Evaluation method for waterfowl food resource availability

Abstract

The invention provides a waterfowl food resource availability evaluation method, and belongs to the field of environmental protection, ecology and remote sensing monitoring. According to the method, a food resource quality evaluation model is obtained according to ground measured data and remote sensing data of a foraging place of a research area, and then plant nutritional ingredients of the foraging place of a to-be-detected area are obtained by utilizing the food resource quality evaluation model. According to the method, the foraging habitat can be accurately identified, the potential food resources of the waterfowl, especially the nutritional status and the availability of wetland vegetation, can be rapidly and accurately evaluated in a large scale, and the method has repeatability and relatively strong generalization ability.

Description

Technical Field

[0001] The invention belongs to the fields of environmental protection, ecology and remote sensing monitoring, and relates to ecological monitoring and protection technology, in particular to an evaluation method for the availability of waterfowl food resources based on remote sensing data and ground measured data. Background Art

[0002] As an important component of wetland ecosystems, waterbirds' habitat selection is closely related to wetland food resources. The quality and availability of waterbird food resources directly affect the reproduction and survival of their populations. Traditional methods for assessing waterbird food resources usually rely on ground surveys, which can accurately obtain characteristic indicators of plants (such as protein content, plant height, biomass, etc.). However, ground surveys have problems such as limited coverage, high destructiveness, and low efficiency. They are difficult to meet the needs of large-scale, high-frequency dynamic monitoring, which limits the effective assessment of waterbird food resources.

[0003] With the rapid development of remote sensing technology, it has become possible to monitor vegetation and wetlands over large areas using satellite data. Remote sensing vegetation indices (such as NDVI, EVI, etc.) can reflect characteristics such as vegetation biomass and health status, but it is currently impossible to directly quantify the relationship between remote sensing vegetation indices and plant characteristics related to the quality of waterfowl food resources (such as protein content, plant height, biomass, etc.), which cannot meet the requirements for effective assessment of waterfowl food resources. Summary of the invention

[0004] The purpose of the present invention is to solve the problem that existing remote sensing technology cannot accurately evaluate the quality of waterfowl food resources. Based on pixel-based foraging habitat classification, a method for evaluating the availability of waterfowl food resources is provided, which can quickly and accurately evaluate the potential food resources of waterfowl, especially the nutritional status and availability of wetland vegetation, on a large scale.

[0005] The present invention is achieved through the following technical solutions:

[0006] The present invention provides a method for evaluating the availability of waterfowl food resources. The method obtains a food resource quality evaluation model based on ground measured data and remote sensing data of foraging grounds in a study area, and then uses the food resource quality evaluation model to obtain plant nutritional components of the foraging grounds in the test area.

[0007] A further improvement of the present invention is:

[0008] The method comprises:

[0009] Step 1: Select the foraging habitat of waterbirds as the study area;

[0010] Step 2: Collect waterfowl food source plant data in the study area as ground measured data;

[0011] Step 3: Obtain remote sensing data of the study area and extract foraging sites in the study area;

[0012] Step 4: Use ground-truth data and remote sensing data to obtain a food resource quality assessment model;

[0013] Step 5: Use the food resource quality assessment model to obtain the plant nutrient composition of the foraging grounds in the test area.

[0014] A further improvement of the present invention is:

[0015] The operation of step 2 includes:

[0016] The measured plant height, measured biomass and measured protein content of each grid pixel in the study area were obtained, namely the ground measured data.

[0017] A further improvement of the present invention is:

[0018] In the step 3, remote sensing data of the study area is obtained based on the Google Earth Engine platform;

[0019] The remote sensing data includes Sentinel-1SAR and Sentinel-2 data.

[0020] A further improvement of the present invention is:

[0021] In step 3, Sentinel-1SAR and Sentinel-2 data are used to divide the study area into permanent water areas, mudflats, new wet grasslands, mature grasslands, overgrown grasslands, permanent vegetation areas, dry land and built-up areas, and new wet grasslands, mature grasslands and overgrown grasslands are extracted as foraging areas.

[0022] A further improvement of the present invention is:

[0023] The operations to divide the study area into permanent water body area, mudflat, new wet grassland, mature grassland, overgrown grassland, permanent vegetation area, and dry land and built-up area include:

[0024] The study area was divided into permanent water body area and non-water body area using Snetinel-1SAR data;

[0025] The study area was divided into permanent vegetation area and non-green vegetation area using Sentinel-2 data;

[0026] If an area is a non-water body area and a non-green vegetation area, it is determined to be a dry land and a built-up area;

[0027] If an area is a permanent water body area and is not a green vegetation area, then the area is determined to be a mudflat;

[0028] New wet grassland, mature grassland and overgrown grassland are determined based on the beach inundation time and vegetation index.

[0029] A further improvement of the present invention is:

[0030] The operations for determining new wet grassland, mature grassland, and overgrown grassland based on the beach flooding time and vegetation index include:

[0031] If the beach flooding time of an area is greater than the time threshold and the MSAVI value is less than the first MSAVI threshold, the area is determined to be a newly formed wet grassland;

[0032] If the beach inundation time of an area is less than or equal to the time threshold, and the MSAVI value is less than the second MSAVI threshold, the area is determined to be mature grassland;

[0033] If the beach flooding time of an area is less than or equal to the time threshold, and the MSAVI value is greater than or equal to the second MSAVI threshold, the area is determined to be over-mature grassland.

[0034] Wherein, the first MSAVI threshold is less than the second MSAVI threshold;

[0035] The beach flooding time is obtained using the annual water level data.

[0036] A further improvement of the present invention is:

[0037] The operation of step 4 includes:

[0038] The following seven vegetation indices were calculated within the sampling time range of the foraging sites in the study area based on the Google Earth Engine platform: MSAVI, MNDWI, NIRv, KNDVI, IRECI, MCARI, and EMBI; the characteristic values ​​of each vegetation index within each grid pixel were obtained: median, standard deviation, minimum, maximum, and interquartile range; one characteristic value of a vegetation index was taken as a variable, and the remote sensing index variable was obtained after deleting all variables with high variance inflation factors;

[0039] The measured plant height, measured biomass, and measured protein content of each grid pixel are matched with the remote sensing index variables of the grid pixel to form a data set, and then the data set is used to train the machine learning model to obtain a food resource quality assessment model.

[0040] A further improvement of the present invention is:

[0041] The machine learning model adopts a random forest model;

[0042] The operations of using the data set to train the machine model to obtain the food resource quality assessment model include:

[0043] Establish plant height random forest model, protein random forest model, and biomass random forest model;

[0044] The data set was used to train the plant height random forest model, protein random forest model, and biomass random forest model respectively to obtain the plant height prediction model, protein content prediction model, and biomass prediction model. These three prediction models constitute the food resource quality assessment model.

[0045] A further improvement of the present invention is:

[0046] The operation of step five includes:

[0047] Extract the foraging sites in the area to be studied, then obtain the remote sensing index variables of the foraging sites in the area to be studied and input them into the food resource quality assessment model;

[0048] The food resource quality assessment model outputs the predicted values ​​of plant height, protein content, and biomass in each grid pixel in the foraging grounds of the study area, that is, the plant nutritional components of the foraging grounds in the study area.

[0049] Compared with the prior art, the beneficial effects of the present invention are: the present invention can accurately identify foraging habitats, quickly and accurately evaluate the potential food resources of water birds on a large scale, especially the nutritional status and availability of wetland vegetation, and has repeatability and strong generalization ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flowchart of the steps of the method of the present invention;

[0051] Figure 2 Distribution map of the study area and sampling points in the embodiments;

[0052] Figure 3 Various index data of moss samples collected in the embodiment;

[0053] Figure 4 Wetland landscape analysis and foraging site extraction in the embodiment

[0054] Figure 5 The remote sensing index characteristic value of the study area in the embodiment;

[0055] Figure 6 Effects of random forest models on plant height, biomass, and protein in the examples

[0056] Figure 7 Contribution rates of different remote sensing indices and random forest model verification results in the embodiment;

[0057] Figure 8 Distribution diagram of plant height, biomass, and protein content in the foraging area to be tested in the embodiment. DETAILED DESCRIPTION

[0058] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0059] The present invention establishes a relationship model between remote sensing vegetation index and plant nutrients (including protein content, plant height, and biomass), thereby achieving spatiotemporal dynamic monitoring of waterfowl food resources and providing a scientific basis for the management of habitat food resources.

[0060] The present invention provides a method for evaluating the availability of waterfowl food resources, wherein the method obtains a food resource quality evaluation model based on ground measured data and remote sensing data of foraging grounds in a study area, and then uses the food resource quality evaluation model to obtain plant nutritional components of the foraging grounds in the test area. The ground measured data includes measured plant height, measured protein content, and measured biomass.

[0061] like Figure 1 As shown, the method of the present invention comprises:

[0062] Step 1. Select the study area: Select the foraging habitat of waterfowl, such as wetlands, lakes, swamps, etc., as the study area.

[0063] Step 2: Collect waterfowl food source plant data in the study area: Conduct ground surveys in the study area and collect waterfowl food source plant data. The waterfowl food source plant data refers to plant characteristic data related to waterfowl food resources, including plant height, coverage, protein content, biomass, etc. In order to ensure the comprehensiveness, representativeness, and typicality of the sampled samples, surveys and samples were conducted on food resources (mosses) of different growth potentials.

[0064] The size of a grid pixel of remote sensing data is 20 meters * 20 meters. In order to match the accuracy of remote sensing data, it is necessary to obtain the water bird food source plant data of each grid pixel (i.e., a sampling point). However, since the area of ​​20 meters * 20 meters is large, it is time-consuming and laborious to collect data on all plants in the entire area of ​​each grid pixel. Therefore, a 1 meter * 1 meter sample square frame is randomly delineated in each grid pixel, and N (for example, 12) mosses are randomly selected in the sample square frame. The plant height of each moss is measured, and then the average plant height of the N moss plants is taken as the measured plant height of the grid pixel. The existing visual method (Braun-Blanquet grade method) was used to estimate the coverage within the grid pixel (i.e., the proportion of the area covered by moss to the total area, which is divided into 5 levels: Level 1: Cover <1% (indicating that the species is rare and there are only a few plants); Level 2: Cover 1%-5% (indicating that the species is relatively rare); Level 3: Cover 5%-25% (indicating that the species is relatively common); Level 4: Cover 25% to 50% (indicating that the species is relatively abundant); Level 5: Cover greater than 50% (indicating that the species is very abundant)) as the coverage of the grid pixel. All the above-ground parts of the moss in the sample square frame within each grid pixel were harvested in the direction flush with the ground as samples and then brought back to the laboratory.

[0065] For each raster pixel, perform the following operations:

[0066] The samples in the grid pixel were sterilized at 105°C, then dried at 80°C to a constant temperature, and then weighed using a balance to obtain the dry weight of the biomass, which was converted to kg / m 2 , that is, the measured biomass of the grid pixel (unit: kg / m2). Since water birds mainly feed on the leaf parts of moss, the leaves and stems of the plants are separated in the laboratory, and only the leaves are crushed and ground with a ball mill. The total nitrogen content of the crushed leaves is measured with an elemental analyzer, and the total nitrogen content is multiplied by a coefficient of 6.25 to obtain the protein content. The protein content is taken as the measured protein content of the grid pixel. The measured plant height, measured biomass, and measured protein content of the grid pixels constitute the ground measured data.

[0067] Step 3: Remote sensing data collection and processing: Based on the Google Earth Engine platform, obtain multi-temporal remote sensing data corresponding to the study area, including Sentinel-1SAR and Sentinel-2 data provided by the Google Earth Engine platform. Use existing methods to pre-process the remote sensing data to remove cloud shadows, noise, and terrain correction and atmospheric correction.

[0068] The present invention proposes a pixel-based foraging habitat classification algorithm. Through the Google Earth Engine platform, 6 days of synthetic data of Sentinel-1SAR (synthetic aperture radar) and 10 days of synthetic data products of Sentinel-2 are synthesized respectively (the existing algorithm of the Google Earth Engine platform is used to obtain the synthetic data, which will not be repeated here). The missing data in the Sentinel-2 time series are filled by linear interpolation and smoothed by Savitzky-Golay filter (Savitzky-Golay filter smoothing can refer to: Cao R, Chen Y, Shen M, et al. A simple method to improve the quality of NDVI time-series data by integrating spatiotemporal information with the Savitzky-Golay filter [J]. Remote Sensing of Environment, 2018, 217: 244-257) to process the filled data to eliminate outliers.

[0069] Then, the 6-day synthetic data of Sentinel-1SAR and the 10-day synthetic data of Sentinel-2 were used to divide the study area into permanent water area, mudflat, new wet grassland, mature grassland, overgrown grassland, permanent vegetation area, dry land and built-up area, and new wet grassland, mature grassland and overgrown grassland were extracted as foraging areas, as follows:

[0070] Permanent water bodies have a very stable low VV state. Using Snetinel-1 SAR data and the existing Edge-OTSU dynamic adaptive threshold method (Edge-OTSU dynamic adaptive threshold method, see: Donchyts G, Schellekens J, Winsemius H, et al. A 30m resolution surface water mask including estimation of positional and thematic differences using landsat 8, srtm and openstreetmap: a case study in the Murray-Darling Basin, Australia [J]. Remote Sensing, 2016, 8 (5): 386.), the study area is divided into permanent water area and non-water area.

[0071] Permanent vegetation also has a very stable state. The study area is divided into permanent vegetation area and non-green vegetation area using Sentinel-2 data, that is, the study area is divided into green vegetation area and non-green vegetation area using the Normalized Difference Vegetation Index (NDVI) (Formula 1), the modified soil adjusted vegetation index (MSAVI) (Formula 2), and the modified normalized water index (MNDWI) (Formula 3). For example, a number of vegetation pixels are randomly selected in the study area to study the distribution characteristics of NDVI, MNDWI and MSAVI values, and the threshold method is used to detect green vegetation. Specifically, the method used in the present invention is as follows: if the three conditions of "NDVI ≥ 0.15, MSAVI ≥ 0.08 and MNDWI> 0" are met, it is determined to be a permanent vegetation area. If any of the above three conditions are not met, it is determined to be a non-green vegetation area.

[0072] Furthermore, if an area is a non-water area and a non-green vegetation area, the area is determined to be dry land and a built-up area; if an area is a permanent water area and a non-green vegetation area, the area is determined to be a mudflat.

[0073] Furthermore, seasonal vegetation zones such as new wet grassland, mature grassland, and overgrown grassland are determined by combining the characteristics of beach inundation time (the time when water bodies turn into non-water bodies) and plant phenology (vegetation index), as follows:

[0074] If the beach flooding time of an area is greater than the time threshold and the MSAVI value is less than the first MSAVI threshold, the area is determined to be a newly formed wet grassland;

[0075] If the beach inundation time of an area is less than or equal to the time threshold, and the MSAVI value is less than the second MSAVI threshold, the area is determined to be mature grassland;

[0076] If the beach flooding time of an area is less than or equal to the time threshold, and the MSAVI value is greater than or equal to the second MSAVI threshold, the area is determined to be over-mature grassland.

[0077] Among them, the first MSAVI threshold is less than the second MSAVI threshold. The flooding time of the beach is obtained using the water level data throughout the year, as follows: extract the date 1 of the highest water level of the year (for example, August 1) and the date 2 of the lowest water level (for example, February 1), use the water level data of Snetinel-1SAR every 6 days, superimpose all the water surface distribution maps from date 1 to date 2 (obtained using the data of the water body area distinguished by Snetinel-1SAR), and make a histogram of the flooding time of the beach. For each pixel, determine the date 3 (for example, May 1) when it transitions from water coverage (i.e., water body) to non-water coverage (i.e., non-water body). Then calculate the difference between date 3 and date 1 (for example, 3 months) as the flooding time of the beach.

[0078] For example:

[0079] New wet grassland = (flooding time > 1 month and MSAVI < 0.4)

[0080] Mature grassland = (flooding time ≤ 1 month and MSAVI < 0.6)

[0081] Over-mature grassland = (flooding time ≤ 1 month and MSAVI ≥ 0.6)

[0082] The following is a brief introduction to the above three indicators:

[0083] NDVI (Normalized Difference Vegetation Index): A commonly used vegetation index used to assess vegetation coverage and health. It is widely used to monitor vegetation growth, ecosystem health, and agricultural applications. The formula is as follows:

[0084] Formula 1: NDVI = (NIR-Red) / (NIR+Red), where NIR represents the reflectance value of the near-infrared band and Red represents the reflectance value of the red band.

[0085] Reference: Tucker CJ, Red and photographic infrared linear combinations for monitoring vegetation. Remote sensing of Environment, 1979 8(2):127–150.

[0086] MSAVI (the modified soil adjusted vegetation index): The modified soil adjusted vegetation index (SAVI) further reduces the impact of bare soil and is more suitable for areas with sparse vegetation or more exposed soil. The formula is as follows:

[0087] Formula 2:

[0088] Among them, NIR is the reflection value of the near infrared band, and Red is the reflection value of the red light band.

[0089] References: Qi J, Chehbouni A, Huete AR, et al. A modified soil adjustedvegetation index [J]. Remote sensing of environment, 1994, 48(2): 119-126.

[0090] MNDWI (The modified NDWI, improved normalized water index): improved NDWI, enhanced the detection ability of water bodies, effectively removed the impact of vegetation and buildings, and is often used in water resource management and water body extraction. Its formula is as follows:

[0091] Formula 3:

[0092] Among them, Green is the reflection value of the green light band, Blue is the reflection value of the blue light band, and SWIR is the reflection value of the shortwave infrared band.

[0093] References: Xu H. Modification of normalized difference water index (NDWI) to enhance open water features in remotely sensed imagery[J]. International journal of remote sensing, 2006, 27(14): 3025-3033.

[0094] Step 4: Machine learning model establishment: Continue to calculate the following seven vegetation indices within the sampling time range of the foraging sites in the study area (for example, the selected time period is February to April) based on the Google Earth Engine platform: MSAVI, MNDWI, Near-Infrared Reflectance of Vegetation (NIRv) (Formula 4) 、 The kernel NDVI (KNDVI) is the normalized difference vegetation index based on the kernel method (Formula 5). 、Inverted Red-Edge Chlorophyll Index (IRECI) (Formula 6), Modified chlorophyll absorption ratio index (MCARI) (Formula 7), Enhanced Modified Bare soil Index (EMBI) (Formula 8).

[0095] Seven vegetation indices are collected for each grid pixel, and harmonic regression analysis is performed (Harmonic Regression Analysis is an existing statistical method based on Fourier series, which is often used to process time series data with periodic changes. This method can detect the periodic fluctuations of vegetation indices in time series. By superimposing multiple harmonic components, time series data can be effectively smoothed to eliminate high-frequency noise. Because remote sensing image data is not available every day, and sometimes cloud cover data is not good, there is not continuous data for a period of time. Through harmonic regression analysis, the remote sensing index of each day during the survey time can be obtained) to obtain the data after harmonic processing, and the characteristic values ​​of the vegetation index of each grid pixel are obtained in turn, as follows:

[0096] During the sampling period, from February to April, as time goes by, each grid pixel has multiple remote sensing data for the vegetation index. For example, a remote sensing image may have one scene every month, and one remote sensing data can be extracted. However, the amount of data is too small, because there are fewer data due to reasons such as clouds and rain. Therefore, harmonic regression analysis is required, which is to fit all the extracted remote sensing data through the harmonic function to obtain the time-varying dynamic data of a vegetation index of the grid pixel. For example, a KNDVI value (i.e., remote sensing data) is obtained every 01 days from February to April for the grid pixel. The multiple remote sensing data of each vegetation index in each grid pixel are calculated to obtain 5 eigenvalues, i.e., median, standard deviation, minimum, maximum, and interquartile range. The calculated 5 eigenvalues ​​are the eigenvalues ​​of the grid pixel. For example, as time goes by, the vegetation index KNDVI of the grid pixel has 10 remote sensing data (i.e., 10 MNDWI values). The maximum and minimum values ​​of these 10 MNDWI values ​​are the maximum and minimum values ​​of the vegetation index, and the median of these 10 MNDWI values ​​is the median of the vegetation index. The standard deviation and interquartile range can also be calculated using existing mathematical methods.

[0097] Then, the eigenvalues ​​of each vegetation index are randomly sampled and collinearity tests are performed, and one eigenvalue of a vegetation index is used as a variable (for example, there are 7 vegetation indices, each vegetation index has 5 eigenvalues, so there are 35 variables in total), and then all variables with high variance inflation factors are deleted to obtain remote sensing index variables, so that the number of remote sensing index variables is less than 35. Specifically, if the variance inflation factor VIF of a variable is greater than 10, the variable is determined to be a variable with a high variance inflation factor, which has a strong correlation with other independent variables, which may cause the model to be unstable or unreliable, so the variable needs to be deleted. After deleting all variables with high variance inflation factors, the remaining variable is the remote sensing index variable. In this way, after the above processing, the remaining eigenvalues ​​of different vegetation indices are not necessarily the same. For example, KNDVI leaves the median, standard deviation, and minimum value, while EMBI leaves the standard deviation, minimum value, maximum value, and interquartile range value.

[0098] The following is a brief introduction to the above five indicators:

[0099] NIRv (Near-Infrared Reflectance of Vegetation, near-infrared vegetation index), the formula is as follows:

[0100] Formula 4: NIRv=NDVI×NIR

[0101] Among them, NIR is the reflectance value of the near-infrared band.

[0102] KNDVI (the kernel NDVI, normalized difference vegetation index based on kernel method): The nonlinear feature extraction capability of NDVI is enhanced by the kernel function. The formula is as follows:

[0103] Formula 5:

[0104] Among them, NIR is the reflectance value of the near-infrared band, red is the reflectance value of the red light band, and σ is the length variable.

[0105] References: Camps-Valls G, Campos-Taberner M, Moreno-Martínez et al.Aunified vegetation index for quantifying the terrestrial biosphere[J].ScienceAdvances,2021,7(9):eabc7447.

[0106] IRECI (Inverted Red-Edge Chlorophyll Index): Identifies the chlorophyll content in vegetation through the red edge band, especially used in agriculture to monitor crop health, nitrogen content and photosynthesis efficiency. The formula is as follows:

[0107] Formula 6:

[0108] Among them, Red is the reflection value of the red light band, and RedEdge is the reflection value of the red edge band.

[0109] References: Frampton, William James, et al. Evaluating the capabilities of Sentinel-2 for quantitative estimation of biophysical variables invegetation. ISPRS journal of photogrammetry and remote sensing 82(2013):83-92.

[0110] MCARI (Modified chlorophyll absorption ratio index): used to detect chlorophyll content in plants, especially suitable for eliminating soil background interference, and improving the sensitivity of vegetation biomass and health status. The formula is as follows:

[0111] Formula 7:

[0112] Among them, Red is the reflection value of the red light band, RedEdge is the reflection value of the red edge band, and Blue is the reflection value of the blue light band.

[0113] Reference: Daughtry, Craig ST, et al. Estimating corn leaf chlorophyllconcentration from leaf and canopy reflectance. Remote sensing of Environment 74.2 (2000): 229-239.

[0114] EMBI (Enhanced Modified Bare Soil Index): Enhanced the detection capability of bare soil and improved the accuracy of soil index. It is often used to identify bare soil in arid areas or fallow areas of farmland. The formula is as follows:

[0115] Formula 8:

[0116]

[0117] Among them, NIR is the reflectance value of the near-infrared band, and SWIR is the reflectance value of the short-wave infrared band.

[0118] References: Zhao, Y. & Zhu, Z. ASI: An artificial surface Index for Landsat8imagery. International Journal of Applied Earth Observation and Geoinformation, 2022, 107, 102703.

[0119] The near infrared band, shortwave infrared band, red light band, red edge band, and blue light band in the above 8 formulas are all obtained from Sentinel-2.

[0120] The measured plant height, measured biomass, and measured protein content of each grid pixel in the foraging area are matched with the remote sensing index variable of the grid pixel to form a data set. The remote sensing index variable in the data set corresponds to the ground measured data (measured plant height, measured biomass, and measured protein content), that is, the remote sensing index variable of the first grid pixel corresponds to the ground measured data of the first grid pixel, and the remote sensing index variable of the second grid pixel corresponds to the ground measured data of the second grid pixel, and so on. Then the data set is divided into a training set and a test set, for example, 80% of the data set is used as a training set and 20% as a test set. Then, the existing random forest algorithm is used to establish nonlinear regression models between protein content, plant height, biomass and remote sensing index variables, namely, a plant height random forest model, a protein random forest model, and a biomass random forest model. Because plant height, protein, and biomass are different indicators reflecting the quality of food resources of moss, and different indicators have different relationships with remote sensing index variables, a plant height random forest model, a protein random forest model, and a biomass random forest model are established respectively.

[0121] The input data of the plant height random forest model, protein random forest model, and biomass random forest model are remote sensing index variables, and the output data are: predicted values ​​of plant height, predicted values ​​of protein content, and predicted values ​​of biomass, respectively.

[0122] The existing training method was used to train the plant height random forest model, protein content random forest model, and biomass random forest model using the training set and the test set, and three optimal models were obtained, namely the plant height prediction model, the protein content prediction model, and the biomass prediction model. These three prediction models constitute the food resource quality assessment model. When training with the training set, the model is trained using the remote sensing index variables and the corresponding measured plant height, measured protein content, and measured biomass. When testing, the remote sensing index variables in the test set are input into the model, and then the predicted values ​​output by the model are compared with the measured plant height, measured protein content, and measured biomass in the test set to evaluate the accuracy of the model.

[0123] Step 5. Food resource quality assessment: extract foraging grounds in the study area (use the same method as step 3 to extract foraging grounds in the study area, i.e., new wet grassland, mature grassland, and overgrown grassland in the study area), then extract remote sensing index variables of the foraging grounds (use the same method as step 4 to obtain 35 variables, and then extract the remote sensing index variables that have been determined in step 4 from the 35 variables), and input them into the food resource quality assessment model. The food resource quality assessment model outputs the predicted values ​​of protein content, plant height, and biomass in each grid pixel (i.e., predicted value of plant height, predicted value of protein content, predicted value of biomass). Based on these predicted values, a spatial distribution map of biomass and protein content in the study area can be generated to determine the availability of food resources.

[0124] Embodiments of the present invention are as follows:

[0125] Step 1. Select the study area: Select the Poyang Lake natural wetland in Jiangxi as the study area;

[0126] Step 2: Ground data collection, i.e. obtaining waterfowl food source plant data: In March 2024, at 85 sampling points in Poyang Lake, Jiangxi Province, samples were collected in one sample box at each sampling point (one sampling point corresponds to a 20m*20m grid pixel), for a total of 85 food resource samples, such as Figure 2 As shown, Figure 2 Represents the study area and sampling point distribution, where a, b, c, d, e, and f represent different areas.

[0127] The collected samples were brought back to the laboratory, and then dried at 80°C after being cured at 105°C. Since wintering geese mainly feed on the leaves of moss, the leaves and stems of the plants were separated, and only the leaves were crushed and ground with a ball mill. The total nitrogen content was measured using an elemental analyzer and converted into protein content. Combined with the on-site survey, the plant height, coverage, biomass, and protein content of the plants at the sampling points were obtained.

[0128] The data of various indicators of moss samples collected are as follows Figure 3 As shown in the figure, the plant height range of the sampling points is 26.18-118.76 cm, the coverage range of the sampling points is 75%-95%; the biomass range is 0.23-0.99 kg / m 2 The protein content ranged from 12.57% to 28.98%, and the coverage was from 75% to 95%, which means that the sampling points were considered to be relatively uniform moss communities.

[0129] Step 3: Remote sensing data collection and processing:

[0130] 3.1 Foraging site extraction

[0131] Remote sensing images of the study area in 2023 were obtained, and vegetation mapping was performed based on the thresholds of NDVI ≥ 0.15, MSAVI ≥ 0.08, and MNDWI > 0, combined with the flooding time of the beach. The area was finally divided into permanent water areas, mudflats, new wet grasslands, mature grasslands, overgrown grasslands, permanent vegetation areas, and dry land and construction areas. New wet grasslands, mature grasslands, and overgrown grasslands were used as foraging areas. Wetland landscape analysis and foraging area extraction are as follows: Figure 4 shown.

[0132] 3.2 Vegetation index extraction

[0133] Based on the remote sensing image data of the plant growth stage (September 2023 to April 2024, 2 scenes per month), the remote sensing index data are extracted and calculated. Due to cloud cover, some data are missing. Based on the harmonic function, the "KNDVI", "MNDWI", "NIRv", "MSAVI", "MCARI", "IRECI", and "EMBI" indices in the region are subjected to harmonic regression analysis. The median, standard deviation, minimum, maximum, and interquartile range of each remote sensing index within the survey sampling time range (February-April 2024) are fitted and calculated to obtain the remote sensing index variables. The characteristic values ​​of the remote sensing index in the study area are as follows: Figure 5 shown.

[0134] Step 4: Machine learning model establishment:

[0135] Based on the plant height, biomass, protein content and remote sensing index variables of the sampling points, random forest models were created respectively, and the data were divided into training sets (80%) and test sets (20%) to determine the three most accurate prediction models. The contribution rate of different remote sensing index feature values ​​to the random forest model is shown in Figure 2. Figure 6 The random forest model validation results are shown in Table 1 and Figure 7The RMSE in Table 1 is the root mean square error of the random forest model, which is an indicator to measure the difference between the model prediction value and the actual value. The smaller the value, the higher the prediction accuracy of the model. The equation in Table 1 represents the relationship between the model prediction value and the measured value, where the x value in the equation is the measured value, the y value is the predicted value, and the RSE is the square root of the error of the equation. The smaller the RSE, the higher the prediction accuracy of the model. In addition, the R 2 They are 0.35, 0.35, and 0.46 respectively, and p is less than 0.05, which is statistically significant, indicating that the performance of the model is good. Figure 6 Refers to the top five characteristic indices that contribute most to each model. For example, on the left, for plant height, KNDVImin, MCARIsd, MCARImedian, IRECImin, and IRECIsd determine plant height; Figure 7 The results show the performance of the model. The upper figure is the RMSE value of the model, which is the root mean square error of the random forest model. The lower figure is the relationship between the predicted value and the measured value. The predicted value and the measured value are positively correlated. 2 The higher it is, the closer the predicted value is to the measured value, and the better the model performance is. p < 0.05 proves that the correlation passes the significance test, which is also the result of showing the performance of the random forest model).

[0136] Table 1 Random forest model performance characterization results

[0137]

[0138] Step 5: Food resource quality assessment:

[0139] According to the extracted remote sensing index variables of the foraging grounds in the study area (i.e., Poyang Lake), based on the above three prediction models, the spatial distribution maps of plant height, biomass, and protein content of the foraging grounds in the survey time range (February 2024-April 2024) were predicted, and the food resource availability indicators were extracted as follows:

[0140] Plant height range is 36.15-99.57cm; biomass is 0.31-0.83kg / m 2 ; Protein content is 15.85-23.64%.

[0141] The grid size of the remote sensing image is 20m×20m (one grid corresponds to one plant height value, one protein content value, and one biomass value). Based on the predicted results, the total biomass of the Poyang Lake foraging area was calculated to be 8.09×10 8 kg, the total protein weight is 1.45×10 8 kg (Total biomass is the area of ​​each grid multiplied by the biomass per unit area of ​​each grid (kg / m 2) values ​​and then sum them up; the total protein mass is calculated by multiplying the area of ​​each grid by the biomass per unit area of ​​each grid (kg / m 2 ) multiplied by the protein per unit area of ​​each grid (%) and then summed). The estimated results obtained based on the prediction results are shown in Table 2. The distribution diagram of plant height, biomass, and protein content obtained based on the prediction results is shown in Figure 8 shown.

[0142] Table 2 Estimation results of biomass and protein in the study area

[0143] index average value median Standard Deviation total Plant height 61.60 62.24 6.29 / Biomass 0.52 0.52 0.05 <![CDATA[8.09×10 8 kg]]> Total protein quality 17.96 17.79 1.18 <![CDATA[1.45×10 8 kg]]>

[0144] Compared with traditional ground survey methods, the method of the present invention can quickly and in real time monitor the dynamic changes of food resources without destroying vegetation. It can achieve accurate assessment of food resources in large-scale areas, breaking the spatial and temporal limitations of traditional ground surveys, and is fast, accurate and efficient. Moreover, the method of the present invention can be repeatedly applied in different geographical regions and time periods to ensure the consistency and reliability of the assessment results, and has repeatability. In addition, since the method of the present invention is a model established based on multiple data sources (remote sensing and measured data), it can be applied to different environments and ecosystems, improving the universality and adaptability of the assessment, and has strong generalization ability.

[0145] In summary, the present invention combines remote sensing data with ground-measured data, uses machine learning methods to establish quantitative relationships between plant growth characteristics and nutritional components and remote sensing indices, and constructs a prediction model for food resources, which is an effective way to improve the accuracy and efficiency of food resource assessment.

[0146] The above technical solution is only one implementation mode of the present invention. For those skilled in the art, it is easy to make various types of improvements or modifications based on the principles disclosed in the present invention, and it is not limited to the technical solution described in the above specific embodiments of the present invention. Therefore, the above description is only preferred and does not have a restrictive meaning.

Claims

1. A method for assessing the availability of waterfowl food resources, characterized in that: The method obtains a food resource quality assessment model based on ground measured data and remote sensing data of foraging grounds in a study area, and then uses the food resource quality assessment model to obtain plant nutrient components of the foraging grounds in the area to be tested.

2. The method for evaluating the availability of waterfowl food resources according to claim 1, characterized in that: The method comprises: Step 1: Select the foraging habitat of waterbirds as the study area; Step 2: Collect waterfowl food source plant data in the study area as ground measured data; Step 3: Obtain remote sensing data of the study area and extract foraging sites in the study area; Step 4: Use ground-truth data and remote sensing data to obtain a food resource quality assessment model; Step 5: Use the food resource quality assessment model to obtain the plant nutrient composition of the foraging grounds in the test area.

3. The method for evaluating the availability of waterfowl food resources according to claim 2, characterized in that: The operation of step 2 includes: The measured plant height, measured biomass and measured protein content of each grid pixel in the study area were obtained, namely the ground measured data.

4. The method for evaluating the availability of waterfowl food resources according to claim 2, characterized in that: In the step 3, remote sensing data of the study area is obtained based on the Google Earth Engine platform; The remote sensing data includes Sentinel-1SAR and Sentinel-2 data.

5. The method for evaluating the availability of waterfowl food resources according to claim 2, characterized in that: In step 3, Sentinel-1SAR and Sentinel-2 data are used to divide the study area into permanent water areas, mudflats, new wet grasslands, mature grasslands, overgrown grasslands, permanent vegetation areas, dry land and built-up areas, and new wet grasslands, mature grasslands and overgrown grasslands are extracted as foraging areas.

6. The method for assessing the availability of waterfowl food resources according to claim 5, characterized in that: The operations to divide the study area into permanent water body area, mudflat, new wet grassland, mature grassland, overgrown grassland, permanent vegetation area, and dry land and built-up area include: The study area was divided into permanent water body area and non-water body area using Snetinel-1SAR data; The study area was divided into permanent vegetation area and non-green vegetation area using Sentinel-2 data; If an area is a non-water body area and a non-green vegetation area, it is determined to be a dry land and a built-up area; If an area is a permanent water body area and is not a green vegetation area, then the area is determined to be a mudflat; New wet grassland, mature grassland and overgrown grassland are determined based on the beach inundation time and vegetation index.

7. The method for evaluating the availability of waterfowl food resources according to claim 6, characterized in that: The operations for determining new wet grassland, mature grassland, and overgrown grassland based on the beach flooding time and vegetation index include: If the flooding time of a beach in an area is greater than the time threshold and the MSAVI value is less than the first MSAVI threshold, the area is determined to be a newly formed wet grassland; If the beach inundation time of an area is less than or equal to the time threshold, and the MSAVI value is less than the second MSAVI threshold, the area is determined to be mature grassland; If the beach inundation time of an area is less than or equal to the time threshold, and the MSAVI value is greater than or equal to the second MSAVI threshold, the area is determined to be over-mature grassland; Wherein, the first MSAVI threshold is less than the second MSAVI threshold; The beach inundation time is obtained using the annual water level data.

8. The method for assessing the availability of waterfowl food resources according to claim 2, characterized in that: The operation of step 4 includes: The following seven vegetation indices were calculated within the sampling time range of the foraging sites in the study area based on the Google Earth Engine platform: MSAVI, MNDWI, NIRv, KNDVI, IRECI, MCARI, and EMBI; the characteristic values ​​of each vegetation index within each grid pixel were obtained: median, standard deviation, minimum, maximum, and interquartile range; one characteristic value of a vegetation index was taken as a variable, and the remote sensing index variable was obtained after deleting all variables with high variance inflation factors; The measured plant height, measured biomass, and measured protein content of each grid pixel are matched with the remote sensing index variables of the grid pixel to form a data set, and then the data set is used to train the machine learning model to obtain a food resource quality assessment model.

9. The method for evaluating the availability of waterfowl food resources according to claim 8, characterized in that: The machine learning model adopts a random forest model; The operations of using the data set to train the machine model to obtain the food resource quality assessment model include: Establish plant height random forest model, protein random forest model, and biomass random forest model; The data set was used to train the plant height random forest model, protein random forest model, and biomass random forest model respectively to obtain the plant height prediction model, protein content prediction model, and biomass prediction model. These three prediction models constitute the food resource quality assessment model.

10. The method for evaluating the availability of waterfowl food resources according to claim 2, characterized in that: The operation of step five includes: Extract the foraging sites in the area to be studied, then obtain the remote sensing index variables of the foraging sites in the area to be studied and input them into the food resource quality assessment model; The food resource quality assessment model outputs the predicted values ​​of plant height, protein content, and biomass in each grid pixel in the foraging grounds of the study area, that is, the plant nutritional components of the foraging grounds in the study area.

Images