A large-scale soil moisture sensor optimization layout method

Abstract

The application provides an optimization layout method of a large-scale soil moisture sensor, comprising the following steps: cutting and extracting remote sensing image data obtained according to the boundary of a to-be-monitored irrigation area and a preset typical period to obtain multi-year soil moisture time series data; constructing a probabilistic semi-variogram function model based on the time series data and a resampling and median aggregation method; generating regular grid points covering the to-be-monitored irrigation area as a candidate point set, randomly selecting N candidate points to form an initial layout scheme of the sensor, constructing an objective function based on the probabilistic semi-variogram function model, and obtaining an optimized layout scheme of the sensor based on a random exchange position optimization method. On this basis, the application further comprises determining an optimal number of sensors based on a prediction uncertainty-sensor number response curve analysis. The application controls the calculation complexity within an acceptable range while ensuring the global search capability, and the designed sensor layout scheme is more basic and long-term applicable.

Description

Technical Field

[0001] This invention relates to the field of soil moisture detection technology, specifically to a method for optimizing the layout of large-scale soil moisture sensors. Background Technology

[0002] Against the backdrop of increasing global water scarcity and intensifying climate change, precision agriculture, by optimizing water and fertilizer management to improve resource utilization efficiency, has become a key strategy for ensuring food security and sustainable agricultural development. Precision irrigation, as its core method, relies on accurate understanding of the spatiotemporal dynamics of soil moisture in the field. However, soil moisture inherently possesses high spatial heterogeneity and temporal variability. How to deploy a ground-based sensor network at a limited cost, enabling it to effectively capture regionally representative information and support irrigation decisions, remains a long-standing scientific and engineering challenge.

[0003] The spatial variability of soil moisture is not constant but strongly dependent on its own humidity state. This physical understanding is a crucial foundation for the scientific design of monitoring networks. Classical studies have shown that the physical processes governing the spatial distribution of soil moisture differ drastically under different moisture states. Under extremely wet conditions (such as after irrigation), soil moisture distribution is mainly controlled by topographically driven gravity drainage and runoff processes; while under extremely dry conditions, strong evapotranspiration leads to patchy distribution, and the spatial structure is disturbed by short-term meteorological and biological processes. The spatial patterns obtained under these two conditions cannot stably reflect the essential spatial heterogeneity determined by the inherent properties of soil texture and structure. Conversely, when soil moisture is at a moderate humidity level between field capacity and wilting point, gravity drainage has essentially ended and strong evapotranspiration has not yet become dominant. The spatial distribution of moisture is mainly controlled by the hydraulic properties of the soil itself, and its semivariogram best represents the persistent spatial dependence structure of the region, unaffected by transient strong disturbances. Therefore, selecting soil moisture data that can represent a "moderate humidity" state to construct a semivariogram model is a key prerequisite for ensuring that the subsequent sensor network design results are robust, reliable, and reflect the regional background heterogeneity.

[0004] Scholars have developed various methods for optimizing sensor deployment. Early studies focused on using prior information such as soil properties and topography to guide sensor placement. For example, in the SMEX02 experiment, Jacobs et al., based on time stability analysis, found that a few stable points selected in a specific field could effectively represent the regional average soil moisture, and pointed out that micro-topographic slope and soil clay content were key indicators for identifying stable points. Such research laid an important foundation for fixed-point observation based on physical laws. With the improvement of data acquisition capabilities, data-driven methods have received increasing attention. Mandal et al. proposed a framework based on spatial correlation of surface soil moisture (SASM), which, by integrating global spatial correlation measures and spatial randomness tests, selected a subset of sensors from densely sampled points that best represented spatial variability. In their study in a blueberry orchard, Dong et al. used a strategy combining random combination methods with time stability analysis to quantify the impact of different numbers of sensors on the accuracy of estimating the average soil moisture of a field, finding that 5-10 sensors could achieve a good balance between cost and accuracy. These advances have significantly improved the scientific rigor of field-scale sensor deployment.

[0005] Existing methods face significant challenges when the research scale expands from single field areas (hectares) to large irrigation districts (thousands of square kilometers). Firstly, there is a lack of data. Most optimization methods rely on high-density ground-based data to construct semivariograms, which is extremely costly and difficult to implement at the vast irrigation district scale. Although remote sensing provides extensive coverage, research on directly using remote sensing data, particularly through probabilistic modeling to quantify spatial structure uncertainty and guide network design, remains largely unexplored. Summary of the Invention

[0006] The present invention aims to at least partially solve one of the technical problems existing in the prior art.

[0007] Therefore, the purpose of this invention is to provide a method for optimizing the layout of large-scale soil moisture sensors. This method can design a layout scheme for a soil moisture sensor monitoring network with optimal cost-effectiveness. It can determine the optimal location of sensors given a certain number of sensors, and also solve a complete cost-benefit trade-off optimization problem: in the absence of limitations, it seeks a number of sensors (N) and their corresponding spatial layout (S) such that the network can achieve the greatest possible reduction in the spatial prediction uncertainty of soil moisture in the entire irrigation area with the lowest possible monitoring cost (which can be considered positively correlated with N).

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] The first aspect of this invention provides a method for optimizing the layout of large-scale soil moisture sensors, comprising:

[0010] Step S1: Based on the boundary of the irrigation area to be monitored and the preset typical time period, the acquired remote sensing image data is cropped and extracted to obtain multi-year soil moisture time series data.

[0011] Step S2: Construct a probabilistic semi-variogram model for the multi-year soil moisture time-series data based on resampling and median aggregation methods;

[0012] Step S3: Generate a set of regular grid points covering the irrigation area to be monitored as candidate points, randomly select N candidate points to form the initial layout scheme of the sensor, construct the objective function based on the probabilistic semi-variogram model, and obtain the optimized layout scheme of the sensor based on the random exchange position optimization method.

[0013] In some embodiments, the preset typical time period refers to the period when soil moisture is in a state that can reveal the intrinsic spatial structure of the soil; the state of the intrinsic spatial structure of the soil refers to the moderate humidity state of soil moisture between field capacity and crop wilting point.

[0014] In some embodiments, step S2 includes:

[0015] A resampling method was used to perform multiple samplings with replacement on the multi-year soil moisture time-series data to generate multiple resampling datasets;

[0016] For each resampled dataset, the median field of its soil moisture spatial distribution is calculated, and the semi-variogram is fitted based on the median field to obtain a set of variogram parameters.

[0017] Based on the set of mutation function parameters obtained from multiple resampling, the probability distribution of the key parameters of the semi-mutation function is generated, and the probabilistic semi-mutation function model is constructed using their mean.

[0018] In some embodiments, the resampling method is the Bootstrap method.

[0019] In some embodiments, the probabilistic semivariogram model is obtained by fitting an exponential model or a spherical model.

[0020] In some embodiments, step S3 includes:

[0021] Within the boundary of the irrigation area to be monitored, generate a first grid point covering the entire area as a candidate point set and a second grid point as a prediction point set. Set the number of sensors to be deployed to N. Randomly initialize a spatial layout containing N sensors from the candidate point set. Construct an objective function based on the probabilistic semi-variogram model and the global average Kriging variance calculated from the prediction point set. Calculate the objective function corresponding to the initial spatial layout.

[0022] The process involves iterating through the following steps: randomly removing one sensor deployment point from the current spatial layout and randomly selecting a new subsequent point from the candidate point set that is not in the current spatial layout to add to the new spatial layout; calculating the global average kriging variance corresponding to the new layout; if it is less than the global average kriging variance of the current spatial layout, the new spatial layout is accepted; otherwise, the current spatial layout is retained.

[0023] Repeat the iteration until the termination condition is met, and output the spatial layout with the minimum global average kriging variance during the iteration process as the optimal spatial layout for that value of N.

[0024] In some embodiments, let the first In the next iteration, the calculated global average kriging variance of the current spatial layout is: :

[0025]

[0026]

[0027] in, Indicates the first The current spatial layout of the N sensors at the nth iteration, where the i-th sensor is located within the... The two-dimensional spatial position of each sensor is ; This represents the two-dimensional spatial location of a prediction point within the irrigation area to be monitored. The area of ​​the irrigation district to be monitored; Current spatial layout of the sensors The Kriging variance at a given point; for The Middle Two-dimensional spatial position of each sensor Two-dimensional spatial position of a certain prediction point The semivariance of soil moisture between the values ​​is calculated using the probabilistic semivariogram model. for The Middle The weight of each sensor, according to The Middle The distance between each sensor and the prediction point is configured to a different value; It is a Lagrange multiplier.

[0028] In some embodiments, the optimized layout method further includes:

[0029] Step S4: Scan the number of sensors within the preset range. For each number of sensors, obtain the optimal objective function that can be achieved according to step S3. Construct a response curve of the optimal objective function as a function of the number of sensors N, which serves as the prediction uncertainty-sensor number response curve. Determine the optimal number of sensors based on the principle of diminishing marginal utility.

[0030] A second aspect of the present invention provides a large-scale soil moisture sensor optimization layout device, comprising:

[0031] The data extraction module is configured to crop and extract the acquired remote sensing image data according to the boundary of the irrigation area to be monitored and the preset typical time period to obtain multi-year soil moisture time series data.

[0032] The model building module is configured to construct a probabilistic semi-variogram model from the multi-year soil moisture time-series data based on resampling and median aggregation methods;

[0033] The iterative optimization module is configured to generate a set of candidate points covering the irrigation area to be monitored, randomly select N candidate points to form the initial layout scheme of the sensor, construct the objective function based on the probabilistic semi-variogram model, and obtain the optimized layout scheme of the sensor based on the random exchange position optimization method.

[0034] In some embodiments, the optimized layout device further includes:

[0035] The quantity optimization module is configured to scan the number of sensors within a preset range. For each number of sensors, the iterative optimization module is called to obtain the optimal objective function that can be achieved. The response curve of the optimal objective function as a function of the number of sensors N is constructed as the prediction uncertainty-sensor number response curve. The optimal number of sensors is determined based on the principle of diminishing marginal utility.

[0036] Compared with the prior art, the present invention has the following characteristics and beneficial effects:

[0037] This invention addresses the challenge of quantifying the cost-benefit trade-off in the deployment of ground-based soil moisture monitoring networks in large irrigation districts. It proposes and validates a comprehensive framework integrating remote sensing spatial information and geostatistical optimization. This framework first utilizes long-term, full-coverage remote sensing soil moisture data to construct a robust probabilistic semivariogram model that characterizes parameter uncertainty through resampling, laying a reliable data foundation for optimization. Based on this, this invention develops a sensor network optimization algorithm that aims to minimize the mean kriging variance (MKV) and combines random replacement iteration with system number scanning. The core output of this algorithm is a precisely quantified "prediction uncertainty - number of sensors" response curve (MKV(N) = 33.6645 - 3.7828 · ln(N)). For the first time, through marginal benefit analysis, it determines the cost-effective monitoring network size (2221 sensors) for the Weishan irrigation district. Robustness testing shows that this optimized layout is highly robust to the uncertainty of key parameters in the semivariogram model, confirming the strong stability of this invention.

[0038] The fundamental logic of this invention lies in synergy rather than replacement, aiming to solve a key problem in smart agriculture observation: how to deploy a ground sensor network at the optimal cost to build an efficient collaborative system, given the wide-area, continuous surface monitoring provided by remote sensing. This invention demonstrates that geostatistical optimization can precisely design a ground network that serves both as a key anchor point for validating and calibrating remote sensing products and directly captures micro-scale field variation information that is difficult to obtain through remote sensing but crucial for irrigation decisions. This invention provides a quantitative and transparent engineering design tool to bridge the gap between macro-level remote sensing monitoring and ground-based truth verification.

[0039] The core advantage of this invention lies in its clear universality and transferability. Its input does not rely on complex prior ground surveys, but only on remote sensing soil moisture data (used to reveal spatial variability structures) and geographic boundary vectors of the target area. Therefore, this invention can be readily applied to irrigation and drainage systems in different climatic and agroecological zones worldwide. Simultaneously, it provides a universal, quantitative decision support tool for optimizing monitoring networks for soil salinity, temperature, or other spatially correlated environmental variables, contributing to the construction of an efficient and collaborative integrated "sky-ground" agricultural hydrological observation system.

[0040] In summary, this study proposes a transferable, robust, and physically sound method for optimizing soil moisture sensor networks in irrigation areas. Through a clear cost-benefit curve, it provides crucial scientific evidence for precise water resource management and lays an important methodological foundation for developing smarter and more efficient multi-scale agricultural hydrological monitoring systems. Its core framework has broad application potential in addressing the challenges of sustainable agricultural water resource management globally. Attached Figure Description

[0041] Figure 1 This is an overall flowchart of a method for deploying large-scale soil moisture sensors provided in the first aspect embodiment of the present invention;

[0042] Figure 2 yes Figure 1 The flowchart shown illustrates the optimization of sensor deployment schemes within an irrigation area under a given number of sensors.

[0043] Figure 3 yes Figure 1 The flowchart shown illustrates the process of determining the optimal number of sensors in the deployment method.

[0044] Figure 4 This is a histogram of the frequency distribution of the Kriging standard deviation obtained from an embodiment of the present invention;

[0045] Figure 5 This is a spatial distribution map of the Kriging standard deviation obtained from an embodiment of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this application clearer, the application will be described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for explaining this application and are not intended to limit this application.

[0047] Conversely, this application covers any alternatives, modifications, equivalent methods, and solutions made within the spirit and scope of this application as defined by the claims. Furthermore, to provide the public with a better understanding of this application, certain specific details are described in detail below. However, those skilled in the art will fully understand this application even without these detailed descriptions.

[0048] See Figure 1 The first aspect of the present invention provides a method for deploying large-scale soil moisture sensors, comprising the following steps:

[0049] Step S1: Based on the boundary of the irrigation area to be monitored and the preset typical time period, the acquired remote sensing image data is cropped and extracted to obtain multi-year soil moisture time series data.

[0050] Step S2: Construct a probabilistic semivariogram model based on the soil moisture time series data obtained in Step S1 using the resampling and median aggregation methods;

[0051] Step S3: Generate a set of regular grid points covering the irrigation area to be monitored as candidate points, and randomly select N candidate points to form the initial layout scheme of the sensor. Construct an objective function based on a probabilistic semi-variogram model, and obtain the optimized layout scheme of the sensor based on a random exchange position optimization method.

[0052] In some embodiments, the area to be monitored by the present invention is at the irrigation district scale, such as the Weishan Irrigation District, whose total area exceeds 5000 km². 2 .

[0053] In one specific embodiment of this application, in step S1, the Weishan Irrigation District is selected as the monitoring area, specifically located in Liaocheng City, Shandong Province, China. According to publicly available monitoring data, from 2005 to 2020, the average annual precipitation in this irrigation area was 540.9 mm, the average annual temperature was 13.4℃, and the average annual relative humidity was 61.4%. The climate pattern of this irrigation area is arid and semi-humid, and the crop planting system is a rotation of winter wheat and summer maize. Winter wheat in this irrigation area is generally sown in October and harvested in June of the following year. Irrigation is carried out during the dry winter and spring seasons depending on soil moisture conditions, generally 2-3 times throughout the growing season, with a total irrigation volume of 200-300 mm. According to the WRD (World Reference Database) soil classification standard, the soil type in this area is silty loam, containing approximately 58% silt, 32% sand, and 10% clay, with an average saturated soil moisture content of 0.43.

[0054] Further, in step S1, the remote sensing data acquired in this embodiment of the invention is the China 1km resolution daily all-weather surface soil moisture dataset (2003-2024). This soil moisture dataset, released by the National Tibetan Plateau Scientific Data Center, features long-term, high-resolution, daily, and all-weather characteristics, effectively capturing the spatiotemporal changes in soil moisture. Based on 36km resolution surface soil moisture retrieved from AMSR-E and AMSR-2 passive microwave inversion, this dataset is downscaled by fusing MODIS optical reflectance and thermal infrared surface temperature data to generate 1km resolution data, achieving "all-weather" coverage. From April to September each year, this soil moisture dataset provides near-complete coverage across the country; the coverage in other months is also significantly improved compared to the original passive microwave observations. The unbiased root mean square error of this soil moisture dataset ranges from 0.053 to 0.056 cm. 3 / cm 3 The data surpasses the widely used SMAP Sentinel active-passive microwave fusion soil moisture dataset at 1 km resolution. Daily all-weather surface soil moisture data for China at 1 km resolution is stored in HDF5 format, with file names in the format "SM_yyyyddd.h5", where "yyyy" represents the year and "ddd" represents the Julian day. For example, "SM_2005001.h5" represents the data for day 1 of 2005 (approximately 01:30 AM). The data is stored as Int16 integers to save space; when using it, it should be multiplied by a scaling factor of 0.001 to convert it to soil volumetric water content units (cm³).3 / cm 3 For example, when the storage value is 327, the corresponding soil moisture is 0.327 cm³. 3 / cm 3 .

[0055] Subsequently, this embodiment of the invention utilizes geographic information system tools and, based on preset typical time periods, spatiotemporally clips the national 1km resolution daily all-weather surface soil moisture dataset (2003–2024) according to the Weishan irrigation district boundary, extracting only soil moisture data for typical time periods within the irrigation district to be monitored over many years. This forms multi-year continuous soil moisture time-series data for subsequent sensor deployment optimization modeling and analysis. The preset typical time period refers to the period corresponding to the soil moisture being in a state that reveals the intrinsic spatial structure of the soil. The state of the intrinsic spatial structure of the soil revealed by the typical time period refers to a moderate humidity state where soil moisture is between field capacity and crop wilting point. Under this state, the spatial variation of soil moisture is mainly controlled by the soil's texture, structure, and hydraulic properties, and is largely unaffected by instantaneous strong interference processes such as irrigation, heavy rainfall, or strong evapotranspiration. Preferably, the preset typical time period is the period before the peak water demand period (large-scale irrigation period) of the main crops in the irrigation district to be monitored. For example, for winter wheat planting areas, the preset typical time period is from February to April each year (before spring irrigation).

[0056] In this embodiment, daily soil moisture remote sensing images from February to April each year from 2020 to 2024 are selected. During this period, soil moisture is mainly controlled by climate factors and soil characteristics, with minimal human irrigation interference. Its spatial pattern is relatively stable and suitable as a design benchmark for a static monitoring network. All remote sensing images are uniformly cropped to the boundary of the Weishan Irrigation District to form a data cube containing hundreds of valid daily values.

[0057] In some embodiments, step S2 includes the following steps:

[0058] Step S21: Using the Bootstrap resampling method, the soil moisture time series data obtained in step S1 is sampled multiple times with replacement to generate multiple resampled datasets.

[0059] Specifically: To quantify the uncertainty of key parameters in the probabilistic semi-variogram model, a non-parametric Bootstrap method is adopted. Assume a total of B = 1000 samplings with replacement are performed. In each sampling, an equal number of sample days are randomly selected with replacement from all valid days obtained in step S1 to form a Bootstrap resampling dataset.

[0060] It should be noted that other resampling methods (such as Jackknife, Subsampling, etc.) can also be used to evaluate parameter uncertainty, but the Bootstrap method has comprehensive advantages in terms of computational efficiency, flexibility in confidence interval construction, and applicability to complex estimators, and is therefore the preferred embodiment of the present invention.

[0061] Step S22: For each resampled dataset, calculate the median field of its soil moisture spatial distribution, and fit the semivariogram based on the median field to obtain a set of semivariogram parameters. That is, each resampled dataset corresponds to a set of semivariogram parameters.

[0062] Specifically, for the effective pixels of a remote sensing image on a given day, they are grouped into a series of point pairs. The distance and semivariogram of each pair are calculated. Then, distance intervals are set (e.g., intervals of 1 km each). All point pairs with distances falling within 0-1 km are selected, and their average value is calculated to obtain the first semivariogram value. Then, the average value of all point pairs with distances falling within 1-2 km is calculated to obtain the second semivariogram value, and so on, calculating the semivariogram values ​​corresponding to each distance space (2-3 km, 3-4 km, 4-5 km, ...). Finally, for M remote sensing images in a resampled dataset, M semivariogram values ​​at 1 km are obtained. The median is taken, and so on, to obtain the median semivariogram values ​​for other intervals. Finally, a fitting is performed based on the semivariogram values ​​obtained from this set of distances.

[0063] It should be noted that this embodiment does not perform time averaging on the pixel values ​​of soil moisture time series data. Instead, it extracts the median of each resampled dataset and independently calculates its spatial empirical semivariogram, and then forms a Bootstrap median empirical variogram curve. The median is chosen instead of the mean in order to obtain a more representative and typical spatial structure that is not sensitive to extreme or abnormal conditions.

[0064] Step S23: Based on the set of semivariogram parameters corresponding to all resampled datasets, generate the probability distribution of key semivariogram parameters (including sill value, range and nugget value, sill value = nugget value + structural variance), and use their mean as a robust estimate of the constructed probabilistic semivariogram model, while obtaining the confidence intervals of key parameters (mainly sill value and range).

[0065] In a specific embodiment of the present invention, for the large-scale sensor layout optimization of irrigation areas, in order to capture the spatial variability of soil moisture in the irrigation areas as much as possible, an exponential model with a relatively small range is selected as the fitting model, and the constructed probabilistic semivariogram model is as follows:

[0066]

[0067] in, Let be the semivariance of soil moisture between the two points. The distance between two points; The nugget value, i.e., the value of the probabilistic semivariogram model at a distance of 0, reflects random variation smaller than the resampling scale, originating from the microscopic heterogeneity of the variable itself or measurement errors. This embodiment, based on the data characteristic that soil moisture at a spatial location should have only one value, will... and Set all to 0; The structural variance is equal to the difference between the sill value and the nugget value. For variable range, the characteristic scale representing spatial autocorrelation, when When the range is exceeded, .

[0068] This embodiment obtained the parameter set through B-order fitting. This constitutes an empirical joint distribution of the model parameters, thereby realizing the transformation from point estimation to probabilistic estimation. The first The sill value and range obtained during the second fitting, or the structural variance and range.

[0069] It should be noted that, in addition to the exponential model mentioned above, other models can also be used when constructing the probabilistic semivariogram model, based on the actual soil moisture data of different irrigation districts and different time periods. For example, a spherical model can be used for fitting, as shown below:

[0070]

[0071] Understandably, this embodiment utilizes multi-year time-series remote sensing data, employing Bootstrap resampling and median aggregation to construct a probabilistic semi-variogram model. This model not only provides robust estimates of key parameters (such as range and sill value) but also quantifies their confidence intervals. This method overcomes the problems of overfitting or insufficient representativeness that may arise from single-period data, improving the model's generalization ability and reliability under interannual variability and climate fluctuations. By selecting a typical time period (such as before spring irrigation), it ensures that soil moisture is at a moderate humidity level. At this time, spatial variability is mainly controlled by the inherent properties of the soil itself, such as texture and structure, rather than instantaneous disturbances such as irrigation or precipitation. This makes the constructed semi-variogram function more reflective of the intrinsic spatial structure of the soil, making the designed sensor network more fundamental and applicable in the long term.

[0072] In some embodiments, step S3 is intended to provide a given number of soil moisture sensors. Under the premise of setting a target function and using a random exchange iterative optimization method, the irrigation area to be monitored is determined. Optimal layout of soil moisture sensors. See [link / reference] Figure 2 Step S3 includes the following steps:

[0073] Step S31, Point Set Generation and Initialization: Within the boundary of the irrigation area to be monitored, generate a first regular grid set covering the entire area as a candidate point set and a second regular grid set as a prediction point set. The number of prediction points in the prediction point set should be greater than or equal to the number of candidate points in the candidate point set (i.e., the density of the second grid points should be greater than or equal to the density of the first grid points). The predicted point locations may or may not coincide with the candidate point locations. The global average Kriging variance (MKV) is used as the objective function. The smaller the MKV, the lower the overall uncertainty of the spatial prediction of the entire area by the sensor's spatial layout scheme. Let the iteration number variable be... and maximum number of iterations For a given number of soil moisture sensors Randomly select from the candidate point set The candidate points constitute the initial spatial layout of the soil moisture sensor. And calculate the initial layout. corresponding The calculation formula is as follows:

[0074]

[0075]

[0076] in, This represents the initial spatial layout of N soil moisture sensors, where the first... The two-dimensional spatial position of each sensor is ; This represents the two-dimensional spatial location of a prediction point within the irrigation area to be monitored. The area of ​​the irrigation area to be monitored can be obtained by dividing the irrigation area into grids and calculating the area, or by using other methods. Let Kriging variance be the variance at a point given the initial spatial layout of the sensors, and let it be the variance at a prediction point. The predicted soil moisture value is , In the current spatial layout of the sensor Candidate points Soil moisture value obtained from actual measurement The following relationship exists between them: , This does not represent a prediction point. Predicted soil moisture values The error is not in the sensor itself, but in the current spatial layout of the sensor, and in the prediction of the point. The inherent, minimum variance of the prediction error when performing kriging interpolation, i.e. It measures the "uncertainty of prediction". The larger the Kriging variance, the less reliable the prediction for that point is based on the existing sensor network. for The Middle Two-dimensional spatial position of each sensor Two-dimensional spatial position of a certain prediction point The semivariance of soil moisture between the two values ​​was calculated using the probabilistic semivariogram model constructed in step S2. This describes the spatial correlation of soil moisture, with its value varying with the distance between two points. It increases with the increase of (until it reaches the sill value), which measures the candidate point. With prediction point Spatial statistical "dissimilarity"; for The Middle The weight of each sensor, Instead of being a fixed value, it is obtained by solving the Kriging equations. The core idea is that the candidate point that is closer to the prediction point and has less spatial correlation with other candidate points in the current spatial layout of the sensor has a greater weight. It is a Lagrange multiplier used in solving for optimal weights. When the constraint condition of unbiased estimation is met, that is... .

[0077] Step S32, Iterative optimization: In the first step... In each iteration, a replacement operation is performed according to the following steps: from the current spatial layout of the sensors. One sensor deployment point (i.e., a candidate point) is randomly removed from the pool, and one sensor not currently deployed in the space is randomly selected from the candidate point set. The addition of new candidate points results in a new spatial layout. .

[0078] Step S33, Evaluation and Acceptance Criteria: Calculate the global mean kriging variance of the new spatial layout. :

[0079]

[0080]

[0081] like If the new spatial layout is better, then accept the replacement. Otherwise, refuse to replace and retain the original layout. .

[0082] Step S34, Convergence and Output: Repeat steps S22 to S23 until the preset maximum number of iterations is reached. The sensor spatial layout with minimum MKV found during the output iteration process. and its corresponding ,but This is the optimal spatial layout scheme for a given number of sensors.

[0083] Understandably, step S3 employs a random swapping iterative optimization strategy to keep computational complexity within an acceptable range while ensuring global search capabilities. It supports optimizing sensor layout schemes in large-scale irrigation areas on ordinary workstations or servers, demonstrating strong engineering feasibility and promotional value.

[0084] In some embodiments, the present invention obtains for any given quantity After finding the optimal layout, the process also includes:

[0085] Step S4: Through a systematic scanning of the number of sensors, construct the prediction uncertainty-sensor number response curve, and determine the optimal number of sensors based on the principle of diminishing marginal benefits, so as to obtain a high-quality solution within an acceptable computational cost.

[0086] Further, see Figure 3 Step S4 specifically includes:

[0087] Step S41: Construct a cost-marginal benefit response curve: within a reasonable physical and cost range , These are the minimum and maximum values ​​for the number of sensors, which need to be set according to the size and actual conditions of the irrigation area, in increments of a certain amount. ( The preferred value is 1, but it can also be set to 2, 5, etc., depending on the actual situation. (Iterate through different numbers of sensors.) For each Then, the position optimization algorithm provided in step S3 above is invoked to obtain the optimal global average kriging variance that it can achieve. Therefore, it is possible to draw... Follow The changing response curve is used as the prediction uncertainty-number of sensors response curve.

[0088] Optionally, the predictive uncertainty-sensor number response curve is in the form of a semi-logarithmic model. Where y is the optimal global average kriging variance, and x is the number of sensors N. and These are the parameters of the prediction uncertainty-number of sensors response curve. This reflects an extreme lack of information (e.g.) The level of prediction uncertainty under the given conditions. This reflects the uncertainty of prediction. The rate of change, also known as marginal benefit.

[0089] Step S42: Determine the optimal number of sensors based on marginal cost benefits: The response curve obtained in step S41 typically exhibits a monotonically decreasing trend with a downward convex shape, meaning that as... The increase, The rate of decline (marginal benefit) gradually slows down. Optimal number of sensors. The optimal number of sensors is determined by identifying the inflection point of the response curve. Specifically, when the reduction in the optimal global average Kriging variance resulting from adding a sensor is lower than a preset marginal benefit threshold, the marginal benefit is considered insufficient to offset the cost increase. The number of sensors before adding a sensor is then taken as the optimal number of sensors. The marginal benefit threshold used is defined as a value between 0.001% and 1% of the total benefit decrease in the initial stage of the cost marginal benefit response curve.

[0090] Finally, the optimal number of sensors is output. And the corresponding optimal spatial layout of sensors that minimizes the prediction uncertainty for the entire region. This provides direct and quantitative decision support for the scientific deployment of irrigation district monitoring networks.

[0091] It is understandable that in step S4 of this embodiment, by constructing a predictive uncertainty-sensor quantity response curve and automatically determining the optimal sensor size based on the principle of diminishing marginal benefits, the synchronous optimization of sensor location and quantity is achieved, a scientific balance is reached between monitoring accuracy and deployment cost, and the subjectivity and resource waste caused by manually setting the quantity are avoided.

[0092] In some embodiments, the deployment method of the present invention further includes robustness verification: fixing the optimal spatial layout, recalculating the prediction uncertainty distribution of the entire area using the upper and lower limits of the confidence intervals of the key parameters of the probabilistic semi-variogram model obtained in step S2, and evaluating the sensitivity of the sensor network performance to parameter uncertainty. Specifically, calculations and analyses can be performed under pessimistic and optimistic conditions respectively. If the sensitivity is low, it indicates that the performance of the current sensor layout scheme designed based on median parameters is not sensitive to parameter fluctuations within a reasonable range and has extremely strong stability. If the sensitivity is high, it indicates that the stability and applicability of the sensor layout scheme in application are poor, and there may be problems such as unreasonable preset typical time periods, which require adjustment.

[0093] The effectiveness of the large-scale soil moisture sensor deployment method provided in this embodiment of the invention is verified below using specific experimental data:

[0094] The optimal layout proposed in this embodiment is based on the Bootstrap median parameter (variable range). sill value To test the robustness of this layout scheme to parameter estimation uncertainties, sensitivity testing was performed in this embodiment. See [link to relevant documentation]. Figure 4 With the optimal layout of 2221 sensors fixed, the parameters of the semi-variogram model are replaced with combinations of the upper and lower limits of the 95% confidence interval (pessimistic scenario). , Optimistic scenario: , The optimal global Kriging standard deviation (KSD) is recalculated. To generate a visually appealing spatial distribution map, the KSD for each point needs to be calculated and plotted. However, this is not suitable for direct comparison, so the average KSD for the entire region is calculated again for a more intuitive data analysis. See [link to relevant documentation] Figure 5 The results show that the average KSD across the entire region is 2.6132% and 2.1891% under pessimistic and optimistic parameter scenarios, respectively, with deviations from the typical scenario (2.3599%) of less than ±0.3%. More importantly, the spatial distribution patterns of KSD are highly consistent across the three scenarios. This demonstrates that the sensor layout scheme designed based on median parameters is insensitive to fluctuations in parameters within a reasonable range and exhibits extremely strong stability.

[0095] A second aspect of the present invention provides a large-scale soil moisture sensor optimization layout device, comprising:

[0096] The data extraction module is configured to crop and extract the acquired remote sensing image data according to the boundary of the irrigation area to be monitored and the preset typical time period to obtain multi-year soil moisture time series data.

[0097] The model building module is configured to construct a probabilistic semi-variogram model from the multi-year soil moisture time-series data based on resampling and median aggregation methods;

[0098] The iterative optimization module is configured to generate a set of candidate points covering the irrigation area to be monitored, randomly select N candidate points to form the initial layout scheme of the sensor, construct the objective function based on the probabilistic semi-variogram model, and obtain the optimized layout scheme of the sensor based on the random exchange position optimization method.

[0099] Furthermore, the optimized layout device in this embodiment also includes:

[0100] The quantity optimization module is configured to scan the number of sensors within a preset range. For each number of sensors, the iterative optimization module is called to obtain the optimal objective function that can be achieved. The response curve of the optimal objective function as a function of the number of sensors N is constructed as the prediction uncertainty-sensor number response curve. The optimal number of sensors is determined based on the principle of diminishing marginal utility.

[0101] It should be noted that the foregoing explanation of the embodiment of the method for optimizing the layout of large-scale soil moisture sensors also applies to the device for optimizing the layout of large-scale soil moisture sensors in this embodiment, and will not be repeated here.

[0102] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0103] Although examples of the invention have been shown and described above, it is understood that the above examples are exemplary and should not be construed as limiting the invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above examples within the scope of the invention.

Claims

1. A method for optimizing the layout of large-scale soil moisture sensors, characterized in that, include: Step S1: Based on the boundary of the irrigation area to be monitored and the preset typical time period, the acquired remote sensing image data is cropped and extracted to obtain multi-year soil moisture time series data. Step S2: Construct a probabilistic semi-variogram model for the multi-year soil moisture time-series data based on resampling and median aggregation methods; Step S3: Generate a regular grid of points covering the irrigation area to be monitored as a candidate point set. Randomly select N candidate points to form the initial sensor layout scheme. Construct an objective function based on a probabilistic semi-variogram model, and obtain the optimized sensor layout scheme based on a random exchange location optimization method, including: Within the boundary of the irrigation area to be monitored, generate a first grid point covering the entire area as a candidate point set and a second grid point as a prediction point set. Set the number of sensors to be deployed to N. Randomly initialize a spatial layout containing N sensors from the candidate point set. Construct an objective function based on the probabilistic semi-variogram model and the global average Kriging variance calculated from the prediction point set. Calculate the objective function corresponding to the initial spatial layout. The process involves iterating through the following steps: randomly removing one sensor deployment point from the current spatial layout and randomly selecting a new subsequent point from the candidate point set that is not in the current spatial layout to add to the new spatial layout; calculating the global average kriging variance corresponding to the new layout; if it is less than the global average kriging variance of the current spatial layout, the new spatial layout is accepted; otherwise, the current spatial layout is retained. Repeat the iteration until the termination condition is met, and output the spatial layout with the minimum global average kriging variance during the iteration process as the optimal spatial layout for that value of N.

2. The optimized layout method according to claim 1, characterized in that, The preset typical time period refers to the period when soil moisture is in a state that can reveal the intrinsic spatial structure of the soil; the state of the intrinsic spatial structure of the soil refers to the moderate humidity state of soil moisture between field capacity and crop wilting point.

3. The optimized layout method according to claim 1, characterized in that, Step S2 includes: A resampling method was used to perform multiple samplings with replacement on the multi-year soil moisture time-series data to generate multiple resampling datasets; For each resampled dataset, the median field of its soil moisture spatial distribution is calculated, and the semi-variogram is fitted based on the median field to obtain a set of variogram parameters. Based on the set of mutation function parameters obtained from multiple resampling, the probability distribution of the key parameters of the semi-mutation function is generated, and the probabilistic semi-mutation function model is constructed using their mean.

4. The optimized layout method according to claim 1, characterized in that, The resampling method used is the Bootstrap method.

5. The optimized layout method according to claim 1, characterized in that, The probabilistic semivariogram model is obtained by fitting an exponential model or a spherical model.

6. The optimized layout method according to claim 1, characterized in that, Let the first In the next iteration, the calculated global average kriging variance of the current spatial layout is: : in, Indicates the first The current spatial layout of the N sensors at the nth iteration, where the i-th sensor is located within the... The two-dimensional spatial position of each sensor is ; This represents the two-dimensional spatial location of a prediction point within the irrigation area to be monitored. The area of ​​the irrigation district to be monitored; Current spatial layout of the sensors The Kriging variance at a given point; for Middle Two-dimensional spatial position of each sensor Two-dimensional spatial position of a certain prediction point The semivariance of soil moisture between the values ​​is calculated using the probabilistic semivariogram model. for Middle The weight of each sensor, according to Middle The distance between each sensor and the prediction point is configured to a different value; It is a Lagrange multiplier.

7. The optimized layout method according to any one of claims 1 to 6, characterized in that, Also includes: Step S4: Scan the number of sensors within the preset range. For each number of sensors, obtain the optimal objective function that can be achieved according to step S3. Construct a response curve of the optimal objective function as a function of the number of sensors N, which serves as the prediction uncertainty-sensor number response curve. Determine the optimal number of sensors based on the principle of diminishing marginal utility.

8. A device for optimizing the layout of large-scale soil moisture sensors, characterized in that, include: The data extraction module is configured to crop and extract the acquired remote sensing image data according to the boundary of the irrigation area to be monitored and the preset typical time period to obtain multi-year soil moisture time series data. The model building module is configured to construct a probabilistic semi-variogram model from the multi-year soil moisture time-series data based on resampling and median aggregation methods; The iterative optimization module is configured to generate a set of candidate points covering the irrigation area to be monitored using a regular grid of points. It randomly selects N candidate points to form the initial sensor layout scheme, constructs an objective function based on a probabilistic semi-variogram model, and obtains the optimized sensor layout scheme based on a random exchange location optimization method, including: Within the boundary of the irrigation area to be monitored, generate a first grid point covering the entire area as a candidate point set and a second grid point as a prediction point set. Set the number of sensors to be deployed to N. Randomly initialize a spatial layout containing N sensors from the candidate point set. Construct an objective function based on the probabilistic semi-variogram model and the global average Kriging variance calculated from the prediction point set. Calculate the objective function corresponding to the initial spatial layout. The process involves iterating through the following steps: randomly removing one sensor deployment point from the current spatial layout and randomly selecting a new subsequent point from the candidate point set that is not in the current spatial layout to add to the new spatial layout; calculating the global average kriging variance corresponding to the new layout; if it is less than the global average kriging variance of the current spatial layout, the new spatial layout is accepted; otherwise, the current spatial layout is retained. Repeat the iteration until the termination condition is met, and output the spatial layout with the minimum global average kriging variance during the iteration process as the optimal spatial layout for that value of N.

9. The optimized layout device according to claim 8, characterized in that, Also includes: The quantity optimization module is configured to scan the number of sensors within a preset range. For each number of sensors, the iterative optimization module is called to obtain the optimal objective function that can be achieved. The response curve of the optimal objective function as a function of the number of sensors N is constructed as the prediction uncertainty-sensor number response curve. The optimal number of sensors is determined based on the principle of diminishing marginal utility.

Images