixels are marked as 0. Remove the image pixels corresponding to the positions with a cloud mask band value of 1;

[0048] Removal of index abnormal pixels: The effective values of indices NDVI, EVI, SAVI and NDWI are between -1 and 1. Remove the image pixels corresponding to the index abnormal values less than -1 or greater than 1;

[0049] Removal of tidal inundation pixels: Find the pixels of the Spartina alterniflora boundary that are inundated and not inundated by tidal water in the image by visual inspection, and obtain the NDWI values of these pixels. Then, obtain the optimal threshold value of the NDWI for separating the inundated and not inundated states by the ROC curve, which is -0.05. Remove the tidal water covered pixels corresponding to the NDWI values greater than or equal to the optimal threshold value, as shown in Figure 2 ;

[0050] (4) Using harmonic functions to fit the effective pixel index time series of NDVI, EVI and SAVI of each effective pixel before Spartina alterniflora removal (i.e. 2020-2021), the observation models of the index NDVI, EVI and SAVI of each effective pixel are established respectively:

[0051] ,

[0052] wherein: represents the model predicted value of the index t of a certain effective pixel at the i th observation, i NDVI, EVI and SAVI respectively; w k is the frequency of the k th harmonic function, which is set as a constant ; k , which is determined according to the number of observations of the effective pixel in the index time series of the effective pixel before Spartina alterniflora removal, is 1 or 2, if the number of observations m ≥10, then m =2, otherwise k =1; k , is the coefficient of the k th harmonic function, which represents the periodic variation trend of the index i ; c i is the intercept, which represents the overall level of the index i variation;

[0053] and are the undetermined parameters in the observation model, which are solved by the LASSO algorithm; c i ;

[0054] (5) Calculate the root mean square error of the observation model:

[0055] ,

[0056] ,

[0057] wherein: RMSE ​​i the index of a certain valid pixel i the root mean square error of the observation model of the index of the valid pixel m the number of observations of the index of the valid pixel in the time series of the index of the valid pixel before Spartina removal p the index of the valid pixel i the number of undetermined parameters in the observation model of the index of the valid pixel residual i,t the residual of the index of the valid pixel at the Tth observation, obtained by the difference between the actual observation value i of the index of the valid pixel at the Tth observation and the model predicted value t i t index i,t t =1, 2, …, m index i,t i t

[0058] Calculate the change range of each valid pixel in the time series of the index of the valid pixel in the Spartina removal period (i.e. 2022-2024) in the first 3 observations (here T=3): n

[0059]

[0060] wherein: magnitude T the change range of a certain valid pixel in the time series of the index of the valid pixel at the Tth observation in the Spartina removal period, T=1, 2, 3 residual i,T the residual of the index of the valid pixel at the Tth observation, obtained by the difference between the actual observation value i of the index of the valid pixel at the Tth observation and the model predicted value i index i,T index i,T i the model predicted value is obtained by substituting the time of the Tth observation into the observation model of step (4);

[0061] ​​​​​​​​​​​​​​​​​​​​​​​The change detection threshold is determined by using chi-square distribution, specifically: the change detection threshold is set to the chi-square distribution value with 3 degrees of freedom at a confidence level of 95%, that is:

[0062] ,

[0063] wherein: threshold change represents the change detection threshold, represents the chi-square distribution, and the number of indexes corresponding to the degrees of freedom 3 is 3, that is: NDVI, EVI, SAVI;

[0064] Taking the pixel with longitude and latitude of 121.158°E and 30.36°N as an example, the first three observation times of the pixel in the Spartina alterniflora removal period are calculated as May 17, July 28, and August 18, 2022, and the corresponding change amplitudes are 5.318, 1.757, and 2.149, respectively. The change detection threshold is 7.815.

[0065] (6) If the change amplitude of a certain effective pixel in the effective pixel index time series in the Spartina alterniflora removal period (i.e. 2022-2024) in the region corresponding to the effective pixel in Ningbo City is greater than the change detection threshold in the first three observations (taking n =3), it is determined that the Spartina alterniflora in the region corresponding to the effective pixel has been removed, and the first observation time of the effective pixel in the Spartina alterniflora removal period is the start time of Spartina alterniflora removal in the region.

[0066] Otherwise, the index of the effective pixel in the Spartina alterniflora removal period is included in the observation model of step (4) to update the model, and steps (4)-(6) are repeated until the change amplitude of the effective pixel in the Spartina alterniflora removal period in the first three observations is greater than the change detection threshold, or all the indexes of the effective pixel in the observation model of the effective pixel index time series before and after the Spartina alterniflora removal are included.

[0067] Taking the pixel with longitude and latitude of 121.158°E and 30.36°N as an example, the change amplitudes of the pixel in the first three observations are calculated to be less than the change detection threshold, and it is determined that the Spartina alterniflora in the region corresponding to the pixel in Ningbo City has not been removed. Then, the index of the pixel observed on May 17, 2022 is included in the observation model of step (4) to update the model, and steps (4)-(6) are repeated until the change amplitudes of the pixel in the observation times of June 11, 2023, July 13, 2023, and September 9, 2023 are calculated to be 11.038, 14.187, and 11.297, respectively, all of which are greater than the change detection threshold 7.815. It is determined that the Spartina alterniflora in the region corresponding to the pixel has been removed, and June 11, 2023 is taken as the start time of Spartina alterniflora removal in the region.

[0068] The remote sensing determination method described in the above embodiments ultimately yielded the spatiotemporal distribution of Spartina alterniflora mowing and tillage clearing in Ningbo City from 2022 to 2024. Figure 3 As shown in the image, the different colors represent the chronological order of Spartina alterniflora eradication. Dark green to light green indicates 2022-2023, light yellow to pale pink indicates 2023-2024, and colors closer to magenta indicate late 2024. The areas in Ningbo City where Spartina alterniflora was eradicated are mainly magenta and beige, indicating that the main eradication period in Ningbo was mid-2023 and late 2024.

[0069] The scope of protection of this invention includes, but is not limited to, the above embodiments. The scope of protection is defined by the claims. Any substitutions, modifications, or improvements to this technology that are easily conceived by those skilled in the art fall within the scope of protection of this invention.

Claims

1. A remote sensing method for determining the clearing time of Spartina alterniflora, characterized in that... The specific steps include the following: (1) Collect Gaofen-1 images covering the monitoring area and within the continuous time period before and after the removal of Spartina alterniflora, and preprocess all Gaofen-1 images respectively. (2) Calculate the Normalized Differential Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Soil-Regulated Vegetation Index (SAVI), and Normalized Differential Water Index (NDWI) corresponding to each pixel after preprocessing. (3) Remove invalid pixels from each image one by one, and sort them according to the order of image observation time to construct an effective pixel index time series; (4) Using harmonic functions to fit the effective pixel index time series before Spartina alterniflora removal, observation models for the indices NDVI, EVI, and SAVI of each effective pixel were established. The observation models are as follows: ， in: This represents the model-predicted value of the index i for a valid pixel at the t-th observation, where i represents NDVI, EVI, and SAVI, respectively; w k Let the frequency of the k-th harmonic function be constant. k takes the value of 1 or 2, and is determined by the number of observations m of the effective pixel in the effective pixel index time series before Spartina alterniflora removal. If the number of observations m ≥ 10, then k = 2; otherwise, k = 1. , Let c be the coefficient of the k-th harmonic function, representing the periodic variation trend of the exponent i; i The intercept represents the overall level of change in the exponent i; The parameters to be determined in the observation model are , and c i The undetermined parameters are solved using the lasso regression algorithm; (5) Calculate the root mean square error of the observation model and the change amplitude of each effective pixel in the effective pixel index time series during the Spartina alterniflora clearing period in the first n observations, and use the chi-square distribution to determine the change detection threshold. (6) By comparing the change range of effective pixels during the Spartina alterniflora removal period with the size of the change detection threshold, it is determined whether the Spartina alterniflora in the area corresponding to the effective pixel has been completely removed. If it has been completely removed, the removal time is determined; otherwise, the observation model is updated and steps (4) to (6) are repeated.

2. The remote sensing method for determining the eradication time of Spartina alterniflora as described in claim 1, characterized in that: In step (1), the preprocessing of the Gaofen-1 image includes geometric correction, radiometric correction, stitching and cropping. At the same time, the cloud mask file in the Gaofen-1 image is geometrically corrected. Specifically, the cloud mask file is geometrically corrected using the same ground feature points from the Gaofen-1 image that are used during the geometric correction of the cloud mask file, so as to ensure that the preprocessed cloud mask file and the spatial position of the Gaofen-1 image are accurately matched.

3. The remote sensing method for determining the eradication time of Spartina alterniflora as described in claim 2, characterized in that: In step (3), invalid pixels include cloud and cloud shadow pixels, exponential anomaly pixels, and tidal flooding pixels, and the removal method is as follows: Removal of cloud and cloud shadow pixels: Removal of pixels covered by clouds and cloud shadows in the image using geometrically corrected cloud mask files; Removal of exponential outliers: The effective values ​​of exponents NDVI, EVI, SAVI and NDWI are between [-1, 1]. Pixels corresponding to exponential outliers that are less than -1 or greater than 1 are removed. Removal of flooded pixels: Visually identify pixels in the image where the boundary of Spartina alterniflora is flooded and those that are not, and obtain the NDWI values ​​of these pixels. Then, obtain the optimal threshold for NDWI to separate the flooded and unflooded states using ROC curves, and remove pixels with NDWI values ​​greater than or equal to the optimal threshold.

4. The remote sensing method for determining the eradication time of Spartina alterniflora as described in claim 1, characterized in that: In step (5), the formula for calculating the root mean square error of the observation model is: ， ， Among them: RMSE i Let be the root mean square error of the observation model for the index i of a certain effective pixel, m be the number of observations of that effective pixel in the effective pixel index time series before Spartina alterniflora clearing, p be the number of undetermined parameters in the observation model for the index i of that effective pixel; and residual(i,t) be the residual of the index i of that effective pixel at the t-th observation, which is the sum of the actual observed value index(i,t) of the index i of that effective pixel at the t-th observation and the model prediction value. The difference is obtained as t=1,2,…,m, and the actual observed value index(i,t) is the value of the index i of the effective pixel at the tth observation.

5. The remote sensing method for determining the clearing time of Spartina alterniflora as described in claim 4, characterized in that: In step (5), the formula for calculating the magnitude of change is: ， Where: magnitude(T) is the change magnitude of a certain effective pixel in the effective pixel index time series during the Spartina alterniflora clearing period, T=1,2,…,n; residual(i,T) is the residual of the index i of the effective pixel at the Tth observation, which is composed of the actual observed value index(i,T) of the index i of the effective pixel at the Tth observation and the model prediction value. The difference is obtained by finding the actual observed value index(i,T), which is the value of the index i of the effective pixel at the Tth observation, and the model predicted value. The observation model in step (4) is obtained by substituting the time of the Tth observation into the observation model.

6. The remote sensing method for determining the eradication time of Spartina alterniflora as described in claim 1, characterized in that: In step (5), the change detection threshold is set to the chi-square distribution value with 3 degrees of freedom at a 95% confidence level, i.e.: ， Where: threshold change Indicates the change detection threshold. This represents a chi-square distribution with 3 degrees of freedom, corresponding to 3 types of indices: NDVI, EVI, and SAVI.

7. The remote sensing method for determining the clearing time of Spartina alterniflora as described in claim 1, characterized in that: In step (6), the method for determining whether Spartina alterniflora has been completely cleared is as follows: If the change magnitude of the effective pixel in the effective pixel index time series within the Spartina alterniflora removal period is greater than the change detection threshold in the first n observations, it is determined that the Spartina alterniflora in the area corresponding to the effective pixel has been completely removed, and the time corresponding to the first observation of the effective pixel within the Spartina alterniflora removal period is the start time of the removal of Spartina alterniflora in the area. Otherwise, the index of the first observation of the effective pixel during the Spartina alterniflora removal period is included in the observation model of step (4) to update the model, and steps (4) to (6) are repeated until the change of the effective pixel during the Spartina alterniflora removal period in the first n observations is greater than the change detection threshold, or the index of all effective pixels before and after the Spartina alterniflora removal is included in the observation model.

Images