Method and system for screening grass seed adaptability of saline-alkali grassland based on multispectral remote sensing

Abstract

The application discloses a salt and alkali grassland grass seed adaptability screening method and system based on multispectral remote sensing, and particularly relates to the technical field of agricultural resource informatization management and evaluation, and is used for solving the problems that the existing salt and alkali grassland grass seed screening method is dependent on manual work, is inefficient, and the evaluation result is greatly affected by local environment variation of a test site; a germplasm database is established and a test plot is arranged; multispectral remote sensing images of the test area are acquired; mixed pixel decomposition is performed on the images to obtain the abundance proportion of different grass seeds in each pixel and pure end-member spectral information of the different grass seeds; an inversion model is established based on the end-member spectral information and ground measured biomass data; a biomass spatial distribution map is generated in combination with the abundance proportion and the inversion model, and grass seeds are adaptability screened according to the biomass spatial distribution map; large-area data are rapidly acquired through remote sensing means, and mixed pixel decomposition technology is used to strip off environmental interference, so that efficient, objective and accurate evaluation on adaptability of multiple grass seeds is realized.

Description

Technical Field

[0001] This invention relates to the field of agricultural resource information management and evaluation technology, and more specifically, to a method and system for screening grass species adaptability in saline-alkali grasslands based on multispectral remote sensing. Background Technology

[0002] In existing technologies, ecological restoration of saline-alkali degraded grasslands requires the selection of superior grass species that can adapt to the local high altitude, low temperature, and saline-alkali complex stress. Traditional screening methods mainly rely on establishing ground germplasm resource nurseries, and conducting long-term, fixed-point artificial field observations and sampling measurements on different grass species to assess their growth adaptability, yield, and stress resistance, which falls under the category of agricultural resource management and evaluation.

[0003] However, the aforementioned existing technologies have drawbacks such as reliance on manual field observation, long evaluation cycles, and significant influence from environmental variations in specific test plots. This makes it difficult to quickly, comprehensively, and objectively evaluate the adaptability of various candidate grass species over a large spatial range, thus restricting the efficient screening of grass species with strong adaptability and excellent performance. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method and system for screening grass species adaptability in saline-alkali grassland based on multispectral remote sensing to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for screening grass species adaptability in saline-alkali grasslands based on multispectral remote sensing includes the following steps:

[0007] S1. Establish a germplasm resource database of salt-tolerant forage grasses in the target area. The germplasm resource database contains identification information of various target candidate grass species.

[0008] S2. Select saline-alkali grassland as the test area within the target area, and arrange test plots containing multiple target candidate grass species within the test area according to the germplasm resource database.

[0009] S3. Acquire multispectral remote sensing images of the experimental area during at least one critical growth period;

[0010] S4. Perform hybrid pixel decomposition on the multispectral remote sensing image to obtain the abundance ratio of target candidate grass species and the corresponding endmember spectral information in each pixel; based on the endmember spectral information and the measured aboveground biomass data of the corresponding experimental plot, establish an aboveground biomass inversion model.

[0011] S5. Combine the abundance ratio and aboveground biomass inversion model to obtain the spatial distribution information of aboveground biomass in the experimental area, and conduct adaptive screening of target candidate grass species based on the spatial distribution information of aboveground biomass.

[0012] Furthermore, S1 includes:

[0013] Collect identification information for various target candidate grass species, including species name, place of origin, salt and alkali tolerance level, cold tolerance level, and key phenological period information;

[0014] The identification information of various target candidate grass species collected will be digitally stored to form a germplasm resource database of salt-tolerant forage grasses.

[0015] Furthermore, S2 includes:

[0016] Within the target area, saline-alkali grasslands with different degrees of salinization were selected as experimental areas;

[0017] Based on the identification information of various target candidate grass species in the salt-tolerant forage germplasm resource database, independent test plots were divided for each target candidate grass species within the test area, and isolation zones were set up between test plots of different target candidate grass species.

[0018] Record the spatial location information of each experimental plot and the identification information of the corresponding target candidate grass species to form spatial distribution data of the experimental plots.

[0019] Furthermore, S3 includes:

[0020] Based on the spatial distribution data of the experimental plots, the acquisition range of multispectral remote sensing images covering all experimental plots was determined;

[0021] During the critical growth period of the target candidate grass species, remote sensing images of the experimental area were acquired through a remote sensing platform to obtain raw multispectral remote sensing images with geographic coordinate information.

[0022] Atmospheric correction and geometric fine correction were performed on the original multispectral remote sensing images to obtain multispectral remote sensing images that accurately match the geographical location of the experimental area.

[0023] Furthermore, the critical reproductive period includes the regrowth period, the rapid growth period, and the maturity period.

[0024] Furthermore, S4 includes:

[0025] Based on the spatial distribution data of the experimental plots and the corresponding identification information of the target candidate grass species, the image regions of each target candidate grass species experimental plot are extracted from the corrected multispectral remote sensing images as end-member references.

[0026] The linear mixed pixel decomposition method was used to decompose the corrected multispectral remote sensing image and solve for the abundance ratio of each target candidate grass species in each pixel and its purified endmember spectral information.

[0027] Aboveground biomass data were collected in experimental plots corresponding to each target candidate grass species.

[0028] Based on the endmember spectral information of each target candidate grass species and the measured aboveground biomass data collected in the corresponding experimental plots, an aboveground biomass inversion model for the target candidate grass species was established using regression analysis.

[0029] Furthermore, linear mixed pixel decomposition methods include constrained least squares decomposition.

[0030] Furthermore, S5 includes:

[0031] Based on the abundance ratio of each target candidate grass species in each cell and its corresponding aboveground biomass inversion model, the aboveground biomass contribution value of each target candidate grass species in each cell is calculated.

[0032] The aboveground biomass contribution values ​​of all target candidate grass species within each pixel are summed to generate aboveground biomass spatial distribution information covering the entire experimental area.

[0033] Based on the spatial distribution information of aboveground biomass, the average aboveground biomass of each target candidate grass species in the corresponding experimental plot was extracted and statistically analyzed.

[0034] The adaptability of multiple target candidate grass species was ranked and evaluated based on the average aboveground biomass of each target candidate grass species.

[0035] Furthermore, the ranking evaluation also incorporates a comprehensive assessment of the salt and alkali tolerance levels and cold tolerance levels of the target candidate grass species.

[0036] On the other hand, the present invention provides a multispectral remote sensing-based system for screening grass species adaptability in saline-alkali grasslands, comprising the following modules:

[0037] The germplasm bank construction module is used to establish a germplasm resource database of salt-tolerant forage grasses in the target area. The germplasm resource database contains identification information of various target candidate grass species.

[0038] The experimental layout module is used to select saline-alkali grassland as the experimental area within the target area, and to arrange experimental plots containing multiple target candidate grass species within the experimental area based on the germplasm resource database.

[0039] The image acquisition module is used to acquire multispectral remote sensing images of the experimental area during at least one critical growth period.

[0040] The model building module is used to perform mixed pixel decomposition on multispectral remote sensing images to obtain the abundance ratio of target candidate grass species and the corresponding endmember spectral information in each pixel; based on the endmember spectral information and the measured aboveground biomass data of the corresponding experimental plots, an aboveground biomass inversion model is established.

[0041] The screening decision module is used to combine the abundance ratio and the aboveground biomass inversion model to obtain the spatial distribution information of aboveground biomass in the experimental area, and to conduct adaptive screening of target candidate grass species based on the spatial distribution information of aboveground biomass.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] By comprehensively utilizing multispectral remote sensing technology and hybrid pixel decomposition model, a complete data-driven screening process for grass species adaptability was constructed. This process can quickly acquire multispectral images covering the entire experimental area and complete simultaneous observation of all experimental plots at once, replacing the time-consuming and laborious manual point-by-point surveys of the traditional method. By associating remote sensing images with specific experimental plots deployed on the ground and using hybrid pixel decomposition technology to extract pure target grass species spectral signals from spectrally mixed pixels, interference from environmental backgrounds such as weeds and bare soil is effectively removed. This achieves efficient and simultaneous data acquisition and preliminary information purification of the growth status of multiple candidate grass species over a large spatial range, significantly improving the breadth, speed, and objectivity of data acquisition. It provides a technical foundation for solving the problems of low efficiency and lengthy cycle of traditional screening methods.

[0044] This method enables precise and quantitative reconstruction of the spatial distribution of grassland biomass. By establishing a dedicated biomass estimation model based on the purified grass species spectrum and weighting the calculation by combining the area proportion of the grass species in each pixel, a high-precision biomass spatial distribution map is generated. This allows the evaluation of the adaptability of candidate grass species to leap from discrete inference based on individual sampling points to a comprehensive evaluation based on continuous spatial information of the entire growth area. This not only significantly improves the representativeness and accuracy of the evaluation results and effectively overcomes the shortcomings of traditional methods that are greatly affected by local plot variations, but also provides reliable data decision support for the final multi-factor comprehensive ranking and screening by combining yield and stress resistance levels. This realizes the transformation of agricultural germplasm resource evaluation and management from experience-based to data-driven and precise. Attached Figure Description

[0045] Figure 1 This is a flowchart of the method for screening grass species adaptability in saline-alkali grasslands based on multispectral remote sensing, as described in this invention.

[0046] Figure 2 This is a schematic diagram of the structure of the multispectral remote sensing-based grass species adaptability screening system for saline-alkali grasslands according to the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] Example 1: Figure 1 This invention presents a method for screening grass species adaptability in saline-alkali grasslands based on multispectral remote sensing, which includes the following steps:

[0049] S1. Establish a germplasm resource database of salt-tolerant forage grasses in the target area. The germplasm resource database contains identification information of various target candidate grass species.

[0050] S2. Select saline-alkali grassland as the test area within the target area, and arrange test plots containing multiple target candidate grass species within the test area according to the germplasm resource database.

[0051] S3. Acquire multispectral remote sensing images of the experimental area during at least one critical growth period;

[0052] S4. Perform hybrid pixel decomposition on the multispectral remote sensing image to obtain the abundance ratio of target candidate grass species and the corresponding endmember spectral information in each pixel; based on the endmember spectral information and the measured aboveground biomass data of the corresponding experimental plot, establish an aboveground biomass inversion model.

[0053] S5. Combine the abundance ratio and aboveground biomass inversion model to obtain the spatial distribution information of aboveground biomass in the experimental area, and conduct adaptive screening of target candidate grass species based on the spatial distribution information of aboveground biomass.

[0054] S1. Establish a germplasm resource database of salt-tolerant forage grasses in the target area. The germplasm resource database contains identification information of various target candidate grass species. The specific implementation is as follows:

[0055] To establish a database of salt-tolerant forage germplasm resources for the target region, the identification information of various candidate forage species was first collected. This collection was conducted through multiple methods, including searching publicly available plant germplasm resource banks, reviewing published academic literature and breeding reports, and investigating local agricultural and pastoral department variety registration records. Each candidate forage species was recorded independently with the collected identification information. The species name refers to the common Chinese name and Latin scientific name of the candidate forage species in plant taxonomy, for example, *Elymus nutans*. The place of origin information refers to the geographical location where the candidate forage species was initially collected or bred, recorded down to the county-level administrative division, and includes the altitude range of the candidate forage species' native habitat, recorded in meters. For example, if the place of origin is Gonghe County, Qinghai Province, the altitude range is 2800 to 3200 meters. The salt tolerance level is a qualitative grading evaluation of the candidate forage species' ability to tolerate soil salinity and alkali stress. The salt tolerance level is determined based on two aspects: first, reviewing existing research literature on the target candidate grass species, which records data on germination rate, survival rate, or biomass reduction percentage in solutions with specific salt concentrations or saline-alkali soils; second, making analogical inferences by referring to reports on the salt tolerance of known grass species of the same genus or with similar ecological habits. Based on literature data or analogical inferences, the salt tolerance level is divided into three levels: high, medium, and low. The threshold for level division is set based on the critical value of biomass response in common studies. For example, if the literature records that the biomass reduction of the target candidate grass species is less than 30% under a total soil salt content of 1%, the salt tolerance level of the target candidate grass species is set as high; if the biomass reduction is between 30% and 60%, it is set as medium; and if the biomass reduction is greater than 60%, it is set as low. When literature data is unavailable, the level is determined by analogy based on the general performance of closely related species of the same genus under the same salt gradient. The cold tolerance level is a qualitative grading evaluation of the target candidate grass species' ability to withstand low-temperature cold stress. The cold hardiness rating is primarily based on historical data of the lowest winter temperatures in the native regions of the target candidate grass species, known overwintering survival rate data, and whether the target candidate grass species possesses persistent or dormant characteristics. For example, grass species with a historical lowest temperature in their native regions below -25 degrees Celsius and a documented overwintering survival rate above 80% are rated as high cold hardiness; those with an overwintering survival rate between 50% and 80% are rated as medium; and those below 50% are rated as low. Key phenological period information refers to the time points or periods during which the target candidate grass species are expected to experience key growth and development stages under the climatic conditions of the experimental area, mainly including the greening period, rapid growth period, and maturity period. Key phenological period information is obtained by analyzing phenological observation records of the target candidate grass species in their native regions or similar climatic zones, combined with historical meteorological data of the experimental area, especially dates when the daily average temperature consistently exceeds 0 degrees Celsius or 5 degrees Celsius, and making estimations accordingly.For example, for a certain grass species, based on the accumulated temperature requirements for the growth of the target candidate grass species, it is estimated that the greening period of the target candidate grass species in the experimental area will be in late April, the rapid growth period will be from May to July, and the maturity period will be in late August.

[0056] After collecting identification information for various target candidate grass species, the identification information is digitized to form a salt-tolerant forage germplasm resource database. This digitization is achieved through the creation of structured electronic spreadsheets. These spreadsheets are created in a database management system software, with each row representing an independent record of a target candidate grass species and each column representing a specific field in the identification information. The fields strictly correspond to the categories of collected identification information, including species name, Latin name, description of place of origin, lower limit of place of origin altitude, upper limit of place of origin altitude, salt tolerance level, cold tolerance level, estimated greening period, estimated start of rapid growth period, estimated end of rapid growth period, and estimated maturity period. The lower and upper limit altitude fields are stored in meters, while phenological fields are stored in date format. The identification information for each target candidate grass species, collected and organized in the previous stage, is then entered into the corresponding rows and columns of the electronic spreadsheet. For example, the Latin name of the target candidate grass species, *Elymus nutans*, is entered in the Latin name field; the high salt tolerance rating is entered in the salt tolerance rating field; and the estimated greening date is entered in the estimated greening date field. After entering all the target candidate grass species records, the electronic spreadsheet is saved. The resulting electronic file containing complete structured data constitutes the salt-tolerant forage germplasm resource database. The salt-tolerant forage germplasm resource database exists as an independent database file, which can be read and accessed in subsequent steps to retrieve the target candidate grass species identification information.

[0057] S2. Select saline-alkali grassland within the target area as the experimental area, and arrange experimental plots containing multiple target candidate grass species within the experimental area according to the germplasm resource database. The specific implementation is as follows:

[0058] Within the target area, saline-alkali grasslands with varying degrees of salinization were selected as experimental sites. In practice, field surveys and historical data investigations were conducted within the target area to identify multiple potential sites with differing soil salinization levels. Soil salinization was assessed by collecting surface soil samples from potential sites and performing laboratory chemical analysis. The analytical indicator was total soil salinity, expressed as a percentage by mass. Based on the analysis results, the salinization levels of potential sites were classified into three grades: mild, moderate, and severe. The thresholds for these grades were based on the common critical values ​​of soil salinity's effect on plant growth: total soil salinity below 0.3% was classified as mild salinization, between 0.3% and 0.6% as moderate salinization, and above 0.6% as severe salinization. At least one representative site was selected from each salinization level to form the final experimental area. The coordinates of the boundary inflection points of the selected test area were determined and recorded using a Global Navigation Satellite System (GNSS) receiver. The coordinate system adopted was the WGS84 coordinate system, and the recording format was latitude and longitude. The total area of ​​the test area was determined based on the actual usable grassland area and the number of target candidate grass species; for example, the total area of ​​the test area was controlled within the range of 5 to 20 hectares.

[0059] Based on the identification information of various target candidate grass species in the salt-tolerant forage germplasm resource database, independent experimental plots were divided for each target candidate grass species within the experimental area, and isolation zones were set up between experimental plots of different target candidate grass species. Specifically, the identification information of all target candidate grass species, especially the species name and salt tolerance level, was retrieved from the established salt-tolerant forage germplasm resource database. According to the salt tolerance level of the target candidate grass species, they were allocated to plots within the experimental area with matching salinity levels. For example, target candidate grass species with high salt tolerance levels were preferentially placed in moderately or severely salinized plots, while those with low salt tolerance levels were placed in slightly salinized plots. Using the field trial design principles of randomized block design or completely randomized design, the specific location of the experimental plot for each target candidate grass species was planned within the determined plots. Each test plot is planned to be a regular rectangle or square, with a consistent area ranging from 50 to 200 square meters (e.g., 100 square meters per plot). Physical buffer zones are established between test plots of different target candidate grass species, with a width between 1 and 2 meters (e.g., 1.5 meters). The buffer zones are left in their natural state or planted with other plants significantly different from the test grass species. Using measuring tools such as tape measures and surveying poles, combined with GPS receivers, the four boundaries of each test plot are marked on-site according to the design drawings, and fixed boundary stakes are driven in as markers.

[0060] The spatial location information of each experimental plot and the corresponding identification information of the target candidate grass species were recorded to form spatial distribution data of the experimental plots. The spatial location information was recorded by measuring the geographic coordinates of the four corner points of each experimental plot. Using a real-time dynamic differential global satellite navigation system (WGS) measuring device, the geodetic coordinates of each corner point of each experimental plot were measured and recorded with centimeter-level accuracy. The coordinate system was consistent with the coordinate system used for recording the boundary of the experimental area, i.e., the WGS84 coordinate system. The measured corner coordinates of each experimental plot were then correlated with the identification information of the target candidate grass species sown or transplanted within that plot. This correlation was performed in spreadsheet software or geographic information system software, creating a new structured data table containing fields for experimental plot number, target candidate grass species name, salinity level of the experimental plot, and corner coordinates. The corner coordinates field recorded the latitude and longitude coordinate sequence of the four corner points of the experimental plot. Each planned experimental plot's number, the name of the target candidate grass species corresponding to the experimental plot, the salinity level of the plot, and the measured coordinates of the four corner points of the experimental plot are entered into the corresponding row of this structured data table. For example, experimental plot A01, with the target candidate grass species being *Leymus chinensis*, is located in a moderately saline-alkali area. The coordinates of the four corner points are the latitude and longitude of point 1, point 2, point 3, and point 4, respectively. This information is entered into different fields of the same row. After completing the information entry and association for all experimental plots, this structured data table is saved. The resulting electronic file containing the spatial location and attribute association information of the experimental plots constitutes the spatial distribution data of the experimental plots. This spatial distribution data file is stored and used in subsequent steps to guide the determination of the remote sensing image acquisition range and assist in locating specific experimental plots from the images.

[0061] S3. Obtain multispectral remote sensing images of the experimental area during at least one critical growth period. The specific implementation is as follows:

[0062] Based on the spatial distribution data of the experimental plots, the acquisition range of multispectral remote sensing images covering all experimental plots is determined. Specifically, the existing spatial distribution data file of the experimental plots is read, containing the corner coordinates of each experimental plot. The latitude and longitude coordinates of the four corner points of each experimental plot in the spatial distribution data file are analyzed to find the minimum, maximum, minimum, and maximum longitude values. Using these four extreme values ​​as boundaries, an initial rectangular range is defined in the electronic map software or geographic information system software. This initial range extends from the minimum longitude to the maximum longitude in the east-west direction and from the minimum latitude to the maximum latitude in the north-south direction. To ensure that the subsequently processed imagery completely covers all experimental plots, the range is extended outwards along the four boundaries of the initial range. The extension distance is determined based on the spatial resolution of the planned remote sensing platform. The extension distance is set as a multiple of the ground dimension corresponding to the platform's spatial resolution. For example, if a UAV remote sensing platform with a spatial resolution of 0.05 meters is planned, the extension distance is set to 400 times the spatial resolution, or 20 meters; if a satellite remote sensing platform with a spatial resolution of 10 meters is planned, the extension distance is set to 30 times the spatial resolution, or 300 meters. Calculations are performed by subtracting the extension distance from the minimum longitude of the initial range to obtain the minimum longitude of the acquisition range, adding the extension distance to the maximum longitude of the initial range to obtain the maximum longitude of the acquisition range, subtracting the extension distance from the minimum latitude of the initial range to obtain the minimum latitude of the acquisition range, and adding the extension distance to the maximum latitude of the initial range to obtain the maximum latitude of the acquisition range. Finally, the rectangular geographical area defined by these four boundary values—minimum longitude, maximum longitude, minimum latitude, and maximum latitude of the acquisition range—is determined as the multispectral remote sensing image acquisition range. The coordinates of the multispectral remote sensing image acquisition area are recorded in latitude and longitude format.

[0063] During the critical growth periods of the target candidate grass species, remote sensing images of the experimental area are acquired using a remote sensing platform to obtain raw multispectral remote sensing images with geographic coordinate information. The critical growth periods include the greening-up stage, the rapid growth stage, and the maturity stage. These time points are derived from the key phenological period information of the target candidate grass species stored in the salt-tolerant forage germplasm resource database. In practice, image acquisition is scheduled during at least one of the three stages: greening-up, rapid growth, and maturity. Based on the climatic conditions of the experimental area, clear, cloudless or minimally clouded conditions with stable sunlight and low wind speeds are selected for image acquisition. The remote sensing platform used can be an unmanned aerial vehicle (UAV) remote sensing platform or a satellite remote sensing platform. When using an UAV remote sensing platform, the boundary coordinates of the multispectral remote sensing image acquisition range are input into the UAV's flight control software to plan an automatic flight path. The flight path planning parameters include flight altitude, forward overlap rate, and lateral overlap rate. The flight altitude is set according to the required ground spatial resolution; the flight altitude is equal to the focal length multiplied by the pixel size and then divided by the ground resolution. For example, to obtain an image with a ground resolution of 0.05 meters, the flight altitude is set to 100 meters. Both forward and lateral overlap rates are set between 70% and 80%, for example, a forward overlap rate of 75% and a lateral overlap rate of 75%. The UAV carries a multispectral camera, which automatically acquires images during flight according to a preset flight path and exposure interval. Each acquired image includes the geographic coordinates of the center point and attitude angle information recorded by the UAV's global navigation satellite system. These images are stitched together to form raw multispectral remote sensing image data covering the acquisition range. When using a satellite remote sensing platform, the coordinates of the multispectral remote sensing image acquisition range are used to search the satellite data supply catalog to order or download satellite multispectral image products covering that range, with cloud cover below, for example, 10%, and taken during the target reproductive period. The acquired satellite image products themselves contain geographic coordinate information that has undergone preliminary geometric correction; this image product serves as the raw multispectral remote sensing image. Regardless of the platform used, the raw multispectral remote sensing image contains reflectance or radiance information for multiple spectral bands, such as the common blue, green, red, and near-infrared bands.

[0064] Atmospheric and geometric corrections were performed on the raw multispectral remote sensing imagery to obtain a multispectral imagery accurately matching the geographical location of the experimental area. Atmospheric correction was performed on the raw multispectral remote sensing imagery to eliminate the influence of atmospheric scattering and absorption on surface reflectance information. When the raw multispectral remote sensing imagery came from a satellite platform and was radiometrically calibrated, an atmospheric correction method based on a radiative transfer model was used. This method requires the image acquisition time, the geographical location of the image center point, and standard atmospheric parameters as input. The radiative transfer model calculates the relationship between the radiance value of the top layer of the atmosphere and the surface reflectance. By iteratively solving for the surface reflectance, the difference between the top layer radiance value calculated by the model and the radiance value observed in the image is minimized, converting the pixel radiance values ​​of the image into surface reflectance values. When the raw multispectral remote sensing imagery came from an UAV platform, atmospheric correction could be achieved by simultaneously acquiring data from a standard reflectance reference board. On the day of the UAV flight, a standard reflectance reference board with known reflectance was placed in the experimental area, and images of the standard reflectance reference board were taken before and after the UAV acquired images of the experimental area. By calculating the linear relationship between the brightness value of the standard reflectance reference plate in the image and its measured reflectance value, gain and bias coefficients are established. These coefficients are then applied to the brightness values ​​of all pixels in the test area, thus converting the brightness values ​​into surface reflectance values. After atmospheric correction, the image undergoes geometric fine correction. The purpose of geometric fine correction is to eliminate geometric distortions caused by sensor attitude, terrain undulations, etc., ensuring accurate registration with the standard map. Geometric fine correction is performed using ground control points. Ground control points are markers deployed within the test area whose precise ground coordinates are known. The coordinates of the ground control points are obtained using real-time dynamic differential global satellite navigation system measurement equipment, achieving centimeter-level accuracy. On the atmospherically corrected image, pixels corresponding to the location of each ground control point are manually identified. The correspondence between the image coordinates of the ground control points and their precise geodetic coordinates is established. A polynomial transformation model is used as the mathematical model for geometric correction, such as a quadratic polynomial model. The image coordinates and geodetic coordinates of at least six ground control points are substituted into the polynomial model, and the polynomial coefficients are solved using the least squares method. After solving for the polynomial coefficients, the entire image is resampled using these coefficients, mapping each pixel to a new geodetic coordinate grid. The resampling method can be nearest neighbor, bilinear interpolation, or cubic convolution; for example, bilinear interpolation can be used. Bilinear interpolation calculates the reflectance value of the target pixel by weighting the reflectance values ​​of the four source pixels surrounding the target pixel and averaging them based on their distances in the row and column directions. After geometric correction, each pixel in the image has accurate geographic coordinates, and the surface reflectance value has been atmospherically corrected. This image is a multispectral remote sensing image that accurately matches the geographical location of the experimental area. This multispectral remote sensing image is stored for subsequent steps in the mixed pixel decomposition process.

[0065] S4. Perform hybrid pixel decomposition on the multispectral remote sensing image to obtain the abundance ratio of target candidate grass species and the corresponding endmember spectral information in each pixel; based on the endmember spectral information and the measured aboveground biomass data of the corresponding experimental plot, establish an aboveground biomass inversion model, specifically implemented as follows:

[0066] Based on the spatial distribution data of the experimental plots and the corresponding identification information of the target candidate grass species, the image regions of each target candidate grass species experimental plot are extracted as endmember references from multispectral remote sensing images that accurately match the geographical location of the experimental area. Specifically, the spatial distribution data file of the experimental plots and the acquired multispectral remote sensing images are first read. The multispectral remote sensing images are obtained after atmospheric correction and geometric fine correction. The spatial distribution data file of the experimental plots contains the corner coordinates of each experimental plot and the species name of the corresponding target candidate grass species. The corner coordinates of the experimental plots are superimposed on the multispectral remote sensing images, and the pixel range covered by each experimental plot is located on the multispectral remote sensing images through geographic coordinate matching. For each experimental plot, the spectral reflectance data of all pixels within the boundary of the experimental plot are extracted. Considering that the edges of the experimental plots may be affected by isolation zones or neighboring plots, a strategy of removing edge pixels or low-activity pixels is adopted when extracting the endmember reference spectrum. The average reflectance and standard deviation of all pixels in the near-infrared band within the experimental plot are calculated. A screening threshold is set to distinguish between high-reflectivity and low-reflectivity pixels. The specific value of the screening threshold is obtained by adding N times the standard deviation of reflectivity to the average reflectivity value, where N is between 0.5 and 2. For example, when N equals 1, the screening threshold equals the average reflectivity value in the near-infrared band plus one standard deviation of reflectivity. Only pixels with near-infrared reflectivity values ​​greater than the screening threshold are retained. The average spectral reflectivity vector of all pixels retained after threshold screening in each test plot is used as the initial endmember spectral information for the target candidate grass species corresponding to that test plot. The average spectral reflectivity vector is obtained by calculating the arithmetic mean of the reflectivity values ​​for each spectral band. Simultaneously, the number of pixels used to calculate the average spectrum in each test plot is recorded. The initial endmember spectral information of each extracted target candidate grass species is associated and stored with its species name.

[0067] A linear mixed pixel decomposition method was used to decompose a multispectral remote sensing image that accurately matched the geographical location of the experimental area, obtaining the abundance ratio of each target candidate grass species and its purified endmember spectral information in each pixel. The linear mixed pixel decomposition method employed a constrained least squares decomposition method. The basic assumption of the constrained least squares decomposition method is that the spectral reflectance of each pixel is the result of a linear mixture of the spectral reflectance of various land cover types contained within it according to their area ratios. In this scenario, land cover types included multiple target candidate grass species extracted from the experimental plots, as well as background land cover such as bare soil. First, an endmember spectral matrix was constructed. Each column of the endmember spectral matrix represents the spectral reflectance vector of a land cover, including the initial endmember spectral information of multiple target candidate grass species extracted from each experimental plot, and the endmember spectral information of bare soil extracted from non-vegetated areas in the image. The bare soil endmember spectral information was obtained by selecting a clean, unvegetated bare soil area on the multispectral remote sensing image and calculating the average spectral reflectance vector of all pixels within that area. For each pixel, its observed spectral reflectance vector is a known quantity in a system of linear equations, approximated by multiplying the endmember spectral matrix by an abundance coefficient vector. Each element in the abundance coefficient vector represents the area proportion of the corresponding land cover within that pixel, i.e., the abundance proportion. The constrained least squares decomposition method requires satisfying two constraints: the first is a non-negativity constraint, requiring the abundance proportion of each land cover to be greater than or equal to zero; the second is a sum-of-abundance constraint, requiring the sum of the abundance proportions of all land covers within a pixel to be equal to one. The solution process involves finding a set of abundance proportion values ​​for each pixel in the image such that the sum of the squares of the differences between the spectral reflectance vector calculated from these abundance proportion values ​​and the endmember spectral matrix and the actual observed spectral reflectance vector of the pixel is minimized, while simultaneously satisfying the two constraints. The sum of the squares of the differences is obtained by calculating the square of the Euclidean distance. This problem is a quadratic programming problem with equality and inequality constraints, which can be solved using iterative algorithms such as the interior-point method. The interior-point method incorporates inequality constraints into the objective function by introducing a barrier function, and uses Newton's method to iteratively solve a series of optimization problems, ultimately approximating the optimal solution that satisfies the constraints. After solving for all pixels, two results are obtained. The first result is the abundance ratio of each target candidate grass species and bare soil in each pixel. The second result is the purified endmember spectral information, which is more consistent with the overall statistical characteristics of the image, obtained by optimizing and adjusting the initial endmember spectral information based on the solved global abundance ratio. The optimization and adjustment process can be achieved by solving a least-squares problem with the endmember spectral matrix as the variable. This problem aims to minimize the overall difference between the observed spectra of all pixels and the spectra calculated based on the current abundance and endmembers.

[0068] Aboveground biomass data were collected in experimental plots corresponding to each target candidate grass species. The collection time for aboveground biomass data was synchronized with or as close as possible to the critical growth period for acquiring multispectral remote sensing images, with an interval of no more than three days. Sampling points were set up in each experimental plot for each target candidate grass species according to standard field sampling methods. The sampling points were set up using a diagonal method or a five-point method; for example, in each 100-square-meter experimental plot, five 1-square-meter quadrats were intersected along two diagonals. In each 1-square-meter quadrat, the aboveground plant stems and leaves were cut at ground level using scissors or mowing equipment. The freshly cut plant samples were placed in sampling bags and immediately labeled with the experimental plot number, target candidate grass species name, sampling date, and quadrat number. The collected fresh plant samples were brought back to the laboratory, and clearly visible non-target weeds and litter were removed. Non-target weeds refer to other plant species whose morphological characteristics are significantly different from those of the target candidate grass species. The purified fresh samples of the target candidate grass species were weighed on an electronic balance, and the fresh weight was recorded in grams. To obtain the dry matter weight, the weighed fresh samples were placed in an oven and dried at 65 degrees Celsius for 48 hours, or until the sample weight differed from the previous weighing by less than 0.5%. The dried samples were then removed, cooled to room temperature in a desiccator, and weighed again using an electronic balance. The dry weight was recorded in grams. The dry weights of all quadrats within the same experimental plot were summed to obtain the total dry weight of that plot. The total dry weight was divided by the total area of ​​all quadrats to obtain the measured aboveground biomass per unit area of ​​that experimental plot, expressed in grams per square meter. The name of the target candidate grass species and its measured aboveground biomass per unit area were recorded for each experimental plot.

[0069] Based on the purified endmember spectral information of each target candidate grass species and the corresponding measured aboveground biomass data collected in the experimental plots, an aboveground biomass retrieval model for the target candidate grass species was established using regression analysis. Specifically, for each target candidate grass species, its purified endmember spectral information was used as the independent variable, and the measured aboveground biomass per unit area of ​​its corresponding experimental plot was used as the dependent variable. The endmember spectral information is a vector containing reflectance values ​​across multiple bands. First, one or more spectral indices need to be constructed from it as feature variables for the regression analysis. Commonly used spectral indices include the Normalized Difference Vegetation Index (NDV). The NDC is calculated by subtracting the red band reflectance from the near-infrared band reflectance, and then dividing by the sum of the near-infrared band reflectance and the red band reflectance. The calculated spectral indices and the measured aboveground biomass values ​​were combined to form a set of sample data. A linear regression model was used for fitting. The form of the linear regression model is that aboveground biomass equals the intercept term plus the slope term multiplied by the spectral indices. The least squares method was used to estimate the parameters of the intercept and slope terms of the linear regression model. The goal of the least squares method is to minimize the sum of squared residuals between the predicted and measured aboveground biomass values ​​for all sample points. The residual is the difference between the predicted and measured values. Optimal estimates of the intercept and slope terms can be obtained by solving the normal equations, which are obtained by taking the partial derivatives of the sum of squared residuals with respect to the intercept and slope terms respectively and setting them equal to zero. To evaluate the model's reliability, the sample data can be randomly divided into two parts, for example, 70% for training the model (parameter estimation) and 30% for validating the model's accuracy. The division ratio is based on the common range of training and test sets used in machine learning. Validation is achieved by calculating the coefficient of determination (COD) or root mean square error (RMSE) between the model's predicted and measured values ​​on the validation dataset to quantify the model's accuracy. The COD is the ratio of the regression sum of squares to the total sum of squares. Finally, the validated linear regression equations for each target candidate grass species, i.e., the specific values ​​of its intercept and slope terms, are saved as the aboveground biomass inversion model for that target candidate grass species. The aboveground biomass inversion model establishes a mathematical relationship for calculating the aboveground biomass per unit area from the purified end-member spectral information of this grass species.

[0070] S5. Combining the abundance ratio with the aboveground biomass inversion model, the spatial distribution information of aboveground biomass in the experimental area was obtained, and the target candidate grass species were adaptively screened based on the spatial distribution information of aboveground biomass. The specific implementation is as follows:

[0071] Based on the abundance proportion of each target candidate grass species in each pixel and its corresponding aboveground biomass inversion model, the aboveground biomass contribution value of each target candidate grass species in each pixel is calculated. Specifically, the result data obtained from the mixed pixel decomposition step is first acquired, which includes the abundance proportion of each target candidate grass species in each pixel. Simultaneously, the aboveground biomass inversion model established for each target candidate grass species is acquired. The aboveground biomass inversion model is a mathematical relationship, such as a linear regression equation, whose input is the purified endmember spectral information of the target candidate grass species, and whose output is the predicted aboveground biomass per unit area of ​​the target candidate grass species, expressed in grams per square meter. To calculate the aboveground biomass contribution value of a certain target candidate grass species in a pixel, the following operations are required: First, the endmember spectral reflectance value of the target candidate grass species is extracted from the purified endmember spectral information corresponding to that pixel. The extracted endmember spectral reflectance values ​​are substituted into the aboveground biomass inversion model for the target candidate grass species to calculate the predicted aboveground biomass per unit area represented by the target candidate grass species under the given pixel conditions. Next, this predicted aboveground biomass per unit area needs to be converted into the actual biomass contribution of the grass species within the pixel. The conversion process needs to consider the actual area of ​​the pixel and the proportion of the grass species within the pixel. Spatial resolution parameters of multispectral remote sensing images accurately matching the geographical location of the experimental area are obtained. The spatial resolution parameter represents the actual ground size corresponding to one pixel; for example, a spatial resolution of 0.05 meters means that one pixel represents a ground area of ​​0.05 meters multiplied by 0.05 meters. The ground area of ​​one pixel is calculated; the ground area is equal to the square of the spatial resolution, for example, 0.05 meters multiplied by 0.05 meters equals 0.0025 square meters. The aboveground biomass contribution of the target candidate grass species within the pixel is equal to the predicted aboveground biomass per unit area of ​​the target candidate grass species multiplied by the ground area of ​​the pixel, and then multiplied by the abundance proportion of the target candidate grass species in the pixel. Abundance ratio is a dimensionless value between 0 and 1. The above multiplication operation yields an absolute value in grams. This calculation process is repeated for each pixel in the image and for each target candidate grass species contained within it, thereby obtaining the aboveground biomass contribution data for each target candidate grass species within each pixel.

[0072] The aboveground biomass contribution values ​​of all target candidate grass species within each pixel are summed to generate aboveground biomass spatial distribution information covering the entire experimental area. Specifically, for each pixel in the image, the aboveground biomass contribution values ​​of all target candidate grass species calculated within it are summed. The mathematical operation of the summation operation is to add the aboveground biomass contribution value of the first target candidate grass species in the pixel to the aboveground biomass contribution value of the second target candidate grass species, and so on, until the aboveground biomass contribution value of the last target candidate grass species is added. If the pixel also contains the abundance proportion of other land cover such as bare soil, but bare soil does not produce aboveground biomass, its contribution value is considered zero. The summation result is a single numerical value representing the total aboveground biomass produced by all vegetation within the pixel, in grams. This summation operation is performed on each pixel in the image to obtain a two-dimensional data matrix of the same size as the original image, where the value at each position in the data matrix corresponds to the total aboveground biomass of that pixel. This two-dimensional data matrix, combined with the geographic coordinates of each pixel, constitutes the spatial distribution information of aboveground biomass covering the entire experimental area. This spatial distribution information of aboveground biomass can be visualized as a thematic map. The information is stored in raster data format, with each raster cell value representing the weight of biomass in grams.

[0073] Based on the spatial distribution information of aboveground biomass, the average aboveground biomass of each target candidate grass species within the corresponding experimental plot is extracted and statistically analyzed. In practice, the spatial distribution data file of the experimental plots is first read, defining the spatial boundaries of each experimental plot and the corresponding target candidate grass species name. Simultaneously, the generated raster data of aboveground biomass spatial distribution information covering the entire experimental area is loaded. Using the spatial analysis function of the geographic information system software, regional statistical operations are performed. For each experimental plot, its polygonal boundary is used as the analysis area, and the aboveground biomass values ​​of all pixels falling within this analysis area are statistically calculated. First, all pixels whose center points fall within the polygonal boundary of the experimental plot are extracted. Then, the arithmetic mean of the aboveground biomass values ​​of these pixels is calculated. The method for calculating the arithmetic mean is to add the aboveground biomass values ​​of all extracted pixels and then divide by the number of extracted pixels. This arithmetic mean represents the average aboveground biomass per unit pixel area within the experimental plot, and its unit is grams per pixel. To convert it to a more general aboveground biomass per unit area, this average needs to be divided by the ground area of ​​a single pixel. For example, if the average value is A grams per pixel and the pixel area is S square meters, then the average aboveground biomass per unit area is equal to A divided by S, in grams per square meter. For each target candidate grass species, there may be multiple replicated experimental plots within the test area. In this case, it is necessary to calculate the average aboveground biomass per unit area of ​​all replicated experimental plots for that grass species, as the representative average aboveground biomass of that target candidate grass species under the test conditions. The calculated representative average aboveground biomass for each target candidate grass species is then associated with its species name and recorded.

[0074] The adaptability of multiple target candidate grass species is evaluated and ranked based on their average aboveground biomass. This ranking also incorporates their salt tolerance and cold hardiness grades for a comprehensive evaluation. In practice, a preliminary ranking is first performed based on average aboveground biomass. The calculated average aboveground biomass values ​​of all target candidate grass species are sorted in descending order to generate a preliminary ranking list based on biomass yield. Then, salt tolerance and cold hardiness grades are introduced for a comprehensive evaluation. Salt tolerance and cold hardiness grade information for each target candidate grass species is retrieved from a salt-tolerant forage germplasm resource database. These qualitative grades need to be quantified into numerical scores for calculation. A quantification rule is established, for example, high grade is quantified as 3 points, medium grade as 2 points, and low grade as 1 point. For each target candidate grass species, its salt tolerance and cold hardiness grade scores are obtained. Finally, the average aboveground biomass, salt tolerance grade score, and cold hardiness grade score are combined into a total score. Since the importance of each indicator may differ, a weight needs to be assigned to each indicator. The weight setting depends on the specific needs of the screening target. A feasible weight allocation scheme is as follows: average aboveground biomass weight 0.5, salt tolerance grade weight 0.3, cold resistance grade weight 0.2, and the sum of all weights is 1. Before comprehensive calculation, the average aboveground biomass value needs to be normalized to eliminate the influence of dimensions. The normalization method uses minimum-maximum normalization, subtracting the minimum biomass among all grass species from the biomass of each grass species, and then dividing by the difference between the maximum and minimum biomass among all grass species, so that the result maps to the interval of 0 to 1. For the grade scores that have already been quantified to 1 to 3, normalization is also required, subtracting 1 and then dividing by 2, so that it also maps to the interval of 0 to 1. A comprehensive score is calculated for each candidate grass species. The comprehensive score equals the normalized average aboveground biomass multiplied by its weight, plus the normalized salt and alkali tolerance score multiplied by its weight, plus the normalized cold hardiness score multiplied by its weight. Finally, all candidate grass species are ranked in descending order based on their comprehensive scores, generating a final comprehensive adaptability ranking list. The ranking evaluation results are output in tabular form, with columns including the candidate grass species name, average aboveground biomass, salt and alkali tolerance level, cold hardiness level, comprehensive score, and final ranking.

[0075] Example 2: Figure 2 A schematic diagram of the multispectral remote sensing-based grass species adaptability screening system for saline-alkali grassland is provided. The system includes the following modules:

[0076] The germplasm bank construction module is used to establish a germplasm resource database of salt-tolerant forage grasses in the target area. The germplasm resource database contains identification information of various target candidate grass species.

[0077] The experimental layout module is used to select saline-alkali grassland as the experimental area within the target area, and to arrange experimental plots containing multiple target candidate grass species within the experimental area based on the germplasm resource database.

[0078] The image acquisition module is used to acquire multispectral remote sensing images of the experimental area during at least one critical growth period.

[0079] The model building module is used to perform mixed pixel decomposition on multispectral remote sensing images to obtain the abundance ratio of target candidate grass species and the corresponding endmember spectral information in each pixel; based on the endmember spectral information and the measured aboveground biomass data of the corresponding experimental plots, an aboveground biomass inversion model is established.

[0080] The screening decision module is used to combine the abundance ratio and the aboveground biomass inversion model to obtain the spatial distribution information of aboveground biomass in the experimental area, and to conduct adaptive screening of target candidate grass species based on the spatial distribution information of aboveground biomass.

[0081] All calculations involved in the embodiments are dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0082] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0083] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0084] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0085] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0086] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0087] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for screening grass species adaptability in saline-alkali grasslands based on multispectral remote sensing, characterized in that, Includes the following steps: S1. Establish a germplasm resource database of salt-tolerant forage grasses in the target area. The germplasm resource database contains identification information of various target candidate grass species. S2. Select saline-alkali grassland as the test area within the target area, and arrange test plots containing multiple target candidate grass species within the test area according to the germplasm resource database. S3. Acquire multispectral remote sensing images of the experimental area during at least one critical growth period; S4. Perform hybrid pixel decomposition on the multispectral remote sensing image to obtain the abundance ratio of target candidate grass species and the corresponding endmember spectral information in each pixel. Based on endmember spectral information and measured aboveground biomass data from corresponding experimental plots, an aboveground biomass retrieval model was established, including: Based on the spatial distribution data of the experimental plots and the corresponding identification information of the target candidate grass species, the image regions of each target candidate grass species experimental plot are extracted from the corrected multispectral remote sensing images as end-member references. The linear mixed pixel decomposition method was used to decompose the corrected multispectral remote sensing image and solve for the abundance ratio of each target candidate grass species in each pixel and its purified endmember spectral information. Aboveground biomass data were collected in experimental plots corresponding to each target candidate grass species. Based on the endmember spectral information of each target candidate grass species and the aboveground biomass measured data collected in the corresponding experimental plots, an aboveground biomass inversion model for the target candidate grass species was established by regression analysis. S5. The spatial distribution information of aboveground biomass in the experimental area was obtained by combining the abundance ratio with the aboveground biomass inversion model. Based on this spatial distribution information, adaptive screening of target candidate grass species was conducted, including: Based on the abundance ratio of each target candidate grass species in each cell and its corresponding aboveground biomass inversion model, the aboveground biomass contribution value of each target candidate grass species in each cell is calculated. The aboveground biomass contribution values ​​of all target candidate grass species within each pixel are summed to generate aboveground biomass spatial distribution information covering the entire experimental area. Based on the spatial distribution information of aboveground biomass, the average aboveground biomass of each target candidate grass species in the corresponding experimental plot was extracted and statistically analyzed. The adaptability of multiple target candidate grass species was ranked and evaluated based on the average aboveground biomass of each target candidate grass species. The ranking evaluation also incorporates a comprehensive assessment of the salt and alkali tolerance levels and cold tolerance levels of the target candidate grass species.

2. The method for screening grass species adaptability in saline-alkali grassland based on multispectral remote sensing according to claim 1, characterized in that, S1 includes: Collect identification information for various target candidate grass species, including species name, place of origin, salt and alkali tolerance level, cold tolerance level, and key phenological period information; The identification information of various target candidate grass species collected will be digitally stored to form a germplasm resource database of salt-tolerant forage grasses.

3. The method for screening grass species adaptability in saline-alkali grassland based on multispectral remote sensing according to claim 1, characterized in that, S2 include: Within the target area, saline-alkali grasslands with different degrees of salinization were selected as experimental areas; Based on the identification information of various target candidate grass species in the salt-tolerant forage germplasm resource database, independent test plots were divided for each target candidate grass species within the test area, and isolation zones were set up between test plots of different target candidate grass species. Record the spatial location information of each experimental plot and the identification information of the corresponding target candidate grass species to form spatial distribution data of the experimental plots.

4. The method for screening grass species adaptability in saline-alkali grassland based on multispectral remote sensing according to claim 1, characterized in that, S3 includes: Based on the spatial distribution data of the experimental plots, the acquisition range of multispectral remote sensing images covering all experimental plots was determined; During the critical growth period of the target candidate grass species, remote sensing images of the experimental area were acquired through a remote sensing platform to obtain raw multispectral remote sensing images with geographic coordinate information. Atmospheric correction and geometric fine correction were performed on the original multispectral remote sensing images to obtain multispectral remote sensing images that accurately match the geographical location of the experimental area.

5. The method for screening grass species adaptability in saline-alkali grassland based on multispectral remote sensing according to claim 4, characterized in that, The critical reproductive periods include the regrowth period, the rapid growth period, and the maturity period.

6. The method for screening grass species adaptability in saline-alkali grassland based on multispectral remote sensing according to claim 1, characterized in that, Linear mixed pixel decomposition methods include constrained least squares decomposition.

7. A multispectral remote sensing-based system for screening grass species adaptability in saline-alkali grasslands, used to implement the multispectral remote sensing-based method for screening grass species adaptability in saline-alkali grasslands as described in any one of claims 1-6, characterized in that, Includes the following modules: The germplasm bank construction module is used to establish a germplasm resource database of salt-tolerant forage grasses in the target area. The germplasm resource database contains identification information of various target candidate grass species. The experimental layout module is used to select saline-alkali grassland as the experimental area within the target area, and to arrange experimental plots containing multiple target candidate grass species within the experimental area based on the germplasm resource database. The image acquisition module is used to acquire multispectral remote sensing images of the experimental area during at least one critical growth period. The model building module is used to perform mixed pixel decomposition on multispectral remote sensing images to obtain the abundance ratio of target candidate grass species and the corresponding endmember spectral information in each pixel; based on the endmember spectral information and the measured aboveground biomass data of the corresponding experimental plots, an aboveground biomass inversion model is established. The screening decision module is used to combine the abundance ratio and the aboveground biomass inversion model to obtain the spatial distribution information of aboveground biomass in the experimental area, and to conduct adaptive screening of target candidate grass species based on the spatial distribution information of aboveground biomass.

Images