Method for calculating effective utilization coefficient of irrigation water based on remote sensing ET

Abstract

This invention discloses a method for calculating the effective utilization coefficient of irrigation water based on remote sensing ET (Emergency Thresholding). The method includes: acquiring the evapotranspiration of each pixel in the study area at high spatiotemporal resolution; extracting multi-source remote sensing features of Sentinel-2 multi-temporal spectral features during the study period, and combining them with a trained ensemble learning classification model to obtain a crop planting structure map for each ten-day period; using the evapotranspiration of the pixel as the crop water requirement for the ten-day period, selecting an effective precipitation model based on the cultivated land type of the crop in the pixel, and calculating the effective precipitation for the pixel based on the crop water requirement and precipitation for each ten-day period; using Sentinel-1 imagery of the study period combined with OTSU classification, marking pixels with increased moisture as moisture increase events in each ten-day period, and correcting the moisture increase events using a scatter plot to obtain a ten-day irrigation distribution map; and calculating the effective utilization coefficient of irrigation water in the study area based on the evapotranspiration, crop planting structure map, effective precipitation, irrigation distribution map, and gross irrigation diversion.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of agricultural water conservancy engineering and remote sensing information technology, specifically to a method for calculating the effective utilization coefficient of irrigation water based on remote sensing ET. Background Technology

[0002] The effective utilization coefficient of irrigation water is a key indicator for measuring the efficiency of water resource utilization in an irrigation district or region and for assessing the level of agricultural water management. Accurately and efficiently obtaining this coefficient is of great significance for optimizing water resource allocation and guiding water-saving agricultural production practices.

[0003] In recent years, with the rapid development of remote sensing technology, the inversion of key elements of the agricultural water cycle using satellite observation data has become a research hotspot. Among these, methods for estimating farmland water consumption based on remote sensing evapotranspiration (ET) models have opened up new technical pathways for irrigation efficiency assessment. Existing research has successfully inverted daily evapotranspiration using remote sensing data through single-source evapotranspiration models such as SEBAL, effectively avoiding the calculation difficulties of complex processes such as soil evaporation and deep seepage in traditional methods, and achieving indirect estimation of net irrigation water consumption for crops. In terms of the application of irrigation water effective utilization coefficient calculation, patent documents 202510844013.5, 201911127287.3, and 202510781504.X have proposed coefficient calculation methods based on multi-source remote sensing information, the SEBAL model, and irrigation area water use process simulation, respectively, exploring the application potential of remote sensing technology in irrigation efficiency assessment.

[0004] However, a comprehensive analysis of existing technologies and research findings reveals the following pressing technical problems that need to be addressed: 1. Mismatch in evapotranspiration data scale and isolation in key parameter inversion: Existing mainstream remote sensing evapotranspiration products generally have coarse spatial resolution (kilometer level), which is insufficient to meet the water use analysis needs at the field scale or fine scale within irrigation districts (such as different crop planting areas), resulting in limited spatial accuracy of irrigation water consumption estimation. Meanwhile, agricultural production areas have complex planting structures, with significant differences in the growth processes and water requirements of different crops. Existing methods are isolated in key aspects such as evapotranspiration estimation, planting structure identification, effective precipitation calculation, and irrigation area identification, lacking an effective coupling mechanism, making it difficult to achieve refined and spatial estimation of irrigation water consumption.

[0005] 2. Insufficient accuracy in identifying irrigation behavior and lack of a collaborative judgment mechanism: When using radar remote sensing to identify irrigation events, both irrigation and precipitation increase soil moisture, resulting in similar signal changes in the radar backscattering coefficient time series. Existing methods lack a stable and efficient mechanism to accurately distinguish between the two, and fail to effectively integrate crop planting structure and phenological information for comprehensive judgment, thus affecting the accurate identification of irrigated area and irrigation water consumption.

[0006] 3. Lack of a spatialized collaborative estimation framework: Most existing methods treat crop planting structure identification, evapotranspiration estimation, and irrigation event monitoring in isolation, failing to form a complete technical solution that can integrate these key parameters at a unified spatial scale (e.g., 10-meter resolution) and directly and spatially calculate the effective irrigation water use coefficient. This results in most current calculations of the effective irrigation water use coefficient being based on regional average estimates using statistical or experimental data, making it difficult to comprehensively reveal the spatial differentiation patterns of irrigation efficiency among different regions and crop types. Summary of the Invention

[0007] To address the aforementioned shortcomings in existing technologies, the method for calculating the effective utilization coefficient of irrigation water based on remote sensing (ET) provided by this invention solves the problem that existing technologies struggle to achieve precise estimation of irrigation water consumption, resulting in insufficient accuracy.

[0008] To achieve the above-mentioned objectives, this invention provides a method for calculating the effective utilization coefficient of irrigation water based on remote sensing (ET), which includes the following steps: S1. Obtain the Sentinel-2 NDVI and low-resolution remote sensing evapotranspiration of the study area during the study period, and use flux decomposition to downscale and obtain the decadal evapotranspiration of each pixel in the study area at high spatiotemporal resolution. S2. Extract multi-source remote sensing features of Sentinel-2 multi-temporal spectral features of the study area during the study period, and combine them with the trained ensemble learning classification model to obtain the crop planting structure map for each ten-day period. S3. The evapotranspiration of the pixel is used as the crop water requirement for the ten-day period. The effective precipitation model is selected according to the cultivated land type of the crop in the pixel. Combined with the crop water requirement and precipitation for each ten-day period, the effective precipitation of the pixel is calculated. S4. Using Sentinel-1 images of the study period combined with OTSU classification, pixels with increased water content in each ten-day period were marked as water increase events. The water increase events were then corrected using scatter plots to obtain the ten-day irrigation distribution map. S5. Based on the ten-day evapotranspiration, ten-day crop planting structure map, ten-day effective precipitation, ten-day irrigation distribution map, and gross irrigation water diversion, calculate the effective utilization coefficient of irrigation water for each ten-day period and / or the entire study period in the study area.

[0009] In one possible implementation, step S4 further includes: S41. Based on the Sentinel-1 images for the study period, generate the VV and VH polarization backscattering coefficients at the decadal scale, and calculate the decadal irrigation response index using the VV and VH polarization backscattering coefficients. ; S42. In the ten-day crop planting structure map, based on the known irrigation information of the pixels, determine irrigated and non-irrigated samples, and extract the samples across all ten-day periods. The optimal segmentation threshold for the i-th ten-day period is calculated by combining OTSU dynamic threshold segmentation. ; S43. For the i-th ten-day period, the study area will satisfy... The pixels were labeled as moisture increase events, and the ten-day cumulative precipitation was extracted for all non-irrigated samples. and the ten-day scale irrigation response index ; S44, with and Using the x and y axes, all non-irrigated samples were used across all ten-day periods. Scatter plots were created using data point pairs; inflection points in the scatter plots were identified using piecewise linear regression, and the corresponding inflection points were assigned to... As the initial ten-day threshold ; S45, will satisfy and Irrigated and non-irrigated samples were labeled as irrigation events, and the misclassification rates for irrigation and non-irrigated samples were calculated separately. S46. When the false positive rate of the irrigation sample is greater than the first proportion, Increase the preset value and return to step S45; when the false positive rate of non-irrigated samples is greater than the second proportion, Decrease the preset value and return to step S45; if the previous two conditions are not met, use... As an effective precipitation event threshold Then proceed to step S47; S47. In each ten-day period, in the cells marked as moisture increase events, the following will be satisfied: The pixels are labeled as irrigation candidate events, and combined with the irrigation period of the crop, the final irrigation events for each ten-day period are determined to form a ten-day irrigation distribution map.

[0010] In one possible implementation, step S41 further includes: Sentinel-1 images for the study period were acquired and preprocessed. All Sentinel-1 images passing through each ten-day period were averaged to generate VV and VH polarization backscattering coefficients at the ten-day scale. Calculate the decadal-scale irrigation response index based on the VV and VH polarization backscattering coefficients: in, is the ten-day scale irrigation response index for the i-th ten-day period; and The images show the VV and VH polarization backscattering coefficients for the i-th tenth period, respectively. and The images show the VV and VH polarization backscattering coefficients for the i-1th tenth of the month.

[0011] In one possible implementation, step S1 further includes: S11. Acquire Sentinel-2 images of the study area during the study period, and reconstruct the daily 10-meter resolution NDVI of each pixel using a method based on linear interpolation and SG filtering. Calculate the average daily 10-meter resolution NDVI of the j-th pixel in month m as the monthly average. ; S12. Calculate the compensation factor based on the historical vegetation index and evapotranspiration of the study area: in, The compensation factor for the j-th pixel in month m; and The average vegetation index of the j-th and l-th pixels in month m is obtained from historical data. and The average evapotranspiration of the j-th and l-th pixels in month m is obtained from historical data. S13. Use compensation factors to adjust monthly average values. After adjustment, the downscaling driving factor is obtained: in, Let be the downscaling driving factor for the j-th pixel in month m; S14. For month m, the evapotranspiration at 1km resolution is calculated using all 10-meter pixels within the 1km resolution pixel of that month. The average value is taken as the driving factor for evapotranspiration at a resolution of 1 km. ; S15. For any ten-day period within month m, according to and Converting evapotranspiration at 1km resolution to evapotranspiration at 10m resolution: in, The evapotranspiration of the j-th pixel at a resolution of 10 meters is given by the given value. Evapotranspiration at a resolution of 1 km.

[0012] The beneficial effects of the above technical solution are as follows: This solution integrates low- and medium-resolution remote sensing evapotranspiration products (1km resolution) with high-resolution optical remote sensing data (Sentinel-2), calculates compensation factors using the vegetation index of different land cover types and the historical statistical relationship of evapotranspiration, and generates spatiotemporally continuous actual evapotranspiration with both high temporal resolution (ten-day period) and high spatial resolution (10 meters) based on the flux decomposition method, effectively solving the problem of mismatch between the scale of evapotranspiration data and the scale of farmland irrigation management.

[0013] In one possible implementation, the arable land type includes paddy fields and dryland crops, and the effective precipitation model for paddy fields is expressed as follows: in, This represents the effective precipitation for the i-th ten-day period; This represents the rainfall in the i-th ten-day period; This represents the crop water requirement for the i-th ten-day period; The expression for the effective precipitation model for dryland crops is as follows: Among them, SF is the soil moisture storage factor; D is 40%-60% of the effective water holding capacity of the crop root zone soil; When dryland crops At that time, then order .

[0014] In one possible implementation, step S2 further includes: Multi-source remote sensing features of each Sentinel-2 image in the study area during the study period were extracted and used to construct the feature vector of each pixel. All feature vectors of each ten-day period of the pixel were stacked to form a multi-dimensional feature vector of each ten-day period of the pixel. The multidimensional feature vector of each pixel is input into the ensemble learning classification models Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), and Extreme Gradient Boosting (XGBoost) to obtain the four class probabilities for each ten-day period. The probabilities of the four categories are concatenated to form a meta-feature, which is then input into a meta-classifier based on a Logistic regression model. This yields the predicted type for each pixel in each ten-day period, forming a ten-day crop planting structure map for the study area.

[0015] In one possible implementation, step S5 further includes: Determine the net irrigation water consumption per ten-day period based on the evapotranspiration of each pixel: in, Let be the net irrigation water consumption of the j-th pixel in the i-th ten-day period; The evapotranspiration data for the j-th pixel in the i-th ten-day period; Let be the effective precipitation of the j-th pixel in the i-th ten-day period; This is a function to find the maximum value. Calculate the effective irrigation water utilization coefficient of crop c in the i-th ten-day period based on the net irrigation water consumption per ten-day period. : Where S represents the study area; The area of ​​a pixel; Let c represent the type of crop planted in the j-th pixel; The gross irrigation water intake for crop c in the study area S during the i-th ten-day period; Calculate the effective irrigation water utilization coefficient of crop c in n ten-day periods based on the net irrigation water consumption per ten-day period: When n=3, Let n be the monthly irrigation water effective utilization coefficient for crop c; when n is the total number of ten days in the study period, denoted as the effective irrigation water utilization coefficient for crop c during the study period.

[0016] In one possible implementation, step S5 further includes calculating the pixel scale coefficient and the coefficients of different spatial units: , in, Let T be the utilization efficiency of irrigation water for the j-th pixel in time window T, where T is a time window consisting of at least one ten-day period. Let J be the net irrigation water consumption of the j-th pixel within time window T. Let be the net irrigation water consumption of the j-th pixel in the i-th ten-day period; To study the utilization efficiency of irrigation water in region S within the time window T; The area of ​​a pixel; The gross irrigation water volume of the study area S during the time window T is calculated.

[0017] In one possible implementation, the optimal segmentation threshold for the i-th tenth time is calculated using the OTSU dynamic threshold segmentation method. The methods include: For the i-th ten-day period, the pixels corresponding to irrigated samples and non-irrigated samples are compared. Sort the data and calculate the global mean. ; Traverse any two adjacent pairs Take the midpoint t, and combine all Divided into two categories: ≤t and >t; Calculate the two types of weights , and mean , and inter-class variance Select to make The largest t is used as the optimal segmentation threshold for the i-th period. .

[0018] In one possible implementation, the multi-source remote sensing features include spectral, vegetation index, texture, and radar features; the spectral features include Blue, Green, Red, NIR, SWIR1, and SWIR2, and the vegetation indices include Normalized Difference Vegetation Index, Enhanced Vegetation Index, Surface Moisture Index, Modified Normalized Difference Water Index, Soil-Regulated Vegetation Index, Red-Edge Normalized Difference Vegetation Index, and Red-Edge Location Index. The texture is a statistical measure calculated from the near-infrared band based on the gray-level co-occurrence matrix: mean, variance, homogeneity, contrast, correlation, and entropy; the radar features are the backscattering coefficients of VV and VH polarization and the radar vegetation index extracted from the contemporaneous Sentinel-1 image.

[0019] Compared with the prior art, the present invention has the following outstanding advantages: This scheme generates spatiotemporally continuous actual evapotranspiration data with both high temporal resolution (ten-day periods) and high spatial resolution (10 meters) based on flux decomposition, effectively solving the problem of mismatch between the scale of evapotranspiration data and the scale of farmland irrigation management. Combined with an ensemble learning classification model, it can accurately extract the crop planting structure and its spatial distribution in the study area. By combining Sentinel-1 imagery with OTSU classification, it can distinguish irrigation signals from precipitation signals, achieving spatial and accurate identification of farmland irrigation events and irrigated areas, solving the technical problem of separating irrigation and precipitation signals in remote sensing monitoring. Through the accurately obtained ten-day evapotranspiration, ten-day crop planting structure map, ten-day effective precipitation, and ten-day irrigation distribution map, it can accurately calculate the effective utilization coefficient of irrigation water, thereby accurately assessing whether the allocation of agricultural irrigation water resources is reasonable, so as to make timely adjustments and ultimately achieve the goal of water conservation.

[0020] An innovative regional-scale measurement model: This invention breaks through the traditional "point-to-area" model that uses field measurements of typical plots and sample irrigation areas to extrapolate regional coefficients. It directly utilizes multi-source remote sensing technology to spatially identify irrigated areas, calculate crop evapotranspiration, and deduct effective precipitation, thereby directly and spatially extrapolating net irrigation water volume. Combined with the gross irrigation water diversion coefficient calculation, this achieves a fundamental shift in irrigation efficiency assessment from a point-scale to an area-scale approach, ensuring the accuracy of the final calculation.

[0021] High-precision irrigation event identification and ET downscaling fusion innovation: By constructing a decadal-scale irrigation response index based on radar backscattering and combining it with a triple discrimination mechanism of "dynamic threshold-precipitation filtering-phenological constraints," the spatiotemporal accuracy of irrigation event identification has been significantly improved. Simultaneously, by using evapotranspiration downscaling technology that integrates optical and thermal infrared remote sensing data, the challenge of acquiring high-resolution evapotranspiration data has been solved, effectively unifying irrigation behavior monitoring and crop water consumption estimation at a 10-meter fine scale, ensuring accurate calculation of irrigation water use efficiency.

[0022] A collaborative estimation framework based on remote sensing evapotranspiration (ET) as a unified metric: This pioneering spatialized computational framework uses high-resolution remote sensing evapotranspiration as its core link. By coupling four core elements—evapotranspiration downscaling, planting structure inversion, effective precipitation calculation, and irrigated area identification—this framework achieves collaborative estimation and comparability analysis of net irrigation water at different time and spatial scales (e.g., plots, irrigation districts). This provides novel technical support and a standardized data foundation for understanding the spatial heterogeneity of irrigation efficiency and for implementing refined irrigation district management. Attached Figure Description

[0023] Figure 1 This is a flowchart of a method for calculating the effective utilization coefficient of irrigation water based on remote sensing (ET). Detailed Implementation

[0024] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0025] refer to Figure 1 , Figure 1 A flowchart illustrating the method for calculating the effective utilization coefficient of irrigation water based on remote sensing ET is shown; for example... Figure 1 As shown, the method S includes steps S1 to S5.

[0026] In step S1, the Sentinel-2 NDVI and low-resolution remote sensing evapotranspiration of the study area during the study period are obtained, and the flux decomposition method is used to downscale and obtain the decadal evapotranspiration of each pixel in the study area at high spatiotemporal resolution. In one embodiment of the present invention, step S1 further includes: S11. Acquire Sentinel-2 images of the study area during the study period, and reconstruct the daily 10-meter resolution NDVI of each pixel using a method based on linear interpolation and SG filtering. Calculate the average daily 10-meter resolution NDVI of the j-th pixel in month m as the monthly average. ; S12. Calculate the compensation factor based on the historical vegetation index and evapotranspiration of the study area: in, The compensation factor for the j-th pixel in month m; and The average vegetation index of the j-th and l-th pixels in month m is obtained from historical data. and The average evapotranspiration of the j-th and l-th pixels in month m is obtained from historical data. S13. Use compensation factors to adjust monthly average values. After adjustment, the downscaling driving factor is obtained: in, Let be the downscaling driving factor for the j-th pixel in month m; S14. For month m, the evapotranspiration at 1km resolution is calculated using all 10-meter pixels within the 1km resolution pixel of that month. The average value is taken as the driving factor for evapotranspiration at a resolution of 1 km. The evapotranspiration at 1km resolution is SSEBop remote sensing ET product.

[0027] S15. For any ten-day period within month m, according to and Converting evapotranspiration at 1km resolution to evapotranspiration at 10m resolution: in, The evapotranspiration of the j-th pixel at a resolution of 10 meters is given by the given value. Evapotranspiration at a resolution of 1 km.

[0028] In step S2, multi-source remote sensing features of Sentinel-2 multi-temporal spectral features of the study area during the study period are extracted and combined with a trained ensemble learning classification model to obtain the crop planting structure map for each ten-day period. The multi-source remote sensing features include spectral, vegetation index, texture, and radar features. The spectral features include Blue, Green, Red, NIR, SWIR1, and SWIR2. The vegetation indices include Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EDI), Surface Moisture Index (SMO), Improved Normalized Difference Water Index (MDDI), Soil-Regulated Vegetation Index (SDI), Red-Edge Normalized Difference Vegetation Index (REDDI), and Red-Edge Location Index (REDCI). The texture features are statistics calculated from the near-infrared band based on the gray-level co-occurrence matrix: mean, variance, homogeneity, contrast, correlation, and entropy. The radar features are backscattering coefficients of VV and VH polarizations and radar vegetation indices extracted from the same period's Sentinel-1 images.

[0029] In one embodiment of the present invention, step S2 further includes: Multi-source remote sensing features of each Sentinel-2 image in the study area during the study period were extracted and used to construct the feature vector of each pixel. All feature vectors of each ten-day period of the pixel were stacked to form a multi-dimensional feature vector of each ten-day period of the pixel. The multidimensional feature vector of each pixel is input into the ensemble learning classification models Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), and Extreme Gradient Boosting (XGBoost) to obtain the four class probabilities for each ten-day period. The probabilities of the four categories are concatenated to form a meta-feature, which is then input into a meta-classifier based on a Logistic regression model. This yields the predicted type for each pixel in each ten-day period, forming a ten-day crop planting structure map for the study area.

[0030] When training the ensemble learning classification model, to capture crop phenological characteristics, each Sentinel-2 image during key growth stages is selected. Features extracted from all growth stages (including 22 features across four aspects: spectral bands, vegetation index, texture features, and radar features) are stacked to form a multidimensional feature vector for each pixel. The crop planting type of the pixel is used as the pixel's category label. Each pixel's multidimensional feature vector and category label constitute a sample, and multiple samples form a sample set. Let the sample set be... ,in Let be the feature vector of the nth sample. For category labels ( (Number of crop types).

[0031] Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), and Extreme Gradient Boosting (XGBoost) were selected as base classifiers. The training samples in the sample set were divided into a 70% training set and a 30% test set, and the hyperparameters of each base classifier were tuned using 5-fold cross-validation.

[0032] A stacking method is used to integrate the results of the base classifiers. This stacking method is a two-layer ensemble strategy. Based on the aforementioned K base classifiers (such as RF, SVM, ANN, XGBoost), the samples are predicted. To avoid overfitting, 5-fold cross-validation is used to generate meta-features: for each base classifier k, it is trained with 4 folds, and the remaining 1 fold is used to predict, resulting in the probability vector for each sample. (The probability vector output by the k-th base classifier for the n-th training sample, with length C and the sum of its components being 1).

[0033] The probabilities of each base classifier are concatenated to form the meta-feature. (The meta-feature vector of the nth sample is formed by concatenating the probability vectors of all base classifiers, with dimensions K×C). (Compared to the true label) Constructing the meta-training set Then, a Logistic regression model was used as the meta-classifier. Conduct training.

[0034] In step S3, the evapotranspiration of a pixel is used as the crop water requirement for that ten-day period. An effective precipitation model is selected based on the cultivated land type of the crop in the pixel, and the effective precipitation for that pixel is calculated by combining the crop water requirement and precipitation for each ten-day period. The cultivated land type includes paddy fields and dryland crops. The expression for the effective precipitation model for paddy fields is as follows: in, This represents the effective precipitation for the i-th ten-day period; This represents the rainfall in the i-th ten-day period; This represents the crop water requirement for the i-th ten-day period; The expression for the effective precipitation model for dryland crops is as follows: Among them, SF is the soil moisture storage factor; D is 40%-60% of the effective water holding capacity of the crop root zone soil; When dryland crops At that time, then order .

[0035] In step S4, Sentinel-1 images of the study period are combined with OTSU classification to mark pixels with increased moisture as moisture increase events in each ten-day period. These events are then corrected using a scatter plot to obtain a ten-day irrigation distribution map. S41. Based on the Sentinel-1 images for the study period, generate the VV and VH polarization backscattering coefficients at the decadal scale, and calculate the decadal irrigation response index using the VV and VH polarization backscattering coefficients. ; S42. In the ten-day crop planting structure map, based on the known irrigation information of the pixels, determine irrigated and non-irrigated samples, and extract the samples across all ten-day periods. The optimal segmentation threshold for the i-th ten-day period is calculated by combining OTSU dynamic threshold segmentation. ; In implementation, this scheme preferably combines the OTSU dynamic threshold segmentation method to calculate the optimal segmentation threshold for the i-th tenth period. The methods include: For the i-th ten-day period, the pixels corresponding to irrigated samples and non-irrigated samples are compared. Sort the data and calculate the global mean. ; Traverse any two adjacent pairs Take the midpoint t, and combine all Divided into two categories: ≤t and >t; Calculate the two types of weights , and mean , and inter-class variance Select to make The largest t is used as the optimal segmentation threshold for the i-th period. .

[0036] S43. For the i-th ten-day period, the study area will satisfy... The pixels were labeled as moisture increase events, and the ten-day cumulative precipitation was extracted for all non-irrigated samples. and the ten-day scale irrigation response index ; S44, with and Using the x and y axes, all non-irrigated samples were used across all ten-day periods. Scatter plots were created using data point pairs; inflection points in the scatter plots were identified using piecewise linear regression, and the corresponding inflection points were assigned to... As the initial ten-day threshold ; S45, will satisfy and Irrigated and non-irrigated samples were labeled as irrigation events, and the misclassification rates for irrigation and non-irrigated samples were calculated separately. S46. When the false positive rate of the irrigation sample is greater than the first proportion, Increase the preset value and return to step S45; when the false positive rate of non-irrigated samples is greater than the second proportion, Decrease the preset value and return to step S45; if the previous two conditions are not met, use... As an effective precipitation event threshold Then proceed to step S47; S47. In each ten-day period, in the cells marked as moisture increase events, the following will be satisfied: Pixels are labeled as candidate irrigation events, and combined with the crop's irrigation period, the final irrigation event for each ten-day period is determined to form a ten-day irrigation distribution map. To further improve the accuracy of identification, it can also be determined whether a pixel is a cultivated land pixel. If so, the pixel is retained because an irrigation event has occurred; otherwise, the pixel is deleted from the irrigation event list.

[0037] In implementation, the preferred step S41 of this scheme further includes: Sentinel-1 images for the study period were acquired and preprocessed. All Sentinel-1 images passing through each ten-day period were averaged to generate VV and VH polarization backscattering coefficients at the ten-day scale. Calculate the decadal-scale irrigation response index based on the VV and VH polarization backscattering coefficients: in, The ten-day scale irrigation response index for the i-th ten-day period; and The images show the VV and VH polarization backscattering coefficients for the i-th tenth period, respectively. and The images show the VV and VH polarization backscattering coefficients for the i-1th tenth of the month.

[0038] In step S5, based on the ten-day evapotranspiration, ten-day crop planting structure map, ten-day effective precipitation, ten-day irrigation distribution map, and gross irrigation diversion, the effective utilization coefficient of irrigation water for each ten-day period and / or the entire study period in the study area is calculated: Determine the net irrigation water consumption per ten-day period based on the evapotranspiration of each pixel: in, Let be the net irrigation water consumption of the j-th pixel in the i-th ten-day period; The evapotranspiration data for the j-th pixel in the i-th ten-day period; Let be the effective precipitation of the j-th pixel in the i-th ten-day period; This is a function to find the maximum value. Calculate the effective irrigation water utilization coefficient of crop c in the i-th ten-day period based on the net irrigation water consumption per ten-day period. : Where S represents the study area; The area of ​​a pixel; Let c represent the type of crop planted in the j-th pixel; To study the gross irrigation water volume of crop c in region S during the i-th ten-day period, such as the ten-day water diversion records at the irrigation canal head and the conversion of ten-day electricity (oil) consumption of agricultural wells into water volume, data can be obtained from field observation data. If there is no ten-day data, the water diversion volume on a longer time scale can be allocated to each ten-day period according to the proportion of irrigated area or crop water requirement.

[0039] Calculate the effective irrigation water utilization coefficient of crop c in n ten-day periods based on the net irrigation water consumption per ten-day period: When n=3, Let n be the monthly irrigation water effective utilization coefficient for crop c; when n is the total number of ten days in the study period, denoted as the effective irrigation water utilization coefficient for crop c during the study period.

[0040] The parameters obtained from steps S1 to S4 can also be used to calculate the pixel scale coefficient and the coefficients of different spatial units. Therefore, step S5 also includes calculating the pixel scale coefficient and the coefficients of different spatial units. , in, Let T be the utilization efficiency of irrigation water for the j-th pixel in time window T, where T is a time window consisting of at least one ten-day period. The net irrigation water consumption of the j-th pixel within time window T; In the calculation formula, the numerator represents the cumulative net irrigation water consumption of the pixel within the time window T; the denominator represents the gross irrigation water consumption per unit area of ​​the pixel within the time window T. It needs to be obtained by using a spatial allocation model to allocate the total gross irrigation water diversion of the upper-level spatial units (such as the control range of the irrigation canal, administrative villages) to each pixel according to the proportion of irrigation area or crop water demand.

[0041] Let be the net irrigation water consumption of the j-th pixel in the i-th ten-day period; To study the utilization efficiency of irrigation water in region S within the time window T; The area of ​​a pixel; To study the gross irrigation water volume of region S within time window T, this data is relatively easy to obtain for units with independent water inlets, such as administrative regions or irrigation districts.

[0042] In summary, the irrigation water effective utilization coefficient calculation method provided in this scheme breaks through the limitations of traditional methods in data acquisition, scale matching, signal recognition, and spatial representation, and realizes a refined, spatial, and dynamic assessment of irrigation efficiency. This ensures the accuracy of the irrigation water effective utilization coefficient calculation, and thus enables accurate assessment of whether agricultural irrigation water resource allocation is reasonable, so as to make timely adjustments and ultimately achieve the goal of water conservation.

Claims

1. A method for calculating the effective utilization coefficient of irrigation water based on remote sensing (ET), characterized in that, Including the following steps: S1. Obtain the Sentinel-2 NDVI and low-resolution remote sensing evapotranspiration of the study area during the study period, and use flux decomposition to downscale and obtain the decadal evapotranspiration of each pixel in the study area at high spatiotemporal resolution. S2. Extract multi-source remote sensing features of Sentinel-2 multi-temporal spectral features of the study area during the study period, and combine them with the trained ensemble learning classification model to obtain the crop planting structure map for each ten-day period. S3. The evapotranspiration of the pixel is used as the crop water requirement for the ten-day period. The effective precipitation model is selected according to the cultivated land type of the crop in the pixel. Combined with the crop water requirement and precipitation for each ten-day period, the effective precipitation of the pixel is calculated. S4. Using Sentinel-1 images of the study period combined with OTSU classification, pixels with increased water content in each ten-day period were marked as water increase events. The water increase events were then corrected using scatter plots to obtain the ten-day irrigation distribution map. S5. Based on the ten-day evapotranspiration, ten-day crop planting structure map, ten-day effective precipitation, ten-day irrigation distribution map, and gross irrigation water diversion, calculate the effective utilization coefficient of irrigation water for each ten-day period and / or the entire study period in the study area.

2. The method for calculating the effective utilization coefficient of irrigation water according to claim 1, characterized in that, Step S4 further includes: S41. Based on the Sentinel-1 images for the study period, generate the VV and VH polarization backscattering coefficients at the decadal scale, and calculate the decadal irrigation response index using the VV and VH polarization backscattering coefficients. ; S42. In the ten-day crop planting structure map, based on the known irrigation information of the pixels, determine irrigated and non-irrigated samples, and extract the samples across all ten-day periods. The optimal segmentation threshold for the i-th ten-day period is calculated by combining OTSU dynamic threshold segmentation. ; S43. For the i-th ten-day period, the study area will satisfy... The pixels were labeled as moisture increase events, and the ten-day cumulative precipitation was extracted for all non-irrigated samples. and the ten-day scale irrigation response index ; S44, with and Using the x and y axes, all non-irrigated samples were used across all ten-day periods. Scatter plots were created using data point pairs; inflection points in the scatter plots were identified using piecewise linear regression, and the corresponding inflection points were assigned to... As the initial ten-day threshold ; S45, will satisfy and Irrigated and non-irrigated samples were labeled as irrigation events, and the misclassification rates for irrigation and non-irrigated samples were calculated separately. S46. When the false positive rate of the irrigation sample is greater than the first proportion, Increase the preset value and return to step S45; when the false positive rate of non-irrigated samples is greater than the second proportion, Decrease the preset value and return to step S45; if the previous two conditions are not met, use... As an effective precipitation event threshold Then proceed to step S47; S47. In each ten-day period, in the cells marked as moisture increase events, the following will be satisfied: The pixels are labeled as irrigation candidate events, and combined with the irrigation period of the crop, the final irrigation events for each ten-day period are determined to form a ten-day irrigation distribution map.

3. The method for calculating the effective utilization coefficient of irrigation water according to claim 2, characterized in that, Step S41 further includes: Sentinel-1 images for the study period were acquired and preprocessed. All Sentinel-1 images passing through each ten-day period were averaged to generate VV and VH polarization backscattering coefficients at the ten-day scale. Calculate the decadal-scale irrigation response index based on the VV and VH polarization backscattering coefficients: in, is the ten-day scale irrigation response index for the i-th ten-day period; and The images show the VV and VH polarization backscattering coefficients for the i-th tenth period, respectively. and The images show the VV and VH polarization backscattering coefficients for the i-1th tenth of the month.

4. The method for calculating the effective utilization coefficient of irrigation water according to claim 1, characterized in that, Step S1 further includes: S11. Acquire Sentinel-2 images of the study area during the study period, and reconstruct the daily 10-meter resolution NDVI of each pixel using a method based on linear interpolation and SG filtering. Calculate the average daily 10-meter resolution NDVI of the j-th pixel in month m as the monthly average. ; S12. Calculate the compensation factor based on the historical vegetation index and evapotranspiration of the study area: in, The compensation factor for the j-th pixel in month m; and The average vegetation index of the j-th and l-th pixels in month m is obtained from historical data. and The average evapotranspiration of the j-th and l-th pixels in month m is obtained from historical data. S13. Use compensation factors to adjust monthly average values. After adjustment, the downscaling driving factor is obtained: in, Let be the downscaling driving factor for the j-th pixel in month m; S14. For month m, the evapotranspiration at 1km resolution is calculated using all 10-meter pixels within the 1km resolution pixel of that month. The average value is taken as the driving factor for evapotranspiration at a resolution of 1 km. ; S15. For any ten-day period within month m, according to and Converting evapotranspiration at 1km resolution to evapotranspiration at 10m resolution: in, The evapotranspiration of the j-th pixel at a resolution of 10 meters is given by the given value. Evapotranspiration at a resolution of 1 km.

5. The method for calculating the effective utilization coefficient of irrigation water according to claim 1, characterized in that, The cultivated land types include paddy fields and dryland crops, and the expression for the effective precipitation model for paddy fields is as follows: in, This represents the effective precipitation for the i-th ten-day period; This represents the rainfall in the i-th ten-day period; This represents the crop water requirement for the i-th ten-day period; The expression for the effective precipitation model for dryland crops is as follows: Among them, SF is the soil moisture storage factor; D is 40%-60% of the effective water holding capacity of the crop root zone soil; When dryland crops At that time, then order .

6. The method for calculating the effective utilization coefficient of irrigation water according to claim 1, characterized in that, Step S2 further includes: Multi-source remote sensing features of each Sentinel-2 image in the study area during the study period were extracted and used to construct the feature vector of each pixel. All feature vectors of each ten-day period of the pixel were stacked to form a multi-dimensional feature vector of each ten-day period of the pixel. The multidimensional feature vector of each pixel is input into the ensemble learning classification models Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), and Extreme Gradient Boosting (XGBoost) to obtain the four class probabilities for each ten-day period. The probabilities of the four categories are concatenated to form a meta-feature, which is then input into a meta-classifier based on a Logistic regression model. This yields the predicted type for each pixel in each ten-day period, forming a ten-day crop planting structure map for the study area.

7. The method for calculating the effective utilization coefficient of irrigation water according to claim 1, characterized in that, Step S5 further includes: Determine the net irrigation water consumption per ten-day period based on the evapotranspiration of each pixel: in, Let be the net irrigation water consumption of the j-th pixel in the i-th ten-day period; The evapotranspiration data for the j-th pixel in the i-th ten-day period; Let be the effective precipitation of the j-th pixel in the i-th ten-day period; This is a function to find the maximum value. Calculate the effective irrigation water utilization coefficient of crop c in the i-th ten-day period based on the net irrigation water consumption per ten-day period. : Where S represents the study area; The area of ​​a pixel; Let c represent the type of crop planted in the j-th pixel; The gross irrigation water intake for crop c in the study area S during the i-th ten-day period; Calculate the effective irrigation water utilization coefficient of crop c in n ten-day periods based on the net irrigation water consumption per ten-day period: When n=3, Let n be the monthly irrigation water effective utilization coefficient for crop c; when n is the total number of ten days in the study period, denoted as the effective irrigation water utilization coefficient for crop c during the study period.

8. The method for calculating the effective utilization coefficient of irrigation water according to claim 1, characterized in that, Step S5 also includes calculating the pixel scale factor and the coefficients of different spatial units: ， in, Let T be the utilization efficiency of irrigation water for the j-th pixel in time window T, where T is a time window consisting of at least one ten-day period. Let J be the net irrigation water consumption of the j-th pixel within time window T. Let be the net irrigation water consumption of the j-th pixel in the i-th ten-day period; To study the utilization efficiency of irrigation water in region S within the time window T; The area of ​​a pixel; The gross irrigation water volume of the study area S during the time window T is calculated.

9. The method for calculating the effective utilization coefficient of irrigation water according to claim 2, characterized in that, The optimal segmentation threshold for the i-th tenth time is calculated using the OTSU dynamic threshold segmentation method. The methods include: For the i-th ten-day period, the pixels corresponding to irrigated samples and non-irrigated samples are compared. Sort the data and calculate the global mean. ; Traverse any two adjacent pairs Take the midpoint t, and combine all Divided into two categories: ≤t and >t; Calculate the two types of weights , and mean , and inter-class variance Select to make The largest t is used as the optimal segmentation threshold for the i-th period. .

10. The method for calculating the effective utilization coefficient of irrigation water according to claim 6, characterized in that, The multi-source remote sensing features include spectral, vegetation index, texture and radar features; the spectral features include Blue, Green, Red, NIR, SWIR1 and SWIR2, and the vegetation indices include normalized difference vegetation index, enhanced vegetation index, surface water index, improved normalized difference water index, soil-regulated vegetation index, red-edge normalized difference vegetation index and red-edge position index. The texture is a statistical measure calculated from the near-infrared band based on the gray-level co-occurrence matrix: mean, variance, homogeneity, contrast, correlation, and entropy; the radar features are the backscattering coefficients of VV and VH polarization and the radar vegetation index extracted from the contemporaneous Sentinel-1 image.

Images