Wetland vegetation phenology monitoring method based on cooperation of ground observation and satellite remote sensing

Abstract

The invention discloses a wetland vegetation phenology monitoring method based on cooperation of ground observation and satellite remote sensing. The method comprises the following steps: arranging sample points in the field to obtain wetland vegetation ground observation phenology data; preprocessing the high / low resolution remote sensing data and generating dense EVI time sequence data by adopting a multi-source remote sensing image space-time fusion algorithm; carrying out smooth denoising on the time sequence EVI data based on an SG filtering algorithm; performing inversion based on a dynamic threshold method to obtain wetland vegetation satellite remote sensing phenological data; the relationship between ground observation phenological data and satellite remote sensing phenological inversion data is analyzed based on a linear regression method, a collaborative inversion model is constructed, and the wetland vegetation phenological inversion precision is improved. According to the method, the problem of scale errors caused by different observation scales of ground observation and remote sensing inversion data modeling is solved through cooperation of ground observation and remote sensing inversion data modeling, the wetland vegetation phenology inversion precision and robustness are remarkably improved, large-scale wetland vegetation phenology remote sensing inversion is achieved, and theoretical and technical supports are provided for health assessment of a wetland ecosystem.

Description

Technical Field

[0001] This invention belongs to the field of wetland vegetation monitoring, specifically referring to a wetland vegetation phenology monitoring method that combines ground observation and satellite remote sensing. Background Technology

[0002] Vegetation phenology refers to the periodic growth phenomena of vegetation in nature caused by the influence of climate, human activities, and genetic factors, including processes such as germination, flowering, leaf yellowing, and withering. Wetland vegetation phenology serves as the most direct and sensitive indicator for assessing changes in the wetland environment. It not only reflects wetland ecological conditions but also helps to reveal the dynamic interaction mechanism between wetland vegetation ecological processes and changes in climate conditions.

[0003] Traditional vegetation phenology monitoring relies on field observations, recording information such as the start time of the growing season (SOS), the end time of the growing season (EOS), and the length of the growing season (LOS). While this method is intuitive and accurate, it consumes a significant amount of manpower, resources, and time, and is limited by the complexity of wetland environments and the difficulty of access due to prolonged flooding, thus failing to meet the needs of large-scale monitoring.

[0004] The development of remote sensing technology has made large-scale vegetation phenology monitoring possible, but existing remote sensing monitoring methods have significant shortcomings:

[0005] Scale error problem: Traditional inversion algorithms do not take into account the scale difference between ground observation data and satellite remote sensing data, resulting in low inversion accuracy and difficulty in meeting the requirements.

[0006] Spatiotemporal resolution contradiction: High temporal resolution data such as MODIS have low spatial resolution (e.g., 500 meters) and are easily affected by mixed pixels; High spatial resolution data such as Landsat (e.g., 30 meters) have a long revisit period (16 days), and the time series data becomes discontinuous after being affected by clouds and rain, increasing the uncertainty of inversion.

[0007] Verification difficulties: Wetland areas are difficult to access, and traditional field surveys are time-consuming and labor-intensive, resulting in a lack of effective means to verify wetland vegetation phenological data. Summary of the Invention

[0008] The technical problems to be solved by this invention are large scale error, easy to affect spatiotemporal resolution and difficulty in data verification.

[0009] To address the above problems, the technical solution adopted by this invention is as follows: The wetland vegetation phenology monitoring method proposed in this invention, which combines ground observation and satellite remote sensing, includes the following steps:

[0010] S1: Determine the monitoring area and conduct field observations to obtain ground phenological data; the ground phenological data includes the start time of the growing season (SOS), the end time of the growing season (EOS), and the length of the growing season (LOS) of vegetation within the monitoring area.

[0011] S2: Acquire high spatial resolution remote sensing data and high temporal resolution remote sensing data of the monitoring area, preprocess the remote sensing data, and generate a time-series vegetation index dataset;

[0012] S3: The preprocessed remote sensing data is fused using a spatiotemporal image fusion model to construct a high spatiotemporal resolution time-series vegetation index dataset, and the dataset is then smoothed and denoised.

[0013] S4: Based on the smoothed high spatiotemporal resolution time-series vegetation index dataset, the dynamic threshold method is used to perform remote sensing inversion of wetland vegetation phenology to obtain remote sensing inversion phenological data.

[0014] S5: Construct a collaborative monitoring model, perform linear regression on the ground phenological data obtained in step S1 and the remote sensing inversion phenological data obtained in step S4 to obtain a linear regression equation, and use the linear regression equation to correct the remote sensing inversion phenological data to obtain the final wetland vegetation phenological monitoring results.

[0015] Further, in step S1, the specific process of field observation includes: setting up observation points within the monitoring area, wherein the setting up of the observation points meets the following conditions: the observation points are evenly distributed in the monitoring area and located in the middle of the vegetation community, ensuring that the remote sensing image pixels corresponding to the observation points are pure pixels; the observation points cover vegetation in various growth states, and the elevation factor of the monitoring area is taken into account; the observation points are observed at a preset frequency, and the germination time, leaf length, color change and flooding status of the vegetation are recorded, and the SOS, EOS and LOS of the vegetation are determined.

[0016] Further, in step S2, the high spatial resolution remote sensing data includes Sentinel-2 data or Landsat data, and the high temporal resolution remote sensing data includes MODIS data; the preprocessing includes: radiometric calibration, atmospheric correction, band recombination to generate Enhanced Vegetation Index (EVI) data, reprojection, resampling, cropping, and image pairing of the remote sensing data; wherein, the calculation formula for EVI data is:

[0017]

[0018] In the formula, G is the gain coefficient, which takes a value of 2.5; ρ NIR ρ red and ρ blue, respectively, represent the surface reflectance in the near-infrared, red, and blue light bands; L is the soil background adjustment parameter, with a value of 1; C1 and C2 are the weighting functions, with values ​​of 6 and 7.5, respectively.

[0019] Further, in step S3, the spatiotemporal image fusion model is the ESTARFM model; the fusion process includes: based on the paired high- and low-resolution images in step S2, using the low-resolution image at the prediction time as a reference, fusing to generate a high-resolution image at the prediction time; combining the fused high-resolution image with cloudless or low-cloud high-resolution images to form a dense temporal high-resolution dataset; and using Savitzky-Golay (SG) filtering to smooth and denoise the dataset, the SG filtering formula being:

[0020]

[0021] In the formula, Y represents the data to be fitted, and Y' represents the fitted value. i is the filter coefficient, representing the weight of the i-th point, and N = 2n + 1 is the size of the sliding window.

[0022] Further, in step S4, the specific process of the dynamic threshold method is as follows: the time when the vegetation index value rises to 20% of the annual vegetation index amplitude is defined as the start of the growing season (SOS), and the time when the vegetation index value falls to 20% of the annual vegetation index amplitude is defined as the end of the growing season (EOS); the length of the growing season (LOS) is the difference between EOS and SOS; the formula for calculating the annual vegetation index amplitude is:

[0023]

[0024] In the formula, VI min and VI max These represent the minimum and maximum vegetation index values ​​during the growing season, respectively, while VI represents the vegetation index value at a specific point in the time series data.

[0025] Further, in step S5, the construction process of the collaborative monitoring model includes: dividing the ground phenological data obtained in step S1 and the remote sensing inversion phenological data obtained in step S4 into a 6:4 ratio, where 60% of the samples are used to construct a linear regression equation, and 40% of the samples are used for model accuracy evaluation; the expression of the linear regression equation is:

[0026] S'=a×S+b

[0027] E'=c×E+d

[0028] In the formula, S′ and E′ are the corrected start and end times of the growing season, respectively; S and E are the start and end times of the growing season retrieved from remote sensing, respectively; and a, b, c, and d are regression coefficients.

[0029] Furthermore, in step S1, the observation sampling points are set up in a square vegetation community area with a side length of 30 meters, and the observation frequency is at least twice a week.

[0030] The beneficial effects achieved by the present invention using the above method are as follows:

[0031] 1. The wetland vegetation phenology monitoring method proposed in this scheme, which combines ground observation and satellite remote sensing data, constructs a linear regression model to correct scale errors and significantly reduces inversion errors.

[0032] 2. The wetland vegetation phenology monitoring method proposed in this scheme, which combines ground observation and satellite remote sensing, uses the ESTARFM model to fuse high and low resolution data to generate a time series dataset with a spatial resolution of 10 meters and a temporal resolution of 8 days, taking into account both spatial details and temporal continuity.

[0033] 3. The wetland vegetation phenology monitoring method proposed in this scheme, which combines ground observation and satellite remote sensing, breaks through the limitations of traditional field surveys. It combines remote sensing technology to achieve large-scale wetland vegetation phenology monitoring, providing data support for wetland ecosystem health assessment.

[0034] 4. The proposed method for monitoring wetland vegetation phenology by combining ground observation and satellite remote sensing has an optimized verification method: by scientifically deploying sampling points and conducting high-frequency observations, reliable ground data can be provided, thus solving the problem of difficulty in verifying wetland phenological data. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the dynamic threshold method of the present invention;

[0036] Figure 2 This is a flowchart illustrating the technical process of the present invention.

[0037] Figure 3 This is a histogram of time-series image composition according to the present invention;

[0038] Figure 4 This is a diagram showing the phenological inversion results for the first quarter of 2023 according to the present invention.

[0039] Figure 5 This is a diagram showing the phenological inversion results for the second quarter of 2023 according to the present invention.

[0040] Figure 6 Linear fitting of phenological phenomena obtained from field observations and remote sensing inversion in the first season of this invention;

[0041] Figure 7 This is a diagram showing the corrected phenological results for the first quarter of 2023 according to the present invention.

[0042] Figure 8 This is a linear fitting diagram of phenology obtained from field observations and remote sensing inversion in the second season of this invention;

[0043] Figure 9 This is a diagram showing the corrected phenological results for the second quarter of 2023, based on the present invention.

[0044] Among them, a) greening period, b) withering period, c) length of growing season, d) maximum value, e) average of left and right minimum values, and f) amplitude.

[0045] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0046] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0047] This invention proposes a method for monitoring wetland vegetation phenology through a combination of ground observation and satellite remote sensing, comprising the following steps:

[0048] S1: Field observation of phenology

[0049] After determining the monitoring area, observation points are set up within the area. Each point must be located in the center of a 30-meter-sided square vegetation community, ensuring that the corresponding remote sensing pixels are clean pixels, evenly distributed, and covering vegetation at various growth stages, while also considering the impact of elevation on flooding. Observations are conducted at least twice a week, recording vegetation germination time, leaf length, color changes, and flooding conditions to obtain ground phenological data (SOS, EOS, LOS).

[0050] S2: Remote Sensing Data Preprocessing

[0051] Download high spatial resolution data (e.g., Sentinel-2L2A data, 10-meter resolution) and high temporal resolution data (e.g., MODISMOD09A1 data, 8-day temporal resolution). After radiometric calibration and atmospheric correction, reassemble the bands to generate EVI data; reproject all images to the same coordinate system (e.g., WGS_1984_UTM_Zone_50N), and resample the low-resolution data to 10-meter resolution; crop the images to the same range and pair them with high- and low-resolution reference images (the time difference must be minimal).

[0052] S3: Spatiotemporal image fusion and temporal data construction

[0053] The ESTARFM model is used to generate high-resolution images for the prediction time by fusing paired high- and low-resolution reference images with low-resolution images at the prediction time. The fused images are then combined with cloud-free high-resolution images to form a dense time-series dataset. SG filtering is used to smooth and denoise the dataset, which serves as the data source for phenological inversion.

[0054] S4: Phenological Remote Sensing Inversion

[0055] Based on smoothed time-series EVI data, a dynamic threshold method is used to extract phenological parameters: the time when the EVI value rises to 20% of the amplitude is defined as SOS, and the time when it falls to 20% of the amplitude is defined as EOS. LOS = EOS - SOS, and the results are expressed in year-order days (DOY).

[0056] S5: Collaborative Monitoring Model Construction and Calibration

[0057] Ground phenology data and remote sensing inversion data were divided in a 6:4 ratio, and 60% of the samples were used to construct a linear regression equation (e.g., ...).

[0058] S′=0.5981×S+13.271,E′=0.7445×E+47.098), 40% sample validation accuracy (assessed by MAE and RMSE). The remote sensing inversion data were corrected using regression equations to obtain the final phenological monitoring results.

[0059] Example 1

[0060] 1. Field observation

[0061] The monitoring area was Poyang Lake wetland, and the research object was *Carex porphyria*. In 2023, 25 observation points were set up in the Nanhu section, Chuankuijia, and Xiquehu areas of the Nanji Wetland National Nature Reserve. Each point was located within a square *Carex porphyria* community with sides of 30 meters. With the assistance of local farmers, observations were conducted twice a week, recording the germination time, leaf length, color, and waterlogging status of the *Carex porphyria*, and determining the SOS, EOS, and LOS at each point.

[0062] 2. Remote sensing data processing

[0063] Download Sentinel-2 L2A data (B8, B4, B2 bands, 10-meter resolution) and MODISMOD09A1 data (red, blue, and near-infrared bands, 500-meter resolution) from 2019 to 2023. Cloud-free images were selected on the GEE platform, and EVI was calculated, followed by mosaicking and cropping to the study area. The MODIS data was resampled to 10-meter resolution to ensure consistent projection with the Sentinel-2 data.

[0064] 3. Spatiotemporal fusion and time sequence construction

[0065] The ESTARFM model was used to fuse the data: Sentinel-2 / MODIS images from January 6 and April 1, 2023, were used as a baseline to generate a high-resolution image from January 31, 2023. This fusion produced 149 images, which, together with 81 cloud-free Sentinel-2 images, formed a 230-image time-series dataset (2019-2023). SG filtering was used to smooth the data, with a window size of N=7.

[0066] 4. Phenological inversion

[0067] Using a dynamic threshold method, the time when the EVI rises to 20% of its amplitude is defined as SOS, and the time when it falls to 20% of its amplitude is defined as EOS. The inversion yielded that the SOS range for the first season of sedge in 2023 was 45-90 DOY, and the EOS range was 115-177 DOY.

[0068] 5. Model Building and Calibration

[0069] The 25 samples were divided into a 6:4 ratio (15 for modeling and 10 for validation). The modeling yielded the first-quarter regression equation: S1′=0.5981×S1+13.27(R²). 2 =0.8337), E1′=0.7445×E1+47.098(R 2 =0.8282). After correction, the MAE of the validation sample SOS decreased from 8.6 days to 4.1 days, and the MAE of EOS decreased from 10.6 days to 6.1 days. Similarly, the second-quarter model was constructed, and after correction, the MAE of SOS decreased from 9.2 days to 4.5 days, and the MAE of EOS decreased from 13.3 days to 7 days.

[0070] 6. Extraction of multi-year phenological data

[0071] The regression equation was applied to correct the inversion data from 2019 to 2023, resulting in the final sedge phenology dataset.

[0072] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention. The actual method is not limited to this. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the present invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for monitoring wetland vegetation phenology by combining ground observation and satellite remote sensing, characterized in that, Includes the following steps: S1: Determine the monitoring area and conduct field observations to obtain ground phenological data; the ground phenological data includes the start time of the growing season (SOS), the end time of the growing season (EOS), and the length of the growing season (LOS) of vegetation within the monitoring area. S2: Acquire high spatial resolution remote sensing data and high temporal resolution remote sensing data of the monitoring area, preprocess the remote sensing data, and generate a time-series vegetation index dataset; S3: The preprocessed remote sensing data is fused using a spatiotemporal image fusion model to construct a high spatiotemporal resolution time-series vegetation index dataset, and the dataset is then smoothed and denoised. S4: Based on the smoothed high spatiotemporal resolution time-series vegetation index dataset, the dynamic threshold method is used to perform remote sensing inversion of wetland vegetation phenology and obtain remote sensing inversion phenology data. S5: Construct a collaborative monitoring model, perform linear regression on the ground phenological data obtained in step S1 and the remote sensing inversion phenological data obtained in step S4 to obtain a linear regression equation, and use the linear regression equation to correct the remote sensing inversion phenological data to obtain the final wetland vegetation phenological monitoring results.

2. The wetland vegetation phenology monitoring method based on ground observation and satellite remote sensing according to claim 1, characterized in that: In step S1, the specific process of field observation includes: setting up observation points within the monitoring area, wherein the setting of the observation points meets the following conditions: the observation points are evenly distributed in the monitoring area and located in the middle of the vegetation community, ensuring that the remote sensing image pixels corresponding to the observation points are clean pixels; the observation points cover vegetation in various growth states, and the elevation factor of the monitoring area is taken into account; the observation points are observed at a preset frequency, and the germination time, leaf length, color change and flooding status of the vegetation are recorded, and the SOS, EOS and LOS of the vegetation are determined.

3. The wetland vegetation phenology monitoring method based on ground observation and satellite remote sensing according to claim 1, characterized in that: In step S2, the high spatial resolution remote sensing data includes Sentinel-2 data or Landsat data, and the high temporal resolution remote sensing data includes MODIS data; the preprocessing includes: radiometric calibration, atmospheric correction, band recombination to generate Enhanced Vegetation Index (EVI) data, reprojection, resampling, cropping, and image pairing of the remote sensing data; wherein, the calculation formula for EVI data is: In the formula, G is the gain coefficient, which takes a value of 2.5; ρ NIR ρ red and ρ blue , respectively, represent the surface reflectance in the near-infrared, red, and blue light bands; L is the soil background adjustment parameter, with a value of 1; C1 and C2 are the weighting functions, with values ​​of 6 and 7.5, respectively.

4. The wetland vegetation phenology monitoring method based on ground observation and satellite remote sensing according to claim 1, characterized in that: In step S3, the spatiotemporal image fusion model is the ESTARFM model; the fusion process includes: based on the paired high- and low-resolution images in step S2, using the low-resolution image of the prediction time as a reference, fusing to generate a high-resolution image of the prediction time; combining the fused high-resolution image with cloudless or low-cloud high-resolution images to form a dense temporal high-resolution dataset; and using Savitzky-Golay (SG) filtering to smooth and denoise the dataset, the SG filtering formula being: In the formula, Y represents the data to be fitted, and Y' represents the fitted value. i is the filter coefficient, representing the weight of the i-th point, and N = 2n + 1 is the size of the sliding window.

5. The wetland vegetation phenology monitoring method based on ground observation and satellite remote sensing according to claim 1, characterized in that: In step S4, the specific process of the dynamic threshold method is as follows: the time when the vegetation index value rises to 20% of the annual vegetation index amplitude is defined as the start of the growing season (SOS), and the time when the vegetation index value falls to 20% of the annual vegetation index amplitude is defined as the end of the growing season (EOS); the length of the growing season (LOS) is the difference between EOS and SOS; the formula for calculating the annual vegetation index amplitude is: In the formula, VI min and VI max These represent the minimum and maximum vegetation index values ​​during the growing season, respectively, while VI represents the vegetation index value at a specific point in the time series data.

6. The wetland vegetation phenology monitoring method based on ground observation and satellite remote sensing according to claim 1, characterized in that: In step S5, the construction process of the collaborative monitoring model includes: dividing the ground phenological data obtained in step S1 and the remote sensing inversion phenological data obtained in step S4 into a 6:4 ratio, where 60% of the samples are used to construct a linear regression equation and 40% of the samples are used for model accuracy evaluation; the expression of the linear regression equation is: S'=a×S+b E'=c×E+d In the formula, S′ and E′ are the corrected start and end times of the growing season, respectively; S and E are the start and end times of the growing season retrieved from remote sensing, respectively; and a, b, c, and d are regression coefficients.

7. The wetland vegetation phenology monitoring method based on ground observation and satellite remote sensing according to claim 2, characterized in that: In step S1, the observation sampling points are set up in a square vegetation community area with a side length of 30 meters, and the observation frequency is at least twice a week.

Images