A method for determining actual irrigated area of a farmland

Abstract

The application discloses a method for determining an actual irrigation area of a farmland, comprising the following steps: extracting farmland in a farmland irrigation research area according to remote sensing images of the farmland irrigation research area, so as to obtain farmland plots in the farmland irrigation research area; determining an irrigation area in the farmland irrigation research area during a crop growth period according to meteorological data observed by a plurality of meteorological stations in the farmland irrigation research area and a neighboring range during the crop growth period; determining farmland NDVI data of the farmland irrigation research area during the crop growth period by using the farmland plots in the farmland irrigation research area and normalized difference vegetation index (NDVI) data of the farmland irrigation research area during the crop growth period; and determining an actual irrigation area of the farmland irrigation research area according to the irrigation area and the farmland NDVI data in the farmland irrigation research area during the crop growth period.

Description

Technical Field

[0001] This invention relates to the field of farmland irrigation technology, and in particular to a method for determining the actual irrigated area of ​​farmland based on remote sensing. Background Technology

[0002] In the process of calculating the effective utilization coefficient of farmland irrigation water in Chongqing, the statistics of the actual irrigated area have always been based on a bottom-up, step-by-step statistical method, and are mainly done manually, which can no longer meet the requirements for the timeliness of data in the calculation process.

[0003] Topographically, Chongqing is a typical "mountain city," dominated by hilly terrain, with arable land relatively scattered. The overall climate is subtropical monsoon, characterized by simultaneous rainfall and heat, and abundant precipitation, forming a rain-fed agricultural system. This has resulted in relatively extensive water use for farmland irrigation in Chongqing, with low water management levels and overall utilization and water-saving efficiency needing improvement. Using remote sensing technology to determine the actual irrigated area of ​​farmland in Chongqing is of value and significance for improving the efficiency of farmland irrigation water use in Chongqing.

[0004] A summary of domestic and international research on extracting actual irrigated farmland area using remote sensing reveals that it primarily relies on multi-temporal remote sensing imagery. Various vegetation indices are used to invert surface moisture and temperature, comparing changes in soil moisture and temperature before and after, and selecting certain thresholds to determine whether farmland is irrigated. In recent years, remote sensing data has been used to invert evapotranspiration to obtain changes in soil moisture content, thereby monitoring farmland irrigation. While these studies have shown some effectiveness in remote sensing extraction of local and regional farmland irrigated areas, several problems remain. First, remote sensing inversion based on soil moisture and temperature is significantly affected by precipitation. Specifically, existing research on soil moisture changes mainly focuses on farmland areas where soil moisture has changed due to artificial irrigation. However, precipitation is another factor contributing to soil moisture changes, and current methods cannot completely eliminate the influence of precipitation. During the extraction process, it cannot be definitively determined whether the soil moisture changes are caused by artificial irrigation. Existing methods and theories have shown some success in arid and semi-arid regions of northern China, but for the humid hilly and mountainous regions of southwestern China, where precipitation is abundant, remote sensing inversion based on soil moisture is more challenging. Secondly, the imagery data selected for existing inversion studies are mostly cloudless or low-cloudy multi-temporal sequences with continuous time or short intervals. Whether for water use efficiency studies or irrigation management studies, the minimum requirements for temporal and spatial resolution are ten days and moderate, respectively, to obtain the irrigated area of ​​farmland. This is feasible for obtaining continuous cloudless or low-cloudy imagery data during relatively favorable climatic conditions in arid and semi-arid regions of the north, but presents challenges for humid and semi-humid regions of the south. Furthermore, high-resolution imagery has long revisit periods and poor continuity, leading to high costs and computational burdens when applied to large-scale irrigation areas (equivalent to county-level scale), resulting in low efficiency. Finally, due to north-south regional differences, the north has a relative advantage in data acquisition and quality, and the greater contiguousness of farmland facilitates water-saving irrigation management. Farmland in the hilly and mountainous areas of southern China is relatively scattered, and irrigation water for farmland under rain-fed agriculture is more arbitrary. Under the conditions of certain difficulties in terms of imaging and methods, it is necessary to select appropriate methods to conveniently and quickly extract the actual irrigated area of ​​farmland in the mountainous areas of southern and southwestern China, especially the hilly and mountainous areas of southern China (such as Chongqing). Summary of the Invention

[0005] This invention provides a method for determining the actual irrigated area of ​​farmland, aiming to conveniently and quickly extract the actual irrigated area of ​​farmland in the mountainous areas of southern and southwestern China, especially the hilly and mountainous areas of southern China (such as Chongqing).

[0006] This invention provides a remote sensing method for the actual irrigated area of ​​farmland. The method includes: extracting cultivated land from remote sensing images of a farmland irrigation study area to obtain cultivated land plots within the farmland irrigation study area; determining the irrigated area within the farmland irrigation study area during the crop growth period based on meteorological data observed from multiple meteorological stations within and adjacent areas of the farmland irrigation study area; determining the cultivated land NDVI data within the farmland irrigation study area during the crop growth period using cultivated land plots and Normalized Difference Vegetation Index (NDVI) data of the farmland irrigation study area during the crop growth period; and determining the actual irrigated area of ​​the farmland irrigation study area based on the irrigated area and cultivated land NDVI data within the farmland irrigation study area during the crop growth period.

[0007] This invention obtains farmland NDVI data based on farmland plots and NDVI data extracted from remote sensing images, and determines the actual irrigated area based on irrigation areas and farmland NDVI data obtained from meteorological data. It can conveniently and quickly extract the actual irrigated area of ​​farmland in the southern and southwestern mountainous areas, especially the hilly and mountainous areas of the south (such as Chongqing). Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating the technical route of the remote sensing method for the actual irrigated area of ​​farmland provided by the present invention.

[0009] Figure 2a and Figure 2b These are flowcharts for the maximum likelihood classification method and the object-oriented classification method, respectively.

[0010] Figure 3a and Figure 3b These are, respectively, a sample irrigation area buffer map and a maximum likelihood classification result map;

[0011] Figure 4a and Figure 4b These are the CART classification decision tree and the classification results diagram of the object-oriented method, respectively.

[0012] Figure 5 This is a flowchart for identifying irrigation areas;

[0013] Figure 6 This is a spatial interpolation map of the average precipitation in the ten-day period of April;

[0014] Figure 7 This is a spatial interpolation plot of the average evaporation in ten days of April;

[0015] Figure 8 This is a spatial interpolation map of the average temperature in the ten-day period of April;

[0016] Figure 9 This is a chart showing the water balance in April;

[0017] Figure 10 This is a schematic diagram of the NDVI data preprocessing results in Chongqing.

[0018] Figure 11 This refers to the NDVI value of arable land in Chongqing in April.

[0019] Figure 12 This is a vectorized image of April temperatures;

[0020] Figure 13 This is a vectorized graph of the moisture surplus / deficit in April;

[0021] Figure 14 This is a map of the irrigation area in April;

[0022] Figure 15 This is a flowchart for extracting the actual irrigated area;

[0023] Figure 16 This is a farmland map of the irrigated area in April;

[0024] Figure 17 This is a map of irrigated farmland in April;

[0025] Figure 18 This is a map showing the distribution of irrigated farmland in Chongqing in 2015.

[0026] Figure 19 This is a flowchart illustrating the method for determining the actual irrigated area of ​​farmland. Detailed Implementation

[0027] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In the following description, suffixes such as "module," "part," or "unit" used to denote elements are used only for the purpose of illustrative purposes and have no inherent meaning. Therefore, "module," "part," or "unit" may be used interchangeably.

[0028] Example 1

[0029] Figure 19 This is a flowchart illustrating the method for determining the actual irrigated area of ​​farmland, such as... Figure 19 As shown, the method may include:

[0030] Step 5101: Based on the remote sensing image of the farmland irrigation study area, extract the cultivated land in the farmland irrigation study area to obtain the cultivated land plots in the farmland irrigation study area.

[0031] Specifically, the maximum likelihood method or object-oriented classification method can be used to extract arable land from the remote sensing images of the farmland irrigation study area to obtain arable land plots in the farmland irrigation study area; alternatively, the maximum likelihood method and object-oriented classification method can be used respectively to extract arable land from the remote sensing images of the farmland irrigation study area, and the optimal arable land extraction result can be selected as the arable land plots in the farmland irrigation study area.

[0032] In addition, prior to step S101, the method further includes: acquiring Landsat 8 remote sensing image data of the farmland irrigation study area, NDVI data of the farmland irrigation study area during the crop growth period, and meteorological data observed by multiple meteorological stations in and around the farmland irrigation study area during the crop growth period; performing data preprocessing on the Landsat 8 remote sensing image data of the farmland irrigation study area, including radiometric correction, geometric correction, and image fusion, to obtain remote sensing images; performing data preprocessing on the NDVI data of the farmland irrigation study area during the crop growth period, including projection transformation and vector clipping, to obtain preprocessed NDVI data; and performing data preprocessing on the meteorological data observed by multiple meteorological stations in and around the farmland irrigation study area during the crop growth period, including outlier removal, to obtain preprocessed meteorological data.

[0033] Step S102: Based on meteorological data observed at multiple meteorological stations within and adjacent to the farmland irrigation study area during the crop growth period, determine the irrigation area within the farmland irrigation study area during the crop growth period.

[0034] In this embodiment, the meteorological data observed by the multiple meteorological stations during the crop growth period are first spatially interpolated to obtain the meteorological spatial interpolation data for each ten-day period. Then, based on the meteorological spatial interpolation data for each ten-day period, the irrigation area in the farmland irrigation study area during the crop growth period is determined.

[0035] Specifically, the meteorological data from each meteorological station includes evaporation, precipitation, and temperature. Accordingly, firstly, based on the evaporation, precipitation, and temperature data for each of the multiple meteorological stations per ten-day period, the average evaporation, precipitation, and temperature for each meteorological station per ten-day period are determined. Then, spatial interpolation is performed on the average evaporation, precipitation, and temperature for each of the multiple meteorological stations per ten-day period to obtain interpolated evaporation, precipitation, and temperature data for the farmland irrigation study area and its adjacent areas per ten-day period. Finally, based on the interpolated evaporation and precipitation data for each ten-day period within the farmland irrigation study area and its adjacent areas, the interpolated water surplus / deficit data for each ten-day period within the farmland irrigation study area and its adjacent areas are determined. Finally, based on the interpolated data of water surplus / deficit and temperature for each ten-day period within the farmland irrigation study area and its adjacent areas, the irrigation areas for each ten-day period within the farmland irrigation study area are identified. Specifically, the interpolated data of water surplus / deficit and temperature for each ten-day period within the farmland irrigation study area and its adjacent areas are vectorized to obtain vector data of water surplus / deficit and temperature for each ten-day period within the farmland irrigation study area and its adjacent areas. The vector data of water surplus / deficit and temperature for each ten-day period within the farmland irrigation study area and its adjacent areas are then cropped using the boundary of the farmland irrigation study area to obtain vector data of water surplus / deficit and temperature for each ten-day period within the coverage area of ​​the farmland irrigation study area. Finally, based on the vector data of water surplus / deficit and temperature for each ten-day period within the coverage area of ​​the farmland irrigation study area, the irrigation areas for each ten-day period within the farmland irrigation study area are identified.

[0036] Step S103: Using the cultivated land plots in the farmland irrigation study area and the Normalized Difference Vegetation Index (NDVI) data of the farmland irrigation study area during the crop growth period, determine the cultivated land NDVI data of the farmland irrigation study area during the crop growth period.

[0037] In this embodiment, the cultivated land plots in the farmland irrigation study area are vectorized to obtain cultivated land vector data of the farmland irrigation study area. The cultivated land vector data of the farmland irrigation study area is used as mask data to extract the NDVI values ​​of cultivated land plots from the NDVI data of the farmland irrigation study area during the crop growth period to form cultivated land NDVI data.

[0038] Step S104: Determine the actual irrigated area of ​​the farmland irrigation study area based on the irrigated area and NDVI data of cultivated land in the farmland irrigation study area during the crop growth period.

[0039] In this embodiment, firstly, irrigated plots with NDVI values ​​are cropped from the irrigated areas within the farmland irrigation study area during the crop growth period to obtain the irrigated plots with NDVI values ​​within the irrigated area during the crop growth period. Then, based on the irrigated plots with NDVI values ​​within the irrigated area during the crop growth period, the actual irrigated area of ​​the farmland irrigation study area is determined. Specifically, the NDVI values ​​of the irrigated plots within the irrigated area are obtained before and after each ten-day period during the crop growth period to obtain the difference in NDVI values ​​before and after each ten-day period. Based on whether the difference in NDVI values ​​of the irrigated plots within the irrigated area during each ten-day period is greater than zero, the irrigated farmland data for each ten-day period during the crop growth period is determined. Based on the irrigated farmland data for each ten-day period during the crop growth period, the actual irrigated area of ​​the farmland irrigation study area is determined.

[0040] This invention can conveniently and quickly extract the actual irrigated area of ​​farmland and is applicable to irrigation areas at the provincial, municipal, and county levels in mountainous areas of southern and southwestern China, especially hilly and mountainous areas in the south, such as determining the actual irrigated area of ​​farmland in Chongqing.

[0041] Example 2

[0042] Existing research primarily relies on multi-temporal remote sensing imagery, employing surface moisture and temperature inversion methods to obtain irrigated farmland area. These studies are mainly concentrated in northern regions, particularly large-scale irrigation areas. In the rain-rich, humid, and semi-humid southern and southwestern regions, taking Chongqing as an example, acquiring multi-temporal remote sensing data is difficult, and methods based on surface moisture and temperature inversion are heavily influenced by precipitation. Irrigation activities mainly occur on farmland plots, and remote sensing extraction of farmland is based on satellite imagery. Stitching together recent Landsat 8 remote sensing images of Chongqing only yielded a barely usable image of Chongqing with low cloud cover in 2015. Therefore, this embodiment uses 2015 as a baseline, fusing images from previous years to obtain a single image of Chongqing with low cloud cover. Using the 2015 remote sensing image as the base for farmland extraction, both maximum likelihood and object-oriented classification methods are employed. The result with better farmland extraction accuracy is used as the subsequent data for extracting the actual irrigated area. The study period for irrigated area was determined by combining government documents and literature. Kriging interpolation was then performed using meteorological data for each ten-day period within the study period. The spatial difference of precipitation and the interpolation results of evaporation were then calculated to obtain the water surplus / deficit data for each ten-day period. Combined with spatial temperature interpolation data, the areas with irrigated activity for each ten-day period were identified. Simultaneously, NDVI data from the study period was used, with extracted farmland as a mask, to extract the NDVI values ​​of farmland plots. Farmland plots with NDVI values ​​were then cropped from the areas with irrigated activity to obtain the farmland plots with NDVI values ​​within the irrigated areas. Finally, based on previous research theories and the actual situation of this study, the NDVI values ​​of farmland before and after each ten-day period were calculated to obtain the irrigated farmland for each ten-day period. These were then combined into a single layer for joint analysis to obtain the irrigated farmland area of ​​Chongqing Municipality in 2015. Geometric calculations were then performed to obtain the actual irrigated area.

[0043] The following combination Figures 1 to 18 , and will be explained in detail.

[0044] I. Overview and Data of the Farmland Irrigation Study Area (hereinafter referred to as the Study Area)

[0045] 1. Overview of the study area

[0046] (1) Natural environmental conditions

[0047] Chongqing Municipality is located in the Sichuan Basin, between 105°17′ and 110°11′ east longitude and 28°10′ and 32°13′ north latitude. Its main topography belongs to the Daba Mountains and Wuling Mountains in the eastern part of the Sichuan Basin. The total area of ​​Chongqing Municipality is 82,400 km². 2The overall provincial boundary resembles an inverted, tilted "mountain" character. Chongqing has a large area of ​​sloping land, dominated by hills and mountains, representing a typical combination of landform types primarily based on mountains.

[0048] Chongqing has a subtropical monsoon climate, characterized by mild winters and hot summers throughout the year, with significant climate variations across different areas of the city. From an atmospheric circulation perspective, Chongqing is less affected by westerly circulation in winter, resulting in smaller daily temperature ranges, weaker winds, and less precipitation. In spring, the westerly circulation weakens, and temperatures rise rapidly. In summer, the influence of the subtropical high pressure over the western Pacific Ocean leads to predominantly descending air currents, often resulting in heavy rains or storms in early summer (June), while the peak summer months of July and August are dominated by sunny and hot weather, a major cause of frequent droughts. Autumn is characterized by prolonged periods of overcast and rainy weather, with temperatures dropping. Due to significant interannual and seasonal variations in meteorological elements, coupled with diverse and complex topography, the climate is highly unpredictable, with droughts occurring in all four seasons, often with inter-seasonal droughts. Summer droughts have the most widespread impact, causing severe drought conditions over extended periods and resulting in substantial crop losses.

[0049] (2) Crop planting structure

[0050] The main crops grown in Chongqing are rice, corn, and rapeseed. Although the planting area of ​​these major crops is relatively large, irrigation for dryland crops is minimal, with virtually no irrigation facilities or large-scale irrigation. Irrigation is only carried out manually in scattered locations during droughts, reflecting a rain-fed farming system. Therefore, rice is the main irrigated crop in the sample irrigation areas, primarily irrigated through flood irrigation. Only about 20% of the typical rice fields use wet irrigation.

[0051] (3) Sampling Irrigation Area Conditions

[0052] The sample irrigation areas are mainly divided into two types: pumping irrigation and gravity irrigation. The main crop irrigated by both is rice. Gravity irrigation areas primarily use pumps to draw water into corresponding irrigation canals, allowing the water to flow naturally into the farmland. Pumping stations, on the other hand, rely on manual labor to draw water. In imagery, the corresponding ditches and pumping stations are only clearly visible in very high-resolution images. Furthermore, due to tree obstruction, pumping stations are easily confused with ordinary houses in images.

[0053] 2. Data Sources and Processing

[0054] (1) Data source

[0055] ①Basic Data. See Table 1 for basic data and its sources. Some data provide resolution, for example, the spatial resolutions of Landsat 8 imagery data and Google Earth imagery data are 30m and 0.5m, respectively; the resolution of DEM data is 30m; the temporal resolution of the vegetation index product NDVI data is 16 days, and the spatial resolution is 250m; the spatial resolution of land use status data is 10m.

[0056] Table 1. Basic Data Source Information

[0057]

[0058] ② Survey Data. See Table 2. This data primarily relies on annual calculations of the effective utilization coefficient of farmland irrigation water in Chongqing, conducting on-site surveys of farmland to understand the main crop structure and irrigation conditions, and also providing a reference for selecting sample data from remote sensing imagery. Through communication with farmers, their irrigation habits were understood.

[0059] Table 2 Survey Results

[0060]

[0061] (2) Data processing

[0062] ① Basic data processing.

[0063] After downloading the 30m resolution DEM data, it was imported into ArcGIS 10.2 software for geometric correction. Then, it was cropped using the Chongqing city vector range and finally stitched together to obtain the Chongqing elevation. A total of 8 Landsat 8 imagery images were required. The images downloaded from the geospatial data cloud were L1 level products, already subjected to system radiometric and geometric corrections. ENVI 5.2 software was used for radiometric calibration and atmospheric correction. The Landsat 8 remote sensing imagery has 9 bands: bands 1-7 and 9 are multispectral bands, and band 8 is a panchromatic band. Since band 9 is mainly used for cloud detection and is not very useful in this example, it was discarded. To improve the spatial resolution of the imagery, the image fusion module in ENVI 5.2 software was used. (Sharpening) fused the 1-7 multispectral bands at 30m resolution and the 8th panchromatic band at 15m resolution into a 15m resolution remote sensing image of Chongqing. The location vector data of 79 sampling points in the irrigation area were imported into the 91 Satellite Imagery software, and training samples were selected according to the 0.5m spatial resolution Google Earth imagery found for comparison. Meteorological data were statistically analyzed for each station using Excel, and outliers were replaced with multi-year averages. The daily average meteorological data were then converted into ten-day average meteorological data by calculating the average value. The data was then imported into ArcGIS 10.2 software, and spatial interpolation was performed using the Kriging interpolation method to obtain the 2015 ten-day average spatial interpolation data for Chongqing. The NDVI data selected was MOD 13Q1 data from MODIS, which required two images to cover the entire Chongqing area. The downloaded image was in HDF format, and was first opened and projected using NEVI 5.2 software. Then, the image was cropped using the Chongqing vector boundary data, and then exported as a TIF image. The exported TIF image was then loaded into ArcGIS 10.2 software, and the image was stretched using the standard deviation. Finally, the data was exported as a GRID image to facilitate the subsequent extraction of NDVI values ​​for cultivated land and other land types from land use data.

[0064] ② Data processing. Locate and organize the necessary research data, such as information on sample irrigation areas and statistical yearbooks, and summarize them to facilitate timely access to and resolution of problems that may arise during subsequent research, as well as to facilitate data visualization.

[0065] II. Farmland Extraction

[0066] Since irrigation occurs on specific plots of arable land, it is first necessary to extract arable land in Chongqing as the basic unit of study. The remote sensing image data is primarily from 2015 and 2016. Combined with literature review and survey data, the planting status of arable land is unlikely to change significantly within a short period, and the extracted arable land area will not differ greatly from the actual area, with the error within a reasonable range. This embodiment is based on stitched remote sensing imagery of Chongqing from 2015. It primarily employs two methods—maximum likelihood estimation and object-oriented classification—to extract arable land, compares the extraction results, and selects the optimal arable land extraction result (or the arable land extraction result with the highest accuracy) as the subsequent research unit for irrigated area.

[0067] 1. Maximum likelihood extraction

[0068] (1) Introduction to basic principles

[0069] The maximum likelihood method, based on the assumption that the grayscale (DN) values ​​of remote sensing images follow a normal distribution, establishes a discrimination mechanism using statistical parameters such as the mean, variance, and covariance matrix of training samples to classify land use types in remote sensing images. Equation 1 identifies unknown land use types as belonging to the land use type with the highest probability.

[0070]

[0071] In the formula, p(w) i ) indicates w i The prior probability in the image, p(x / w) i ) indicates w i The conditional probability of a type having x is the posterior probability under the condition of x. Under the three assumptions of the maximum likelihood method, given the prior probability of the occurrence of the land cover type and the conditional probability of x, the posterior probability of the land cover type is calculated. Then, based on the minimum confusion probability and combined with the mathematical mechanism of the minimum risk algorithm, supervised classification of the image to be classified can be achieved.

[0072] The process generally involves interactive human-computer identification of ground features based on imagery and human subjective experience. Different numbers of samples are selected for different land cover types, known as training samples. These samples need to be evenly distributed across the imagery to ensure reasonable selection. Finally, the computer software's built-in algorithm calculates the spectral characteristics and other information of all sample data to establish discrimination rules or functions, yielding the classification result. Specific steps include: sample selection, training samples, classification execution, accuracy verification, and result output. See [link to detailed process] for more information. Figure 2a .

[0073] (2) Sample selection

[0074] Maximum likelihood estimation, as a supervised classification method, follows the principle of establishing a discriminant function based on the spectral characteristics of training samples, and then classifying land features. The accuracy of the final classification result depends on the quality of the training samples; therefore, when selecting training samples, characteristics such as uniformity, discriminativeness, and representativeness must be comprehensively considered. In terms of quantity, the number of sample points for each land feature class should be controlled between 10 and 100 to ensure sufficient quantity for each land feature, thereby minimizing the impact of errors. In terms of distribution, localized concentrations of sample points should be avoided; they should be evenly distributed throughout the study area to reduce the influence of environmental differences on the spectral characteristics of the same sample point. Finally, samples should be selected from large, contiguous land features, and mapped from their central locations. This ensures the accuracy of the samples in terms of spectral characteristics, maximizes the differentiation of different land features, and makes the training samples more representative.

[0075] In the 2015 study on cultivated land extraction in Chongqing, the training sample was selected centered on 79 sampling points in irrigation districts of Chongqing. Multiple buffer zones were created outwards using ArcGIS 10.2 software. Figure 3a As shown, the distances are 10km, 20km, 30km, and so on, and the results are exported as shapefiles. Large areas of various land features with significant characteristics are uniformly selected within the buffer zone to ensure the quality of the training samples.

[0076] The training samples were categorized into six land types: cultivated land, forest land, grassland, water area, construction land, and unused land. The exported buffer shapefiles were imported into the 9L satellite imagery software. Google Earth imagery of Chongqing from 2015 was selected, scaled to the highest resolution of 0.26m, and placed within the sample irrigation area buffer zone. Using interactive methods and referencing field survey data, training samples were selected for each of the six land types. At least one training sample was selected for each land type within each irrigation area buffer zone, prioritizing the selection of training samples with the highest probability of matching the corresponding land features. Finally, 641 training samples were randomly selected, and the training samples for each land type were exported as shapefiles for subsequent maximum likelihood extraction.

[0077] (3) Classification and extraction

[0078] The classification was primarily based on the maximum likelihood method using ENVI 5.2 software. The processed 2015 Chongqing imagery was opened in ENVI 5.2. First, bands 4, 3, and 2 were selected for natural true color. Later, different band combinations were used to validate different land cover types in the training samples. For example, bands 7, 6, and 4 highlighted urban and other built-up areas; bands 5, 4, and 3 highlighted vegetation information; and bands 5, 6, and 4 better distinguished between water and land. The Region of Interest (ROI) tool in ENVI 5.2's Basic Tools was used. The shapefile data (shp) of different land features drawn using 91 Satellite Imagery software was imported into ENVI 5.2. Then, by clicking the "file" button in the import training sample window, the data was converted into ROI files for each land cover type. The data was automatically imported into the ROI Tool, conforming to the ENVI training sample format, facilitating subsequent farmland extraction.

[0079] Then, the selected training sample data needs to be verified for data separation and relevance. Only those meeting the standards can proceed to the classification process. This mainly involves calculating and checking the mean, variance, and other matrices, as well as related statistical indicators, to ensure the usability of the sample data. In ENVI 5.2 software, the option to select Compute ROI Separability is used. If the separation of each landform in the classification samples is above 1.8, the training samples are considered qualified and can be used for classification extraction; otherwise, new training samples need to be selected. The calculation of the separation of each sample shows that the separation of each landform is above 1.8. After the samples pass the verification, the Maximum Likelihood classifier in ENVI 5.2 software is selected for classification to obtain the final classification results. Figure 3b As shown.

[0080] (4) Accuracy Evaluation

[0081] The results used the 2015 land use status of Chongqing (accuracy 10m) as validation data. Validation samples were randomly selected within a buffer zone centered on 79 sample irrigation areas to ensure that the validation samples did not overlap with the training samples. The main indicators for evaluating classification accuracy included overall classification accuracy, Kappa coefficient, mapping accuracy, and user accuracy. Finally, a confusion matrix was generated to verify the accuracy of the maximum likelihood classification results, as shown in Table 3. The results were exported to an Excel spreadsheet.

[0082] Table 3. Accuracy Evaluation of Maximum Likelihood Classification Results

[0083]

[0084] The table shows an overall classification accuracy of 74.91% and a Kappa coefficient of 0.64. The overall extraction results are generally average. The highest accuracy was found in construction land, cultivated land, and water areas, while the lowest accuracy was found in grassland and unused land. Since the maximum likelihood method uses the grayscale (DN) value of remote sensing images as the basis for discrimination, construction land and water areas, with their relatively high and low grayscale values ​​respectively, are easily distinguishable. However, cultivated land, forest land, and grassland show little difference during the crop planting season, leading to misclassification by the maximum likelihood method and thus unsatisfactory classification results. Unused land mainly consists of bare rock and bare soil. Bare rock has a high grayscale value in the image and is easily classified as construction land, while bare soil is easily confused with cultivated land. The table also shows that unused land is primarily classified into these two land categories.

[0085] 2. Object-oriented extraction

[0086] Object-oriented classification replaces traditional pixels with the homogeneity of objects as the basic unit of analysis, increasing the information content of remote sensing images and improving classification accuracy. As a classification method for distinguishing ground features, compared to traditional methods based on single-pixel spectral values, object-oriented classification uses remote sensing images as objects and employs classification algorithms to divide different ground features into objects of varying sizes. It leverages the different spectral information, shapes, textures, and other features that different ground objects possess, segmenting highly correlated ground objects into the same category, thereby distinguishing ground objects.

[0087] Object-oriented classification is primarily based on eCognition version 9.0 software. eCognition is a relatively advanced commercial software for object-oriented image classification. It mimics human thinking to perform intelligent analysis and information extraction, thereby transforming Earth observation remote sensing image data into spatial geographic information more accurately, efficiently, and intelligently. The main steps of object-oriented classification include image segmentation, feature selection, sample selection, classification execution, and accuracy verification. For detailed procedures, see [link to documentation]. Figure 2b .

[0088] (1) Segmentation scale

[0089] The segmentation methods used in eCognition software mainly include checkerboard segmentation, quadtree segmentation, and multi-scale segmentation, among which multi-scale segmentation is the most effective algorithm. Multi-scale segmentation starts from any pixel and proceeds step-by-step, dividing the image into objects of varying sizes at different scales. Smaller objects are then merged into larger objects, resulting in homogeneous ground feature objects. After segmentation, the segmentation ratio is directly proportional to the size of the segmented objects.

[0090] The choice of segmentation scale is a crucial foundation and prerequisite for the success of object-oriented classification, significantly impacting subsequent classification accuracy. The heterogeneity of an object's spectral color and shape determines the heterogeneity of the image, calculated according to formula (2).

[0091] f = w c ·h c +w s ·h s ....................................(Equation 2)

[0092] In the formula, w c The weighting value for spectral information; w s h represents the weight value for shape information. c h is the value of spectral heterogeneity. s Let w be the value of shape heterogeneity; where w c +w s =1.

[0093] The spectral heterogeneity index h of the object c The formula is:

[0094]

[0095] In the formula, w b Here are the weight values ​​for each band; s b is the standard deviation of each band; i is the number of layers.

[0096] The shape heterogeneity index of an object includes two indices: smoothness and compactness, as shown in Equation 4. In the formula, h... s For compactness; h c For smoothness; w s and w c These represent the weights of the two values, and w s +w c =1.

[0097] h shape =w s ·h s +w c ·h c ....................................(Equation 4)

[0098] After multiple experimental setups, the segmentation scale parameters for levels 1-4 were selected as 100, 50, 25, and 20, respectively, which can satisfy the extraction of the six types of land features in this example. Multi-scale segmentation parameters were set in the eCognition software.

[0099] The images were segmented at scales of 100, 50, 25, and 20 to obtain the extraction results of six land features at different scales. After comparing the segmentation results at different scales, at a scale of 100, large areas of vegetation and water were well distinguished. For water extraction, the central part of the water area was well identified, and water areas with different water qualities were also well distinguished. However, there were many misclassifications at the intersections of woodland and other land features, and misclassifications still existed at the edges of water areas. Roads between farmlands were not well distinguished, and green vegetation in urban construction land was sometimes misclassified as construction land. Sparse construction land in rural areas was severely missed. At a segmentation scale of 50, the distinction between woodland and grassland and the edges of construction land was improved compared to the 100-scale segmentation, but minor misclassifications still existed. Misclassifications still existed between rural farmland and construction land, while green space and water areas in urban areas were extracted well. At a segmentation scale of 25, further refined segmentation enabled the differentiation of farmland and grassland edges from woodland edges. Most sparse rural settlements were well identified, and vegetation and water features were extracted effectively. At a segmentation scale of 20, the focus was primarily on differentiating scattered settlements and the edges of unused land. Different farmland plots were distinguished, and the differentiation of river systems within fields was also good. Vegetation between urban areas was well differentiated. At this scale, the object segmentation is very fine, facilitating precise identification of different land features. Table 4 summarizes the main land features extracted at different segmentation scales.

[0100] Table 4. Segmentation scales for different land cover extractions

[0101]

[0102] (2) Feature selection

[0103] Because land cover spatial information is diverse, and each land cover has different texture and geometric features, their characteristic attributes are the main basis for image classification extraction using object-oriented classification methods. In the eCognition software, object features required for classification can be selected and defined, mainly including spectral features, geometric features, and texture features. Specific indicators are shown in Table 5.

[0104] Table 5. Selection of Object Features

[0105]

[0106] ① Spectral characteristics

[0107] Normalized Difference Vegetation Index (NDVI): This index reflects crop growth and the crop's photosynthetic effective radiation absorption capacity, and is one of the most widely used vegetation indices in crop research. It is extensively used in ground cover extraction studies, and its main calculation formula is as follows:

[0108]

[0109] In the formula, the numerator and denominator represent the difference and sum of the reflectance values ​​in the near-infrared and infrared bands of the remote sensing image, respectively, corresponding to bands 5 and 4 in the Landsat 8 remote sensing image. Different ground objects have different reflectance values ​​in the near-infrared and infrared bands. Vegetation has high reflectance in the near-infrared band and low reflectance in the infrared band, therefore its NDVI value is positive. Water has a negative NDVI value due to its high reflectance in the near-infrared band, while rocks and bare soil have the same reflectance in both the near-infrared and infrared bands, therefore their value is 0.

[0110] The improved Normalized Difference Water Index (MNDWI): This was proposed by Xu based on the Normalized Difference Water Index (NDWI), with modifications to the band combination. The main calculation formula is as follows:

[0111]

[0112] The numerator and denominator in the formula represent the reflectance values ​​of the green (GREEN) and shortwave infrared (SWIR) bands in the image, respectively. Unlike the original Normalized Difference Water Index (NDWI), the SWIR band has been replaced with the near-infrared (NIR) band. This not only facilitates the accurate extraction of water body information in urban areas but also reveals subtle characteristics of different water qualities. Landsat 8 has two corresponding SWIR bands, SWIR1 and SWIR2. Their application in eCognition software for water body extraction is largely similar; therefore, SWIR1 is selected as the default SWIR band in the formula.

[0113] Normalized Building Index (NDBI): Proposed by Zha Yong et al., this index is based on the characteristic that impermeable surfaces such as construction land have higher reflectance in the mid-infrared than in the near-infrared. It is used to extract information about bare land and construction land. NDBI values ​​range from -1 to 1, with higher values ​​indicating denser buildings. The formula is as follows:

[0114]

[0115] Among them, SWIR and NIR are the shortwave infrared band and the near-infrared band, respectively. Landsat8 still uses SWIR1 by default.

[0116] Band mean: The average value of the layer is calculated from the pixel values ​​of all pixels constituting the image. The mean value of each band from 1 to i is obtained by the following formula:

[0117]

[0118] Among them, V i This represents the cell value in the object, where n represents the number of cells.

[0119] ② Geometric features

[0120] Length / Width: Primarily used to extract linear features; the calculation formula is the ratio of the eigenvalues ​​of the covariance matrix.

[0121] Shape Index: It is four times the boundary length of an image object divided by the square root of its area.

[0122] Area: Represents the total number of pixels that make up the segmented object.

[0123] ③Texture features

[0124] The Gray Co-occurrence Matrix (GLCM) includes eight texture parameters: Mean, Variance, Entropy, Angular Second Moment, Correlation, Dissimilarity, Contrast, and Homogeneity.

[0125] (3) Establishing a classification

[0126] The classification operation primarily utilizes the CART (Classification and Regression Tree) algorithm built into the eCognition software. The CART algorithm iteratively divides the test and target variables into two subsets, ensuring each non-leaf node has two branches, forming a binary tree-like decision tree structure. Segmented land features are selected as sample data, with random selection based on the segmented objects. Land types are categorized into six types: cultivated land, forest land, grassland, water area, construction land, and unused land, forming the basis for the decision tree classification rules. Spectral, textural, and geometric features are used as the basis for classification, with different band combinations or features used for different land types.

[0127] CART decision trees fully leverage the characteristics between different land types, maximizing the advantages of the CART algorithm to construct optimal classification rules. Before building the CART decision tree, the segmentation process is completed. The software automatically calculates the spectral, shape, and texture features of the sample data, and then automatically establishes classification rules based on these feature indices. The system calculates the separability between training samples using features including Normalized Difference Vegetation Index (NDVI), Modified Normalized Difference Water Index (MNDWI), Normalized Difference Building Index (NDBI), band mean, aspect ratio (Length / Width), shape index, area, and gray-level co-occurrence matrix (GLCM). The final optimal land classification decision tree constructed by the system is shown below. Figure 4a As shown, the classification results based on object-oriented classification are obtained by constructing a decision tree. Figure 4b As shown.

[0128] (4) Accuracy Evaluation

[0129] The accuracy evaluation method is the same as the maximum likelihood method. The 2015 land use status of Chongqing (accuracy 10m) was used as validation data, with a buffer zone centered on 79 sample irrigation areas. Validation samples were randomly selected within this buffer zone. Similarly, overall classification accuracy, Kappa coefficient, mapping accuracy, and user accuracy were selected as indicators to measure the accuracy of the classification results. Finally, the accuracy evaluation results based on the confusion matrix of the facet object classification method are shown in Table 6.

[0130] Table 6. Accuracy Evaluation of Object-Oriented Classification Results

[0131]

[0132] The object-oriented classification method achieved an overall accuracy of 86.35% and a Kappa coefficient of 0.83, indicating good land use extraction accuracy. However, the extraction of unused land was less effective, with lower accuracy. Construction land, water areas, and forest land showed the best extraction results. Since object-oriented classification primarily classifies land features based on spectral, texture, and shape characteristics, it is highly distinguishable from forest land, water areas, and construction land in terms of both spectral and shape features, making these three types easier to extract with higher accuracy. Unused land, mainly consisting of bare land and impermeable materials such as stone around towns with excavated foundations awaiting construction, is easily confused with construction land. Similar to the maximum likelihood method, the object-oriented classification method also exhibits some confusion in the extraction of forest land, grassland, and cultivated land. These three types have low distinguishability in imagery, and as the study area scale increases, the distinguishability between them decreases further, leading to greater confusion. However, the object-oriented classification method uses multiple features to distinguish between these three types of land features, which can effectively extract these three types of land features and achieve satisfactory extraction results.

[0133] 3. Comparative Analysis of Farmland Extraction Results

[0134] By comparing the mapping accuracy, user accuracy, overall accuracy, and Kappa coefficient of the two classification results (see Table 7), it is easy to see that the object-oriented method has higher classification accuracy.

[0135] Table 7. Comparison of classification accuracy between maximum likelihood method and object-oriented method

[0136]

[0137] Based on the accuracy verification results of both methods, the object-oriented classification method shows a significant advantage in extraction accuracy for the other five land categories, except for construction land, where the accuracy is slightly lower than that of the maximum likelihood method. The closer the overall accuracy is to 100% and the closer the Kappa coefficient is to 1, the better the accuracy. The object-oriented classification method surpasses the maximum likelihood method in both overall accuracy and Kappa coefficient, and is closest to 100% and 1. Both methods perform best in extracting water bodies and construction land among the six land categories, but their extraction results for unused land are not very good. The object-oriented classification method outperforms the maximum likelihood method in both institutional and user accuracy for arable land extraction, yielding more accurate results.

[0138] This example compares two land use extraction methods: maximum likelihood estimation and object-oriented classification, to extract arable land parcels. Maximum likelihood estimation, a commonly used remote sensing classification method for extracting land use information, has a relatively well-established operational procedure and methodology. Object-oriented classification, on the other hand, emerged with the rise of high-resolution remote sensing imagery and has developed rapidly due to its significant extraction advantages. The results are examined for accuracy, and the more accurate results are selected for the next step of extracting irrigated arable land and its area. This comparative analysis can also provide a reference for the rapid extraction of arable land in the future. Firstly, in the extraction of various land use types, except for construction land which is slightly less accurate than maximum likelihood estimation, object-oriented classification shows a significant accuracy advantage in extracting other land use types. The accuracy comparison reveals that object-oriented classification is superior to maximum likelihood estimation in both overall accuracy and Kappa coefficient. Maximum likelihood estimation performs better for extracting construction land, arable land, and water bodies than other land use types, while both methods are less effective for extracting unused land. Object-oriented classification has significant advantages over other land types in extracting both water bodies and built-up land, with mapping and user accuracy for cultivated land extraction approaching 80%. Furthermore, existing research has demonstrated that object-oriented classification outperforms maximum likelihood estimation at the county scale, and the comparative results of this invention also show advantages at the provincial and municipal scales. Object-oriented classification is an emerging method developed from extracting ground features from high-resolution remote sensing imagery. This method is also applicable to extracting ground objects from medium-resolution remote sensing imagery, and it can achieve better extraction results than maximum likelihood estimation.

[0139] III. Extraction from Irrigation Areas

[0140] Based on the Chongqing Statistical Yearbook and previous research, the main crops and their growing seasons in Chongqing have been determined. The main crops and their growing seasons are as follows: rice (paddy) from April to August, wheat from November to May of the following year, corn from April to October, potatoes from September to December, sweet potatoes from May to November, and vegetables whose growing seasons are seasonal.

[0141] The main crops requiring irrigation in the city include rice, vegetables, oilseeds, and corn, with their growing season primarily from April to December. Unlike crops such as corn and potatoes, which are irrigated only during planting or severe drought, rice requires multiple irrigations throughout its growing season, maintaining a relatively constant water level in the fields. Furthermore, Chongqing's agriculture is primarily rain-fed, meaning irrigation practices are significantly influenced by climate conditions, posing a considerable challenge to determining the irrigated area. Therefore, using relatively objective meteorological data as the primary basis for determining the irrigated area is scientifically sound. Although the imagery data for Chongqing is mainly a composite of 2015 and 2016, varying climatic conditions at different times can influence irrigation practices. Literature review indicates that the climatic conditions in 2015 and 2016 were not significantly different, thus the corresponding impact was relatively small (as shown in Table 8). The rice growing season from April to August was selected as the irrigation study period, and meteorological data from April to August were also used for the study. Since Chongqing's summer drought period is concentrated from June to September, meteorological data from September was included in the study data as needed.

[0142] Table 8. Basic Meteorological Conditions in 2015 and 2016

[0143]

[0144] 1. Principles and Theories

[0145] The processed temperature vector data and water surplus / deficit vector data were mapped one-to-one according to month and time, then overlaid and analyzed to identify high-temperature and water-deficient areas, which were used as data for subsequent irrigation of farmland. First, the water surplus / deficit data was processed. When its value is greater than 0, indicating a positive number, it means there is abundant rainfall, a water surplus, and humid conditions; when it is less than 0, it means strong evaporation, insufficient rainfall to meet the water needs of crop growth, and a water-deficient state, indicating a dry climate. The magnitude of the water surplus / deficit reflects the degree of water surplus / deficit and the dryness / humidity of the climate. Drought is mainly caused by reduced rainfall or increased temperature leading to crop water deficit. Furthermore, Chongqing's agriculture is primarily rain-fed, and rainfall has a significant impact on irrigation. Therefore, the magnitude of the water surplus / deficit was considered a primary factor. Data for each ten-day period was filtered, and areas with values ​​less than 0 were selected as potential drought areas. Second, the temperature vector data was processed. Because temperatures vary depending on the time of year when drought occurs, with spring droughts generally having lower temperatures than summer droughts, specific considerations are needed for each period. The primary basis should be the water balance, followed by the influence of temperature. Based on field surveys of farmers' irrigation practices, and adhering to the principle of irrigating during droughts, areas prone to drought were selected as irrigation study areas. Irrigated farmland was then identified based on the NDVI values ​​of cultivated land within these areas.

[0146] The following is combined Figure 5 The flowchart shown illustrates the irrigation area identification process, detailing the process of extracting irrigation areas.

[0147] 2. Meteorological data interpolation

[0148] Spatial interpolation is a calculation method that extrapolates data from other unknown points in the same region based on data from known points. This invention employs ordinary kriging interpolation. Ordinary kriging utilizes the structural characteristics of known variation data and variograms in a known region to predict variations in an unknown region using a linear, unbiased, optimal estimation method. The calculation formula is as follows:

[0149]

[0150] In the formula: Z represents the meteorological value at the unknown point; n represents the number of interpolation stations; λ i The weighting coefficients for meteorological elements at unknown points in the interpolation. x i This refers to the site location.

[0151] All meteorological data are sourced from the China Meteorological Administration (http: / / www.cma.gov.cn / ) and the National Meteorological Center (http: / / data.cma.cn / ). Meteorological station coordinates are from the Chongqing Meteorological Bureau (http: / / cq.cma.gov.cn / ). Data corrections are based on the Chongqing Climate Impact Assessment Report (http: / / cq.weather.com.cn / ). The data time resolution is daily. Since the NDVI data for cultivated land is based on ten-day periods, daily average meteorological data were converted to ten-day average data through calculation before spatial interpolation. Data from 22 national-level meteorological stations were selected from the 699 national-level benchmark and basic stations in China, including those in Chongqing and neighboring Hubei, Hunan, Guizhou, and Sichuan provinces. The meteorological station data information is shown in Table 9.

[0152] Table 9 Meteorological Station Information

[0153]

[0154] All downloaded meteorological data is stored in a single physical file directory, with one file per month for each feature and item, along with a corresponding data description file. The data file names consist of the dataset code (SURF_CLI_CHN_MUL_DAY), feature code (XXX), item code (XXXXX), year identifier (YYYY), and month identifier (MM). Here, SURF represents surface meteorological data, CLI represents surface climate data, CHN represents China, MUL represents multi-feature data, and DAY represents daily data. The file name is “SURF_CLI_CHN_MUL_DAY-XXX-XXXXX-YYYYMM.TXT”, where “XXX” is the feature code, “XXXXX” is the item code, and YYYYMM represents the year and month of the data. The date and time system are Beijing time, with the date set at 20:00. XXX-XXXXX Explanation: PRS-10004 represents the station's air pressure, TEM-12001 represents the air temperature, RHU-13003 represents the relative humidity, PRE-13011 represents precipitation, EVP-13240 represents evaporation, WIN-11002 represents wind direction and speed, SSD-14032 represents sunshine, and GST-12030-0cm represents the ground temperature at 0cm depth.

[0155] (1) Spatial interpolation of precipitation

[0156] The precipitation data is named SURF_CLI_CHN_MUL_DAY-PRE-13011-201501.TXT. The Chinese name of the data is "China Surface Climate Data Daily Values ​​Dataset (v3.0)". SURF_CLI_CHN_MUL_DAY represents the dataset code, PRE represents the feature code, 13011 represents the project code, and 201501 represents the year and month of the data. The data format is explained below:

[0157] Table 10. Precipitation Data Format Description

[0158]

[0159] The data is in text format and includes data from 699 national-level basic meteorological stations in China. It can be directly dragged into an Excel spreadsheet for processing. Using the filtering function, data from 22 meteorological stations in Chongqing and its surrounding areas were extracted based on the station number. According to the instructions in the accompanying data file, a precipitation value of 32700 indicates "trace" precipitation, which can be replaced with 0.01 or simply 0; this article uses 0.01. The quality control code 0 indicates correct data, 1 indicates questionable data, 2 indicates incorrect data, 8 indicates missing data or no observation task, and 9 indicates no quality control was performed. After processing all the data, the raw spatial interpolation data was obtained.

[0160] Because the spatially interpolated data is in TIFF format, cropping it using the Chongqing municipal boundary can result in jagged edges or even missing sections, leaving some farmland uncovered. To avoid this, the interpolation result display range was changed to cover the entire Chongqing municipality. The daily average precipitation was arithmetically averaged to obtain the ten-day average precipitation. Each month was divided into three ten-day periods (early, middle, and late). All data from April to September was processed in an Excel spreadsheet, with location and elevation information for each station added. This data was then imported into ArcGIS 10.2 software, converting the tabular data into vector station data. The projected coordinate system was set to UTM Zone 48 North, and the geographic coordinate system to GCS WGS 1984. Finally, using the geostatistical analysis tools in the software, ordinary kriging was selected, with other settings default. The search area was set to standard tools, the variogram to semi-variogram, and the model type to stable. The classification method in the interpolation results is set to the natural discontinuity classification method by default, dividing the data into 10 categories. The color of the bands is determined based on the data type, research needs, and aesthetics of the graph. Low values ​​are selected with a reddish tint, indicating less precipitation, while high values ​​are selected with a bluish tint, indicating more precipitation. For example, in April, the final data processing results can be found in [link to data]. Figure 6 .

[0161] (2) Spatial interpolation of evaporation

[0162] The evaporation data file is named: SURF_CLI_CHN_MUL_DAY-EVP-13240-201501.TXT. The file naming convention is consistent with the previous precipitation data description. A value of 32700 indicates evaporator freezing; +1000 indicates all water injected into the evaporator has evaporated, requiring an increment of 1000 from the injected water data. The quality control code 0 indicates correct data, 1 indicates questionable data, 2 indicates incorrect data, 8 indicates missing data or no observation task, and 9 indicates data without quality control. Values ​​of 32700 were replaced with 0.01, and after checking the quality control codes, no questionable, relatively erroneous, missing observation, or quality-controlled data were found within the study period of April to August.

[0163] Table 11. Evaporation Data Format Description

[0164]

[0165] The text data was also dragged into an Excel spreadsheet for processing. Using the filtering function, the raw data for spatial interpolation of evaporation was obtained. The processing procedure was the same as for precipitation. Data for each month was divided into three periods: early, middle, and late. This data was also imported into ArcGIS 10.2 software, with the projected coordinate system being UTM Zone 48North and the geographic coordinate system being GCS WGS1984, respectively. Ordinary Kriging was used in ArcGIS 10.2 geostatistical analysis, with other settings default. The interpolation results were also classified into 10 categories using the natural discontinuity classification method by default. Since areas with high evaporation may experience water shortages, the color of the evaporation data bands was set to the opposite color to that of precipitation for research purposes. Low values ​​were selected with a bluish tint, indicating low evaporation, while high values ​​were selected with a reddish tint, using April as an example. The final data processing results are shown in [link to data]. Figure 7 .

[0166] (3) Spatial interpolation of air temperature

[0167] The temperature file is named SURF_CLI_CHN_MUL_DAY-TEM-13240-201501.TXT. The file naming convention is consistent with the previous explanation of the precipitation data. The temperature data format is shown in Table 12. There are no characteristic values ​​in the temperature data. Similarly, the quality control code 0 indicates correct data, 1 indicates questionable data, 2 indicates incorrect data, 8 indicates missing data or no observation task, and 9 indicates that the data was not quality controlled and no erroneous values ​​were found through Ecxel filtering.

[0168] Table 12. Temperature Data Format Instructions

[0169]

[0170] Meteorological text data was processed in an Excel spreadsheet. After filtering by April 19th, the raw data for spatial interpolation of temperature was obtained. The processing procedure was the same as for precipitation. Data for each month was divided into three periods: early, middle, and late. The data was imported into ArcGIS 10.2 software, with the projected coordinate system being UTM Zone 48North and the geographic coordinate system being GCS WGS1984, respectively. Ordinary Kriging was used in ArcGIS 10.2 geostatistical analysis, with other settings default. The interpolation results were also classified into 10 categories using the natural discontinuity classification method by default. Areas with higher temperatures have a higher probability of drought. For research purposes, the zone colors were set to a bluish tint for low values ​​(indicating lower average temperatures) and a reddish tint for high values ​​(indicating higher average temperatures). April was used as an example. The final data processing results are shown in [link to data]. Figure 8 .

[0171] (4) Spatial interpolation of moisture deficit

[0172] The main method involves using a raster calculator to subtract evaporation from precipitation to determine the water surplus / deficit areas in Chongqing from April to September, identifying farmland plots potentially prone to drought. The water surplus / deficit areas are determined using the SpatialAnalyst tool in ArcGIS 10.2 software. The mathematical analysis function subtracts the values ​​of the second and third input raster cells sequentially from the values ​​of the first and second input raster cells. The first input is spatial interpolated precipitation data, and the second is spatial interpolated evaporation data, corresponding to each ten-day period. The water surplus / deficit situation for each ten-day period is obtained by subtracting the values ​​of the two raster cells.

[0173] To ensure aesthetic appeal and meet research requirements, a natural discontinuity classification method was used to categorize the water deficit data (the difference between precipitation and evaporation raster data) into 10 classes. The band colors maintain the same hue as before; high values ​​are selected with a bluish tint, indicating precipitation exceeding evaporation, suggesting abundant rainfall and a low likelihood of irrigation; low values ​​are selected with a reddish tint, indicating evaporation exceeding precipitation, suggesting a higher probability of drought. (April was used as an example.) The final data processing results are shown in [link to data]. Figure 9 .

[0174] 3. NDVI data processing

[0175] Water is one of the most important factors affecting vegetation growth. The peak value of the normalized difference vegetation index (NDVI) within a region can reflect differences in vegetation growth. This index can be used as an important basis to determine whether farmland is irrigated. If the peak value of NDVI for farmland is relatively large compared to other land types within a small area, it indicates that the crops or vegetation in that area are growing well, and the farmland is more likely to be irrigated than rain-fed farmland. This embodiment uses the changes in NDVI values ​​of farmland under different climatic conditions to determine the irrigation status of crops planted on farmland. Farmland with a large NDVI peak shows better crop growth. For example, under high temperature and water shortage conditions, farmland with an increased NDVI value compared to the previous period is more likely to be irrigated. Extracting the NDVI values ​​of farmland for each period from April to September is a key step in this study.

[0176] (1) MODIS NDVI Data Analysis

[0177] NDVI data was downloaded from NASA's Distributed Active Archive (DAAC) (https: / / ladsweb.modaps.eosdis.nasa.gov / ), with a temporal resolution of 16 days and a spatial resolution of 250 meters. This embodiment uses MOD13Q1 data (MODIS / Terra Vegetation Index 16-day L3 global 250-meter grid), one of the MODIS products for the terrestrial theme, which also includes data products for the atmosphere, snow and ice, and ocean themes. Since its data time format is based on January 1st of each year as the first day, accumulating sequentially, it needs to match the irrigation period of the study. The number of days corresponding to April to August is shown in Table XX. Therefore, NDVI data with the corresponding number of days was selected for processing. The NDVI data covering the study period mainly consists of twelve periods.

[0178] Table 13. Dates Corresponding to NDVI Data

[0179]

[0180] Two NDVI images are needed to cover Chongqing, with track numbers h27v05 and h27v06. The data can be imported into ENVI 5.0 software for processing. First, projection correction is performed, with the projection coordinate system being UTM Zone 48 North and the geographic coordinate system being GCS WGS 1984. After stitching the two NDVI images, they are imported into a Chongqing municipal administrative boundary shapefile for vector cropping to obtain the Chongqing NDVI value map. The result is then exported as a TIFF file that can be opened with ArcGIS software. This completes the preprocessing of the MODIS imagery. The preprocessed result is shown below. Figure 10 As shown.

[0181] (2) NDVI extraction from cultivated land

[0182] The main steps are as follows: first, convert the NDVI data in tif format to grid format; second, extract the cultivated land vector data using the land use type results extracted by the object-oriented classification method; and third, use the cultivated land vector data as mask data to extract the cultivated land NDVI value.

[0183] First, based on the preprocessed NDVI data in TIFF format, import it into ArcGIS 10.2 software for NDVI extraction of cultivated land. Right-click the data, select Attributes, click Display in the Symbol System, select Stretch, and choose Standard Deviation Stretch as the type. This type has the same effect as percentage clipping stretch. This stretching method first trims the extreme values ​​of the image, and then performs linear stretching on the other pixel values ​​of the image to increase the image contrast. The maximum, minimum, and standard deviation information of the image can be viewed through the attribute ratio, which allows for statistical analysis of the image information. This facilitates the subsequent determination of NDVI values ​​for cultivated land units and is a key step in extracting irrigated cultivated land. After performing standard deviation stretching, right-click the data, select Data, and export the data to convert the TIFF format data to grid format data.

[0184] Secondly, the 2015 land use status data of Chongqing Municipality, extracted using object-oriented classification, was loaded into ArcGIS 10.2 software in TIFF format. Then, using the conversion tools in ArcGIS 10.2, the raster was converted to polygons, and then to shapefile (SHP) format. Polygon simplification was not selected, and the boundaries of features in the converted SHP file were consistent with the boundaries of features in the original TIFF image. The SHP data was then imported into ArcGIS 10.2 software, and the cultivated land types were located using the built-in attribute codes. A cultivated land layer was created based on the selected features, and then the cultivated land layer data was exported as SHP data to obtain cultivated land vector data.

[0185] Mask data was extracted using NDVI values ​​from cultivated land data based on current land use status. The 12 periods of NDVI data corresponding to the irrigation periods in Chongqing were used as the extraction targets. The extraction tool in the spatial analysis tools was selected, and then mask extraction was chosen. The 12 periods of NDVI data were sequentially input into the raster, and the cultivated land shapefile was selected as the mask file, defining the save path and name. The extraction of NDVI values ​​for 12 periods of cultivated land in Chongqing was completed sequentially. To ensure consistency with the interpolation classification of the previous meteorological interpolation data, the 12 periods of NDVI data were loaded again into ArcGIS 10.2 software. In the symbology, the natural breakpoint classification method, consistent with spatial interpolation, was selected, dividing the cultivated land NDVI values ​​into 10 categories and using a color band system consistent with the spatial interpolation data. NDVI was used to judge the crop growth status of cultivated land; cultivated land with higher NDVI values ​​showed better crop growth and was represented by a blue tint, while cultivated land with lower NDVI values ​​showed a higher probability of crop water shortage and was represented by a red tint. Taking April as an example, the final results are shown in [reference needed]. Figure 11 .

[0186] 4. Irrigation area identification

[0187] (1) Vectorization of interpolation data

[0188] This operation primarily relies on ArcGIS 10.2 software, mainly processing the results of temperature interpolation data, water balance interpolation data, and NDVI value data for cultivated land. Since the results of these data differences are raster data, they need to be converted into vector data (SHP format) that can be used for attribute joining, overlaying, and statistical analysis. The previously processed temperature, water balance, and NDVI value data are floating-point data, while ArcGIS 10.2 requires integer data for raster to vector conversion. Therefore, the mathematical analysis tool in spatial analysis is needed to convert the floating-point data to integers, and then use the conversion tool to convert the raster data to polygon data.

[0189] First, the vectorization of temperature data involves importing the spatially interpolated temperature data into ArcGIS 10.2 software. This process consists of two steps: first, converting the floating-point data to integers, and then exporting the integer data as a vector. Since the pixel values ​​of floating-point data are generally less than 0.5, they need to be enlarged to prevent the values ​​from becoming zero after conversion, which would affect the accuracy of the data. This operation is performed using the multiplication function in Mathematical Analysis, followed by converting the data to integers using the convert-to-integer function within Mathematical Analysis. Because the pixel values ​​of temperature data are relatively large, pixel enlargement is not necessary; the temperature data is directly converted from floating-point to integer and then to shapefile (SHP) data. Since the interpolated data covers the Chongqing municipality, the next step involves vector clipping using the Chongqing boundary. Because the vector clipping reduces the coverage area and alters the classification system with varying numbers of categories, for data requirements and aesthetic appeal, the eight most common classification systems are selected, and consistent color bands are maintained for the output. In the symbol system, the color band display uses the natural breakpoint grading method for classification, dividing the values ​​of each period's vector map into 8 categories. A reddish hue indicates higher temperatures, while a bluish hue indicates lower temperatures. For example, the processed temperature vector data for April can be found here. Figure 12 .

[0190] Secondly, regarding the vectorization of water balance data. Since the data type is floating-point, like temperature data, it becomes smaller due to the difference between precipitation and evaporation data. To facilitate subsequent calculations, this process differs from temperature conversion in that it uses mathematical multiplication to enlarge the pixel values ​​by a factor of 10 before converting them to integer data. After conversion to integer data, the processing method is the same as for temperature data. After converting to shapefile data, it is cropped using the Chongqing city boundary. Because the water balance data values ​​are not significantly different in some months, large-scale classification is not possible. Therefore, classification is performed according to the actual situation, resulting in the vectorized and cropped water balance data. Taking April as an example, the processed water balance vector data can be found here. Figure 13 .

[0191] (2) Irrigation area identification

[0192] Based on the results of water balance measurements, except for the periods of early April, late May, and mid-September, there were areas in Chongqing with water balances less than 0. The lowest value was greater than 1 in late May, while the lowest values ​​for early April and mid-September were both greater than 2. This indicates that during these two periods, Chongqing experienced abundant rainfall and a relatively humid climate. Based on the survey of farmers' irrigation behavior and the characteristics of rain-fed agriculture in Chongqing, these two periods were not included in the overall assessment.

[0193] The operation was performed in ArcGIS 10.2 software, primarily using vector results with water deficits less than 0 to trim the temperature vector data. The trimmed results showed that in areas with water deficits less than 0, the average temperature for the ten-day period in April and May was above 18℃, and the average temperature for the ten-day period from June to September was above 20℃. The difference in average temperature between months was not significant, with the highest values ​​occurring in late July and early August, at 30.42℃ and 30.32℃ respectively, consistent with the timing of the drought in Chongqing. Therefore, except for the three periods of abundant water (early April, late May, and mid-September), areas with water deficits less than 0 can be considered irrigation areas. For example, the irrigated areas after data processing are shown in the image below. Figure 14 The final results, when combined, cover the entire Chongqing municipality, ensuring no areas are missed.

[0194] In summary, this example section first determines the growth period of major crops in Chongqing and then identifies the main irrigation periods for farmland crops, thereby determining the time periods for selecting meteorological and NDVI data. The data from the selected time periods are then processed separately: spatial interpolation is performed on the meteorological data, and NDVI data for cultivated land is extracted. Finally, the processed meteorological and NDVI data are analyzed. After converting the meteorological data from raster to vector format, water surplus / deficit data is obtained through the difference between precipitation and evaporation data. Combined with temperature data analysis, areas with irrigation activity are identified. These overlapping areas cover the entire Chongqing area without omissions, indicating that the selected irrigation areas are reasonable.

[0195] IV. Extraction of Actual Irrigated Area

[0196] Based on previous research, areas with high NDVI data are more likely to be irrigated farmland. Following this principle, NDVI values ​​were extracted from farmland units within irrigated areas for each ten-day period. Analysis was performed using the NDVI difference between later and earlier periods. Farmland with a difference greater than 0 showed better growth than in the previous period, indicating irrigation under high-temperature and water-scarce conditions. This process was repeated to obtain irrigated farmland plots from April to September. Using the co-analysis tool in ArcGIS 10.2, all farmland plots were overlaid, with overlapping areas retained only once and non-overlapping areas overlaid. This yielded the actual irrigated farmland plots in Chongqing in 2015. Finally, computational geometry was used to obtain the actual irrigated area.

[0197] The following is combined Figure 15 The process of extracting the actual irrigated area is explained in detail.

[0198] 1. Principles and Theories

[0199] Introducing multi-temporal Normalized Difference Vegetation Index (NDVI) into the extraction of arable land data can improve extraction accuracy. NDVI values ​​are relatively high when crops are growing well. Among the same or partially different crops, arable land pixels with large NDVI peaks are more likely to be irrigated arable land, even regardless of farming patterns or crop type, especially during periods of insufficient rainfall. Therefore, the determination of the NDVI threshold mainly involves assigning time-series NDVI data values ​​to each arable land plot and using the difference in NDVI values ​​before and after each ten-day period to determine the irrigation status of the arable land. Arable land with relatively increasing NDVI values ​​even when there is a water deficit is considered irrigated arable land.

[0200] Table 14. Number of NDvI data days and dates corresponding to irrigation times for each ten-day period.

[0201]

[0202] The NDVI of vegetation shows a relatively weak lag response to temperature, but a significant lag response to precipitation. Cultivated land's NDVI is more sensitive to both temperature and precipitation than forest land. Analyzing the number of days in NDVI data corresponding to dates with irrigation activity, since the time resolution of each ten-day period is 10 days while the NDVI data time resolution is 17 days, satisfies the theoretical requirement for lag. The difference between the date with lag or that was irrigated and the previous date is used. If the difference is greater than 0, it indicates that the cultivated land is growing better than the previous date; in the case of water shortage and drought, this indicates that the cultivated land has been irrigated; otherwise, it is considered not irrigated. Table 14 shows the number of days and dates in NDVI data corresponding to each ten-day period.

[0203] 2. Extraction from irrigated farmland

[0204] (1) Extraction of arable land in irrigated areas

[0205] This step primarily involves extracting the arable land in the irrigated areas for each ten-day period. Using the clipping tool in ArcGIS 10.2, the input features are selected as NDVI vector data, and the area is clipped according to the region where irrigation occurs within each ten-day period. Since each ten-day period requires two map outputs to calculate the difference between the preceding and following dates, the map is output using the number of days closest to the date of the ten-day period each month as an example, thus obtaining the distribution of arable land with irrigated activity for each ten-day period. For example, in mid-to-late April, the arable land in the irrigated area can be found [see attached image]. Figure 16 .

[0206] (2) Extraction from irrigated farmland

[0207] Because the logic operations and raster calculator in ArcGIS 10.2, which are mathematical analysis tools, require all data to be raster data, vector data cannot be processed. Therefore, after performing subtraction logic operations on the NDVI raster data of cultivated land, the clipping function in the data management tool is used to clip the data according to the irrigation vector area for each ten-day period to obtain the raster data results of the difference between cultivated land before and after irrigation. The raster is then converted to a vector polygon, following the same steps as the above-mentioned vectorization steps for meteorological data difference data. First, the logic operations tool in mathematical analysis is used to convert the raster floating-point data to integer data, and then the raster to polygon function in the data management tool is used to convert it to vector data for subsequent attribute analysis. Taking mid-to-late April as an example, irrigated cultivated land with a difference greater than 0 is selected. See [link to relevant documentation]. Figure 17 .

[0208] 3. Extraction of actual irrigated area

[0209] First, filter farmland with an NDVI difference greater than 0 by attribute in the attribute table, export and save the file, and then import it into ArcGIS 10.2 software. Create a new double-precision field (area) for each ten-day period of vector data, setting the precision to 20 and the decimal places to 2. Perform computational geometry calculations in the same projected coordinate system (UTM Zone 48 North) and geographic coordinate system (GCS WGS1984), selecting square kilometers as the area unit.

[0210] Secondly, the irrigated farmland data with calculated area information from April to September were jointly processed using the union tool in ArcGIS 10.2 geoprocessing software. By calculating the geometric union of the input features, all features and their attributes were written into the output feature class. If the appropriate option is not selected when connecting attributes using the union function, all attributes of the input features will be passed to the output feature class, resulting in an increase in redundant fields and hindering subsequent operations. Therefore, NO_FID is selected in the join attribute bar, which reduces the number of useless fields passed from the input features to the output feature class. Finally, the actual distribution of irrigated farmland in Chongqing in 2015 is obtained as follows: Figure 18 As shown.

[0211] Using the attribute table statistics function, the total actual irrigated area in Chongqing in 2015 was found to be 4687.66 km². 2 Converted to hectares in the statistical yearbook, this is 468,766 hm. 2 According to the statistical results of the 2015 Chongqing Municipal Farmland Irrigation Water Effective Utilization Coefficient Calculation and Analysis Report, Chongqing Municipality has a total of 25,670 irrigation districts, including 124 medium-sized irrigation districts and 25,546 small irrigation districts, with an actual irrigated area of ​​6,369,600 mu (424,661 hectares). 2 .

[0212] Table 15. Comparison of Accuracy in Extracting Actual Irrigated Area

[0213]

[0214] This embodiment selects areas with high temperatures and low rainfall as irrigated areas. Within these areas, farmland experiencing water deficit but with a relatively increasing NDVI value is considered irrigated. Therefore, the difference in NDVI value between different periods is used as the criterion. A difference greater than 0 indicates improved growth compared to the previous day, suggesting irrigation in a drought-stricken environment; otherwise, it's considered unirrigated. Using the extracted irrigated areas as a mask, the cropping function extracts farmland within the irrigated areas for each ten-day period. Based on the NDVI interpolation results, the irrigated farmland results for each ten-day period of each month are selected. After exporting to vector data, a new double-precision area field is created, setting the precision and decimal places to 20 and 2 respectively, with the geometric unit being square kilometers. The final result shows that the actual irrigated area in Chongqing in 2015 was 468,766 hectares. 2 The statistical result in the 2015 Chongqing Municipal Water Resources Bureau's report on the effective utilization coefficient of farmland irrigation water was 424,661 hectares. 2 The relative error was 10.4%, and the accuracy reached 89.6%. This proves that the method of extracting irrigated farmland using the NDVI difference of cultivated land is simple and has high accuracy.

[0215] The beneficial effects of this application example are as follows:

[0216] 1. Using remote sensing to obtain the actual irrigated area of ​​farmland can save time for manual statistics and provide a reference for the investigation and verification of the actual irrigated area of ​​farmland in Chongqing, thus helping to ensure the accuracy of the measurement work.

[0217] 2. Existing technologies for remote sensing monitoring of farmland irrigation area are mostly based on large and medium-sized irrigation districts, with large irrigation districts equivalent to the county level. However, there are few studies at the provincial and municipal levels. This example of remote sensing extraction of the actual irrigated area of ​​farmland at the provincial and municipal levels, such as Chongqing, can provide a certain reference for provincial and municipal level studies.

[0218] 3. In terms of topography, Chongqing is a typical mountainous city, with mountains and hills accounting for over 94% of its total area, forming a unique combination of landform types dominated by mountains. Farmland is fragmented and scattered. In terms of climate, Chongqing experiences a subtropical monsoon climate with mild winters and hot summers, abundant rainfall, and simultaneous rain and heat, leading to a rain-fed agriculture model. Under similar conditions in the hilly and mountainous areas of Southwest China, extracting the actual irrigated area of ​​farmland in Chongqing can serve as a representative study for the Southwest region and has certain applicability.

[0219] The preferred embodiments of the present invention have been described above with reference to the accompanying drawings, but this does not limit the scope of the invention. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and spirit of the present invention should be within the scope of the present invention.

Claims

1. A method for determining the actual irrigated area of ​​farmland, characterized in that, The method includes: Based on remote sensing images of the farmland irrigation study area, the cultivated land in the farmland irrigation study area is extracted to obtain the cultivated land plots in the farmland irrigation study area; Based on meteorological data observed at multiple meteorological stations within and adjacent areas of the farmland irrigation study area during the crop growth period, the irrigation areas within the farmland irrigation study area during the crop growth period are determined. Using cultivated land plots in the farmland irrigation study area and normalized differential vegetation index (NDVI) data of the farmland irrigation study area during the crop growth period, the cultivated land NDVI data of the farmland irrigation study area during the crop growth period are determined. Based on the NDVI data of irrigation areas and cultivated land in the farmland irrigation study area during the crop growth period, the actual irrigated area of ​​the farmland irrigation study area is determined; The determination of the actual irrigated area of ​​the farmland irrigation study area based on NDVI data of irrigated areas and cultivated land within the farmland irrigation study area during the crop growth period includes: Farmland plots were cut out from the irrigated areas within the farmland irrigation study area during the crop growth period; Based on the NDVI data of cultivated land plots in the irrigated area for the period before and after each ten-day period during the crop growth period, the difference in NDVI data of cultivated land plots in the irrigated area for the period before and after each ten-day period during the crop growth period is obtained. From the cultivated land plots within the irrigation area, select cultivated land plots where the difference between the cultivated land NDVI data before and after each ten-day period during the crop growth period is greater than zero, and determine them as irrigated cultivated land for each ten-day period during the crop growth period. The actual irrigated area of ​​the farmland in the research area was determined based on the amount of irrigated farmland per ten-day period during the crop growth cycle.

2. The method according to claim 1, characterized in that, Before extracting the cultivated land in the farmland irrigation study area based on remote sensing images of the farmland irrigation study area to obtain the cultivated land plots in the farmland irrigation study area, the method further includes: Acquire Landsat 8 remote sensing image data of the farmland irrigation study area, NDVI data of the farmland irrigation study area during the crop growth period, and meteorological data observed by multiple meteorological stations in the farmland irrigation study area and its vicinity during the crop growth period; The Landsat 8 remote sensing image data of the farmland irrigation study area were preprocessed, including radiometric correction, geometric correction and image fusion, to obtain remote sensing images. The NDVI data of the farmland irrigation study area during the crop growth period were preprocessed, including projection transformation and vector clipping, to obtain the preprocessed NDVI data. Meteorological data observed at multiple meteorological stations in the farmland irrigation study area and its vicinity during the crop growth period were preprocessed, including the removal of outliers, to obtain preprocessed meteorological data.

3. The method according to claim 1, characterized in that, The process involves extracting arable land from remote sensing images of the farmland irrigation study area to obtain arable land plots within the farmland irrigation study area, including: The maximum likelihood method or object-oriented classification method is used to extract cultivated land from the remote sensing images of the farmland irrigation study area to obtain cultivated land plots in the farmland irrigation study area. Alternatively, the maximum likelihood method and object-oriented classification method can be used respectively to extract cultivated land from the remote sensing images of the farmland irrigation study area, and the optimal cultivated land extraction result can be selected as the cultivated land plot in the farmland irrigation study area.

4. The method according to claim 1, characterized in that, The determination of the irrigated areas within the farmland irrigation study area during the crop growth period, based on meteorological data observed at multiple meteorological stations within and adjacent areas of the farmland irrigation study area, includes: Spatial interpolation is performed on the meteorological data of each ten-day period observed by the multiple meteorological stations during the crop growth period to obtain the spatial interpolated meteorological data for each ten-day period; Based on the meteorological spatial interpolation data for each ten-day period, the irrigation area for each ten-day period in the farmland irrigation study area during the crop growth period is determined.

5. The method according to claim 4, characterized in that, Meteorological data from each meteorological station includes evaporation, precipitation, and temperature. Spatial interpolation is performed on the meteorological data observed at multiple meteorological stations during the crop growing season for each ten-day period to obtain the spatially interpolated meteorological data for each ten-day period, which includes: Based on the evaporation, precipitation, and temperature of each of the multiple meteorological stations per ten-day period, the average evaporation, precipitation, and temperature of each meteorological station per ten-day period are determined. Spatial interpolation was performed on the average evaporation, precipitation, and temperature of the multiple meteorological stations for each ten-day period to obtain interpolated evaporation, precipitation, and temperature data for each ten-day period in the farmland irrigation study area and its adjacent areas.

6. The method according to claim 5, characterized in that, The determination of the irrigation area for each ten-day period within the farmland irrigation study area based on the meteorological spatial interpolation data for each ten-day period includes: Based on the interpolated data of evaporation and precipitation for each ten-day period in the farmland irrigation study area and its adjacent areas, the interpolated data of water surplus / deficit for each ten-day period in the farmland irrigation study area and its adjacent areas are determined. Based on the interpolated data of water surplus / deficit and temperature for each ten-day period within the farmland irrigation study area and its vicinity, the irrigation areas within each ten-day period of the farmland irrigation study area are identified.

7. The method according to claim 6, characterized in that, The step of identifying the irrigated areas within the farmland irrigation study area for each ten-day period based on the interpolated data of water surplus / deficit and temperature for each ten-day period in the farmland irrigation study area and its adjacent areas includes: The interpolated data of water surplus / deficit and temperature for each ten-day period in the farmland irrigation study area and its adjacent area are vectorized to obtain vector data of water surplus / deficit and temperature for each ten-day period in the farmland irrigation study area and its adjacent area. Using the boundary of the farmland irrigation study area, the vector data of water surplus / deficit and temperature for each ten-day period in the farmland irrigation study area and its adjacent areas are cropped to obtain the vector data of water surplus / deficit and temperature for each ten-day period within the coverage area of ​​the farmland irrigation study area. Based on the vector data of water surplus / deficit and temperature within the coverage area of ​​the farmland irrigation study area for each ten-day period, the irrigation areas within the farmland irrigation study area for each ten-day period are identified.

8. The method according to claim 1, characterized in that, The process of determining the NDVI data of cultivated land in the farmland irrigation study area during the crop growth period using cultivated land plots and normalized differential vegetation index (NDVI) data of the farmland irrigation study area includes: The cultivated land plots in the farmland irrigation study area are vectorized to obtain the cultivated land vector data of the farmland irrigation study area. Using the cultivated land vector data of the farmland irrigation study area as mask data, the NDVI values ​​of cultivated land plots are extracted from the NDVI data of the farmland irrigation study area during the crop growth period to form cultivated land NDVI data.

Images