A multi-source data fusion-based fine fertilization decision method and system for phyllostachys edulis

Abstract

The present application relates to the technical field of intelligent management of agriculture and forestry, and specifically discloses a fine fertilization decision-making method and system for Phyllostachys edulis based on multi-source data fusion, which collects soil available nutrients, canopy spectral reflectance, growth response data and environmental covariates of each management grid in the target management area; the grids are divided into several environment-homogeneous micro-areas through fuzzy clustering; the micro-area centroids are used as the reference to search for similar plots in the historical plot database, and to construct individualized prior information pools for each micro-area; hierarchical Bayesian inference is performed in combination with the growth response data of each grid to output the localized corrected characteristic parameters of each micro-area; the characteristic parameters are embedded in the decision-making data of each grid, and the differentiated fertilization ratio and dosage are calculated grid by grid to generate fertilization prescriptions with spatial positioning attributes, which are applied to intelligent fertilization management of Phyllostachys edulis forests.

Description

Technical Field

[0001] This invention relates to the field of intelligent management technology in agriculture and forestry, specifically to a method and system for precise fertilization decision-making for moso bamboo based on multi-source data fusion. Background Technology

[0002] Moso bamboo is characterized by its rapid growth, short timber production cycle, and sustainable management after a single afforestation project. Fertilization management of moso bamboo forests is an important management measure to improve bamboo shoot yield and mature bamboo quality. Moso bamboo forests with different terrain conditions, soil textures, and climatic backgrounds have significant spatial differences in their nutrient requirements.

[0003] In recent years, with the gradual promotion of soil sensors, drone remote sensing, meteorological monitoring, and geographic information technology in agriculture and forestry, various types of data, including soil nutrients, canopy spectra, meteorological observations, and topographic environment, have been accumulated in the management of moso bamboo forests, providing a data foundation for refined nutrient management. At the same time, historical management records preserve a large amount of phenological response data for moso bamboo in different production areas and years, which reveals the growth patterns and nutrient requirements of moso bamboo under different environmental conditions.

[0004] The existing technology has the following shortcomings: Given the lack of long-term local observation data in newly established bamboo bases, how can we effectively integrate limited local regional data with external historical prior knowledge to eliminate the systematic estimation bias introduced by direct transplantation of phenological parameters across ecological zones? At the same time, how can we overcome the shortcomings of the uniform fertilization scheme that does not match the actual needs of the plot due to environmental heterogeneity such as topography, light, and soil within the same management area? This would provide a basis for refined fertilization decisions that are adapted to different microenvironmental conditions on a spatial scale.

[0005] In actual bamboo management, fertilization operations rely on manual spreading or simple mechanical spreading, lacking mechanized and precise execution methods that match the spatially differentiated prescriptions. Traditional operation methods make it difficult to accurately implement the differentiated fertilization plans generated by grids, resulting in an execution gap between prescription decision-making and field operations, which restricts the full realization of the effectiveness of refined fertilization management. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for precision fertilization decision-making for moso bamboo based on multi-source data fusion, which is applicable to mechanized variable fertilization operation scenarios, in order to solve the problems existing in the above-mentioned background technology.

[0007] The objective of this invention can be achieved through the following technical solutions: A method for refined fertilization decision-making for moso bamboo based on multi-source data fusion includes the following steps: S1. Collect soil available nutrient content, canopy spectral reflectance and growth response data for each management grid in the target management area, and obtain environmental covariates for each grid, including accumulated temperature, precipitation coefficient of variation, topographic humidity index and slope aspect. S2 takes the environmental covariates of each grid as input and divides all management grids into several environmentally homogeneous micro-regions through fuzzy clustering, so that the characteristics of environmental covariates within each micro-region tend to be consistent. S3, using the centroid of the environmental covariate of each micro-region as the retrieval benchmark, retrieves the records of plots in the historical plot database whose cosine similarity with each centroid exceeds a preset threshold, extracts the statistical values ​​of the phenological response parameters of each plot record, and constructs a personalized prior information pool for each micro-region. S4. The personalized prior information pool of each micro-region is used as the prior constraint of the phenological response parameters of each micro-region. Combined with the growth response data of each grid, hierarchical Bayesian inference is performed on the phenological response parameters of each micro-region, and the characteristic parameters of each micro-region after local data correction are output. S5 embeds the characteristic parameters of each micro-region into the basic decision data of each grid, and calculates the differentiated fertilization ratio and dosage grid by grid based on the soil available nutrient content and canopy spectral reflectance of each grid, generating a differentiated fertilization prescription with spatial positioning attributes.

[0008] As a further aspect of the present invention: the division of the entire management grid into several environmentally homogeneous micro-regions through fuzzy clustering specifically includes: The range normalization process is performed on each environmental covariate vector according to its feature dimension to unify the value range of each dimension to a preset interval; Using the Euclidean distance between the environmental covariate vectors of each grid after normalization as the difference measure, and based on the preset number of micro-regions, fuzzy membership degree assignment is performed on each grid, and the membership degree value of each grid belonging to each candidate micro-region is calculated one by one. Based on the membership degree values ​​of each grid to each candidate microregion, each grid is assigned to the candidate microregion with the largest membership degree value, thus completing the initial division of all grids; After the initial division, the intra-class average distance of all grid environmental covariate vectors in each micro-region is calculated. Grids with intra-class average distances exceeding a preset threshold are marked. Then, the corresponding grids are repositioned based on the environmental covariate similarity with the centroid of each micro-region until the intra-class average distance of all micro-regions is lower than the preset threshold. Finally, the environmental homogeneous micro-region division result is output.

[0009] As a further aspect of the present invention: the step of calculating the membership value of each grid cell belonging to each candidate micro-region specifically includes: Using the centroid of the environmental covariate vector of each candidate microregion as a reference, the Euclidean distance between the environmental covariate vector and the centroid of each grid is calculated one by one, and the grids are arranged in order of proximity to each candidate microregion from the closest to the furthest. Using the Euclidean distance between the grid and the centroid of each candidate microregion as the independent variable and the initial influence intensity of the grid on each candidate microregion as the dependent variable, a set of initial influence intensity values ​​of the grid on each candidate microregion are determined according to the principle that the closer the distance, the greater the influence intensity. The initial influence intensity value of the grid on each candidate microregion is compressed proportionally, so that the sum of the initial influence intensity values ​​of the grid on all candidate microregions is normalized to a fixed value, and the membership value of the grid to each candidate microregion is obtained. After traversing all grids, output the set of membership values ​​of each candidate microregion corresponding to each grid, so that the grid can be assigned to the candidate microregion with the largest membership value in the subsequent process.

[0010] As a further aspect of the present invention: the construction of personalized prior information pools for each micro-region specifically includes: Using the centroid of the environmental covariate of each micro-region as the retrieval vector, the cosine similarity between the environmental covariate record vector of each plot in the historical plot database and the retrieval vector is calculated one by one. All plot records with a cosine similarity exceeding a preset threshold are selected to form a candidate similar plot set for the micro-region. The average value and dispersion of phenological response parameters recorded by each parcel in the candidate similar parcel set of the micro-region are calculated separately. The average value is used as the central estimate of the phenological response parameters of the micro-region, and the dispersion is used to determine the confidence boundary of the phenological response parameters of the micro-region. The central estimates of the phenological response parameters corresponding to each micro-region are associated with and stored with the confidence boundary, forming a personalized prior information pool for each micro-region for subsequent hierarchical Bayesian inference.

[0011] As a further aspect of the present invention: the determination of the confidence boundary of the micro-region phenological response parameters based on the degree of discreteness specifically includes: The maximum and minimum values ​​of phenological response parameters are extracted from the candidate similar plots in the micro-region, and the maximum and minimum values ​​are used to define the coarse boundary of the phenological response parameter values ​​in the micro-region. The distribution frequency of phenological response parameter values ​​in each value interval of the candidate similar plot set in the micro-region is statistically analyzed. Value intervals with distribution frequencies exceeding a preset frequency threshold are considered as densely distributed intervals. The minimum and maximum values ​​of the densely distributed intervals are used as high-density intervals for the phenological response parameter values ​​in the micro-region. The high-density region is used as the narrow boundary, and the difference between the coarse boundary and the narrow boundary is used as the elastic relaxation amount of the confidence boundary of the micro-region phenological response parameter. The narrow boundary and the elastic relaxation amount are superimposed as the confidence boundary output of the micro-region phenological response parameter.

[0012] As a further aspect of the present invention: the hierarchical Bayesian inference of the phenological response parameters of each micro-region specifically includes: An initial distribution of phenological response parameters for each micro-region is set, with the central estimate in the micro-region's personalized prior information pool used as the location parameter of the initial distribution, and the confidence boundary of the micro-region used as the scale constraint of the initial distribution, thus forming the prior distribution of the micro-region. Using the growth response data of each grid as the observation input, the local likelihood contribution value of the phenological response parameters of each grid under the constraint of the prior distribution of the micro-region to which each grid belongs is calculated grid by grid according to the prior distribution of the micro-region to which each grid belongs; The local likelihood contribution values ​​of all grids within the same micro-region are merged, and the combined likelihood value is used to jointly adjust the location parameters and scale constraints of the prior distribution of the micro-region to obtain the posterior distribution of the phenological response parameters of the micro-region. The location parameters of the posterior distribution are then output as the characteristic parameters of the micro-region after correction by local data.

[0013] As a further aspect of the present invention: the calculation process of the local likelihood contribution value is as follows: The degree of deviation between the measured response value in the growth response data of the grid and the theoretically expected position of the prior distribution of the micro-region is used as the initial measure of the grid response deviation. The initial measure of grid response bias is scaled according to the discrete magnitude of the observed response values ​​of all grids within the microregion to which the grid belongs, to obtain a relative measure of grid response bias; The local likelihood contribution value of the grid under the prior distribution constraint of its microregion is obtained by multiplying the relative measure of the grid response deviation with the prior probability density of the corresponding position within the confidence boundary of the microregion to which the grid belongs. After completing the traversal of all grids, the local likelihood contribution value of each grid is output.

[0014] As a further aspect of the present invention: S5 specifically includes: The nitrogen, phosphorus and potassium requirement baseline ratios for each grid are determined by combining the available soil nutrients content and canopy spectral reflectance. The amplitude of the requirement baseline ratios is modulated by the characteristic parameters of the micro-region to which the grid belongs, and the initial fertilizer application ratio and initial application amount for the grid are obtained. The average initial fertilization ratio and initial application rate of all grids within the same micro-region are used as the common reference of the micro-region. The deviation of the initial fertilization ratio and initial application rate of each grid from the common reference of the micro-region is calculated one by one. The initial fertilization ratio and initial application rate of the grid are reversed by the deviation to obtain the secondary adjustment ratio and secondary adjustment amount of the grid. The secondary adjustment ratio and dosage of the grid are linked and bound to the geographical coordinates of the grid's spatial location to generate a differentiated fertilization prescription with spatial positioning attributes.

[0015] As a further aspect of the present invention: the calculation of the deviation of the initial fertilization ratio and the initial application rate of each grid from the common reference of the micro-region specifically includes: The initial differences between the grids in terms of ratio and dosage are obtained by subtracting the common ratio and dosage in the common reference of the micro-region to which the grid belongs from the initial fertilization ratio and initial dosage of each grid. Determine the sign of the initial difference between the grids in the proportion and dosage dimensions, classify the grids into positive or negative deviation groups according to the sign, and calculate the absolute magnitude of the initial difference of all grids in each group. Divide the absolute magnitude of the initial difference of the grid by the maximum absolute magnitude within the group to obtain the normalized relative deviation value of the grid in the proportion dimension and the usage dimension. Then, assign the direction attribute of the normalized relative deviation value with the positive or negative sign of the grid, and output the final deviation of the grid with the direction attribute.

[0016] A precision fertilization decision-making system for moso bamboo based on multi-source data fusion includes: The multi-source data acquisition and gridded input module collects soil available nutrient content, canopy spectral reflectance and growth response data for each management grid in the target management area, and obtains environmental covariates for each grid, including accumulated temperature, precipitation coefficient of variation, topographic humidity index and slope aspect. The homogeneous micro-region division module takes the environmental covariates of each grid as input and divides all management grids into several homogeneous micro-regions through fuzzy clustering, so that the environmental covariate characteristics within each micro-region tend to be consistent. The personalized prior information pool construction module uses the centroid of the environmental covariate of each micro-region as the retrieval benchmark, retrieves the records of land parcels in the historical land parcel database whose cosine similarity with each centroid exceeds a preset threshold, extracts the statistical values ​​of phenological response parameters of each land parcel record, and constructs a personalized prior information pool for each micro-region. The hierarchical Bayesian inference and parameter correction module uses the personalized prior information pool of each micro-region as the prior constraint of the phenological response parameters of each micro-region. Combined with the growth response data of each grid, it performs hierarchical Bayesian inference on the phenological response parameters of each micro-region and outputs the feature parameters of each micro-region after local data correction. The differentiated fertilization prescription generation module embeds the characteristic parameters of each micro-region into the basic decision data of each grid, and calculates the differentiated fertilization ratio and dosage grid by grid based on the soil available nutrient content and canopy spectral reflectance of each grid, generating a differentiated fertilization prescription with spatial positioning attributes.

[0017] The beneficial effects of this invention are: (1) By constructing environmentally homogeneous micro-regions and using hierarchical Bayesian inference to localize and correct historical phenological parameters, the phenological response estimation bias caused by direct cross-regional transplantation of mechanism parameters can be effectively reduced, the management cost of repeated manual trial and error adjustment of parameters can be reduced, and the newly built bamboo base can still obtain decision-making basis that matches the local environment even in the absence of long-term local observation data.

[0018] (2) By dividing the homogeneous micro-regions of the environment through fuzzy clustering and combining it with the generation of gridded differentiated prescriptions, the plots in the same management area with different terrain and light conditions can obtain the appropriate fertilization ratio and amount. This helps to reduce the management risk of both over-fertilization and under-fertilization at the spatial level, reduce the local nutrient accumulation or deficiency caused by empirical uniform fertilization, provide operators with traceable differentiated operation basis under different years and growth cycles, and improve the stability of forest land yield and the spatial balance of nutrient management. Attached Figure Description

[0019] The invention will now be further described with reference to the accompanying drawings.

[0020] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a flowchart illustrating how fuzzy clustering is used in this invention to divide the entire management grid into several homogeneous micro-regions. Figure 3 This is a flowchart illustrating the process of constructing personalized prior information pools for each micro-region in this invention. Figure 4 This is a system block diagram of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 1 As shown, this invention is a method for refined fertilization decision-making for moso bamboo based on multi-source data fusion, comprising the following steps: S1. Collect soil available nutrient content, canopy spectral reflectance and growth response data for each management grid in the target management area, and obtain environmental covariates for each grid, including accumulated temperature, precipitation coefficient of variation, topographic humidity index and slope aspect. S2 takes the environmental covariates of each grid as input and divides all management grids into several environmentally homogeneous micro-regions through fuzzy clustering, so that the characteristics of environmental covariates within each micro-region tend to be consistent. S3, using the centroid of the environmental covariate of each micro-region as the retrieval benchmark, retrieves the records of plots in the historical plot database whose cosine similarity with each centroid exceeds a preset threshold, extracts the statistical values ​​of the phenological response parameters of each plot record, and constructs a personalized prior information pool for each micro-region. S4. The personalized prior information pool of each micro-region is used as the prior constraint of the phenological response parameters of each micro-region. Combined with the growth response data of each grid, hierarchical Bayesian inference is performed on the phenological response parameters of each micro-region, and the characteristic parameters of each micro-region after local data correction are output. S5 embeds the characteristic parameters of each micro-region into the basic decision data of each grid, and calculates the differentiated fertilization ratio and dosage grid by grid based on the soil available nutrient content and canopy spectral reflectance of each grid, generating a differentiated fertilization prescription with spatial positioning attributes.

[0023] In S1, soil available nutrient content, canopy spectral reflectance, and growth response data were collected for each management grid in the target management area. Environmental covariates for each grid were also obtained, including accumulated temperature, precipitation coefficient of variation, topographic humidity index, and slope and aspect. Specifically, these included: The available nutrient content of the soil in each management grid of the target management area was collected in the following manner: Within the target management area, management grids were divided into 10-meter by 10-meter grids. During the leaf expansion stage of moso bamboo, mixed soil samples were collected from the 0-20 cm soil layer and the 20-40 cm soil layer in each management grid using the five-point sampling method. After the soil samples were air-dried and passed through a 2 mm sieve, the available nitrogen content of the soil was determined by the alkaline hydrolysis diffusion method, the available phosphorus content of the soil was determined by the molybdenum antimony colorimetric method, and the available potassium content of the soil was determined by the flame photometry method. The available nitrogen content, available phosphorus content, and available potassium content of the soil in each grid were recorded as the available nutrient content of the soil in that grid.

[0024] The canopy spectral reflectance of each management grid in the target management area was collected in the following manner: During the bamboo leafing period and on a clear, cloudless day (e.g., from 10:00 AM to 2:00 PM), a fixed-wing UAV equipped with a multispectral imager was used to fly over the target management area at a relative flight altitude of 120 meters, a forward overlap of 80%, and a lateral overlap of 60%. The canopy reflectance values ​​of the blue, green, red, red-edge, and near-infrared bands at the corresponding spatial locations of each management grid were obtained. The five band reflectance values ​​of each grid were recorded as the canopy spectral reflectance of that grid.

[0025] Growth response data for each management grid in the target management area were collected, specifically as follows: During the leaf-spreading stage of moso bamboo, five standard upright bamboos were selected within a 10-meter radius around the center point of each management grid. The diameter at breast height (DBH) of each standard upright bamboo was measured using a DBH ruler and the average value was recorded. The relative chlorophyll content of the functional leaves in the middle of the canopy of each standard upright bamboo was measured using a handheld chlorophyll meter and the average value was recorded. At the same time, after the moso bamboo matured in the same year, the number of new bamboos and the average DBH were recorded by management grid. The average DBH and average relative chlorophyll content measured during the leaf-spreading stage, as well as the number of new bamboos and the average DBH recorded after the bamboo matured, together constitute the growth response data for each grid.

[0026] The accumulated temperature and precipitation variation coefficients in the environmental covariates of each grid in the target management area are obtained as follows: A small automatic weather station is set up at the center of the target management area to continuously record the daily average temperature and monthly precipitation. The cumulative daily average temperature with a daily average temperature greater than or equal to 10 degrees Celsius during the annual growth period of moso bamboo is used as the annual effective accumulated temperature shared by each grid. The ratio of the standard deviation to the average monthly precipitation during the annual growth period of moso bamboo is used as the precipitation variation coefficient shared by each grid. The annual effective accumulated temperature and precipitation variation coefficients are assigned to each management grid.

[0027] To obtain the topographic humidity index and slope and aspect from the environmental covariates of each grid in the target operating area, the following method is used: Extract the elevation, slope, and aspect values ​​of the center point of each management grid from the digital elevation model data of the target operating area. Use a single flow direction algorithm to calculate the runoff accumulation of each grid based on the elevation value of each grid. Then, use the natural logarithm of the runoff accumulation of each grid and the tangent of the slope value as the topographic humidity index of each grid. Record the slope value, aspect value, and topographic humidity index of each grid as the slope, aspect, and topographic humidity index of that grid, respectively.

[0028] Please see Figure 2 As shown in S2, the environmental covariates of each grid are used as input, and fuzzy clustering is used to divide all management grids into several environmentally homogeneous micro-regions, so that the environmental covariate characteristics within each micro-region tend to be consistent. Specifically, this includes: Fuzzy clustering is used to divide all management grids into several environmentally homogeneous micro-regions. Specifically, the environmental covariance vector of each management grid is composed of accumulated temperature, precipitation coefficient of variation, topographic humidity index, slope, and aspect. The environmental covariance vector of each management grid is normalized according to the feature dimension. Specifically, for each feature dimension, the maximum and minimum values ​​of all management grids in that dimension are taken. The original value of each grid in that dimension is subtracted from the minimum value of that dimension, and then divided by the difference between the maximum and minimum values ​​of that dimension, so that the value range of each dimension is unified to the interval between 0 and 1. The normalized environmental covariance vector of each grid is then output.

[0029] Based on a preset number of micro-regions, fuzzy membership assignment is performed on each grid. Specifically, this includes: setting the number of micro-regions according to the area of ​​the target operating area and the complexity of the environmental gradient; this number is an integer between 3 and 5; using the centroid of the environmental covariate vector of each candidate micro-region as a reference, calculating the Euclidean distance between the normalized environmental covariate vector of each grid and the centroid of each candidate micro-region; the Euclidean distance is calculated by subtracting the values ​​of the two vectors along the same dimension and taking the square; then summing the squares of all dimensions and taking the square root to obtain the relationship between the grid and each candidate micro-region. The distance between centroids is used as the initial influence intensity of the grid on each candidate microregion. The smaller the distance, the greater the initial influence intensity of the candidate microregion. The initial influence intensity of the grid on each candidate microregion is divided by the sum of the initial influence intensity values ​​of the grid on all candidate microregions, so that the sum of the initial influence intensity values ​​of the grid on all candidate microregions is normalized to 1, and the membership degree value of the grid on each candidate microregion is obtained. After traversing all grids, the set of membership degree values ​​of each grid corresponding to each candidate microregion is output.

[0030] The initial division of all grids is completed based on the membership values ​​of each grid to each candidate micro-region. Specifically, this includes: for each management grid, assigning it to the candidate micro-region with the largest membership value; if the membership value of a grid to all candidate micro-regions is lower than a preset assignment threshold, then the grid is not assigned to any micro-region for the time being. The assignment threshold is a preset value between 0.3 and 0.5. After the initial assignment of all assignable grids is completed, the number of grids contained in each micro-region is counted. If the number of grids contained in a micro-region is less than a preset minimum grid number threshold, then the micro-region is disbanded, and all grids originally belonging to the micro-region are reassigned to the micro-region with the largest membership value among the remaining micro-regions. The minimum grid number threshold is determined based on the total number of management grids in the target operating area.

[0031] The intra-class average distance of all grid environmental covariate vectors within each micro-region after the initial partitioning is calculated, and secondary repositioning is performed until the intra-class average distance of all micro-regions is lower than a preset threshold. Specifically, this is achieved as follows: For each micro-region, the Euclidean distance between each pair of environmental covariate vectors within that micro-region is calculated. All pairwise distance values ​​are summed and divided by the total number of distance values ​​to obtain the intra-class average distance. The intra-class average distance threshold is set to a value between 0.15 and 0.25. Micro-regions with intra-class average distances exceeding this threshold are marked as micro-regions to be split. For each grid in a micro-region to be split, the Euclidean distance between the grid's environmental covariate vector and the centroid of each micro-region is calculated, and the grid is assigned to the micro-region with the smallest distance. After each repositioning of all grids to be adjusted, the centroid of each micro-region and the intra-class average distance of each micro-region are recalculated. The above judgment and adjustment process is repeated until the intra-class average distance of all micro-regions is lower than the intra-class average distance threshold. The final environmental homogeneous micro-region partitioning result and the micro-region assignment information of each grid are then output.

[0032] The initial value and iterative update of the centroid of the environmental covariate vector of each candidate microregion specifically include: in the first iteration, the environmental covariate vector of a preset number of grids is randomly selected as the initial centroid of each candidate microregion; after each subsequent grid assignment adjustment, for each microregion, the values ​​of the environmental covariate vectors of all grids in the same dimension are added together and divided by the total number of grids in the microregion to obtain the mean of each dimension. The mean of each dimension is used to reconstruct the new centroid of the microregion for the next round of distance calculation and assignment determination.

[0033] Please see Figure 3 As shown, in S3, using the centroid of the environmental covariate of each micro-region as the retrieval benchmark, the historical land parcel database is searched for land parcel records whose cosine similarity to each centroid exceeds a preset threshold. The statistical values ​​of phenological response parameters for each land parcel record are extracted, and personalized prior information pools for each micro-region are constructed, specifically including: Using the centroid of the environmental covariate in each micro-region as the retrieval vector, the cosine similarity between the environmental covariate record vector of each parcel in the historical parcel database and the retrieval vector is calculated one by one, specifically as follows: Each record in the historical land database contains the accumulated temperature, precipitation coefficient of variation, topographic humidity index, slope, and aspect values ​​of the land. These five values ​​are arranged in a fixed order to form the environmental covariate record vector of the land. The centroids of the environmental covariates of the micro-region to be searched are arranged in the same fixed order of the five dimensions mentioned above to form the search vector; For each record in the historical land database, the dot product of the record vector and the retrieval vector along the same dimension is summed to obtain the dot product of the two vectors. Calculate the length of the record vector and the length of the retrieval vector respectively. The vector length is calculated by taking the square of the values ​​of each dimension of the vector, summing them, and then taking the square root. Divide the dot product value by the product of the record vector length and the retrieval vector length to obtain the cosine similarity value between the plot record and the centroid of the micro-region; The cosine similarity value ranges from -1 to +1. The closer the value is to +1, the higher the consistency of the two vectors' directions. A preset threshold is set between 0.80 and 0.95. Plots with cosine similarity values ​​greater than or equal to the preset threshold are selected to form a set of candidate similar plots for the micro-region.

[0034] The average value and dispersion of phenological response parameters recorded for each parcel in the candidate similar parcel set of the micro-region are calculated separately, specifically as follows: Each record in the candidate similar plot set contains the values ​​of the phenological response parameters of moso bamboo recorded in the plot over many years of observation. The phenological response parameters include the maximum nutrient dilution rate of moso bamboo and the accumulated temperature at the inflection point of dry matter accumulation. For each type of phenological response parameter, the average value of the parameter is obtained by summing the values ​​of all plots in the set and dividing by the total number of plot records in the set. For each type of phenological response parameter, the absolute value of the difference between each recorded value in the set and the average value is calculated. The average absolute deviation of the parameter is obtained by summing all the absolute deviations and dividing by the total number of records. The average absolute deviation is used as a measure of the dispersion of the parameter. The above average value is output as the central estimate of the phenological response parameter of this microregion.

[0035] Determining the confidence boundaries of micro-region phenological response parameters based on their degree of dispersion includes the following steps: From all the phenological response parameter values ​​contained in the candidate similar plot set of the micro-region, find the maximum and minimum values ​​of each type of phenological response parameter, and use the maximum and minimum values ​​to define the coarse boundary of the value of the type of phenological response parameter; For each type of phenological response parameter, the range of values ​​from minimum to maximum is divided into several smaller intervals at equal intervals. The frequency of phenological response parameter values ​​falling into each interval in the candidate similar plot set is counted. Intervals with frequencies exceeding a preset frequency threshold are marked as dense intervals. The minimum and maximum values ​​among all dense intervals constitute the high-density interval for this type of parameter. The preset frequency threshold is set to 1 / 10 of the total number of records in the candidate similar plot set. However, for sets with less than 30 records, this threshold is reduced to 3 records. The left relaxation distance is obtained by subtracting the minimum value of the coarse boundary from the minimum value of the high-density region; the right relaxation distance is obtained by subtracting the maximum value of the high-density region from the maximum value of the coarse boundary; when determining the confidence boundary, the elastic relaxation amount of the phenological response parameters of each micro-region is calculated using the following formula: ; in, Indicates the first The elastic relaxation amount of the phenological response parameters of each micro-region to the right of the coarse boundary. This represents the maximum value of this type of phenological response parameter in the set of candidate similar plots within the micro-region. This represents the maximum value of this type of phenological response parameter in the high-density interval of the candidate similar plot set in the micro-region. This represents the preset elastic weighting coefficient, which takes a value between 0.3 and 0.7. Correspondingly, the elastic relaxation amount to the left of the coarse boundary is calculated using the same formula, by subtracting the minimum value of the high-density region from the minimum value of the coarse boundary and then multiplying it by the elastic weighting coefficient.

[0036] By multiplying the elastic relaxation amount on the left side of the coarse boundary and the elastic relaxation amount on the right side of the coarse boundary by a preset relaxation reduction coefficient, the left side reduction relaxation amount and the right side reduction relaxation amount are obtained. The relaxation reduction coefficient is a value between 0.4 and 0.6. When calculating the lower confidence boundary, the minimum value of the high-density interval is subtracted from the left-side reduction relaxation to obtain the lower confidence boundary. When calculating the upper confidence boundary, the maximum value of the high-density interval is added to the right-side reduction relaxation to obtain the upper confidence boundary. Based on this, the upper confidence boundary is fine-tuned using the following formula: ; in, Indicates the first The upper confidence boundary of the phenological response parameters of this type in a microregion after fine-tuning Indicates the first The fine-tuning correction value for each micro-region is calculated as follows: multiply the reciprocal of the total number of records in the candidate similar parcel set by the baseline correction constant. The baseline correction constant is 5, that is, when the total number of records in the candidate similar parcel set is N, the fine-tuning correction value is equal to 5 divided by N. The lower confidence boundary is fine-tuned in the same way: the minimum value of the high-density interval is subtracted from the left elastic relaxation value, and then the fine-tuning correction value is subtracted. The lower confidence boundary and the upper confidence boundary after fine-tuning are used as the confidence boundary outputs for this type of phenological response parameter.

[0037] The central estimates of the phenological response parameters corresponding to each micro-region are associated with and stored along with the confidence boundary, specifically implemented as follows: For each microregion, the central estimate, lower confidence boundary, and upper confidence boundary of each type of phenological response parameter in the microregion are stored in the form of triples. The triples of each type of phenological response parameter contained in the microregion together constitute the personalized prior information pool of the microregion. This personalized prior information pool is used as a location reference and scale constraint for the prior distribution of the phenological response parameters of the micro-region in the subsequent hierarchical Bayesian inference steps.

[0038] In S4, the personalized prior information pools of each micro-region are used as prior constraints for their respective phenological response parameters. Combined with the growth response data of each grid, hierarchical Bayesian inference is performed on the phenological response parameters of each micro-region, and the characteristic parameters of each micro-region after local data correction are output, specifically including: For each microregion, an initial distribution of phenological response parameters is set, specifically as follows: The central estimate of the phenological response parameters stored in the personalized prior information pool of each micro-region is used as the location parameter of the initial distribution of the phenological response parameters of that micro-region, and half of the difference between the lower limit and the upper limit of the confidence boundary of the phenological response parameters stored in the personalized prior information pool of that micro-region is used as the initial scale parameter of the initial distribution. The prior distribution of the phenological response parameters of the micro-region is constructed using the location parameter and the initial scale parameter. The prior distribution is a symmetrical distribution centered on the location parameter and measured by the initial scale parameter to determine the dispersion range. When the total number of records of candidate similar plots in a micro-region is less than 10, the initial scale parameter of the micro-region is multiplied by a predetermined expansion coefficient, which is between 1.5 and 2.0. The expanded scale parameter replaces the original initial scale parameter to form the prior distribution of the micro-region.

[0039] Using the growth response data of each grid as the observation input, and according to the prior distribution of the microregion to which each grid belongs, the local likelihood contribution value of the phenological response parameter of each grid under the prior distribution constraint of the microregion to which it belongs is calculated grid by grid. Specifically, it is implemented in the following way: For each management grid, the measured relative chlorophyll content and measured diameter at breast height (DBH) of bamboo in the growth response data of that grid are compared with the location parameters of the prior distribution of the microregion to which the grid belongs. The deviation values ​​of the relative chlorophyll content and the DBH of bamboo are calculated respectively. Specifically, the difference is obtained by subtracting the location parameter from the measured value, and then the difference is divided by the initial scale parameter of the microregion to obtain the initial deviation measure of the grid in the chlorophyll dimension and the DBH dimension. The average of the initial deviation measures of chlorophyll dimension and diameter at breast height dimension is taken as the initial measure of the grid response deviation. The maximum and minimum values ​​of the initial response deviation metric of all grids within the micro-region to which the grid belongs are calculated. The discrete magnitude of the response deviation of all grids in the micro-region is obtained by subtracting the minimum value from the maximum value. The relative measure of the grid response deviation is obtained by subtracting the median value of the discrete amplitude from the initial measure of the grid response deviation, and then dividing by 1 / 4 of the discrete amplitude. The median value of the discrete amplitude is the value in the middle after arranging the initial measures of the response deviation of all grids in the micro-region in terms of numerical magnitude, and 1 / 4 of the discrete amplitude is the discrete amplitude value multiplied by 0.25. In the prior distribution of the micro-region to which the grid belongs, find the prior probability density value at the position corresponding to the relative metric of the response deviation of the grid. Multiply the relative metric of the response deviation by the prior probability density value to obtain the local likelihood contribution value of the grid under the prior distribution constraint of the micro-region to which it belongs. After traversing all management grids of the target operating area, output the local likelihood contribution value corresponding to each grid.

[0040] The local likelihood contributions of all grids within the same micro-region are merged, and the merged joint likelihood value is used to jointly adjust the location parameters and scale constraints of the prior distribution of the micro-region. This is implemented as follows: For each microregion, the local likelihood contribution values ​​of all grids within that microregion are summed to obtain the joint likelihood value of that microregion. Using the joint likelihood value as a weight, the position parameters of the prior distribution of the micro-region are offset and adjusted. Specifically, the difference between the average value of the measured growth response data of all grids in the micro-region and the position parameters of the prior distribution is multiplied by the joint likelihood value to obtain the position offset. The original position parameters of the prior distribution are added to the position offset to obtain the updated position parameters. For adjusting the scale constraints, the scale reduction factor is obtained by multiplying the reciprocal of the joint likelihood value by the square of the initial scale parameter of the micro-region. The updated scale parameter is obtained by subtracting the product of the scale reduction factor and the initial scale parameter from the initial scale parameter. The distribution of phenological response parameters of the microregion is reconstructed using the updated location parameters and the updated scale parameters, and is used as the posterior distribution of the phenological response parameters of the microregion. The location parameters of the posterior distribution are then output as the feature parameters of the microregion after correction by local data.

[0041] In S5, the characteristic parameters of each micro-region are embedded into the basic decision data of each grid, and the differentiated fertilization ratio and dosage are calculated grid by grid based on the soil available nutrient content and canopy spectral reflectance of each grid, generating differentiated fertilization prescriptions with spatial positioning attributes, specifically including: The baseline ratio of nitrogen, phosphorus, and potassium requirements for each grid is determined by combining the available soil nutrients and the canopy spectral reflectance, specifically as follows: For each management grid, the soil available nitrogen content, soil available phosphorus content, and soil available potassium content of that grid are multiplied by preset soil nutrient weighting coefficients to obtain the soil nutrient weighted value of that grid. The weighting coefficient for soil available nitrogen content is 0.4, the weighting coefficient for soil available phosphorus content is 0.3, and the weighting coefficient for soil available potassium content is 0.3. Divide the red edge reflectance value of the canopy spectral reflectance of the grid by the near-infrared reflectance value to obtain the vegetation red edge index of the grid. Then divide the vegetation red edge index by the average value of the vegetation red edge indices of all grids in the micro-region to which the grid belongs to obtain the spectral relative index of the grid. The product of the soil nutrient weighted value of the grid and the spectral relative index of the grid is used as the comprehensive fertilizer requirement index of the grid. The comprehensive fertilizer requirement index is decomposed into nitrogen, phosphorus and potassium contribution components. The ratio of the three contribution components is the benchmark ratio of nitrogen, phosphorus and potassium requirements of the grid. Using the maximum nutrient dilution rate value in the characteristic parameters of the micro-region to which the grid belongs as the amplitude modulation coefficient, the three contribution components in the nitrogen, phosphorus and potassium demand baseline ratio of the grid are multiplied by the amplitude modulation coefficient to obtain the nitrogen element ratio, phosphorus element ratio and potassium element ratio in the initial fertilizer application ratio of the grid. The initial application rate of the grid is obtained by multiplying the total content of nitrogen, phosphorus and potassium in the available soil nutrients of the grid by the preset basic fertilization coefficient. The basic fertilization coefficient is between 0.05 and 0.15. When the total content of available soil nutrients increases by 100 mg per kilogram, the basic fertilization coefficient is reduced by 0.02.

[0042] The average initial fertilization ratio and initial application rate of all grids within the same micro-region are used as the common reference for that micro-region, specifically implemented as follows: For each micro-region, the nitrogen element ratio in the initial fertilization ratio of all grids in the micro-region is added together and then divided by the total number of grids in the micro-region to obtain the common nitrogen ratio of the micro-region. The common phosphorus ratio, common potassium ratio, and common dosage of this microregion were calculated in the same manner. The common nitrogen ratio, common phosphorus ratio, common potassium ratio, and common dosage of this microregion constitute the common reference for this microregion.

[0043] The deviation of the initial fertilization ratio and initial application rate from the common reference of the micro-region is calculated for each grid cell, specifically as follows: For each grid, the initial nitrogen ratio difference of the grid is obtained by subtracting the common nitrogen ratio of the micro-region to which the grid belongs from the initial nitrogen ratio of the grid. The initial differences in phosphorus ratio, potassium ratio, and dosage for this grid were calculated in the same manner. Determine whether the initial difference of the grid in the four dimensions of nitrogen ratio, phosphorus ratio, potassium ratio and dosage is greater than zero. Mark the dimension with the initial difference greater than zero as the positive deviation dimension and the dimension with the initial difference less than zero as the negative deviation dimension. For each dimension marked as a positive deviation dimension, the initial difference of that dimension is divided by the maximum positive initial difference of all grids in that dimension within that micro-region to obtain the positive normalized relative deviation of that dimension on that grid. For each dimension marked as a negative deviation dimension, the absolute value of the initial difference of that dimension is divided by the maximum value of the absolute values ​​of the negative initial differences of all grids in that dimension within that micro-region to obtain the negative normalized relative deviation value of that dimension on that grid; for dimensions with an initial difference of zero, the normalized relative deviation value of that dimension is recorded as zero. The normalized relative deviation values ​​obtained in each dimension are assigned values ​​according to the original positive and negative direction attributes of each dimension. When the initial difference of a dimension is positive, the final deviation of that dimension is positive; when the initial difference of a dimension is negative, the final deviation of that dimension is negative; when the initial difference of a dimension is zero, the final deviation of that dimension is zero. The final deviation of each dimension is output as the final deviation of the grid.

[0044] The initial fertilization ratio and dosage for this grid are reversed using the deviation amount to obtain the secondary adjustment ratio and dosage for this grid. This is achieved in the following way: For each dimension of each grid, the initial difference of that dimension is multiplied by a preset reverse shrinkage coefficient to obtain the shrinkage adjustment amount of that dimension. The reverse shrinkage coefficient is a value between 0.3 and 0.7. Subtract the shrinkage adjustment amount of the dimension from the initial adjustment value of the dimension to obtain the secondary adjustment value of the dimension; If the secondary adjustment value of a certain dimension is less than zero, then the secondary adjustment value of that dimension is reset to 0; the secondary adjustment values ​​obtained from each dimension are combined in the order of nitrogen ratio, phosphorus ratio, potassium ratio and dosage to form the secondary adjustment ratio and dosage of the grid.

[0045] The secondary adjustment ratio and amount of the grid are linked and bound to the geographic coordinates of the grid's spatial location, specifically as follows: For each management grid, the nitrogen, phosphorus, and potassium ratios and the amount of secondary adjustment in the grid are associated and stored with the longitude and latitude of the grid's center point. The longitude and latitude of each grid are used as the spatial positioning attribute of the grid, and the secondary adjustment ratio and amount of secondary adjustment are used as the fertilization attribute of the grid. The spatial positioning attributes and fertilization attributes of all grids are combined to form a differentiated fertilization prescription with spatial positioning attributes.

[0046] Please see Figure 4 As shown, a precision fertilization decision-making system for moso bamboo based on multi-source data fusion includes: The multi-source data acquisition and gridded input module collects soil available nutrient content, canopy spectral reflectance and growth response data for each management grid in the target management area, and obtains environmental covariates for each grid, including accumulated temperature, precipitation coefficient of variation, topographic humidity index and slope aspect. The homogeneous micro-region division module takes the environmental covariates of each grid as input and divides all management grids into several homogeneous micro-regions through fuzzy clustering, so that the environmental covariate characteristics within each micro-region tend to be consistent. The personalized prior information pool construction module uses the centroid of the environmental covariate of each micro-region as the retrieval benchmark, retrieves the records of land parcels in the historical land parcel database whose cosine similarity with each centroid exceeds a preset threshold, extracts the statistical values ​​of phenological response parameters of each land parcel record, and constructs a personalized prior information pool for each micro-region. The hierarchical Bayesian inference and parameter correction module uses the personalized prior information pool of each micro-region as the prior constraint of the phenological response parameters of each micro-region. Combined with the growth response data of each grid, it performs hierarchical Bayesian inference on the phenological response parameters of each micro-region and outputs the feature parameters of each micro-region after local data correction. The differentiated fertilization prescription generation module embeds the characteristic parameters of each micro-region into the basic decision data of each grid, and calculates the differentiated fertilization ratio and dosage grid by grid based on the soil available nutrient content and canopy spectral reflectance of each grid, generating a differentiated fertilization prescription with spatial positioning attributes.

[0047] The working principle of this invention is as follows: Soil available nutrient content, canopy spectral reflectance, and growth response data are collected from each management grid within the target management area. Environmental covariates for each grid are obtained, including accumulated temperature, precipitation coefficient of variation, topographic humidity index, and slope and aspect. Using these environmental covariates as input, fuzzy clustering is used to divide all management grids into several environmentally homogeneous micro-regions. Using the centroids of the environmental covariates of each micro-region as the retrieval benchmark, records of plots with a cosine similarity exceeding a preset threshold to each centroid are retrieved from the historical plot database. Phenological responses from each plot record are then extracted. Based on the statistical values ​​of the parameters, a personalized prior information pool is constructed for each micro-region. This pool serves as the prior constraint for the phenological response parameters of each micro-region. Combined with the growth response data of each grid, hierarchical Bayesian inference is performed on the phenological response parameters of each micro-region, outputting the characteristic parameters of each micro-region after correction using local data. The characteristic parameters of each micro-region are embedded into the basic decision data of each grid. Based on the available soil nutrient content and canopy spectral reflectance of each grid, differentiated fertilization ratios and amounts are calculated grid-by-grid, generating differentiated fertilization prescriptions with spatial positioning attributes. This method achieves spatial differentiation and precision in fertilization decisions for bamboo forests by fusing multi-source data.

[0048] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for refined fertilization decision-making for moso bamboo based on multi-source data fusion, characterized in that, Includes the following steps: S1. Collect soil available nutrient content, canopy spectral reflectance and growth response data for each management grid in the target management area, and obtain environmental covariates for each grid, including accumulated temperature, precipitation coefficient of variation, topographic humidity index and slope aspect. S2 takes the environmental covariates of each grid as input and divides all management grids into several environmentally homogeneous micro-regions through fuzzy clustering, so that the characteristics of environmental covariates within each micro-region tend to be consistent. S3, using the centroid of the environmental covariate of each micro-region as the retrieval benchmark, retrieves the records of plots in the historical plot database whose cosine similarity with each centroid exceeds a preset threshold, extracts the statistical values ​​of the phenological response parameters of each plot record, and constructs a personalized prior information pool for each micro-region. S4. The personalized prior information pool of each micro-region is used as the prior constraint of the phenological response parameters of each micro-region. Combined with the growth response data of each grid, hierarchical Bayesian inference is performed on the phenological response parameters of each micro-region, and the characteristic parameters of each micro-region after local data correction are output. S5 embeds the characteristic parameters of each micro-region into the basic decision data of each grid, and calculates the differentiated fertilization ratio and dosage grid by grid based on the soil available nutrient content and canopy spectral reflectance of each grid, generating a differentiated fertilization prescription with spatial positioning attributes.

2. The method for refined fertilization decision-making for moso bamboo based on multi-source data fusion according to claim 1, characterized in that, The process of dividing the entire management grid into several homogeneous micro-regions through fuzzy clustering specifically includes: The range normalization process is performed on each environmental covariate vector according to its feature dimension to unify the value range of each dimension to a preset interval; Using the Euclidean distance between the environmental covariate vectors of each grid after normalization as the difference measure, and based on the preset number of micro-regions, fuzzy membership degree assignment is performed on each grid, and the membership degree value of each grid belonging to each candidate micro-region is calculated one by one. Based on the membership degree values ​​of each grid to each candidate microregion, each grid is assigned to the candidate microregion with the largest membership degree value, thus completing the initial division of all grids; After the initial division, the intra-class average distance of all grid environmental covariate vectors in each micro-region is calculated. Grids with intra-class average distances exceeding a preset threshold are marked. Then, the corresponding grids are repositioned based on the environmental covariate similarity with the centroid of each micro-region until the intra-class average distance of all micro-regions is lower than the preset threshold. Finally, the environmental homogeneous micro-region division result is output.

3. The method for refined fertilization decision-making for moso bamboo based on multi-source data fusion according to claim 2, characterized in that, The step of calculating the membership value of each grid cell belonging to each candidate micro-region specifically includes: Using the centroid of the environmental covariate vector of each candidate microregion as a reference, the Euclidean distance between the environmental covariate vector and the centroid of each grid is calculated one by one, and the grids are arranged in order of proximity to each candidate microregion from the closest to the furthest. Using the Euclidean distance between the grid and the centroid of each candidate microregion as the independent variable and the initial influence intensity of the grid on each candidate microregion as the dependent variable, a set of initial influence intensity values ​​of the grid on each candidate microregion are determined according to the principle that the closer the distance, the greater the influence intensity. The initial influence intensity value of the grid on each candidate microregion is compressed proportionally, so that the sum of the initial influence intensity values ​​of the grid on all candidate microregions is normalized to a fixed value, and the membership value of the grid to each candidate microregion is obtained. After traversing all grids, output the set of membership values ​​of each candidate microregion corresponding to each grid, so that the grid can be assigned to the candidate microregion with the largest membership value in the subsequent process.

4. The method for refined fertilization decision-making for moso bamboo based on multi-source data fusion according to claim 1, characterized in that, The construction of personalized prior information pools for each micro-region specifically includes: Using the centroid of the environmental covariate of each micro-region as the retrieval vector, the cosine similarity between the environmental covariate record vector of each plot in the historical plot database and the retrieval vector is calculated one by one. All plot records with a cosine similarity exceeding a preset threshold are selected to form a candidate similar plot set for the micro-region. The average value and dispersion of phenological response parameters recorded by each parcel in the candidate similar parcel set of the micro-region are calculated separately. The average value is used as the central estimate of the phenological response parameters of the micro-region, and the dispersion is used to determine the confidence boundary of the phenological response parameters of the micro-region. The central estimates of the phenological response parameters corresponding to each micro-region are associated with and stored with the confidence boundary, forming a personalized prior information pool for each micro-region for subsequent hierarchical Bayesian inference.

5. The method for refined fertilization decision-making for moso bamboo based on multi-source data fusion according to claim 4, characterized in that, The determination of the confidence boundary for the micro-region phenological response parameters based on the degree of dispersion specifically includes: The maximum and minimum values ​​of phenological response parameters are extracted from the candidate similar plots in the micro-region, and the maximum and minimum values ​​are used to define the coarse boundary of the phenological response parameter values ​​in the micro-region. The distribution frequency of phenological response parameter values ​​in each value interval of the candidate similar plot set in the micro-region is statistically analyzed. Value intervals with distribution frequencies exceeding a preset frequency threshold are considered as densely distributed intervals. The minimum and maximum values ​​of the densely distributed intervals are used as high-density intervals for the phenological response parameter values ​​in the micro-region. The high-density region is used as the narrow boundary, and the difference between the coarse boundary and the narrow boundary is used as the elastic relaxation amount of the confidence boundary of the micro-region phenological response parameter. The narrow boundary and the elastic relaxation amount are superimposed as the confidence boundary output of the micro-region phenological response parameter.

6. The method for refined fertilization decision-making for moso bamboo based on multi-source data fusion according to claim 1, characterized in that, The hierarchical Bayesian inference of the phenological response parameters of each micro-region specifically includes: An initial distribution of phenological response parameters for each micro-region is set, with the central estimate in the micro-region's personalized prior information pool used as the location parameter of the initial distribution, and the confidence boundary of the micro-region used as the scale constraint of the initial distribution, thus forming the prior distribution of the micro-region. Using the growth response data of each grid as the observation input, the local likelihood contribution value of the phenological response parameters of each grid under the constraint of the prior distribution of the micro-region to which each grid belongs is calculated grid by grid according to the prior distribution of the micro-region to which each grid belongs; The local likelihood contribution values ​​of all grids within the same micro-region are merged, and the combined likelihood value is used to jointly adjust the location parameters and scale constraints of the prior distribution of the micro-region to obtain the posterior distribution of the phenological response parameters of the micro-region. The location parameters of the posterior distribution are then output as the characteristic parameters of the micro-region after correction by local data.

7. The method for refined fertilization decision-making for moso bamboo based on multi-source data fusion according to claim 6, characterized in that, The calculation process for the local likelihood contribution value is as follows: The degree of deviation between the measured response value in the growth response data of the grid and the theoretically expected position of the prior distribution of the micro-region is used as the initial measure of the grid response deviation. The initial measure of grid response bias is scaled according to the discrete magnitude of the observed response values ​​of all grids within the microregion to which the grid belongs, to obtain a relative measure of grid response bias; The local likelihood contribution value of the grid under the prior distribution constraint of its microregion is obtained by multiplying the relative measure of the grid response deviation with the prior probability density of the corresponding position within the confidence boundary of the microregion to which the grid belongs. After completing the traversal of all grids, the local likelihood contribution value of each grid is output.

8. The method for refined fertilization decision-making for moso bamboo based on multi-source data fusion according to claim 1, characterized in that, S5 specifically includes: The nitrogen, phosphorus and potassium requirement baseline ratios for each grid are determined by combining the available soil nutrients content and canopy spectral reflectance. The amplitude of the requirement baseline ratios is modulated by the characteristic parameters of the micro-region to which the grid belongs, and the initial fertilizer application ratio and initial application amount for the grid are obtained. The average initial fertilization ratio and initial application rate of all grids within the same micro-region are used as the common reference of the micro-region. The deviation of the initial fertilization ratio and initial application rate of each grid from the common reference of the micro-region is calculated one by one. The initial fertilization ratio and initial application rate of the grid are reversed by the deviation to obtain the secondary adjustment ratio and secondary adjustment amount of the grid. The secondary adjustment ratio and dosage of the grid are linked and bound to the geographical coordinates of the grid's spatial location to generate a differentiated fertilization prescription with spatial positioning attributes.

9. A method for refined fertilization decision-making for moso bamboo based on multi-source data fusion according to claim 8, characterized in that, The calculation of the deviation of the initial fertilization ratio and initial application rate from the micro-region common reference for each grid specifically includes: The initial differences between the grids in terms of ratio and dosage are obtained by subtracting the common ratio and dosage in the common reference of the micro-region to which the grid belongs from the initial fertilization ratio and initial dosage of each grid. Determine the sign of the initial difference between the grids in the proportion and dosage dimensions, classify the grids into positive or negative deviation groups according to the sign, and calculate the absolute magnitude of the initial difference of all grids in each group. Divide the absolute magnitude of the initial difference of the grid by the maximum absolute magnitude within the group to obtain the normalized relative deviation value of the grid in the proportion dimension and the usage dimension. Then, assign the direction attribute of the normalized relative deviation value with the positive or negative sign of the grid, and output the final deviation of the grid with the direction attribute.

10. A precision fertilization decision-making system for moso bamboo based on multi-source data fusion, characterized in that, A method for implementing a refined fertilization decision-making method for moso bamboo based on multi-source data fusion as described in any one of claims 1-9, comprising: The multi-source data acquisition and gridded input module collects soil available nutrient content, canopy spectral reflectance and growth response data for each management grid in the target management area, and obtains environmental covariates for each grid, including accumulated temperature, precipitation coefficient of variation, topographic humidity index and slope aspect. The homogeneous micro-region division module takes the environmental covariates of each grid as input and divides all management grids into several homogeneous micro-regions through fuzzy clustering, so that the environmental covariate characteristics within each micro-region tend to be consistent. The personalized prior information pool construction module uses the centroid of the environmental covariate of each micro-region as the retrieval benchmark, retrieves the records of land parcels in the historical land parcel database whose cosine similarity with each centroid exceeds a preset threshold, extracts the statistical values ​​of phenological response parameters of each land parcel record, and constructs a personalized prior information pool for each micro-region. The hierarchical Bayesian inference and parameter correction module uses the personalized prior information pool of each micro-region as the prior constraint of the phenological response parameters of each micro-region. Combined with the growth response data of each grid, it performs hierarchical Bayesian inference on the phenological response parameters of each micro-region and outputs the feature parameters of each micro-region after local data correction. The differentiated fertilization prescription generation module embeds the characteristic parameters of each micro-region into the basic decision data of each grid, and calculates the differentiated fertilization ratio and dosage grid by grid based on the soil available nutrient content and canopy spectral reflectance of each grid, generating a differentiated fertilization prescription with spatial positioning attributes.

Images