A method for generating a soybean variable fertilization prescription map based on remote sensing inversion of soil nutrient distribution
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-15
- Publication Date
- 2026-08-11
AI Technical Summary
解决现有技术中施肥决策模型适用性差、遥感影像分区可解释性弱、土壤养分信息冗余、管理分区算法稳定性低以及无法实现网格级精准施肥的技术问题，实现田块尺度遥感反演土壤养分空间数据在大豆起垄施肥环节的高效应用
本发明以土壤养分卫星遥感反演技术为基础，建立了基于土壤肥力指数和SOM-K-Means聚类算法的农田土壤肥力管理分区模型，构建了适用于黑龙江省尖山农场大豆起垄施肥环节的复合肥变量施肥处方决策体系，为大田变量施肥作业提供决策依据。弥补了以往研究大多停留在农田精准管理分区层面，缺乏结合具体种植作物进行复合肥变量施肥管理的不足，提高了变量施肥处方图的生成效率，有助于卫星遥感变量施肥技术在规模化农业生产中得到推广应用。与基于遥感影像数据的分区方法相比，融合碱解氮、有效磷、速效钾分布数据的土壤肥力指数在反映土壤供肥能力方面具有更强的农学可解释性，保证了农田精准管理分区与变量施肥处方决策的衔接性。SOM-K-Means两阶段聚类算法克服了传统K-Means聚类算法对初始质心敏感和SOM算法聚类结果稳定性较差的问题，能够更有效地识别田块内部土壤肥力指数的空间异质性，提升了农田管理分区模型的稳定性与可靠性。
Smart Images
Figure CN122551201A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of large-scale agricultural production technology, specifically to a method for generating soybean variable fertilization prescription maps based on remote sensing inversion of soil nutrient distribution. Background Technology
[0002] Fertilization is a core agricultural technology measure to ensure soybean yield and quality. Currently, large-scale soybean planting areas in Northeast my country (such as farms under the Beidahuang Group) generally adopt a uniform compound fertilizer application model, which involves pre-mixing urea, diammonium phosphate, and potassium sulfate in a fixed ratio and then applying the fertilizer uniformly to the entire field using a tractor-mounted ridge-forming fertilizer applicator. This model ignores the spatial heterogeneity of soil nutrients, and long-term application will lead to low fertilizer utilization and soil nutrient imbalance, failing to achieve the goals of improving the quality of black soil farmland and achieving high-yield and high-quality grain production.
[0003] Prescription map-based variable fertilization technology, due to its well-developed performance and high reliability, is becoming an important means of balanced soil nutrient management in farmland. Precise farmland management zoning is a key step in realizing this technology, aiming to identify independently manageable areas within a plot of land with similar attributes. Currently, most studies rely on point-like information such as soil nutrients and crop yields obtained from field sampling, converting it into areal data through spatial interpolation, and then using fuzzy clustering algorithms to divide management zones. While this method can reflect the spatial heterogeneity within a field, it suffers from time-consuming, costly, and untimely results, making it difficult to meet the needs of efficient management of large-scale farmland. With the development of remote sensing technology, existing studies have begun to use remote sensing imagery for zoning, improving timeliness; however, this neglects the mapping relationship between remote sensing spectral information and soil nutrient indicators, resulting in weak agronomic interpretability of the zoning results.
[0004] Using machine learning algorithms to construct a soil nutrient inversion model can generate a spatial distribution map of soil nutrient content, and further combine it with the nutrient balance method to establish a variable fertilization prescription map. However, the existing methods still have the following shortcomings: (1) The nutrient balance method has many undetermined parameters, and it is difficult to accurately determine the target yield and soil nutrient correction coefficient, resulting in poor model stability and universality; (2) Soil nutrient indicators are usually regarded as independent zoning variables, ignoring the correlation between different nutrient indicators and their comprehensive impact on fertility; (3) Traditional clustering algorithms are sensitive to the initial cluster centers and are easily affected by noise, affecting the accuracy of zoning; (4) The management zoning map only provides management suggestions at the regional level and cannot achieve precise adjustment of fertilization amount at the grid level. The above problems restrict the promotion and application of variable fertilization technology in large-scale agricultural production. Summary of the Invention
[0005] To address the aforementioned shortcomings of existing technologies, this invention provides a method for generating soybean variable fertilization prescription maps based on remote sensing inversion of soil nutrient distribution. This method solves the technical problems of poor applicability of fertilization decision models, weak interpretability of remote sensing image zoning, redundancy of soil nutrient information, low stability of management zoning algorithms, and the inability to achieve grid-level precision fertilization in existing technologies. It enables the efficient application of field-scale remote sensing inversion of soil nutrient spatial data in the soybean ridging and fertilization process.
[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: A method for generating soybean variable fertilization prescription maps based on remote sensing inversion of soil nutrient distribution is provided, which includes the following steps: S1: Obtain raster data of soil available nitrogen, available phosphorus, and available potassium content generated by satellite remote sensing inversion; determine the weights of each soil nutrient factor based on soybean fertilizer formula and analytic hierarchy process; and construct a soil fertility index integrating the three nutrient factors using a weighted summation method. SFI ; S2: Based on the operating width of the compound fertilizer variable fertilizer application ridging machine, the soil fertility index is resampled; the resampled soil fertility index is used as the zoning index to determine the number of management zones; S3: Based on the SOM-K-Means two-stage clustering algorithm and the number of management zones, a soil fertility management zoning model is constructed. The soil fertility index is clustered using the soil fertility management zoning model to obtain cluster labels. The SOM-K-Means two-stage clustering algorithm first uses the self-organizing map algorithm for preliminary clustering to determine the initial cluster centers, and then uses the K-Means algorithm for secondary clustering. S4: Combining the pixel coordinate information of farmland plots, the clustering labels are filled into a two-dimensional raster matrix that matches the shape of the target plots, the areas outside the plots are removed and visualized to generate a soil fertility management zoning map. S5: Determine the benchmark fertilizer application rate for compound fertilizer in the soybean ridging and fertilization process, formulate a graded fertilization strategy with fluctuations above and below the benchmark value, convert the soil fertility management zoning map into vector format, generate plot grid vector data with corresponding operation width, match fertility level and assign corresponding fertilizer application rate, adjust plot boundaries and transform coordinate system, and export SHP format prescription map that can be executed by variable fertilizer applicator.
[0007] Furthermore, in step S1, the formula for calculating the soil fertility index is: ; in, For the first i Soil fertility index of each raster cell; The normalized soil available nitrogen content for the i-th raster cell; For the first iSoil available phosphorus content after normalization of individual raster pixels; For the first i Soil available potassium content after normalization of individual raster pixels; , and The values of all are in the range of [0, 1]. The weighting coefficient for soil alkaline nitrogen hydrolysis; This is the weighting coefficient for available phosphorus in the soil; is the weighting coefficient for available potassium in the soil, and .
[0008] Furthermore, in step S1, the soil nutrient factor weights are obtained by normalizing the pure nutrient input ratio of N:P:K in the soybean fertilizer formula. , and .
[0009] Furthermore, in step S2, the zoning indicators are standardized using Z-scores, the range of cluster numbers is set, and the appropriate number of farmland management zones is determined by combining the elbow rule with field production experience.
[0010] Furthermore, in step S3, the specific steps of the SOM-K-Means two-stage clustering algorithm are as follows: S31: Set the number of neurons, learning rate, and number of iterations for the SOM network, and initialize the weight vectors of the output layer neurons; S32: Through the neuron competition mechanism, a winning node is matched for each soil fertility index sample point, the weight vector of the winning node and its neighboring neurons is updated, and the sample points are mapped to the output node to obtain the preliminary clustering results. S33: Using the initial cluster centers, which are the same number as the number of management partitions output by the SOM algorithm, as the initial cluster centers for the K-Means algorithm, calculate the distance between the unassigned sample points and each cluster center and assign them to the nearest cluster; S34: Recalculate the cluster centers of each cluster, iterate until the change in cluster centers is less than a preset threshold or the preset maximum number of iterations is reached, and save the cluster labels of the soil fertility index.
[0011] Furthermore, in step S5, the benchmark fertilizer application rate is calculated based on the unit area pure fertilizer application rate and fertilizer formula ratio in the soybean ridging and fertilization process; and the fertilizer application rate corresponding to different fertility levels is determined by adopting a fertilization strategy that allows for graded fluctuations above and below the benchmark value.
[0012] Furthermore, in step S5, the soil fertility management zoning map is reclassified and rasterized using geographic information system software to generate plot grid vector data with twice the working width. The corresponding units of the fertility level and the plot grid vector data are spatially connected. A fertilizer application amount field is added and attribute values are assigned. After being converted to a general geographic coordinate system, an SHP format prescription map that can be executed by the variable fertilizer applicator is exported.
[0013] The beneficial effects of this invention are as follows: This invention, based on satellite remote sensing inversion technology for soil nutrients, establishes a farmland soil fertility management zoning model based on soil fertility indices and the SOM-K-Means clustering algorithm. It constructs a compound fertilizer variable application prescription decision-making system applicable to the soybean ridging and fertilization stage at Jianshan Farm in Heilongjiang Province, providing a decision-making basis for variable application fertilization operations in large-scale fields. This addresses the shortcomings of previous studies, which mostly focused on precise farmland management zoning and lacked integration with specific crop-specific compound fertilizer variable application management. It improves the efficiency of variable application prescription map generation and facilitates the widespread application of satellite remote sensing variable application technology in large-scale agricultural production. Compared with zoning methods based on remote sensing image data, the soil fertility index, which integrates data on the distribution of available nitrogen, available phosphorus, and available potassium, has stronger agronomical interpretability in reflecting soil nutrient supply capacity, ensuring the connection between precise farmland management zoning and variable application prescription decisions. The SOM-K-Means two-stage clustering algorithm overcomes the problems of the traditional K-Means clustering algorithm being sensitive to the initial centroid and the poor stability of the SOM algorithm's clustering results. It can more effectively identify the spatial heterogeneity of soil fertility index within a field, thus improving the stability and reliability of the farmland management zoning model. Attached Figure Description
[0014] Figure 1 This is a schematic diagram of the process of the present invention; Figure 2 This is a soil fertility management zoning map of the test plots in the embodiment; Figure 3 This is a diagram of the variable fertilizer prescription for soybeans in the experimental plots of the example. Detailed Implementation
[0015] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.
[0016] Example This embodiment uses a bare soil field after autumn crop harvest at Jianshan Farm, Jiuzhou Administration Bureau, Heilongjiang Province as the test object, and takes the variable application of basal fertilizer in the soybean ridging and fertilization process of the following year as the application scenario, adopting the following... Figure 1 The process shown generates a soybean variable fertilizer prescription map. The specific implementation steps are as follows: S1: Select basic farmland plots in the autumn bare soil period of Jianshan Farm, Jiuzhou Administration Bureau, Heilongjiang Province, and acquire soil nutrient spatial distribution data generated by satellite remote sensing inversion technology, including distribution maps of available nitrogen, available phosphorus, and available potassium. The data type is raster data, the spatial resolution is 10m, and the storage format is TIF. Based on soybean fertilizer formulas and the analytic hierarchy process (AHP), the weights of each soil nutrient factor are determined. A soil fertility index integrating the three nutrient factors is constructed using a weighted summation method. The calculation formula for the soil fertility index is as follows: ; in, For the first i Soil fertility index of each raster cell; The normalized soil available nitrogen content for the i-th raster cell; For the first i Soil available phosphorus content after normalization of individual raster pixels; For the first i Soil available potassium content after normalization of individual raster pixels; , and The values of all are in the range of [0, 1]. The weighting coefficient for soil alkaline nitrogen hydrolysis; This is the weighting coefficient for available phosphorus in the soil; is the weighting coefficient for available potassium in the soil, and .
[0017] Based on the ratio of the three pure nutrients in the farm's soybean fertilizer formula N:P:K = 1:1.38:0.4, the relative importance of nitrogen, phosphorus, and potassium for soybean production was determined. This ratio was then normalized to obtain the soil nutrient factor weights. =0.3597、 =0.4964、 =0.1439.
[0018] The formula for calculating the soil fertility index is: ; To verify the rationality of this weighting, a judgment matrix was constructed using the analytic hierarchy process (AHP) and a consistency test was performed. Specifically, pairwise comparisons were made based on the importance of nitrogen, phosphorus, and potassium in soybean fertilizer formulations to construct a judgment matrix and calculate the AHP weights. =0.3196、 =0.5584、 =0.1220). By performing a consistency test on the judgment matrix, the consistency ratio CR was calculated to be 0.0158. This value is less than 0.1, indicating that the judgment matrix meets the internal consistency requirement. Comparing the results obtained from the two methods, it can be seen that the normalized weights obtained from the fertilizer formula are basically consistent with the weights obtained from the analytic hierarchy process (AHP), indicating that the weights assigned to each factor by the fertilizer formula are reasonable and feasible in terms of hierarchical structure. Therefore, the weights determined using the fertilizer formula are more reliable and effective, which is beneficial to improving the practical application significance of the soil fertility index in farmland fertilization management zoning.
[0019] S2: Based on the operating width of the compound fertilizer variable fertilization ridging machine (6.6m), the spatial resolution of the soil fertility index distribution map is resampled to twice the operating width (13.2m) using bilinear interpolation to buffer the spatial deviation caused by GPS positioning error and terrain undulation, and reduce map fragmentation.
[0020] With the K value set to a range of 2-10, the soil fertility index of the resampled soil fertility index distribution map was standardized using Z-score processing. The sum of squared errors under different K values was calculated based on the K-means clustering algorithm, and elbow curves were plotted. Considering the operational complexity, efficiency, and cost of variable operation of the fertilizer applicator, and combining the fertilization production experience of Jianshan Farm over many years, the appropriate number of farmland management zones was determined to be 5.
[0021] Scale adjustment of raster data not only helps to buffer spatial deviations caused by GPS positioning errors and terrain undulations during field operations, but also effectively reduces the fragmentation of map patches caused by excessively small zoning scales, thereby improving the spatial accuracy and operational stability of variable fertilization.
[0022] S3: Based on the SOM-K-Means two-stage clustering algorithm and the number of management zones, a soil fertility management zoning model is constructed. The soil fertility index is clustered using the soil fertility management zoning model to obtain cluster labels. Specifically, the SOM network parameters are set, including the number of neurons, learning rate, and number of iterations, and the weight vectors of the output layer neurons are initialized. Through competitive learning, a winning node is matched for each soil fertility index sample point, and the weights of the winning node and its neighboring neurons are updated to complete preliminary clustering and obtain an initial cluster center number equal to the number of management partitions.
[0023] Next, based on the five initial cluster centers output by SOM, the Euclidean distance between each sample point and the cluster center is calculated, and the sample is assigned to the nearest cluster. The cluster centers of each cluster are recalculated, and the iteration continues until a preset termination condition is met, finally obtaining the cluster label for each raster cell.
[0024] The partitioning performance of four algorithms, FCM, K-Means, SOM, and SOM-K-Means, was compared using three metrics: SC (profile coefficient), CHI (Kalinsky-Harabass index), and DBI (Davidsonburgin index). The results are shown in Table 1.
[0025] Table 1
[0026] Table 1 shows that the SOM-K-Means model has an SC value of 0.5643 and a DBI value of 0.5213. Compared with the K-Means, SOM, and FCM models, the SC values increased by 0.13, 3.23, and 0.75 percentage points, respectively. Compared with the SOM and FCM models, the CHI values increased by 1273.5473 and 34.4894, respectively. This indicates that the SOM-K-Means model has a better clustering effect and higher intra-cluster sample similarity. The SOM model has an SC value of 0.5320, lower than the SOM-K-Means model, but a DBI value of 0.5465, higher than the SOM-K-Means model, indicating that the SOM model has a poorer clustering effect and lower intra-cluster sample similarity. In summary, the SOM-K-Means model can more accurately capture the intrinsic structural characteristics of soil fertility index, providing more accurate and reliable clustering results for farmland management zoning.
[0027] S4: Create a two-dimensional raster matrix based on the boundary coordinates of the test plots. Fill the corresponding cell positions with the clustering labels obtained in step S3, filling non-plot areas with NaN values and removing them. Use the symbology of ArcGIS 10.8 software to perform differentiated coloring on the five management zones, generating results as follows: Figure 1 The diagram shows the soil fertility management zones. The average soil fertility of each management zone, from highest to lowest, is as follows: Zone 1, Zone 2, Zone 5, Zone 4, Zone 3. Zone 3 represents the lowest soil fertility, i.e., fertility level 1, while Zone 1 represents the highest soil fertility, i.e., fertility level 5.
[0028] S5: In the soybean ridging and fertilization process at Jianshan Farm in Heilongjiang Province, the compound fertilizer mainly consists of three single-element fertilizers: urea (containing 46% N), diammonium phosphate (containing 18% N and 46% P2O5), and potassium sulfate (containing 50% K2O). In agricultural production, fertilizer application rates are calculated based on the statistical indicators of pure NPK fertilizer. Therefore, based on the pure fertilizer application rate of 18.53 catties / mu for the three single-element fertilizers in the soybean ridging and fertilization process at Jianshan Farm, and combined with the fertilizer formula N:P:K = 1:1.38:0.4, the calculated application rates are 6.67 catties / mu for urea, 20.00 catties / mu for diammonium phosphate, and 5.33 catties / mu for potassium sulfate, equivalent to a standard compound fertilizer application rate of 240 kg / hm².2 .
[0029] At 240 kg / hm 2 Based on the baseline, a fertilizer application management strategy was adopted, with the application rate fluctuating by 5% and 10% above and below the baseline value, respectively. The fertilizer fertility level 1 was 264 kg / hm². 2 Grade 2 is 252 kg / hm 2 Grade 3 is 240 kg / hm 2 Grade 4 is 228 kg / hm 2 Grade 5 is 216 kg / hm 2 .
[0030] The soil fertility management zoning map was reclassified using ArcGIS 10.8 software, sorted by fertility level from low to high, and then converted to vector format. Based on the vector format of the soil fertility management zoning map sorted by fertility level, 13.2m × 13.2m plot grid vector data was generated using the fishing net tool. The spatial join function was used to associate fertility levels with corresponding cells in the plot grid vector data. A "VALUE" field was added to the attribute table, and the field join function was used to map fertility levels to corresponding fertilizer application rates. Prescription cells in the plot boundary areas were adjusted using UAV synthetic imagery, and the coordinate system was converted from WGS1984 UTM Zone 51N to the WGS 1984 geographic coordinate system. The final output is shown below. Figure 2 The soybean variable fertilizer prescription diagram in SHP format shown can be directly imported into the intelligent control system of the compound fertilizer variable fertilizer applicator for operation.
Claims
1. A method for generating a variable fertilization prescription map of soybean based on remote sensing inversion of soil nutrient distribution, characterized in that, Includes the following steps: S1: Obtain raster data of soil available nitrogen, available phosphorus, and available potassium content generated by satellite remote sensing inversion; determine the weights of each soil nutrient factor based on soybean fertilizer formula and analytic hierarchy process; and construct a soil fertility index integrating the three nutrient factors using a weighted summation method. SFI ; S2: Based on the operating width of the compound fertilizer variable fertilizer application ridging machine, the soil fertility index is resampled; the number of management zones is determined using the resampled soil fertility index as the zoning index. S3: Based on the SOM-K-Means two-stage clustering algorithm and the number of management zones, a soil fertility management zoning model is constructed. The soil fertility index is clustered using the soil fertility management zoning model to obtain cluster labels. The SOM-K-Means two-stage clustering algorithm first uses the self-organizing map algorithm for preliminary clustering to determine the initial cluster centers, and then uses the K-Means algorithm for secondary clustering. S4: Combining the pixel coordinate information of farmland plots, the clustering labels are filled into a two-dimensional raster matrix that matches the shape of the target plots, the areas outside the plots are removed and visualized to generate a soil fertility management zoning map. S5: Determine the benchmark fertilizer application rate for compound fertilizer in the soybean ridging and fertilization process, formulate a graded fertilization strategy with fluctuations above and below the benchmark value, convert the soil fertility management zoning map into vector format, generate plot grid vector data with corresponding operation width, match fertility level and assign corresponding fertilizer application rate, adjust plot boundaries and transform coordinate system, and export SHP format prescription map that can be executed by variable fertilizer applicator.
2. The method of claim 1, wherein, In step S1, the formula for calculating the soil fertility index is: ; in, For the first i Soil fertility index of each raster cell; The normalized soil available nitrogen content for the i-th raster cell; For the first i Soil available phosphorus content after normalization of individual raster pixels; For the first i Soil available potassium content after normalization of individual raster pixels; , and The values of all are in the range of [0, 1]. The weighting coefficient for soil alkaline nitrogen hydrolysis; This is the weighting coefficient for available phosphorus in the soil; is the weighting coefficient for available potassium in the soil, and .
3. The method of claim 2, wherein, In step S1, the pure nutrient input proportion of the soybean fertilization formula N:P:K is normalized to obtain soil nutrient factor weights , and .
4. The method of claim 1, wherein, In step S2, the zoning indicators are standardized using Z-score, the range of cluster numbers is set, and the appropriate number of farmland management zones is determined by combining the elbow rule and field production experience.
5. The method of claim 1, wherein, In step S3, the specific steps of the SOM-K-Means two-stage clustering algorithm are as follows: S31: Set the number of neurons, learning rate, and number of iterations for the SOM network, and initialize the weight vectors of the output layer neurons; S32: Through the neuron competition mechanism, a winning node is matched for each soil fertility index sample point, the weight vector of the winning node and its neighboring neurons is updated, and the sample points are mapped to the output node to obtain the preliminary clustering results. S33: Using the initial cluster centers, which are the same number as the number of management partitions output by the SOM algorithm, as the initial cluster centers for the K-Means algorithm, calculate the distance between the unassigned sample points and each cluster center and assign them to the nearest cluster; S34: Recalculate the cluster centers of each cluster, iterate until the change in cluster centers is less than a preset threshold or the preset maximum number of iterations is reached, and save the cluster labels of the soil fertility index.
6. The method of claim 1, wherein, In step S5, the benchmark fertilizer application rate is calculated based on the pure fertilizer application rate per unit area and the fertilizer formula ratio in the soybean ridging and fertilization process; the fertilizer application rate corresponding to different fertility levels is determined by adopting a fertilization strategy that allows for graded fluctuations above and below the benchmark value.
7. The method of claim 6, wherein, In step S5, the soil fertility management zoning map is reclassified and rasterized using geographic information system software to generate plot grid vector data with twice the working width. The corresponding cells of the fertility level and the plot grid vector data are spatially connected. A fertilizer application amount field is added and attribute values are assigned. After being converted to a general geographic coordinate system, an SHP format prescription map that can be executed by the variable fertilizer applicator is exported.