A method for identifying artificial grassland

Abstract

The application discloses a method for identifying artificial grassland. The method comprises the following steps: A, extracting a temperature suitable ploughing area; B, extracting a slope suitable ploughing area; C, performing intersection calculation on the temperature suitable ploughing area and the slope suitable ploughing area, and taking the intersection as a final artificial grassland suitable ploughing area; and D, introducing an artificial grassland identification index AGDI in the final artificial grassland suitable ploughing area, the index representing a ratio of a normalized vegetation index in a growth peak period of the artificial grassland to a normalized vegetation index in a seeding and seedling period; when the ratio meets a threshold requirement, it is considered that the artificial grassland is a potential artificial grassland ploughing area; and further, when the normalized vegetation index in the growth peak period also meets a certain threshold requirement, it is considered that the artificial grassland is an actual artificial grassland ploughing area. The artificial grassland identification method can obtain the space-time distribution of the artificial grassland with high precision and high efficiency through a small amount of data input.

Description

TECHNICAL FIELD

[0001] The present application relates to the cross field of remote sensing and ecology, and particularly relates to an artificial grassland identification method. BACKGROUND

[0002] Through the planting of artificial grassland (AG), a total of 5,592 km 2 of black beach has been restored and managed. However, artificial grassland will degenerate again after 3-5 years of planting, and needs to be planted repeatedly. Fertilization and plowing during planting will significantly increase the total nitrogen concentration in the water body, and may further affect the water quality downstream of these rivers. Therefore, it is of great significance to determine the spatiotemporal distribution of artificial grassland for evaluating the long-term repair effect of artificial grassland on black beach and its negative impact on water quality.

[0003] Remote sensing technology is the most effective method for obtaining the spatiotemporal distribution of artificial grassland. Many methods have combined remote sensing data and machine learning to identify the spatial distribution of grassland. However, these methods only stop at identifying natural grassland rather than artificial grassland. So far, the related research on remote sensing monitoring of land cover types in the Qinghai-Tibet Plateau has not involved the identification of artificial grassland. SUMMARY

[0004] The purpose of the present application is to provide an artificial grassland identification method, especially a method for identifying artificial grassland in plateau areas. The method comprises the following steps:

[0005] Step A, extraction of temperature suitable for cultivation area: according to multi-year ground air temperature observation data, the spatially continuous average air temperature is obtained by interpolation, and based on the average value, the area suitable for artificial grassland growth in terms of temperature is selected as the temperature suitable for cultivation area;

[0006] Step B, extraction of slope suitable for cultivation area: using an object-oriented segmentation method to segment the ground into multi-scale objects, and combining with a digital elevation model, the objects that meet the planting requirements of artificial grassland in terms of slope are selected as the slope suitable for cultivation area;

[0007] Step C, intersection calculation of temperature suitable for cultivation area and slope suitable for cultivation area, and taking their intersection as the final artificial grassland suitable for cultivation area;

[0008] Step D, in the final artificial grassland suitable for cultivation area, an artificial grassland identification index AGDI is introduced, which represents the ratio of the normalized vegetation index in the growth peak period to the normalized vegetation index in the seeding and germination period. When the ratio meets the threshold requirement, it is considered as a potential artificial grassland cultivation area. Further, when the normalized vegetation index in the growth peak period also meets a certain threshold requirement, it is considered as an actual artificial grassland cultivation area.

[0009] Preferably, in step A, the ground air temperature observation data comes from the National Meteorological Science Data Center, the interpolation method is the local thin plate spline interpolation method, and the temperature suitable for cultivation refers to the area with a multi-year average temperature greater than 0℃.

[0010] Preferably, in step B, the object-oriented segmentation method is a multi-scale segmentation algorithm. A multi-scale object refers to a land feature patch composed of multiple pixels. The digital elevation model is slope data calculated by SRTM elevation. A slope-suitable arable area refers to a number of objects with a slope of less than 6° and accounting for more than 60%.

[0011] Preferably, the method includes step E: based on the extraction of artificial grassland, further identifying annual and perennial artificial grasslands according to the time trajectory method.

[0012] Preferably, in step D,

[0013] ① Calculate the Normalized Difference Vegetation Index (NDVI) for each remote sensing image;

[0014] ② Based on the data quality control layer, remove snow, clouds, cloud shadows, etc. from the NDVI image;

[0015] ③ The Normalized Difference Vegetation Index (NDVI) at the sowing and emergence periods was obtained using the maximum value synthesis method. se Normalized Difference Vegetation Index (NDVI) during peak growth period gp The normalized differential vegetation index (NDVI) during the yellowing and withering period was obtained using the average value method. wilt ;

[0016] ④ Affected by cloudy weather, NDVI se With NDVI gp There will be some missing values, which means that the time since NDVI was used is far away. se Recent valid value NDVI se_c Supplement NDVI se Missing values, time elapsed since NDVI gp Recent valid value NDVI gp_c Supplement NDVI gp Missing values;

[0017] ⑤ Calculate the Artificial Grassland Identification Index (AGDI), AGDI = NDVI gp / NDVI se ;

[0018] ⑥ Potential artificial grassland cultivation areas are obtained by the threshold method. The calculation method is as follows: if AGDI>T, the land cover is judged to be a potential artificial grassland cultivation area, where T is the threshold for judging whether the land cover is a potential artificial grassland.

[0019] ⑦ Use NDVI gpThe actual artificial grassland cultivated area is screened out, and the calculation method is: if NDVI gp F, further determining that the land cover is artificial grassland, wherein F is a threshold value for further determining whether the land cover is actual artificial grassland.

[0020] Preferably, in step D, further comprising: using 3*3 window mode filtering to eliminate the influence of noise in the data processing process.

[0021] Preferably, step E further comprises:

[0022] ①General year judgment: According to the change trajectory of a certain pixel in the time series, whether the artificial grassland is annual or perennial is distinguished, if a certain pixel is continuously identified as artificial grassland, the artificial grassland on the pixel is considered to be annual artificial grassland, otherwise it is perennial artificial grassland;

[0023] ②Special year judgment: If a pixel is identified as artificial grassland in the last year of the time series, threshold method is used to distinguish annual and perennial artificial grassland according to the NDVI of the yellow and dry period of perennial artificial grassland; the calculation method is as follows: if NDVI wilt F, determining that the land cover is annual artificial grassland, wherein F is a threshold value for further determining whether the land cover is annual artificial grassland.

[0024] The artificial grassland identification method of the embodiment of the application does not need training data such as field samples and agricultural meteorological data, and does not need to establish the NDVI phenology profile of the entire vegetation growth cycle. Therefore, it can be used for automatic identification of artificial grassland in places such as the Qinghai-Tibet Plateau. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 The preprocessing of remote sensing data, temperature data and terrain data is shown. Wherein, NDVI is the normalized vegetation index; NDVI se and NDVI gp are the normalized vegetation indexes of the artificial grassland seeding and emergence period (SE) and growth peak period (GP), respectively; NDVI wilt is the normalized vegetation index of the yellow and dry period of artificial grassland; SWIR1 and SWIR2 are short-wave infrared data; SAVG is the texture of NDVI se obtained by calculating the gray level co-occurrence matrix (GLCM); NDVI se_c is the NDVI value closest to NDVI se in time, which is used to supplement the missing of NDVI se ; NDVI gp_c is the NDVI closest to NDVI gp in time, which is used to supplement the missing of NDVI gp .

[0026] Figure 2 Illustration of artificial grassland suitable for cultivation area extraction. Among them, NDVI se is the normalized difference vegetation index of sowing and emergence period (SE) artificial grassland; SWIR1 and SWIR2 are short-wave infrared data; SAVG is the texture of NDVI se calculated by the gray level co-occurrence matrix.

[0027] Figure 3 Illustration of artificial grassland identification. Among them, AG is artificial grassland, and OLU is other land use type. AGDI is artificial grassland identification index.

[0028] Figure 4 Illustration of determination of annual and perennial artificial grassland. Among them, AG is artificial grassland; NDVI wilt is the NDVI of perennial artificial grassland in the yellow withered period.

[0029] Figure 5 Illustration of the planting area of annual artificial grassland, perennial artificial grassland and their total in Maqin County from 2017 to 2021.

[0030] Figure 6 Illustration of the spatial location of artificial grassland planting area in Maqin County from 2017 to 2021, mainly distributed in the southeast, northwest and southwest of Maqin County.

[0031] Figure 7 Illustration of the spatial distribution of artificial grassland planting area in the southeast of Maqin County from 2017 to 2021.

[0032] Figure 8 Illustration of the spatial distribution of artificial grassland planting area in the northwest of Maqin County from 2017 to 2021.

[0033] Figure 9 Illustration of the spatial distribution of artificial grassland planting area in the southwest of Maqin County from 2017 to 2021. DETAILED DESCRIPTION

[0034] Considering that artificial grassland and cultivated land have some similar management measures, such as sowing, harvesting (only for annual artificial grassland) and fertilization, etc. Therefore, farmland remote sensing identification method can be used for the identification of artificial grassland. Farmland identification method can be divided into: (1) classification method; (2) information extraction method. Classification method uses various classification algorithms supported by ground training samples or prior knowledge to identify the spatial distribution of farmland, and information extraction method identifies the spatial distribution of certain specific farmland according to spectral characteristics or time spectral curve.

[0035] Classification methods can be further divided into two categories: unsupervised classification and supervised classification. Unsupervised classification automatically constructs classification labels without training samples and determines whether data can be labeled as the same class based on their similarity. This method does not require ground truth data and prior information, and the cost is relatively low, but it is vulnerable to outliers, high dimensions, and noise. The accuracy of this classification method is usually lower than that of supervised classification. Supervised classification requires manual definition of specific classification labels and establishes a corresponding mathematical model between remote sensing spectral features and user-defined classes, and then performs feature classification based on the mathematical model. Common methods include maximum likelihood, support vector machines, and random forests. These methods are theoretically feasible, but require a large number of training samples and are difficult to repeat over time and space.

[0036] Information extraction methods can be divided into time trajectory methods and specific phenology methods. Time trajectory methods are based on the fact that each crop type has special phenological characteristics and spectral behavior, and the characteristics of the same crop are usually more similar than those of other crops. Therefore, the time spectrum profile of remote sensing images can be compared with standard remote sensing images of different feature types to determine the feature type. Main methods include parallelepiped method, dynamic time warping method, and harmonic analysis, etc. However, the time resolution of open source and free data sets such as Landsat and Sentinel is limited, so it is easy to miss the key node information of the vegetation growth cycle and cannot be effectively compared with the standard spectral curve. Therefore, it is difficult to use time trajectory methods to identify artificial grasslands.

[0037] Specific phenology methods extract crops through specific features at certain specific phenological stages during crop development. Specific features can be physiology-based or knowledge-based. Although these methods cannot completely eliminate ground observation data and require more knowledge than classification methods, they are more efficient without the need for a large number of training samples. In addition, since this method only requires the most accurate phenological data, the time resolution requirement of remote sensing data is lower than that of time trajectory methods. Therefore, this method is more suitable for the Qinghai-Tibet Plateau, where the natural environment is harsh, field sampling is difficult, and remote sensing data is severely affected by clouds.

[0038] The artificial grassland recognition method according to the embodiment of the present application is, for example, an artificial grassland recognition method based on vegetation phenology knowledge. The artificial grassland recognition method uses multi-year ground air temperature observation data, interpolates to obtain a spatially continuous air temperature average value, and screens a temperature suitable cultivation area based on the average value. An object-oriented method is used to segment the ground into multi-scale objects, and the average slope of each object is calculated in combination with a digital elevation model, and the object whose slope satisfies artificial grassland cultivation is taken as a slope suitable cultivation area. The intersection of the temperature suitable cultivation area and the slope suitable cultivation area is taken as the final artificial grassland suitable cultivation area. Within the scope of the suitable cultivation area, the ratio of the normalized difference vegetation index (NDVI) of the growth peak period, the sowing and emergence period of the artificial grassland is calculated, and when the ratio meets certain threshold requirements, it is considered that the potential cultivation area of the artificial grassland is the actual cultivation area of the artificial grassland. On this basis, the method further identifies the annual and perennial artificial grasslands according to the time trajectory method. The method of the embodiment of the present application can obtain the spatio-temporal distribution of the artificial grassland with high precision and high efficiency through very little data input.

[0039] As shown in Figure 3 The artificial grassland recognition method according to the embodiment of the present application identifies or extracts the artificial grassland by introducing an artificial grassland recognition index AGDI (artificial grassland detection index). The index is determined by the ratio of the normalized difference vegetation index (NDVI) of the growth peak period (GP) of the artificial grassland and the normalized difference vegetation index (NDVI) of the sowing and emergence period (SE). The artificial grassland recognition method of the embodiment of the present application does not need training data such as field samples and agricultural meteorological data, and does not need to establish the NDVI phenological profile of the entire vegetation growth period. Therefore, it can be used to automatically identify the artificial grassland in places such as the Qinghai-Tibet Plateau.

[0040] According to one embodiment of the present application, the annual average temperature of the artificial grassland planting area should be higher than 0℃, the slope should be lower than 6°, and the remaining areas are considered to be unsuitable for cultivating artificial grassland. In order to avoid the uncertainty brought by these non-cultivation areas, the area suitable for artificial grassland cultivation is first extracted.

[0041] Specifically, the artificial grassland recognition method according to the embodiment of the present application includes the following four steps:

[0042] Step 1, extract the temperature suitable cultivation area. Specifically, it includes the following two small steps:

[0043] (1.1) Obtain ground air temperature observation data through National Centers for Environmental Prediction (NCEP) or other ways; use ANUSPLIN toolbox (http: / / fennerschool.anu.edu.au) to interpolate the multi-year monthly temperature data using local thin-plate splines interpolation method; and calculate the average value of the interpolated grid data to obtain a multi-year average temperature data. One scene refers to a grid image covering the ground area.

[0044] (1.2) According to the local standard of Qinghai Province (DB / T60-2006) or other set standards or rules, the average temperature of artificial grassland planting area should be greater than 0℃ or meet other temperature requirements. Therefore, first of all, exclude the areas where the annual average temperature does not meet the growth requirements of artificial grassland. Advantageously, considering the possible errors and uncertainties in temperature data interpolation, therefore, take the annual average temperature-1℃ as the temperature threshold, and retain the areas with temperature greater than-1℃, so that the artificial grassland planting area with suitable temperature will not be underestimated.

[0045] Step 2, extract the slope suitable for planting area. Specifically, it includes the following three sub-steps:

[0046] (2.1) ① Download the digital elevation model (DEM) with a spatial resolution of 30 meters from the space shuttle radar topography mission (SRTM) database or other sources, and calculate the slope from the DEM data; ② Based on the multi-band image composed of NDVI of SE period (NDVI se ), sum average (SAVG) obtained from gray level co-occurrence matrix, shortwave infrared reflectance 1 (SWIR1) and shortwave infrared reflectance 2 (SWIR2), use multi-scale segmentation algorithm to divide the entire area into multiple ground objects.

[0047] (2.2) According to the corresponding standards or specifications to determine the slope, for example, according to the local standard of Qinghai Province (DB / T60-2006), artificial grassland planting should be in relatively flat area, and the slope should be less than 6°. Therefore, exclude the areas where the slope does not meet the planting requirements. Considering that the DEM data has errors, the elevation change of flat area is relatively slow, which can easily lead to overestimation of the slope value, and further lead to underestimation of the slope suitable for planting artificial grassland. Therefore, the object-oriented method in (2.1) is used to improve the screening of slope suitable for planting artificial grassland. As long as more than 60% of the area within the multi-scale segmentation object has a slope less than 6°, it is considered that the object is suitable for planting artificial grassland, and these objects are extracted as the slope suitable for planting area.

[0048] (2.3) Intersect the temperature suitability area with the slope suitability area, and take their intersection as the final artificial grassland suitability area.

[0049] Step 3, Artificial grassland identification in the final suitability area. The land use types in the Qinghai-Tibet Plateau mainly include grassland, forest land and other vegetation land, and water body, unused land (glacier, bare land, etc.), city, industry and mining, airport and other non-vegetation land. Therefore, the process of identifying artificial grassland mainly includes two steps: ① distinguish artificial grassland from natural grassland, forest land and other vegetation land; ② distinguish artificial grassland from city, industry and mining, airport and other non-vegetation land. Considering that artificial grassland has similar management measures as cultivated land, artificial grassland can be identified according to the growth characteristics of artificial grassland at different stages under management measures.

[0050] Artificial grassland has low NDVI in SE period when it is just planted or just turning green, and has high NDVI in GP period when it is growing vigorously. Natural grassland, forest land and other vegetation land have high NDVI in SE period when they have turned green, and have very high NDVI in GP period when they are growing vigorously. Water body, city, industry and mining, airport and other non-vegetation land have very low NDVI in SE period when there is no vegetation growth, and have low NDVI in GP period when there is little vegetation growth. Therefore, according to the NDVI ratio of the two periods, artificial grassland can be distinguished from natural grassland, forest land and other vegetation land, and according to the NDVI value in GP period, artificial grassland can be distinguished from water body, unused land and construction land. Therefore, artificial grassland extraction in the suitability area includes the following seven steps:

[0051] (3.1) Calculate the NDVI of each period. NDVI comes from Landsat-8 Operational Land Imager (OLI) and Sentinel-2A. First, calculate the normalized vegetation index NDVI of each remote sensing image using the following equation: NDVI = (NIR-Red) / (NIR+Red). Where NIR and Red are the reflectance of near-infrared and red bands, respectively.

[0052] (3.2) Quality control of NDVI image. Refer to Figure 1 , remove bad observations on NDVI image according to data quality control layer. Here, bad observation refers to the pixel that cannot identify the ground object type due to snow, cloud, cloud shadow blocking. For Landsat-8 OLI, NDVI pixels covered by snow, cloud and cloud with confidence higher than 67% are considered as bad observations. For Sentinel-2A, pixels covered by opaque cloud or cirrus cloud are considered as bad observations, and in addition, according to the cloud probability layer of Sentinel-2A, NDVI pixels with cloud confidence higher than 60% are also considered as bad observations. Image, for example, comes from Sentinel-2 satellite system.

[0053] (3.3) Calculate the seeding and emergence period (NDVI) of artificial grassland se ), peak growth period (NDVI) gp ) and the Yellow and Dry Season (NDVI) wilt NDVI in these three periods. Among them, NDVI se and NDVI gp It is obtained by using the Maximum Value Composite (MVC) method to synthesize the NDVI of all quality-controlled products within the corresponding period. wilt The NDVI values ​​were obtained by averaging all quality-controlled values ​​during the dry season. All data were collected from the Google Earth Engine (GEE) platform and generated using the platform's JavaScript language.

[0054] (3.4) Missing value completion. Due to the limited temporal resolution of remote sensing data and severe cloudy weather conditions, even NDVI... se and NDVI gp The images were synthesized from the maximum values ​​of Landsat-8OLI and Sentinel-2A images within the corresponding phenological period, and they still contain some missing values. Therefore, further processing of these missing values ​​is necessary. This method uses temporal distance NDVI... se Recent valid value NDVI se_c (closed NDVI se Supplement NDVI se For missing values, use distance NDVI gp Recent valid value NDVI gp_c (closed NDVI gp Supplement NDVI gp Missing values.

[0055] (3.5) Differentiate between artificial grasslands and natural grasslands, woodlands, and other vegetated land using the AGDI method. The calculation method is: AGDI = NDVI gp / NDVI se Among them, NDVI gp NDVI during the peak growth period of artificial grassland se This refers to the NDVI (NDensivity and Viability Index) during the sowing and emergence stages of artificial grassland. Generally, the NDVI of artificial grassland is lower than that of natural grassland during the sowing and emergence stages, but similar to that of natural grassland during the peak growth period. Therefore, the AGDI (Augmented Grammar Index) of artificial grassland is significantly greater than that of natural grassland. The two can be distinguished using a threshold method, calculated as: AGDI > T. T is the threshold for determining whether land cover is artificial grassland. Areas meeting the threshold condition are considered potential artificial grassland planting areas.

[0056] (3.6) Distinguishing artificial grasslands from non-vegetated land such as urban areas, industrial and mining areas, and airports based on threshold methods. NDVI of some water bodies, unused land, and construction land. se It will be very low. At this point, even NDVI gp A low NDVI can also lead to a high AGDI value. In this case, according to the threshold method, these areas are easily misclassified as artificial grassland, resulting in an overestimation of artificial grassland planting areas. Therefore, it is necessary to implement further restrictions on the identification of artificial grassland. Generally, artificial grassland has a high NDVI. gp These land use types have lower NDVI. gp Therefore, NDVI can be used. gp Further adjustments are made to remove the overestimation of artificial grassland caused by these land use types. The calculation formula is: NDVI gp >F. Where F is the threshold for further determining whether land cover is artificial grassland. Potential artificial grassland planting areas that meet the threshold condition are considered actual artificial grassland planting areas.

[0057] (3.7) After the spatial distribution of artificial grassland is determined, the mode filter of 3×3 window is used to eliminate the influence of noise in the remote sensing data processing.

[0058] 4. Distinguish between annual and perennial artificial grasslands. Annual artificial grasslands are mainly used for silage production, while perennial artificial grasslands are mainly used for ecological restoration. Their management practices and ecological functions differ, therefore further subdivision is necessary. This mainly involves two steps:

[0059] (4.1) Determining the typical year: Due to infrastructure requirements such as irrigation, annual artificial grasslands are generally replanted the following year, while perennial artificial grasslands, even if they need replanting due to degradation, are replanted several years later. Therefore, annual or perennial artificial grasslands can be distinguished based on the changes in land cover trajectory over time at the pixel scale. Based on the artificial grassland identification results, if a pixel is continuously identified as artificial grassland over time, the artificial grassland in that pixel is considered an annual artificial grassland; otherwise, it is considered a perennial artificial grassland.

[0060] (4.2) Special year judgment: If a pixel is identified as artificial grassland in the last year of the time series, it cannot be determined whether it is annual or perennial by the time series trajectory method. Considering that perennial artificial grassland is mainly planted with grasses, similar to natural grassland, it has withered in autumn. Annual artificial grassland varieties are mainly oats, which are mainly used for silage feed and will not wither when perennial grassland withers. At this time, the NDVI of annual artificial grassland is higher than that of perennial artificial grassland. Therefore, here we use the threshold method to distinguish annual and perennial artificial grassland according to the NDVI of the yellowing and withering period of perennial artificial grassland. The calculation method is as follows: NDVI wilt F. Wherein F is the threshold value for judging annual and perennial artificial grassland. Artificial grassland that meets the threshold condition is considered annual artificial grassland.

[0061] Accuracy assessment. In one embodiment, the following eight indicators are used to verify the effectiveness of the method, including overall accuracy (OA), Cohen’s Kappa (CK), F-score (FS), precision (PC) and recall (RC). OA refers to the proportion of all correct identifications from all sample points. CK represents the balance between missed and wrong scores between artificial grassland and other land use (OLU). PC represents the missed score, and the higher the PC, the fewer the missed scores. RC indicates the wrong score, and the higher the RC, the fewer the wrong scores. FS takes into account the overall accuracy of PC and RC. PC, RC, FS, OA, CK are calculated as follows:

[0062]

[0063]

[0064]

[0065]

[0066]

[0067]

[0068]

[0069]

[0070]

[0071] Wherein AG and OLU represent artificial grassland and other land use; TP, FP, FN and TN are true, false positive, false negative and true negative, respectively.

[0072] Based on the artificial grassland identification method in this invention, the following example is given: Taking Maqin County, Guoluo Tibetan Autonomous Prefecture, Qinghai Province, China as an example, using remote sensing data, temperature data and topographic data, the spatial distribution of artificial grassland planting areas from 2017 to 2021 is extracted.

[0073] Maqin County lies between 33°43'N and 35°16'N in latitude and between 98°48'E and 100°56'E in longitude. Its area is approximately 1.35 × 10⁻⁶. 4 The county covers an area of ​​square kilometers, with an average altitude of over 4300 meters. More than 87.8% of the land is alpine grassland, with an average annual temperature of -3.8℃ to 3.5℃. The average annual precipitation is 423 to 565 mm. Maqin County is located in the middle reaches of the Yellow River source area, where alpine grassland degradation is severe. Extensive efforts have been made over the decades to restore black soil areas and plant artificial grasslands. Therefore, this invention uses this county as an example to explore a method for identifying the spatial distribution of artificial grasslands on the Qinghai-Tibet Plateau. The method includes the following steps:

[0074] 1. Extraction of temperature-suitable cultivation zones. See also... Figure 2 The process of establishing a suitable temperature range for cultivation involves the following two steps:

[0075] (1.1) Using ground observation data from 55 meteorological stations in Qinghai Province and surrounding provinces, spatial interpolation was performed using the local thin-plate spline interpolation method with the ANUSPLIN toolbox (http: / / fennerschool.anu.edu.au) to obtain monthly temperature distribution data from 2010 to 2020. The average temperature for the multi-year period from 2011 to 2020 was obtained by mosaicking the data using the "Mosaic to new Raster" toolbox in ArcGIS. Based on the vector boundary of Maqin County, the mosaicking results were clipped using the "Extract by mask" toolbox in ArcGIS to obtain the multi-year spatial distribution data of temperature in Maqin County.

[0076] (1.2) In order to avoid the underestimation of the evaluation results due to the error of temperature interpolation data, the “Raster calculator” toolbox in ArcGIS was used to extract the areas with temperatures greater than -1℃ in the temperature spatial distribution data of Maqin County.

[0077] 2. Identify suitable arable land areas based on slope. This includes the following three steps:

[0078] (2.1) Download the Digital elevation model (DEM) data with spatial resolution of 30 meters from the Shuttle Radar Topography Mission (SRTM) database (http: / / srtm.csi.cgiar.org / ). In ArcGIS, calculate the slope according to the DEM data using the "Slope" tool; according to the vector boundary of Maqin County and the slope data, use the "Extract by mask" tool to clip the slope of Maqin County;

[0079] (2.2) In Google Earth Engine, collect the artificial grassland SE period covering Maqin County Landsat-8 OLI and Sentinel-2A and their corresponding quality control data. Considering the actual geographical location and climate environment of Maqin County, the SE period is set as June of each year. From 2017 to 2022 June, a total of 33 Landsat-8 OLI images and 277 Sentinel-2A images were collected. First, calculate the normalized vegetation index NDVI = (NIR-Red) / (NIR+Red) of each image. Where NIR and Red are the reflectivity of near-infrared and red bands, respectively. They correspond to band 4 and band 3 of Landsat-8 OLI, and band 8 and band 4 of Sentinel-2A. Then, define the quality control function to remove the bad observations of NDVI. And use the "max" function to synthesize the maximum value of these NDVI data to obtain the NDVI of artificial grassland seeding and emergence period (NDVI se ). In GEE, use the "glcmTexture" function to calculate the texture of NDVI se according to the gray level co-occurrence matrix method, use the "select" function to select the texture data SAVG. And through the high-quality synthesis function "qualityMosaic" to calculate the corresponding short-wave infrared 1 (SWIR1) and short-wave infrared 2 (SWIR2) of NDVI se . In GEE, combine the above four bands into an ImageCollection, and synthesize a multi-band data through the "toBands" function.

[0080] (2.3) After downloading the multi-band data, use the object-oriented multi-resolution segmentation algorithm in Trimble eCognition Developer to perform multi-scale segmentation on the multi-band image composed of the above four bands.

[0081] (2.4) Input the multi-scale segmentation results into ArcGIS, use the "Zonal Statistics" toolbox to extract the proportion of slope greater than 6° within each multi-scale segmentation object; use the "Polygon to Raster" toolbox to convert the extracted proportion data from vector format to raster format. And further use the "Raster calculator" toolbox, keep the objects with slope proportion greater than 60% as the slope arable land.

[0082] (2.5) In ArcGIS, use the "Extract By Mask" toolbox to intersect the temperature arable land and the slope arable land, and take their intersection as the final artificial grassland arable land.

[0083] 3. Use artificial grassland identification method AGDI and threshold method to extract artificial grassland in arable land.

[0084] (3.1) Collect images and calculate NDVI. Including Landsat-8 OLI and Sentinel-2A and their corresponding quality control data. From June to September 2017-2022, a total of 156 Landsat-8 OLI images and 1120 Sentinel-2A images were collected in Maqin County. First, calculate the normalized vegetation index NDVI of each image using the following equation: NDVI = (NIR-Red) / (NIR+Red). Where NIR and Red are the reflectance of near-infrared and red bands, respectively. They correspond to band 4 and band 3 of Landsat-8 OLI, and band 8 and band 4 of Sentinel-2A.

[0085] (3.2) According to the data quality control layer, define a quality control function to remove bad observations on the NDVI image. For Landsat-8 OLI, NDVI pixels covered by snow, cloud and cloud with confidence higher than 67% are considered as bad observations. For Sentinel-2A, pixels covered by opaque cloud or cirrus cloud are considered as bad observations, and according to the cloud probability layer of Sentinel-2A, NDVI pixels with confidence higher than 60% of cloud are also considered as bad observations.

[0086] (3.3) Calculate the NDVI of sowing and emergence stage (NDVI se ), growth peak stage (NDVI gp ) and yellowing stage (NDVI wilt ) based on the above NDVI data. Wherein, NDVI se and NDVI gp are obtained by using the "qualityMosaic" function in GEE from the cloud-free NDVI, and NDVI wiltIt is obtained in GEE using the "mean" function from the cloud-removed NDVI. Based on the actual geographical environment, the sowing and emergence period in Maqin County is June each year. The peak growth period is August each year, and the yellowing and withering period is September each year.

[0087] (3.4) In GEE, filter out the time closest to NDVI. se and NDVI gp NDVI image se_c and NDVI gp_c Combine them into a new imageCollection and use the "mosaic" function to supplement NDVI. se and NDVI gp Missing values ​​in the image.

[0088] (3.5) Calculate the Artificial Grassland Identification Index (AGDI). In GEE, the "divide" function is used to calculate the AGDI. The actual artificial grassland is then extracted using the "updateMask" and "where" functions. The NDVI of the artificial grassland in Maqin County is considered. gp The NDVI of artificial grassland is approximately 0.7–0.8. se The value is approximately 0.3 to 0.4; therefore, the threshold for AGDI is set to 2.0, and the NDVI... gp The threshold is set to 0.5.

[0089] (3.6) Download the data from the GEE platform to the local machine, and use the “Majority Filter” toolbox in ArcGIS to perform mode filtering in a 3×3 window.

[0090] 4. Distinguish between annual and perennial artificial grasslands.

[0091] From 2017 to 2021, the changes in artificial grasslands in Maqin County can be roughly divided into three categories, as shown in Table 1. The type of artificial grassland can be determined using the "RasterCalculator" toolbox in ArcGIS.

[0092] For artificial grasslands classified as "undetermined", NDVI is used. wilt The threshold condition is used for judgment. Artificial grasslands that meet the threshold adjustment criteria are considered annual artificial grasslands; otherwise, they are considered perennial artificial grasslands.

[0093] Table 1. Time Series Variation Trajectory of Artificial Grassland

[0094]

[0095] 5. Collect sample points to verify the accuracy of artificial grassland identification results.

[0096] In ArcGIS, the "Create random point" toolbox was used to randomly create 50 sample points of artificial grassland and 100 sample points of other land use types for each year from 2017 to 2021. Then, through visual interpretation of remote sensing images and field investigation, the method of

[0001] was used to determine whether these sample points were classified correctly.

[0097] 6、Results showed that from 2017 to 2021, the accuracy of artificial grassland identification in Maqin County was relatively high, and a good balance between missed classification and misclassification was maintained. The average OA over the years was 96.2%, CK was 0.91, and the FS of artificial grassland and other types were 94.0% and 97.3%, respectively. The RC of artificial grassland was 89.6%, and the PC was 99.2%; the PC of other land use was 95.1%, and the RC was 99.6%. During the study period, the changes of each accuracy index were relatively stable. The maximum and minimum values of OA were 97.3% and 93.3%, respectively, with a coefficient of variation of only 1.8%. The minimum and maximum values of CK were 0.84 and 0.94, respectively, with a coefficient of variation of 4.7%. The maximum and minimum FS of artificial grassland were 95.9% and 88.9%, respectively, and the FS of other land use was 95.2% to 98.0%, with coefficients of variation of 3.2% and 1.3%, respectively (data results as shown in Table 2). In Table 2, PC, RC, FS, OA, and CK are accuracy, recall, F-score, overall accuracy, and Cohen's Kappa, respectively.

[0098] Table 2. Accuracy assessment of artificial grassland maps from 2017 to 2021.

[0099]

[0100] 7、From 2017 to 2021, the total area of artificial grassland planted in Maqin County was 98.29 km 2 . The largest planting area was in 2019, with an area of 30.51 km 2 , and the smallest total planting area was in 2021, with an area of 5.11 km 2 . The annual planting area of annual artificial grassland was 15.43 km 2 , accounting for 15.7%; the perennial artificial grassland planting area was 82.86 km 2 , accounting for 84.3%. The planting area of perennial artificial grassland fluctuated greatly during the study period, with the largest planting area in 2019, at 25.98 km 2 , and the smallest perennial planting area in 2021, at 1.73 km 2 . The annual planting area of annual artificial grassland was relatively stable. The largest annual planting area of annual grassland was in 2019, with an area of 4.53 km 2The area planted with annual grassland was the smallest in 2017, at 2.12 km². 2 ( Figure 5 ).

[0101] 8. The planting areas of artificial grassland show obvious clustering, mainly distributed in the southeastern part of Maqin County. Figure 6 a) Northwest Figure 6 .b) and the southwest ( Figure 6 .c). Among them, the artificial grassland planting area in the southeast is the largest ( Figure 7 A total of 64.0 km 2 , accounting for 85.3%. Northwest ( Figure 8 The planting area is the smallest, totaling 3.6 km. 2 , accounting for 4.7%. Southwest ( Figure 9 Planting area 7.5km 2 , accounting for 10.0%. The planting area refers to the area that was actually cultivated between 2017 and 2021, and the same planting area can be planted multiple times.

[0102] The AGDI method of this invention overcomes the problems faced in identifying artificial grasslands in regions such as the Qinghai-Tibet Plateau. First, this method can easily identify artificial grasslands, unaffected by the spectral similarity between artificial and natural grasslands. Second, this method does not require training data such as field samples and agro-meteorological data, overcoming the difficulty of obtaining ground training samples due to the harsh natural conditions of the Qinghai-Tibet Plateau. Finally, this method only uses remote sensing data from the sowing and emergence periods and peak growth stages, without needing to establish an NDVI phenological profile of the entire vegetation growth cycle, reducing reliance on high temporal resolution optical remote sensing data, which is crucial for identifying artificial grasslands under harsh weather conditions such as the Qinghai-Tibet Plateau. Therefore, the method of this invention can obtain detailed spatiotemporal distributions of artificial grasslands with high accuracy and efficiency with minimal data input.

Claims

1. A method for identifying artificial grassland, characterized in that, The method includes the following steps: A. Extraction of temperature-suitable cultivation zones: Based on years of ground air temperature observation data, interpolation is used to obtain the spatially continuous average air temperature, and areas with temperatures suitable for artificial grassland growth are selected based on this average value as temperature-suitable cultivation zones. B. Extraction of suitable arable areas: The ground is divided into multi-scale objects using an object-oriented segmentation method. Combined with a digital elevation model, objects with slopes suitable for artificial grassland planting are selected as suitable arable areas. C. The intersection of the temperature-suitable cultivation zone and the slope-suitable cultivation zone is calculated, and their intersection is taken as the final artificial grassland suitable cultivation zone. D. Within the final suitable cultivated area of ​​artificial grassland, the Artificial Grassland Identification Index (AGDI) is introduced. This index represents the ratio of the Normalized Difference Vegetation Index (NDI) during the peak growth period of the artificial grassland to the NDI during the sowing and emergence periods. When the ratio meets a threshold requirement, it is considered a potential artificial grassland cultivated area. Furthermore, when the NDI during the peak growth period also meets a certain threshold requirement, it is considered an actual artificial grassland cultivated area. In step D, ① Calculate the Normalized Difference Vegetation Index (NDVI) for each remote sensing image; ② Based on the data quality control layer, remove snow, clouds, and cloud shadows from the NDVI image; ③ The Normalized Difference Vegetation Index (NDVI) at the sowing and emergence periods was obtained using the maximum value synthesis method. se Normalized Difference Vegetation Index (NDVI) during peak growth period gp The normalized differential vegetation index (NDVI) during the yellowing and withering period was obtained using the average value method. wilt ; ④ Affected by cloudy weather, NDVI se With NDVI gp There will be some missing values, which means that the time since NDVI was used is far away. se Recent valid value NDVI se_c Supplement NDVI se Missing values, time elapsed since NDVI gp Recent valid value NDVI gp_c Supplement NDVI gp Missing values; ⑤ Calculate the Artificial Grassland Identification Index (AGDI), AGDI = NDVI gp / NDVI se ; ⑥ Potential artificial grassland cultivation areas are obtained by the threshold method. The calculation method is as follows: if AGDI>T, the land cover is judged to be a potential artificial grassland cultivation area, where T is the threshold for judging whether the land cover is a potential artificial grassland. Using NDVI gp Further determine the cultivated area of ​​artificial grassland, calculated as follows: if NDVI gp >F, further determining whether the land cover is an actual artificial grassland cultivated area, where F is the threshold for further determining whether the land cover is artificial grassland.

2. The artificial grassland identification method as described in claim 1, characterized in that, In step A, the ground air temperature observation data comes from the National Meteorological Science Data Center, and the interpolation method is the local thin plate spline interpolation method. The temperature suitable for cultivation area refers to the area with an average annual temperature greater than 0℃.

3. The artificial grassland identification method as described in claim 1, characterized in that, In step B, the object-oriented segmentation method is a multi-scale segmentation algorithm. Multi-scale objects refer to land feature patches composed of multiple pixels. The digital elevation model is the slope calculated from the elevation. The slope suitable for cultivation refers to a number of objects with a slope of less than 6° and accounting for more than 60%.

4. The artificial grassland identification method as described in claim 1, characterized in that, The method includes step E: based on the extraction of artificial grassland, further identifying annual and perennial artificial grasslands according to the time trajectory method.

5. The artificial grassland identification method as described in claim 1, characterized in that, Step D further includes: using a 3×3 window mode filter to eliminate the influence of noise, wherein noise refers to the recognition error caused by sensor parameters, atmospheric refraction and reflection, and environmental factors during data processing.

6. The artificial grassland identification method as described in claim 1, characterized in that, Step E further includes: ① Judgment of general years: Based on the change trajectory of a certain pixel in the time series, distinguish whether it is an annual or perennial artificial grassland. If a certain pixel is continuously identified as artificial grassland, it is considered that the artificial grassland on that pixel is an annual artificial grassland; otherwise, it is a perennial artificial grassland. ② Judgment of Special Years: If a pixel is identified as artificial grassland only in the last year of the time series, then annual and perennial artificial grasslands are identified based on the NDVI of the yellowing and withering period of perennial artificial grasslands; the calculation method is as follows: if NDVI wilt F is the threshold for determining whether land cover is an annual artificial grassland.

Images