Method for removing interference of submerged emergent vegetation in remote sensing monitoring of submerged vegetation

Abstract

This invention relates to a method for removing interference from submerged emergent vegetation in remote sensing monitoring of submerged vegetation, belonging to the field of ecological and environmental remote sensing monitoring technology. This invention comprehensively utilizes the spectral recognition capabilities of multispectral remote sensing imagery and the surface moisture and structure detection capabilities of radar imagery, eliminating interference by leveraging the differences in the ecological responses of vegetation to continuous flooding. First, underwater vegetation areas containing both target and interfering vegetation are preliminarily identified based on multispectral imagery. Then, the cumulative flooding duration for each pixel is calculated using time-series radar imagery. Finally, by comparing the flooding duration with a locally determined emergent vegetation decay threshold, pixels that do not exceed the threshold are identified as submerged emergent vegetation and removed, thus obtaining an accurate distribution of submerged vegetation. This invention significantly improves the accuracy and reliability of remote sensing identification of submerged vegetation in waters with drastic seasonal water level changes, such as flooded lakes. It is suitable for large-scale, long-term dynamic monitoring and has the advantages of high automation, wide applicability, and low cost.

Description

Technical Field

[0001] This invention relates to the field of remote sensing monitoring technology for ecological environment, specifically to a remote sensing monitoring method for submerged vegetation in flooded lakes, and particularly to a method that utilizes radar imagery and the difference in response of emergent and submerged vegetation to flooding to eliminate interference from submerged emergent vegetation in multispectral remote sensing monitoring. Background Technology

[0002] Submerged vegetation is an important component of lake ecosystems, and its degradation will have a serious impact on the stability of the aquatic ecological structure, biodiversity, and ecosystem services. Therefore, effective monitoring of its spatial distribution and dynamics is of great significance.

[0003] Traditional field survey methods are highly accurate but costly and inefficient, making them unsuitable for large-scale, long-term monitoring. Monitoring methods based on multispectral remote sensing imagery offer advantages such as wide coverage, lower cost, and the ability to trace historical changes, and have become an important tool. However, this method faces significant challenges when applied to floodplains (lakes with significant seasonal water level fluctuations).

[0004] During the high-water season, floodplains experience rapid water level rises, submerging large areas of emergent vegetation (such as sedge and sedge). The underwater spectral characteristics of these completely submerged emergent plants are highly similar to those of submerged vegetation, making them difficult to distinguish in multispectral imagery. This severely interferes with the accurate identification and extraction of submerged vegetation, limiting the application of remote sensing technology in monitoring submerged vegetation in floodplains.

[0005] In the existing technology, there are no publicly reported methods for remote sensing monitoring of submerged vegetation in flooded lakes that effectively solve the above-mentioned interference problems. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to overcome the shortcomings of the existing multispectral remote sensing methods in monitoring submerged vegetation in floodplains, which are easily affected by the spectral interference of submerged emergent vegetation underwater. The present invention provides a method that can effectively eliminate such interference and achieve accurate, stable and low-cost monitoring of submerged vegetation in floodplains.

[0007] To address the aforementioned technical problems, this invention provides a method for removing interference from submerged emergent vegetation in remote sensing monitoring of submerged vegetation. This method comprehensively utilizes the spectral recognition capabilities of multispectral images and the detection capabilities of radar images for surface moisture and structural information. It also combines the diametrically opposed ecological response patterns of emergent and submerged vegetation under continuous flooding conditions to remove submerged emergent vegetation from the submerged vegetation identification results.

[0008] The method for removing interference from submerged emergent vegetation in remote sensing monitoring of submerged vegetation, as described in this invention, includes the following steps:

[0009] Step S1: Remote Sensing Image Selection and Preprocessing

[0010] Acquire temporally matched multispectral remote sensing and radar images covering the target floodplain lake study area. Landsat series (e.g., TM, ETM+, OLI) or Sentinel-2 MSI images are preferred for multispectral imagery; Sentinel-1 SAR images are preferred for radar imagery. Perform radiometric correction, atmospheric correction (for multispectral imagery), geometric fine correction, resampling to a uniform spatial resolution, and cropping according to the study area on the acquired images.

[0011] Step S2: Preliminary identification of underwater vegetation based on multispectral imagery

[0012] Using preprocessed multispectral imagery, a decision tree model was constructed to preliminarily identify "underwater vegetation" areas containing submerged vegetation and potentially submerged emergent vegetation. Specifically, these include:

[0013] (2.1) Calculate the improved Normalized Difference Water Index (MNDWI) and set a threshold (e.g., MNDWI > 0) to distinguish between water bodies and non-water bodies (terrestrial and emergent / floating leaf vegetation).

[0014] (2.2) For the identified water areas, calculate the Normalized Difference Vegetation Index (NDVI). Combined with the brightness index (average reflectance of red and green light bands), further differentiation is achieved by setting thresholds.

[0015] - Areas with NDVI less than a certain threshold (e.g., < 0) are identified as open water.

[0016] - Areas with NDVI greater than or equal to this threshold and brightness greater than a certain threshold (e.g., > 7.5%) are identified as shallow water.

[0017] - Areas with NDVI greater than or equal to the threshold and brightness less than or equal to the threshold (e.g., ≤ 7.5%) are initially identified as "underwater vegetation areas", which include submerged vegetation and completely submerged emergent vegetation.

[0018] The above thresholds can be determined by the statistical distribution of the training samples (such as kernel density curves).

[0019] Step S3: Calculation of flood duration based on radar imagery and elimination of disturbance from emergent vegetation

[0020] (3.1) Preprocess the time-series radar images (covering the period from the high water season to the target monitoring date). Set a backscattering coefficient threshold (e.g., on Sentinel-1 VV polarimetric images, less than -21 dB is considered water, otherwise it is non-water / emergent vegetation), and binarize each radar image into "water body" and "non-water body" maps.

[0021] (3.2) Perform overlay analysis on the binarized image sequence, count the number of days that each pixel is continuously or cumulatively identified as a "water body" during the observation period, and obtain the cumulative flooding duration of each pixel.

[0022] (3.3) Obtain the flooding duration threshold T for the complete death of the aboveground parts of emergent vegetation in the target area (e.g., determined experimentally to be 60 days).

[0023] (3.4) Determine the value of each pixel in the "underwater vegetation area" initially identified in step S2:

[0024] If the cumulative flooding duration of a pixel is greater than the threshold T, then the pixel is determined to be submerged vegetation.

[0025] If the cumulative flooding duration of a pixel is less than or equal to the threshold T, then the pixel is determined to mainly reflect the spectral signal of emergent vegetation that has been submerged but has not yet completely died, and it should be excluded from the classification results of submerged vegetation.

[0026] Step S4: Post-processing and validation of classification results

[0027] After eliminating interference, the submerged vegetation distribution map was subjected to necessary morphological processing (such as removing small patches and merging fragments). Independent field survey data, historical records, or high-resolution UAV imagery were used as validation datasets to construct a confusion matrix and calculate indicators such as overall accuracy (OA), Kappa coefficient, user accuracy (UA), and producer accuracy (PA) to assess the reliability of the monitoring results.

[0028] The basic principle of this invention is as follows:

[0029] In flooded lakes, emergent and submerged vegetation respond to prolonged flooding in fundamentally different ways:

[0030] (1) Emergent vegetation: Its above-ground parts (stems and leaves) need to be exposed above the water surface for gas exchange. When completely submerged, the above-ground parts will gradually die and decompose due to lack of oxygen. The degree of death is positively correlated with the duration of submersion. The longer the submersion time, the smaller the biomass and height of the above-ground parts.

[0031] (2) Submerged vegetation: Its entire life cycle is completed underwater. It begins to germinate and grow under continuous flooding conditions. The flooding time provides it with a growth cycle. The longer the flooding time, the greater the biomass and height are usually.

[0032] (3) The unique role of radar imagery: The microwaves emitted by synthetic aperture radar (SAR, such as Sentinel-1) can penetrate clouds and fog, enabling all-weather, all-day observation. The backscattering intensity of radar waves is sensitive to surface roughness and dielectric constant. The rough surface of emergent vegetation canopy, compared to the smooth surface of water, shows a significant difference in backscattering coefficient in radar images, thus effectively distinguishing emergent vegetation areas from open water areas.

[0033] Based on the above principles, when a sandbar area is submerged, high-frequency radar imagery can continuously track the process of the area changing from "emergent vegetation" to "water," thereby accurately calculating the cumulative submersion duration for each pixel. Through controlled field experiments or continuous observation, the submersion duration threshold required for the complete demise of the aboveground parts of the local emergent vegetation can be determined (for example, in the Poyang Lake area, this threshold is approximately 60 days). If the submersion duration of a pixel exceeds this threshold, the emergent vegetation at that location is considered to have disappeared, and the underwater vegetation identified by the multispectral imagery in that pixel should be submerged vegetation; if the submersion duration does not exceed the threshold, the emergent vegetation is considered not to have completely disappeared, and its spectral signal interferes with the identification, so this pixel should be excluded from the submerged vegetation results.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] (1) Significantly improve monitoring accuracy: The innovative introduction of radar imagery to calculate flood duration, combined with vegetation ecological response threshold, fundamentally solves the problem of spectral confusion between underwater emergent vegetation and submerged vegetation in flooded lakes, and greatly improves the extraction accuracy of submerged vegetation spatial distribution information.

[0036] (2) Achieve long-term dynamic monitoring: This method makes full use of long-term open-source remote sensing data, which can reconstruct and analyze the historical spatiotemporal evolution of submerged vegetation in flooded lakes without the need for large-scale field surveys, providing long-term data support for ecological assessment and management.

[0037] (3) High applicability and low cost: The method is based on widely available satellite remote sensing data, has a high degree of automation, and is applicable to large flooded lakes in different regions. It greatly reduces the manpower and financial costs of large-scale, routine monitoring and has high application and promotion value.

[0038] (4) The technical solution is complete and reliable: It provides a complete technical process from data preprocessing, feature extraction, model building to interference elimination and accuracy verification, and has been specifically verified in the embodiments, which confirms its effectiveness and reliability. Attached Figure Description

[0039] Figure 1 This is a flowchart illustrating the single-day submerged vegetation extraction process of the method described in one embodiment of the present invention.

[0040] Figure 2 This is a schematic diagram of a decision tree model for extracting submerged vegetation, constructed using Poyang Lake as the study area, in one embodiment of the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of protection of this invention.

[0042] Example: Remote Sensing Monitoring Method for Submerged Vegetation in Poyang Lake, China

[0043] This embodiment details the specific process of applying the method of the present invention to Poyang Lake, a typical floodplain lake.

[0044] 1. Overview of the study area and data preparation

[0045] Poyang Lake is China's largest freshwater lake, characterized by its vast expanse of water during floods and its razor-thin water level during droughts, exhibiting significant seasonal fluctuations in water level. This study selected the period from 2015 to 2024 as the observation period.

[0046] Multispectral data: Download Landsat 8 OLI and Sentinel-2 MSI cloudless or low-cloud imagery for the Poyang Lake receding water period (September to January of the following year).

[0047] Radar data: Download Sentinel-1 SAR (IW mode, VV polarization) images covering the Poyang Lake area.

[0048] Supporting data: Historical field survey records from 2018 and UAV survey data from 2022-2023 were collected for verification; a digital elevation model (DEM) of Poyang Lake was acquired. All image data underwent radiometric correction, geometric fine correction, and spatial registration.

[0049] 2. Construction of Submerged Vegetation Extraction Model and Threshold Determination

[0050] A decision tree classification model was constructed based on multi-source features, and the key thresholds were determined through statistical analysis of 4,692 training sample points (including 2,666 water bodies, 657 shoals, and 1,369 underwater vegetation areas).

[0051] MNDWI threshold: Areas with MNDWI > 0 are classified as water bodies.

[0052] NDVI and brightness threshold: Statistical analysis shows that areas with NDVI ≥ 0 include submerged vegetation and shoals; further differentiation is achieved using the brightness index (mean of red and green light bands), with brightness > 7.5% classified as shoals and brightness ≤ 7.5% classified as "underwater vegetation area" (including target submerged vegetation and disturbing vegetation).

[0053] 3. Calculation of flood duration and interference elimination based on radar imagery (core step)

[0054] (1) Water body identification: For Sentinel-1 images, water bodies are defined as having a backscattering coefficient of < -21 dB, and a time series of binarized water body distribution maps are generated.

[0055] (2) Calculation of flood duration: Overlay analysis is performed on radar image sequences that are pushed forward by a sufficient number of days from the target monitoring date to calculate the cumulative number of days that each pixel is continuously identified as a water body.

[0056] (3) Application of ecological response threshold: According to previous studies, typical emergent plants in Poyang Lake (such as podzolium gracilis) almost completely die off above ground after being completely submerged for about 60 days. Therefore, the flooding duration threshold T is set to 60 days.

[0057] (4) Decision-making: For each pixel of the "underwater vegetation area" initially identified from the multispectral image:

[0058] Retrieve radar image sequences for the pixel location within 60 days prior to the target date.

[0059] If the pixel is identified as a body of water in all radar images within these 60 days (i.e., continuous flooding for more than 60 days), it will ultimately be determined to be submerged vegetation.

[0060] If, within these 60 days, it is identified as emergent vegetation once or multiple times (i.e., the flooding is discontinuous or lasts less than 60 days), it is determined to be submerged emergent vegetation and is excluded.

[0061] Final decision tree model as Figure 2 As shown.

[0062] 4. Accuracy Verification and Effect Analysis

[0063] The classification results were validated using 590 independent validation points (430 from historical records and 160 from UAV observations). The resulting confusion matrix is ​​shown in Table 1.

[0064] Table 1. Confusion Matrix for Submerged Vegetation Validation

[0065]

[0066] The verification results show that the method of the present invention has high classification accuracy and reliability.

[0067] 5. Comparison of interference elimination effects

[0068] To visually demonstrate the interference-eliminating effect of this invention, the submerged vegetation area of ​​Poyang Lake obtained using the traditional multispectral method (without interference elimination) and the method of this invention (after radar flooding duration correction) from 2015 to 2024 were compared. The results are shown in Table 2.

[0069] Table 2. Comparison of submerged vegetation area obtained by different treatment methods (km²)

[0070]

[0071] As clearly shown in Table 2, the submerged vegetation area is significantly reduced after correction using the method of this invention. For example, the uncorrected area in 2018 was 822.26 km², while the corrected area was 375.78 km², eliminating approximately 446.48 km² of interference pixels from non-submerged vegetation. This fully demonstrates that traditional methods, unable to distinguish underwater vegetation types, severely overestimate the submerged vegetation area, while the method of this invention effectively eliminates the interference from submerged emergent vegetation, yielding monitoring results closer to the actual situation.

[0072] 6. Examples of Long-Sequence Applications

[0073] By applying the method of this invention to process historical images of Poyang Lake from 2015 to 2024, a 10-year spatiotemporal evolution sequence of submerged vegetation was successfully reconstructed. Analysis shows that the annual maximum distribution area of ​​submerged vegetation in Poyang Lake exhibited a trend of first increasing and then decreasing between 2015 and 2024, reaching a peak of 375.78 km² in 2018. Spatially, the submerged vegetation is mainly distributed in the saucer-shaped lake, and elevationally concentrated in the 12-13m region. This application demonstrates the effectiveness and value of the method of this invention in long-term, large-scale dynamic monitoring.

[0074] In summary, this embodiment, using Poyang Lake as an example, fully demonstrates the operational process, key technical parameter determination methods, and excellent application effects of the method of the present invention. Those skilled in the art will understand that by adjusting the spectral index threshold, radar backscattering threshold, and emergent vegetation decay duration threshold T, the method of the present invention can be directly applied to other floodplains with similar hydrological characteristics to achieve accurate monitoring of submerged vegetation.

Claims

1. A method for removing disturbance from submerged emergent vegetation in remote sensing monitoring of submerged vegetation, characterized in that, Includes the following steps: S1. Remote sensing image selection and preprocessing: Acquire multispectral remote sensing images and radar images that cover the target study area and are time-matched, and perform preprocessing. S2. Preliminary identification of underwater vegetation based on multispectral images: Using preprocessed multispectral images, a decision tree model is constructed to preliminarily identify underwater vegetation areas containing submerged vegetation and emergent vegetation that may be submerged. S3. Calculation of flood duration based on radar imagery and elimination of disturbance from emergent vegetation: S31. Preprocess the time-series radar images covering the period from the high water season to the target monitoring date, set the backscattering coefficient threshold, and binarize each period's image into water body and non-water body maps. S32. Perform overlay analysis on the binarized image sequence, count the cumulative number of days that each pixel was identified as a water body during the observation period, and obtain the cumulative flooding duration. S33. Obtain the threshold T of the flooding duration required for the aboveground parts of emergent vegetation in the study area to completely die off. S34. For each pixel in the underwater vegetation area initially identified in step S2, make a judgment: if the cumulative flooding time of the pixel is greater than the threshold T, it is determined to be submerged vegetation; if the cumulative flooding time is less than or equal to the threshold T, it is determined to be submerged emergent vegetation and excluded from the submerged vegetation classification results. S4. Post-process and accuracy verification of the submerged vegetation distribution map after interference elimination.

2. The method according to claim 1, characterized in that, In step S2, constructing the decision tree model specifically includes: The improved Normalized Difference Water Index (MNDWI) is calculated, and water bodies are distinguished from non-water bodies by a first threshold. For the identified water areas, calculate the Normalized Difference Vegetation Index (NDVI) and the Brightness Index. The NDVI and the second threshold are used to distinguish the open water surface, and the shoals and the underwater vegetation area are distinguished by combining the brightness index and the third threshold.

3. The method according to claim 2, characterized in that, The brightness index is the average reflectance of the red and green light bands.

4. The method according to claim 1, characterized in that, In step S33, the flooding duration threshold T is determined through field control experiments or continuous observation.

5. The method according to claim 1, characterized in that, In step S1, the multispectral remote sensing image is a Landsat series image or a Sentinel-2 MSI image; the radar image is a Sentinel-1 SAR image.

6. The method according to claim 1, characterized in that, In step S4, the post-processing includes morphological processing; the accuracy verification includes constructing a confusion matrix using field survey data, historical records, or high-resolution UAV imagery, and calculating overall accuracy, Kappa coefficient, user accuracy, and producer accuracy.

7. The method according to claim 1, characterized in that, The method is applicable to flooded lakes with significant seasonal water level fluctuations.

Images