Remote sensing-based agricultural medium and short term water shortage intelligent estimation method

Abstract

The invention belongs to the technical field of agricultural remote sensing and monitoring, and provides an agricultural medium-short-term water shortage intelligent estimation method based on remote sensing, which comprises the steps of data collection and preprocessing, crop spatial distribution and current growth cycle extraction in a research area, medium-short-term meteorological element prediction based on a WRF model, and medium-short-term crop evapotranspiration prediction and medium-short-term agricultural water deficit dynamic intelligent estimation are realized. According to the method, the time-space continuous monitoring advantage of the remote sensing image is fully utilized, and rapid identification and early warning of the water shortage condition of the crops in a large range are achieved; meanwhile, by means of simulation of a numerical weather forecasting model on complex meteorological conditions, the future weather trend is effectively predicted, and a scientific basis is provided for agricultural water resource management.

Description

Technical Field

[0001] The present invention belongs to the technical field of agricultural remote sensing and monitoring, and in particular relates to an intelligent estimation method for agricultural medium- and short-term water shortage based on remote sensing. Background Art

[0002] In agricultural production systems, the rational allocation and efficient use of water resources are key factors in ensuring healthy crop growth and increasing crop yields. However, traditional agricultural water shortage prediction methods rely heavily on manual experience and on-site soil moisture monitoring. Their accuracy and efficiency are no longer sufficient to meet the high demands of modern agricultural precision management.

[0003] An in-depth analysis of existing technologies reveals the following three significant problems:

[0004] (1) Data acquisition has limitations. Traditional agricultural water shortage predictions often rely on observation data from ground stations, such as precipitation, soil moisture, and meteorological conditions. However, the collection of these data is greatly restricted by the density and geographical location of the stations. In vast areas or remote areas with inconvenient transportation, data acquisition is difficult, which seriously affects the comprehensiveness and integrity of the data.

[0005] (2) Traditional agricultural water shortage forecasting is seriously ineffective. Due to the lack of large amounts of real-time data support, traditional forecasting methods are unable to quickly and accurately present the current actual situation of agricultural water use and its dynamic changing trends. This makes it difficult to formulate effective response strategies in the face of sudden water shortages.

[0006] (3) Traditional agricultural water shortage predictions are inaccurate. Traditional agricultural water shortage predictions typically only consider staple crops, failing to fully account for the significant differences in water requirements among different crop types. They also fail to consider the differences in water requirements among different crops during the same growth cycle. These factors combine to create significant deviations between predictions and actual conditions, making it difficult to meet the development needs of precision agriculture.

[0007] Given this, it is imperative to develop an efficient and accurate method for surveying agricultural water shortages using cutting-edge technologies such as remote sensing and precipitation forecasting models. This is of vital practical significance for accurately forecasting large-scale, medium- and short-term agricultural water shortages, improving agricultural drought prevention and response capabilities, and promoting sustainable agricultural development. Summary of the Invention

[0008] To overcome the limitations of existing technologies, this paper proposes a remote sensing-based intelligent method for estimating short- and medium-term agricultural water shortages, combining the powerful information acquisition capabilities of remote sensing technology with the predictive potential of numerical models. This method leverages the spatiotemporal and temporal continuity of remote sensing imagery to rapidly identify and provide early warnings of large-scale crop water shortages. Furthermore, by leveraging the simulation of complex meteorological conditions using numerical weather forecast models, this method effectively predicts future weather trends, providing a scientific basis for agricultural water resource management.

[0009] It should be noted that the present invention is completed based on the following findings of the inventors:

[0010] With the continuous advancement of remote sensing technology and numerical models, their application in crop water demand forecasting and agricultural water resources management is deepening. Remote sensing imagery, with its high temporal and spatial resolution, can provide detailed and timely crop information across the monitored area, accurately capturing subtle changes throughout the crop growth cycle. This information is crucial for accurately estimating agricultural water scarcity. Through in-depth analysis of remote sensing data, we can accurately determine key indicators such as crop type and growth status, providing a solid scientific basis for agricultural drought warning and water resources management. Furthermore, advanced numerical models, such as the Weather Research and Forecasting Model (WRF), simulate complex geophysical processes, providing the essential physical foundation for mesoscale numerical weather forecasting. By coupling multi-scale and multi-physics processes and employing a flexible parallel computing architecture, WRF achieves high-resolution and detailed simulation of physical processes, demonstrating its broad application value in areas such as severe weather warning, climate simulation, and environmental impact assessment.

[0011] Based on this, the inventors used long-sequence, high-resolution satellite remote sensing data to construct a two-layer random forest model to classify crops and identify growing periods; further based on the WRF model, by inputting high-temporal and spatial resolution meteorological driving data, they realized medium- and short-term precipitation forecasts; created and trained an SVR regression model for evapotranspiration forecasts, and combined with the characteristic data of the application area to realize spatial intelligent estimation of medium- and short-term water shortages in agriculture.

[0012] The object of the present invention is achieved like this:

[0013] The present invention provides an intelligent estimation method for medium- and short-term agricultural water shortage based on remote sensing, comprising the following steps:

[0014] Step 1, data collection and preprocessing:

[0015] Determine the scope of the study area and collect long-sequence, high-resolution satellite remote sensing data within the study area, including optical remote sensing data and radar remote sensing data;

[0016] The optical remote sensing data were subjected to geometric precision correction, atmospheric correction and cloud mask processing, and the radar remote sensing data were subjected to geometric precision correction to obtain satellite remote sensing images of the study area.

[0017] Step 2: Extract the spatial distribution and current growth cycle of crops in the study area:

[0018] S21, Crop Area Delineation

[0019] Comparative analysis of satellite remote sensing images of the study area was conducted to identify crop planting areas in the study area by observing surface features. Texture features of the images were further extracted to distinguish the planting locations and ranges of different crop types, and finally the crop planting area layers or vector data were delineated.

[0020] S22, based on a two-layer random forest model, determines the spatial distribution and current growth cycle of crops

[0021] The first layer of the model is used to distinguish crop types in space; the second layer of the model identifies the crop growth period for the identified crop types. The specific construction steps include:

[0022] Extract key vegetation indices and color features from the satellite remote sensing images of the study area obtained in step 1; use the Bootstrap sampling method to extract samples with replacement from the original training set as the training subsets for each decision tree, and randomly select some features from each training subset as candidate features for the decision tree branch nodes; use the feature impurity index to determine the optimal nodes and branches, and establish a decision tree regression model;

[0023] For the first layer model, a feature set including color features, texture features, and shape features is extracted to classify crop types, which is expressed as:

[0024] C=RandomForest(F,L)

[0025] Where C is the crop classification result; F is the feature set extracted from the remote sensing image; L is the label set of the crop type in the training data;

[0026] For the second-layer model, we use crop remote sensing image data from different periods to learn the characteristic changes of different crops in different growth periods for growth period identification, which can be expressed as:

[0027] S:F′→L′

[0028] Where S is the result of crop growth period recognition; F' is the feature set for the growth period; L' is the label set of the crop growth period in the training data.

[0029] Step 3: Medium- to short-term weather forecast based on the WRF model:

[0030] The WRF model is used to predict the short-term and medium-term weather in the study area, and the following meteorological elements are obtained:

[0031]

[0032] Where, For the future Temperature conditions during the time period; For the future Precipitation conditions during the period; For the future Humidity during the period; For the future Wind speed during the period.

[0033] Step 4: Prediction of short-term crop evapotranspiration in the study area based on the SVR model:

[0034] S41, Evapotranspiration prediction based on SVR:

[0035] (1) Generate training set / test set

[0036] Randomly sort the samples and extract the first half as the training set and the second half as the test set;

[0037] (2) Create / train an SVR regression model

[0038] Collect meteorological and hydrological data, crop types, crop growth cycle data, and evapotranspiration data from different historical periods in the study area for model training:

[0039] F[n(x,y,z),t,p,h,w,c,s]→ET

[0040] Where n(x, y, z) is the location of the crop, and its characteristics include longitude (x), latitude (y), and altitude (z). t is temperature, p is precipitation, h is humidity, w is wind speed, c is crop type, s is the crop growth cycle, and ET is crop evapotranspiration, i.e., crop water consumption.

[0041] S42, Forecast of future short- to medium-term evapotranspiration

[0042] The prediction results of the medium- and short-term precipitation in the study area in the future obtained based on the WRF model in step 3 are input into the SVR regression model. The meteorological and hydrological data are combined with the crop spatial distribution and crop growth cycle obtained by the two-layer random forest model to predict the crop growth cycle in the future and estimate future water consumption:

[0043]

[0044] Where, For the future Crop evapotranspiration during the period; For the future Temperature conditions during the time period; For the future Precipitation conditions during the period; For the future Humidity during the period; For the future The wind speed of the time period, c is the crop type; For the future The growth cycle of crops in that period.

[0045] Step 5: Dynamic and intelligent estimation of short- to medium-term agricultural water shortage:

[0046] The WRF model and SVR model constructed in steps 3 and 4 are applied to a large residential area. By inputting the characteristic data of each location, spatial intelligent estimation of short-term and medium-term agricultural water shortage is achieved.

[0047] The formula for calculating the short-term water shortage in regional agriculture is:

[0048]

[0049] Where W is the future The agricultural water shortage during the period, i.e., the amount of water required for irrigation; c is the crop type; α is the unit conversion coefficient; For the future Crop evapotranspiration during the period; For the future The precipitation in the period; β is the crop water use coefficient; A c is the area of ​​each crop.

[0050] The advantages and beneficial effects of the present invention are:

[0051] 1. Efficient and Accurate Monitoring and Early Warning: The estimation method described in this invention utilizes multi-source, high-resolution remote sensing data, fully leveraging the unique advantages of continuous spatiotemporal monitoring of remote sensing imagery. This method overcomes the limitations of traditional data acquisition in terms of spatial scope and temporal frequency, enabling rapid and accurate identification of water shortages across a wide range of crops and timely issuance of early warning signals. Furthermore, the method utilizes a two-layer random forest model to achieve high-precision crop classification and detailed identification of growth periods, improving the accuracy and efficiency of crop identification. Compared to traditional monitoring methods that rely on ground-based stations, this method significantly enhances the comprehensiveness and timeliness of monitoring, buying valuable response time for agricultural producers and effectively reducing the risk of crop yield reductions due to water shortages.

[0052] 2. Scientific and reliable weather forecast assistance: By applying the WRF climate model and combining it with high-resolution driving data, this invention can precisely simulate complex meteorological conditions and achieve accurate predictions of short- and medium-term precipitation and evaporation. Combining these precise weather forecasts with crop water shortages provides a more scientific and comprehensive basis for decision-making in agricultural water resource management. For example, irrigation schedules and water volumes can be rationally planned based on the probability and intensity of future precipitation, avoiding water waste and maximizing water resource utilization efficiency.

[0053] 3. Improving Disaster Response Effectiveness: By enabling accurate prediction and early warning of large-scale crop water shortages, the system significantly enhances the effectiveness of agricultural drought response. Based on these early warnings, agricultural practitioners can formulate targeted drought mitigation measures in advance, such as timely adjusting irrigation plans, adopting water-saving irrigation techniques, and prioritizing water for crops during critical growth periods, thereby effectively mitigating the adverse impact of drought on agricultural production.

[0054] 4. Optimize the spatial allocation of water resources: The research results of this invention can provide scientific and detailed data support for relevant government departments, helping them to optimize the spatial allocation of agricultural water resources in the region. By comprehensively considering the crop planting structure, water shortage conditions and future meteorological conditions in different regions, water resources can be rationally allocated to ensure that water resources are fairly and efficiently distributed among different regions and different crops, thereby improving the overall agricultural water use efficiency. In this process, the SVR regression model is used to process various types of data, combined with high-resolution driving data, to further improve the accuracy of the analysis of water resource-related factors, providing a stronger guarantee for the scientific allocation of water resources.

[0055] 5. Enhancing Agricultural Drought Resistance and Food Security: Overall, the application of this technology is of great significance for improving agricultural resilience to drought disasters, providing advanced and reliable technical means for agricultural drought management. Stable agricultural production is the cornerstone of food security. By ensuring the normal growth of crops in water-scarce conditions, this invention provides solid technical support for food security and effectively maintains the stability and security of the national grain supply. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The present invention will be further described below with reference to the accompanying drawings and examples.

[0057] Figure 1 This is a flow chart of the remote sensing-based intelligent estimation method for medium- and short-term agricultural water shortages according to an embodiment of the present invention. DETAILED DESCRIPTION

[0058] To make the purpose, technical solutions and advantages of the present invention more clear, the following will further describe the disclosed embodiments of the present invention in detail with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0059] Example 1:

[0060] like Figure 1 As shown, this embodiment provides an intelligent estimation method for agricultural medium- and short-term water shortage based on remote sensing, including the following steps:

[0061] Step 1, data collection and preprocessing:

[0062] Determine the scope of the study area and collect long-sequence, high-resolution satellite remote sensing data in the study area, including optical remote sensing data and radar remote sensing data; the optical remote sensing data include Gaofen-1, Gaofen-2, Gaofen-6, Sentinel-2, MODIS and Landsat series remote sensing data; the radar remote sensing data include Gaofen-3 and Sentinel-1 remote sensing data.

[0063] For optical remote sensing data, high-quality cloud-free and snow-free data should be selected, and geometric correction, atmospheric correction, and cloud mask processing should be performed; for radar remote sensing data, geometric correction should be performed to obtain satellite remote sensing images of the study area. The specific methods are as follows:

[0064] (1) Geometric correction

[0065] 1:10000 topographic maps and medium- and high-resolution satellite remote sensing data covering the image area were collected. Based on the visual interpretation method, points with obvious features on the image (road intersections, building highlights, etc.) were selected as control points, and they were distributed as evenly as possible in the space covered by the remote sensing image. A rational function model constrained by control points was used to perform geometric precision correction on the medium- and high-resolution remote sensing data.

[0066] (2) Atmospheric correction

[0067] According to the type of satellite remote sensing data, the corresponding algorithm or tool is selected for atmospheric correction. The FLAASH atmospheric correction model is used to perform atmospheric correction on the Gaofen series and Landsat series satellite remote sensing images to eliminate the influence of factors such as atmosphere and light on the reflection of ground objects.

[0068] (3) Cloud mask processing

[0069] Based on the relevant cloud and cloud shadow coverage information provided by the image quality inspection (QA) band, the areas covered by clouds and their shadows on the remote sensing image are masked to obtain satellite remote sensing images of cloud-free areas.

[0070] Step 2: Extract the spatial distribution and current growth cycle of crops in the study area:

[0071] S21, Crop Area Delineation

[0072] Comparative analysis of satellite remote sensing imagery of the study area was conducted. Surface features were observed to identify crop planting areas within the study area. A gray-level co-occurrence matrix method was then used to extract texture features from the imagery, distinguishing the planting locations and ranges of different crop types. Post-processing of the classification results, such as removing small patches and smoothing boundaries, was performed to improve their accuracy and reliability. Finally, an accuracy assessment was conducted to verify the accuracy and reliability of the delineation results, ultimately delineating the crop planting area as a layer or vector data.

[0073] Among them, the surface features include remote sensing vegetation indices such as the Normalized Difference Vegetation Index (NDVI), the Difference Vegetation Index (DVI), the Ratio Vegetation Index (RVI) and the Enhanced Vegetation Index (EVI), which are used to reflect the growth and coverage of vegetation and identify crop planting areas.

[0074] S22, based on a two-layer random forest model, determines the spatial distribution and current growth cycle of crops

[0075] The two-layer random forest model aims to accurately classify crops and identify their growing periods using remote sensing data. The first layer of the model distinguishes crop types spatially. This requires extracting multi-color features reflecting crop growth characteristics from remote sensing imagery and using a random forest algorithm to learn and classify these features. The second layer identifies the growing period of each identified crop type, which typically involves analyzing the dynamic changes in vegetation indices throughout the crop's growth cycle.

[0076] The specific construction steps include:

[0077] Extract key vegetation indices and color features from the satellite remote sensing images of the study area obtained in step 1; use the Bootstrap sampling method to extract samples with replacement from the original training set as the training subset of each decision tree, and randomly select some features from each training subset as candidate features for the branch nodes of the decision tree; use the feature impurity index to determine the optimal nodes and branches, and establish a decision tree regression model.

[0078] For each sample set and selected candidate features, the feature impurity index (such as mean square error, absolute deviation, etc.) is used to determine the best nodes and branches, and a decision tree regression model is established. For each decision tree T i , its construction process is expressed as:

[0079] T i =BuildTree(B i ,M)

[0080] Where B i is the i-th training subset obtained by self-sampling, and M is the model parameter: including the maximum depth of the tree, the minimum number of node samples, etc.

[0081] Through the above steps, k different decision tree regression models are obtained, each of which is trained based on a different Bootstrap sample and feature subset; the final prediction result P is the average of these k regression prediction results:

[0082]

[0083] Where, P is the final prediction result; P i is the prediction result of the i-th training subset.

[0084] By randomly selecting features and samples during training, Random Forest can comprehensively consider the opinions of multiple decision trees, effectively reducing the risk of overfitting and improving the generalization performance of the overall model and overall prediction accuracy. This ensemble learning approach enables the model to more comprehensively capture the complex relationships between features, achieving accurate estimation of crop types and growth cycles.

[0085] When constructing a random forest, we first use the bootstrap sampling method. Bootstrap sampling is a method of randomly selecting samples from the original training set with replacement, with the same number of samples as the original training set. This step aims to introduce randomness into the samples. By sampling the training samples of each decision tree with replacement, we increase the diversity of the model and avoid over-reliance on specific samples.

[0086] For each bootstrap sample set, m features are randomly selected as candidate features for decision tree branch nodes, where M > m. This step is to introduce randomness into the features, increase model diversity, and reduce model variance by limiting the number of features used in each decision tree.

[0087] These models are trained on the sample sets to obtain their own regression prediction results.

[0088] Specifically, for each node t, a feature f and a threshold θ are selected to divide the sample set S into two subsets S 左 and S 右 This segmentation process is accomplished by minimizing the impurity index. For regression problems, the MSE is calculated as follows:

[0089]

[0090] Where, P pred,t represents the predicted value of the t-th validation sample, P obs,t represents the true value of the t-th validation sample, and z represents the size of the validation sample.

[0091] Through continuous iteration, the decision tree is recursively built until the predetermined tree depth is reached or other stopping conditions are met.

[0092] In the application of remote sensing data, the two-layer random forest model extracts a feature set including color features (such as RGB components, HSV, NDVI, etc.), texture features (such as gray-level co-occurrence matrix GLCM, gray value standard deviation, etc.) and shape features (such as Hu invariant moment, regional area, etc.) for crop type classification in the first layer model, which is expressed as:

[0093] C=RandomForest(F,L)

[0094] Where C is the crop classification result; F is the feature set extracted from the remote sensing image; L is the label set of the crop type in the training data.

[0095] Building on the first-layer model, the second-layer model further refines the growing period of the identified crop types. This layer also uses the random forest algorithm, trained using crop remote sensing imagery data from different periods. During training, the model learns the characteristics of different crops and how they change over time. The second-layer model uses crop remote sensing imagery data from different periods to learn the characteristic changes of different crops during different growing periods for growing period identification, expressed as:

[0096] S:F′→L′

[0097] Where S is the result of crop growth period recognition; F' is the feature set for the growth period; L' is the label set of the crop growth period in the training data.

[0098] Finally, the model application and results were validated: The two-layer random forest model was used for automated crop classification and growth period monitoring, providing important decision support for agricultural management. After the model was deployed, the classification results were compared with ground truth data or high-precision imagery to verify accuracy and ensure the model's reliability and effectiveness. Ultimately, the crop classification maps and growth period information generated by the model can be used to assess crop growth status, predict water requirements, and guide agricultural practices.

[0099] Step 3: Medium- to short-term weather forecast based on the WRF model:

[0100] The WRF model (Weather Research and Forecasting Model) is used to predict the medium- and short-term precipitation in the study area. The WRF model is used to predict the medium- and short-term weather in the study area to obtain the following future meteorological elements:

[0101]

[0102] Where, For the future Temperature conditions during the time period; For the future Precipitation conditions during the period; For the future Humidity during the period; For the future Wind speed during the period.

[0103] Step 4: Prediction of short-term crop evapotranspiration in the study area based on the SVR model:

[0104] S41, evapotranspiration prediction based on SVR (support vector machine, referred to as SVM):

[0105] (1) Generate training set / test set

[0106] The generated training and test sets should be interchangeable. This means randomly sorting the samples and extracting the first half as the training set and the second half as the test set. To do this, first sort the original data from 1 to x, then set a random sequence. Each run randomly sorts the data from 1 to x to form a new sequence.

[0107] (2) Create / train an SVR regression model

[0108] The key to this step is to ensure the diversity and adequacy of samples to reflect the water consumption under different climate conditions and crop growth periods. Meteorological and hydrological data, crop types, crop growth cycle data, and evapotranspiration data from different historical periods in the study area are collected for model training:

[0109] F[n(x,y,z),t,p,h,w,c,s]→ET

[0110] Where n(x, y, z) is the location of the crop, and its characteristics include longitude (x), latitude (y), and altitude (z). t is temperature, p is precipitation, h is humidity, w is wind speed, c is crop type, s is the crop growth cycle, and ET is crop evapotranspiration, i.e., crop water consumption.

[0111] During the design process, the effects of normalization, the type of kernel function, and the parameter values ​​on the regression model must be comprehensively considered. The selection of the kernel function is a crucial step in establishing an SVR model. While new kernel functions are constantly emerging, the four most widely accepted basic kernel functions are the linear kernel function, the d-order polynomial kernel function, the radial basis kernel function, and the sigmoid kernel function. Because the radial basis kernel function (RBF) involves fewer parameters and is computationally simple, this embodiment uses the radial basis kernel function. The model parameters are determined using a cross-validation method to find the optimal parameters, which are then used to train the model.

[0112] (3) Model parameter optimization analysis

[0113] After the model is established, the relevant parameters are input. The first parameter returned is the predicted value mapped according to the multidimensional space, and the second parameter is the mean square error (MSE) and the coefficient of determination R of the test set. 2 The smaller the MSE value, the better the accuracy of the simulated numerical value of the constructed prediction model. 2 The closer the value is to 1, the higher the reference value of the related equation; conversely, the closer the value is to 0, the lower the reference value. The calculation formulas for both are as follows:

[0114]

[0115] Where: y i ——The true value of the i-th sample; ——The predicted value of the i-th sample; l is the total number of samples.

[0116] Use the mean square error MSE and determination coefficient R returned by the model 2 , the performance of the established SVR regression model can be evaluated. If the mean square error MSE is too large or the determination coefficient R 2 If it is too small, the model parameters can be modified or the kernel function type can be reselected until the requirements are met.

[0117] S42, Forecast of future short- to medium-term evapotranspiration

[0118] The prediction results of the medium- and short-term precipitation in the study area in the future obtained based on the WRF model in step 3 are input into the SVR regression model. The meteorological and hydrological data are combined with the crop spatial distribution and crop growth cycle obtained by the two-layer random forest model to predict the crop growth cycle in the future and estimate future water consumption:

[0119]

[0120] Where, For the future Crop evapotranspiration during the period; For the future Temperature conditions during the time period; For the future Precipitation conditions during the period; For the future The humidity of the time period; c is the crop type; For the future The growth cycle of crops in that period.

[0121] Step 5: Dynamic and intelligent estimation of short- to medium-term agricultural water shortage:

[0122] The WRF and SVR models constructed in steps 3 and 4 are applied to a large residential area. By inputting characteristic data from each location, spatially intelligent estimation of short- and medium-term agricultural water shortages is achieved. This process makes the estimation results more interpretable by comprehensively considering multidimensional characteristics such as geographic location, crop type, crop growth period, and future weather conditions.

[0123] The formula for calculating the short-term water shortage in regional agriculture is:

[0124]

[0125] Where W is the future The agricultural water shortage during the period, that is, the amount of water required for irrigation; α is the unit conversion coefficient; β is the crop water utilization coefficient; A c is the area of ​​each crop.

[0126] To ensure the reliability of the model, accuracy verification is necessary. This verification process can include comparisons with actual observational data and comprehensive evaluation of model performance using methods such as cross-validation. This series of verification steps ensures the model's generalization performance across different datasets, providing reliable scientific support for its practical application. The construction and verification process of this intelligent estimation model provides a reasonable and practical solution for predicting short-term and medium-term water shortages in large-scale agriculture.

[0127] Finally, it should be noted that the above is only used to illustrate the technical solution of the present invention and is not limiting. Although the present invention is described in detail with reference to the preferred arrangement scheme, ordinary technicians in this field should understand that the technical solution of the present invention (such as the use of various formulas, the sequence of steps, etc.) can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A remote sensing-based intelligent estimation method for agricultural medium- and short-term water shortages, characterized by: The method comprises the following steps: Step 1, data collection and preprocessing: Determine the scope of the study area and collect long-sequence, high-resolution satellite remote sensing data within the study area, including optical remote sensing data and radar remote sensing data; Perform geometric precision correction, atmospheric correction and cloud mask processing on optical remote sensing data, and geometric precision correction on radar remote sensing data to obtain satellite remote sensing images of the study area; Step 2: Extract the spatial distribution and current growth cycle of crops in the study area: S21, Crop Area Delineation Comparative analysis of satellite remote sensing images of the study area was conducted to identify crop planting areas in the study area by observing surface features. Texture features of the images were further extracted to distinguish the planting locations and ranges of different crop types, and finally the crop planting area layers or vector data were delineated. S22, based on a two-layer random forest model, determines the spatial distribution and current growth cycle of crops The first layer of the model is used to distinguish crop types in space; the second layer of the model identifies the crop growth period for the identified crop types. The specific construction steps include: Extract key vegetation indices and color features from the satellite remote sensing images of the study area obtained in step 1; use the Bootstrap sampling method to extract samples with replacement from the original training set as the training subsets for each decision tree, and randomly select some features from each training subset as candidate features for the decision tree branch nodes; use the feature impurity index to determine the optimal nodes and branches, and establish a decision tree regression model; For the first layer model, a feature set including color features, texture features, and shape features is extracted to classify crop types, which is expressed as: C=RandomForest(F,L) Where C is the crop classification result; F is the feature set extracted from the remote sensing image; L is the label set of the crop type in the training data; For the second-layer model, we use crop remote sensing image data from different periods to learn the characteristic changes of different crops in different growth periods for growth period identification, which can be expressed as: S:F'→L' Where, S is the result of crop growth period identification; F ’ is a feature set for the reproductive period; L ’ is the label set of crop growth periods in the training data; Step 3: Medium- to short-term weather forecast based on the WRF model: The WRF model is used to predict the short-term and medium-term weather in the study area, and the following meteorological elements are obtained: Where, For the future Temperature conditions during the time period; For the future Precipitation conditions during the period; For the future Humidity during the period; For the future Wind speed during the time period; Step 4: Prediction of short-term crop evapotranspiration in the study area based on the SVR model: S41, Evapotranspiration prediction based on SVR: (1) Generate training set / test set Randomly sort the samples and extract the first half as the training set and the second half as the test set; (2) Create / train an SVR regression model Collect meteorological and hydrological data, crop types, crop growth cycle data, and evapotranspiration data from different historical periods in the study area for model training: F[n(x,y,z),t,p,h,w,c,s]→ET Where n(x, y, z) is the location of the crop, and its characteristics include longitude (x), latitude (y), and altitude (z). t is temperature, p is precipitation, h is humidity, w is wind speed, c is crop type, s is the crop growth cycle, and ET is crop evapotranspiration, i.e., crop water consumption. S42, Forecast of future short- to medium-term evapotranspiration The prediction results of short-term precipitation in the study area in the future obtained by the WRF model in step 3 are input into the SVR regression model. The meteorological and hydrological data are combined with the crop spatial distribution and crop growth cycle obtained by the two-layer random forest model to predict the crop growth cycle in the future and estimate future water consumption: Where, For the future Crop evapotranspiration during the period; For the future Temperature conditions during the time period; For the future Precipitation conditions during the period; For the future Humidity during the period; For the future Wind speed during the time period; For the future The growth cycle of the crops in that period; Step 5: Dynamic and intelligent estimation of short- to medium-term agricultural water shortage: The WRF model and SVR model constructed in steps 3 and 4 are applied to a large residential area. By inputting the characteristic data of each location, spatial intelligent estimation of short-term and medium-term agricultural water shortage is achieved. The formula for calculating the short-term water shortage in regional agriculture is: Where W is the future The agricultural water shortage during the period, i.e., the amount of water required for irrigation; c is the crop type; α is the unit conversion coefficient; For the future Crop evapotranspiration during the period; For the future The precipitation in the period; β is the crop water use coefficient; A c is the area of ​​each crop.

2. The remote sensing-based intelligent estimation method for agricultural short-term and medium-term water shortage according to claim 1 is characterized in that: In step 1, the optical remote sensing data includes remote sensing data from Gaofen-1, Gaofen-2, Gaofen-6, Sentinel-2, MODIS and Landsat series; the radar remote sensing data includes remote sensing data from Gaofen-3 and Sentinel-1.

3. The remote sensing-based intelligent estimation method for agricultural medium- and short-term water shortage according to claim 1 is characterized in that: In step 2, a decision tree regression model is established. For each decision tree T i , its construction process is expressed as: T i =BuildTree(B i ,M) Where B i is the i-th training subset obtained by bootstrapping, and M is the model parameter: Through the above steps, k different decision tree regression models are obtained, each of which is trained based on a different Bootstrap sample and feature subset; the final prediction result P is the average of these k regression prediction results: Where, P is the final prediction result; P i is the prediction result of the i-th training subset.

4. The remote sensing-based intelligent estimation method for agricultural short-term and medium-term water shortage according to claim 1 is characterized in that: In step 2, the color features include RGB components, color model HSV, and normalized vegetation index NDVI, and the texture features include gray level co-occurrence matrix GLCM and gray value standard deviation; Shape features include Hu invariant moments and area.

5. The remote sensing-based intelligent estimation method for agricultural short-term and medium-term water shortage according to claim 1 is characterized in that: In step 2, the surface characteristics include remote sensing vegetation indices including the Normalized Difference Vegetation Index (NDVI), the Difference Vegetation Index (DVI), the Ratio Vegetation Index (RVI), and the Enhanced Vegetation Index (EVI).

Images