Rice nitrogen fertilizer application amount prediction method

Abstract

The invention relates to a rice nitrogen fertilizer application amount prediction method, which comprises the following steps: firstly, constructing each rice test data sample based on multi-point nitrogen fertilizer gradient test data, then constructing a secondary mixing effect yield response model containing a fixed effect coefficient and a random effect item, and fitting to obtain a yield-nitrogen application amount response model; then, based on a yield-nitrogen application amount response model and an economic objective function, solving the highest yield nitrogen application amount and the economic optimal nitrogen application amount of each test point, then, carrying out Monte Carlo resampling and training to obtain a highest yield nitrogen application model group and an economic optimal nitrogen application model group, and finally, aiming at each position of a target to-be-tested area, carrying out optimal nitrogen application on each position of the target to-be-tested area. According to the method, the nitrogen application amount related to the highest yield target and the nitrogen application amount related to the economic optimal target are determined, then the high-resolution regional optimal nitrogen application map is formed, the design scheme has both the interpretability of an empirical model and the generalization ability of machine learning, and the decision precision of farmland nitrogen fertilizer management recommendation can be remarkably improved.

Description

Technical Field

[0001] This invention relates to a method for predicting nitrogen fertilizer application rates in rice, belonging to the fields of farmland nitrogen fertilizer management, crop models, and precision agriculture big data modeling technology. Background Technology

[0002] Currently, nitrogen fertilizer application rates in farmland largely rely on empirical recommendations or linear / quadratic response models based on single-point experiments. These recommendations lack spatiotemporal extrapolation capabilities and fail to reflect the differences in climate-soil-variety interactions, leading to under- or over-application of nitrogen fertilizer, thereby reducing nitrogen fertilizer utilization efficiency (NUE) and economic returns. While existing machine learning methods can perform spatial interpolation predictions, they often lack a clear economic objective function, making it difficult to use "yield revenue minus nitrogen fertilizer cost" as a direct optimization indicator. Therefore, there is an urgent need for a coupled prediction technology that centers on the economically optimal nitrogen application objective and combines interpretability (empirical quadratic response) with generalization ability (machine model). Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a method for predicting the amount of nitrogen fertilizer applied to rice, which can effectively improve the efficiency of nitrogen application, farmers' economic benefits, and the efficiency of nitrogen fertilizer resource utilization, and realize the engineering implementation and policy promotion of optimized nitrogen application in rice under the conditions of precision agriculture.

[0004] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention designs a method for predicting nitrogen fertilizer application rate in rice, comprising the following steps:

[0005] Step A. Based on the analysis time of each test site, collect field yield data of nitrogen fertilizer gradient at multiple points, and simultaneously collect preset soil physicochemical property data and preset climate characteristic data of each test site to construct data samples for each rice experiment, and then proceed to Step B;

[0006] Step B. Construct a quadratic mixed-effect yield response model that includes fixed-effects coefficients and random-effects terms, and fit the yield-nitrogen application rate response model based on the data samples from each rice experiment, and then proceed to step C;

[0007] Step C. Based on the yield-nitrogen application rate response model, solve for the highest yield nitrogen application rate corresponding to each test point, and construct the economic objective function to solve for the economically optimal nitrogen application rate corresponding to each test point, and then proceed to step D;

[0008] Step D. Based on each test site, Monte Carlo resampling is used. According to the soil physicochemical properties data and climate characteristics data of the test sites, the highest yield nitrogen application rate and the economically optimal nitrogen application rate of the test sites are used as target values. The network to be trained is trained separately to obtain the highest yield nitrogen application model group and the economically optimal nitrogen application model group. Then proceed to step E.

[0009] Step E. Based on the highest yield nitrogen application model group and the economically optimal nitrogen application model group, calculate and determine the nitrogen application amount for each location to be tested in the target area with respect to the highest yield target and the nitrogen application amount with respect to the economically optimal target.

[0010] As a preferred technical solution of the present invention: step A includes the following steps A1 to A2;

[0011] Step A1. Based on a preset number of test points, and each test point includes rice test areas with preset nitrogen application rates, for each analysis time, the nitrogen application rate and rice yield of each rice test area in each test point corresponding to the analysis time are statistically analyzed, and preset soil physicochemical property data and preset climate characteristic data of each test point corresponding to the analysis time are collected, and then proceed to step A2.

[0012] Step A2. Based on each rice experimental area and each analysis time in each experimental site, construct rice experimental data samples using the nitrogen application rate, rice yield, and preset soil physicochemical properties and climate characteristics data of the corresponding analysis time for each rice experimental area, and then obtain each rice experimental data sample.

[0013] As a preferred technical solution of the present invention: step A further includes the following steps A3 to A5, after step A2 is completed, step A3 is entered;

[0014] Step A3. For each rice experiment data sample, if there are missing values, use the inverse distance weighted interpolation method to fill in the missing values, update the rice experiment data sample, and then proceed to step A4;

[0015] Step A4. Based on the data samples from each rice experiment, determine the normal value range for each target data type, including nitrogen application rate, rice yield, preset soil physicochemical properties data, and preset climate characteristic data, and then proceed to step A5.

[0016] Step A5. For each rice experiment data sample, determine whether there are any outliers in the data values ​​of each target data type corresponding to the rice experiment data sample that are outside the corresponding normal value range. If so, delete the rice experiment data sample; otherwise, do not perform any further processing.

[0017] As a preferred technical solution of the present invention: In step A4, based on each rice test data sample, for each target data type of nitrogen application rate, rice yield, preset soil physicochemical property data, and preset climate characteristic data, the following steps A4-1 to A4-2 are performed;

[0018] Step A4-1. Sort the data values ​​of the target data type in ascending order, and obtain the data value that is in the 25th position of the sort. And to obtain the data value that is in the 75th position of the sorted list. Then press Calculate the interquartile range Then proceed to step A4-2;

[0019] Step A4-2. Calculation The result constitutes the lower limit of the normal numerical range corresponding to the target data type, and the calculation The result constitutes the upper limit of the normal numerical range corresponding to the target data type, and thus constitutes the normal numerical range corresponding to the target data type.

[0020] As a preferred technical solution of the present invention: step B includes the following steps B1 to B2;

[0021] Step B1. Construct a quadratic mixed-effects output response model including fixed-effects coefficients and random-effects terms as follows:

[0022] ;

[0023] in, , Indicates the number of test sites. , Indicates the first The number of rice experimental areas in each experimental site Indicates the first The first test site The rice experimental area corresponds to the first Rice yield at the time of analysis Indicates the first The first test site The rice experimental area corresponds to the first Nitrogen application rate at each analysis time point , , All are fixed effects coefficients to be trained. Indicates baseline rice yield. This represents the slope of the linear term of the nitrogen fertilizer effect. This represents the quadratic effect of the linear effect of nitrogen fertilizer. , , All represent random effects to be trained. Indicates the first Random intercepts at each test point Follows a normal distribution , Indicates the random effects to be trained variance Indicates the first The random slope of each test point Follows a normal distribution , Indicates the random effects to be trained variance Indicates the first The test point corresponds to the first The random intercept of each analysis time period, Follows a normal distribution , Indicates the random effects to be trained variance Indicates the first The first test site The rice experimental area corresponds to the first The analysis time follows a normal distribution. random error, express The variance; then proceed to step B2;

[0024] Step B2. Based on the data samples from each rice experiment, use the R language... In the bag The function is fitted to the quadratic mixed-effects output response model using the restricted maximum likelihood estimation method to obtain the fixed effects coefficients. , , The value of the random effect at each test site , The value of the random effect at each experimental point corresponding to each analysis time. The value of, and the variance of random effects , , , The value of is obtained, that is, the yield-nitrogen application rate response model.

[0025] As a preferred technical solution of the present invention: step C includes the following steps C1 to C2;

[0026] Step C1. Eliminate yield-nitrogen application rate response models for the test sites. and The expected rice yield model is as follows, where, Indicates the first Nitrogen application rate at each test site Indicates the first Rice yield at each experimental site express Expected value;

[0027]

[0028] Execute the expected rice yield model with respect to nitrogen application rate The first derivative of , and set the derivative to zero:

[0029]

[0030] Determine the maximum nitrogen application rate as follows:

[0031]

[0032] Then, based on step B, the yield-nitrogen application rate response model is obtained. and The value is determined for each test point, based on the corresponding value of the test point. The nitrogen application rate corresponding to the highest yield at the test site was calculated. , Indicates the first The highest yield nitrogen application rate corresponding to each test point is obtained, and then the highest yield nitrogen application rate corresponding to each test point is obtained; then proceed to step C2;

[0033] Step C2. Construct the economic benefit function for the test sites. As follows, among which, This indicates the unit price of rice. This indicates the unit price of nitrogen fertilizer;

[0034]

[0035] Execute the economic return function Regarding nitrogen application rate The first derivative of , and set the derivative to zero:

[0036]

[0037] Will Substituting into the above formula, the update is as follows:

[0038]

[0039] Determine the economically optimal nitrogen application rate as follows:

[0040]

[0041] Then, based on step B, the yield-nitrogen application rate response model is obtained. and The value is determined for each test point, based on the corresponding value of the test point. The economically optimal nitrogen application rate corresponding to the test site was calculated. , Indicates the first The economically optimal nitrogen application rate corresponding to each test site is determined, and then the economically optimal nitrogen application rate corresponding to each test site is obtained.

[0042] As a preferred embodiment of the present invention: In step D, the nitrogen application rate with the highest yield and the economically optimal nitrogen application rate are used as target values, respectively, and the total number of Monte Carlo resampling times is calculated based on a preset target. ,initialization Perform steps D1 to D4 to obtain the nitrogen application model group corresponding to the target amount, namely the highest yield nitrogen application model group and the economically optimal nitrogen application model group.

[0043] Step D1. For each test point, calculate the average value of the preset soil physicochemical properties data corresponding to each analysis time, and construct the average value of each soil physicochemical property corresponding to the test point. At the same time, calculate the average value of the preset climate characteristic data corresponding to each analysis time, and construct the average value of each climate characteristic corresponding to the test point. Then proceed to step D2.

[0044] Step D2. Based on each test point, use the Bootstrap resampling method to sample test points with replacement, and construct the first... Sub-sample subset, and proceed to step D3;

[0045] Step D3. Based on the first The subsample, with the mean values ​​of soil physicochemical properties and climate characteristics of each experimental point as input and the target quantity of each experimental point as output, is used to train the network to be trained, resulting in the trained network, which constitutes the first subsample corresponding to the target quantity. First, select a nitrogen application model, and then proceed to step D4;

[0046] Step D4. Judgment Is it equal to That is, obtaining the target quantity corresponding to One nitrogen application model constitutes the nitrogen application model group corresponding to the target amount; otherwise, for... The value is incremented by 1 and the process returns to step D1.

[0047] As a preferred embodiment of the present invention: in step D, when the nitrogen application rate with the highest yield and the economically optimal nitrogen application rate are respectively used as target rates to obtain the nitrogen application model group corresponding to the target rate, each time steps D1 to D3 are executed, the nitrogen application model group corresponding to the target rate is obtained. After implementing the nitrogen application model, the increase in mean square error for each soil physicochemical property and each climate characteristic was calculated, and then... After executing steps D1 to D3, the corresponding data for each soil physicochemical property and each climate characteristic are obtained. The mean square error was increased by each time; then, based on the mean square error increase, the data were sorted from largest to smallest, and the data on each soil physicochemical property and each climate characteristic were sorted to represent the contribution of each soil physicochemical property and each climate characteristic to the output of the nitrogen application model group corresponding to the corresponding target quantity from largest to smallest.

[0048] Then, determine each target data in any of the following ways, and then execute step D. Based on each test point, resample through Monte Carlo. According to each target data of the test point, use the highest yield nitrogen application rate and the economically optimal nitrogen application rate of the test point as the target values, and train the network to be trained respectively to obtain the highest yield nitrogen application model group and the economically optimal nitrogen application model group.

[0049] Method 1: The average contribution of each soil physicochemical property data and each climate characteristic data is used to form a contribution threshold. Data that is less than the contribution threshold in each soil physicochemical property data and each climate characteristic data are removed, and the remaining data constitute each target data.

[0050] Method 2: Based on the soil physicochemical properties data and climate characteristic data, sort them from largest to smallest contribution, remove the data after the preset percentile in sequence, and the remaining data constitute the target data.

[0051] As a preferred technical solution of the present invention: Step E includes obtaining preset soil physicochemical property data and preset climate characteristic data for each test location in the target test area, applying each nitrogen application model in the highest yield nitrogen application model group to obtain the highest yield nitrogen application amount corresponding to each nitrogen application model for each test location, and calculating the average value to form the nitrogen application amount of the test location with respect to the highest yield target, thereby obtaining the nitrogen application amount of each test location in the target test area with respect to the highest yield target;

[0052] Simultaneously, for each detection location in the target area, preset soil physicochemical property data and preset climate characteristic data are obtained for each detection location. Then, each nitrogen application model in the economically optimal nitrogen application model group is applied to obtain the economically optimal nitrogen application amount for each nitrogen application model corresponding to the detection location, and the average value is calculated to form the nitrogen application amount of the detection location with respect to the economically optimal target. Thus, the nitrogen application amount of each detection location in the target area with respect to the economically optimal target is obtained.

[0053] As a preferred technical solution of the present invention: In step E, after applying each nitrogen application model in the highest yield nitrogen application model group to obtain the highest yield nitrogen application amount corresponding to each nitrogen application model at the position to be detected, the method further includes sorting each highest yield nitrogen application amount from smallest to largest, obtaining the highest yield nitrogen application amount ranked at the 2.5% position to form the lower limit of the confidence interval, obtaining the highest yield nitrogen application amount ranked at the 97.5% position to form the upper limit of the confidence interval, and thus forming the confidence interval corresponding to the highest yield nitrogen application.

[0054] Similarly, by applying each nitrogen application model in the economically optimal nitrogen application model group, after obtaining the economically optimal nitrogen application amount corresponding to each nitrogen application model for the location to be detected, the process also includes sorting each economically optimal nitrogen application amount from smallest to largest, obtaining the economically optimal nitrogen application amount ranked at the 2.5% position to form the lower limit of the confidence interval, obtaining the economically optimal nitrogen application amount ranked at the 97.5% position to form the upper limit of the confidence interval, and thus forming the confidence interval corresponding to the economically optimal nitrogen application.

[0055] The method for predicting nitrogen fertilizer application rate in rice described in this invention, compared with existing technologies, has the following technical advantages:

[0056] (1) This invention designs a method for predicting nitrogen fertilizer application rate in rice. First, based on multi-point nitrogen fertilizer gradient test data, a sample of rice test data is constructed. Then, a quadratic mixed effect yield response model containing fixed effect coefficients and random effect terms is constructed and fitted to obtain a yield-nitrogen application rate response model. Then, based on the yield-nitrogen application rate response model and the economic objective function, the highest yield nitrogen application rate and the economically optimal nitrogen application rate at each test point are solved. Then, through Monte Carlo resampling, the highest yield nitrogen application model group and the economically optimal nitrogen application model group are trained. Finally, for each location in the target test area, the nitrogen application rate for the highest yield target and the nitrogen application rate for the economically optimal target are determined, thereby forming a high-resolution regional optimal nitrogen application map. The design scheme of this invention combines the interpretability of empirical models and the generalization ability of machine learning, which can significantly improve the decision accuracy of farmland nitrogen fertilizer management recommendations and provide quantitative, regional, and interannually variable economically optimal nitrogen application guidance for rice production. It has significant agricultural production application value and promotion potential.

[0057] (2) This invention designs a method for predicting nitrogen fertilizer application rate in rice. It considers the nitrogen application rate for maximum yield and the economically optimal nitrogen application rate, and then combines machine learning to extrapolate the model. This can refine the nitrogen application recommendation to the field level, rather than the regional average, which significantly improves the accuracy of nitrogen application and the precision of field-level fertilization. Furthermore, for the economically optimal nitrogen application rate, the "economic fertilization theory" is used for constraint, and the fixed effect and random effect decomposition are retained in the model, so that the model has a clear agronomic meaning, rather than a pure statistical black box, and realizes an "interpretable + generalizable" fertilization decision-making mode. Attached Figure Description

[0058] Figure 1 This is a flowchart illustrating the method for predicting nitrogen fertilizer application rate in rice according to the present invention.

[0059] Figure 2 This is a box plot showing the distribution of the target nitrogen application rate obtained from solving the solution at each test site in the application embodiment of this invention.

[0060] Figure 3 This is a ranking map of feature variable importance calculated using a random forest model in an application embodiment of the present invention.

[0061] Figure 4 This is an exemplary spatial distribution map obtained by spatially predicting the target nitrogen application rate under different future climate scenarios in the design and application embodiments of the present invention. Detailed Implementation

[0062] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0063] This invention designs a method for predicting nitrogen fertilizer application rates in rice. In practical applications, such as... Figure 1 As shown, the specific design process involves the following steps:

[0064] Step A. Based on the analysis time of each test site, collect field yield data of nitrogen fertilizer gradient at multiple points, and simultaneously collect preset soil physicochemical property data and preset climate characteristic data of each test site to construct data samples for each rice experiment, and then proceed to Step B.

[0065] In practical applications, the above step A is specifically designed to be executed as follows: steps A1 to A5.

[0066] Step A1. Based on a predetermined number of test sites, each of which includes rice test areas with different predetermined nitrogen application rates, for each analysis time, statistically analyze the nitrogen application rate and rice yield of each rice test area at each test site for the corresponding analysis time. Also, collect predetermined soil physicochemical property data and predetermined climate characteristic data for each test site at the corresponding analysis time. Then proceed to Step A2. The predetermined soil physicochemical property data includes at least one of pH, soil organic matter, total nitrogen content, total phosphorus, texture, and cation exchange capacity; the predetermined climate characteristic data includes at least one of daily rainfall, daily temperature, accumulated temperature ≥10℃, radiation, and evapotranspiration.

[0067] Step A2. Based on each rice experimental area and each analysis time in each experimental site, construct rice experimental data samples using the nitrogen application rate, rice yield, and preset soil physicochemical properties and climate characteristics data of the corresponding analysis time for each rice experimental area, and then proceed to step A3.

[0068] Step A3. For each rice experiment data sample, if there are missing values, use the inverse distance weighted interpolation method in the "sp" package of R software to fill in the missing values, update the rice experiment data sample, and then proceed to step A4.

[0069] Step A4. Based on the data samples from each rice experiment, for each target data type of nitrogen application rate, rice yield, preset soil physicochemical properties data, and preset climate characteristics data, perform the following steps A4-1 to A4-2 to determine the normal value range under the target data type, and then proceed to step A5.

[0070] Step A4-1. Sort the data values ​​of the target data type in ascending order, and obtain the data value that is in the 25th position of the sort. And to obtain the data value that is in the 75th position of the sorted list. Then press Calculate the interquartile range Then proceed to step A4-2.

[0071] Step A4-2. Calculation The result constitutes the lower limit of the normal numerical range corresponding to the target data type, and the calculation The result constitutes the upper limit of the normal numerical range corresponding to the target data type, and thus constitutes the normal numerical range corresponding to the target data type.

[0072] Step A5. For each rice experiment data sample, determine whether there are any outliers in the data values ​​of each target data type corresponding to the rice experiment data sample that are outside the corresponding normal value range. If so, delete the rice experiment data sample; otherwise, do not perform any further processing.

[0073] Step B. Following steps B1 to B2, construct a quadratic mixed-effect yield response model that includes fixed-effect coefficients and random-effect terms, and fit the yield-nitrogen application rate response model based on the data samples from each rice experiment, and then proceed to step C.

[0074] Step B1. Construct a quadratic mixed-effects output response model including fixed-effects coefficients and random-effects terms as follows:

[0075] ;

[0076] in, , Indicates the number of test sites. , Indicates the first The number of rice experimental areas in each experimental site Indicates the first The first test site The rice experimental area corresponds to the first Rice yield at the time of analysis Indicates the first The first test site The rice experimental area corresponds to the first Nitrogen application rate at each analysis time point , , All are fixed effects coefficients to be trained. Indicates baseline rice yield. This represents the slope of the linear term of the nitrogen fertilizer effect. This represents the quadratic effect of the linear effect of nitrogen fertilizer. , , All represent random effects to be trained. Indicates the first Random intercepts at each test point Follows a normal distribution , Indicates the random effects to be trained variance Indicates the first The random slope of each test point Follows a normal distribution , Indicates the random effects to be trained variance Indicates the first The test point corresponds to the first The random intercept of each analysis time period, Follows a normal distribution , Indicates the random effects to be trained variance Indicates the first The first test site The rice experimental area corresponds to the first The analysis time follows a normal distribution. random error, express The variance; then proceed to step B2.

[0077] Step B2. Based on the data samples from each rice experiment, use the R language... In the bag The function is fitted to the quadratic mixed-effects output response model using the restricted maximum likelihood estimation method to obtain the fixed effects coefficients. , , The value of the random effect at each test site , The value of the random effect at each experimental point corresponding to each analysis time. The value of, and the variance of random effects , , , The value of is obtained, that is, the yield-nitrogen application rate response model.

[0078] Further analysis based on the above design reveals the following model, which only includes the fixed effects component in the yield-nitrogen application response model:

[0079]

[0080] By substituting the nitrogen application rates from each rice experiment data sample, the rice yield under each corresponding single fixed effect was obtained. And further obtain the corresponding variance as the fixed effects variance. .

[0081] The following model is based on the yield-nitrogen application rate response model, which only includes the random effects component:

[0082]

[0083] Substitute the nitrogen application rate from each rice experiment data sample, and the random effect at each analysis time for each experimental site. The values ​​were used to obtain the rice yield under each single random effect. And further obtain the corresponding variance as the variance of random effects. .

[0084] Further, follow the formula below:

[0085] ;

[0086] ;

[0087] get , constitute the margin and obtain Constituting conditions .

[0088] If the condition A high value (e.g., > 0.6) indicates that the obtained yield-nitrogen application response model can capture the variation in the data well, and the yield-nitrogen application response model is effective. We can then confidently proceed to step C to solve for the nitrogen application target.

[0089] If the condition If the value is very low, it indicates that the obtained yield-nitrogen application response model fails to effectively explain the yield change and may have problems (such as poor data quality or incorrect model structure). In this case, it is necessary to go back for inspection or adjustment, such as re-examining the data preprocessing in step A, or modifying the structure of the secondary mixed effect yield response model in step B1 (such as adding or removing random effect terms), rather than directly using an unreliable model for subsequent calculations.

[0090] By comparing margins and conditions This allows for the quantification of the importance of random effects. A significant difference between the two indicates that the variations in experimental location and analysis time have a substantial impact on rice yield. This, in turn, demonstrates the necessity and superiority of the quadratic mixed-effects yield response model (rather than a conventional fixed-effects model) used in this invention.

[0091] The yield response model for a quadratic mixed effect constructed by this invention has the following conditions. It reached 0.72, far exceeding the marginal value of using only the fixed effects model. (0.41), which proves that the design method of the present invention can significantly improve the explanatory power of yield variation, thus laying a solid foundation for accurate recommendation. The application of the final yield-nitrogen application response model can improve the field-level fertilization accuracy and has both interpretability and generalization ability.

[0092] Step C. Following steps C1 to C2 below, based on the yield-nitrogen application rate response model, solve for the highest yield nitrogen application rate YHN (Yield Highest N rate) corresponding to each test point, and construct the economic objective function to solve for the economically optimal nitrogen application rate ONR (Optimal N Rate for economic return) corresponding to each test point, and then proceed to step D.

[0093] Step C1. Eliminate yield-nitrogen application rate response models for the test sites. and The expected rice yield model is as follows, where, Indicates the first Nitrogen application rate at each test site Indicates the first Rice yield at each experimental site express Expected value;

[0094]

[0095] Execute the expected rice yield model with respect to nitrogen application rate The first derivative of , and set the derivative to zero:

[0096]

[0097] Determine the maximum nitrogen application rate as follows:

[0098]

[0099] Then, based on step B, the yield-nitrogen application rate response model is obtained. and The value is determined for each test point, based on the corresponding value of the test point. The nitrogen application rate corresponding to the highest yield at the test site was calculated. , Indicates the first The highest yield nitrogen application rate corresponding to each test point is obtained, and then the highest yield nitrogen application rate corresponding to each test point is obtained; then proceed to step C2.

[0100] Step C2. Construct the economic benefit function for the test sites. As follows, among which, This indicates the unit price of rice. This indicates the unit price of nitrogen fertilizer;

[0101]

[0102] Execute the economic return function Regarding nitrogen application rate The first derivative of , and set the derivative to zero:

[0103]

[0104] Will Substituting into the above formula, the update is as follows:

[0105]

[0106] Determine the economically optimal nitrogen application rate as follows:

[0107]

[0108] Then, based on step B, the yield-nitrogen application rate response model is obtained. and The value is determined for each test point, based on the corresponding value of the test point. The economically optimal nitrogen application rate corresponding to the test site was calculated. , Indicates the first The economically optimal nitrogen application rate corresponding to each test site is determined, and then the economically optimal nitrogen application rate corresponding to each test site is obtained.

[0109] Step D. Based on each test site, Monte Carlo resampling is used. According to the soil physicochemical properties data and climate characteristics data of the test sites, the highest yield nitrogen application rate and the economically optimal nitrogen application rate of the test sites are used as target values ​​to train the network to be trained, and the highest yield nitrogen application model group and the economically optimal nitrogen application model group are obtained. Then proceed to step E.

[0110] In practical applications, the nitrogen application rate with the highest yield and the economically optimal nitrogen application rate are used as target values, respectively, based on the preset total number of Monte Carlo resampling times. ,initialization Perform steps D1 to D4 to obtain the nitrogen application model group corresponding to the target amount, namely the highest yield nitrogen application model group and the economically optimal nitrogen application model group.

[0111] Step D1. For each test point, calculate the average value of the preset soil physicochemical properties data corresponding to each analysis time, and construct the average value of each soil physicochemical property corresponding to the test point. At the same time, calculate the average value of the preset climate characteristic data corresponding to each analysis time, and construct the average value of each climate characteristic corresponding to the test point. Then proceed to step D2.

[0112] Step D2. Based on each test point, use the Bootstrap resampling method to sample test points with replacement, and construct the first... The sample subset is then processed, and the process proceeds to step D3.

[0113] Step D3. Based on the first The subsample, with the mean values ​​of soil physicochemical properties and climate characteristics of each experimental point as input and the target quantity of each experimental point as output, is used to train the network to be trained, resulting in the trained network, which constitutes the first subsample corresponding to the target quantity. A nitrogen application model is established, and then proceed to step D4.

[0114] Step D4. Judgment Is it equal to That is, obtaining the target quantity corresponding to One nitrogen application model constitutes the nitrogen application model group corresponding to the target amount; otherwise, for... The value is incremented by 1 and the process returns to step D1.

[0115] In practical applications, regarding the further analysis and design of step D, step D uses the highest yield nitrogen application rate and the economically optimal nitrogen application rate as target values ​​to obtain the nitrogen application model group corresponding to the target values. During each execution of steps D1 to D3, the corresponding target value is obtained. After implementing the nitrogen application model, the increase in mean square error for each soil physicochemical property and each climate characteristic was calculated, and then... After executing steps D1 to D3, the corresponding data for each soil physicochemical property and each climate characteristic are obtained. The mean square error was calculated as follows: the mean square error was calculated as follows; then, based on the mean square error increase, the data were sorted from largest to smallest, and the data on each soil physicochemical property and each climate characteristic were sorted. The order represents the contribution of each soil physicochemical property and each climate characteristic to the output of the nitrogen application model group corresponding to the corresponding target quantity, from largest to smallest.

[0116] Then, determine the target data in any of the following ways, and then execute step D. Based on each test point, resample using Monte Carlo. According to the target data of each test point, use the highest yield nitrogen application rate and the economically optimal nitrogen application rate of the test point as the target values, and train the network to be trained respectively to obtain the highest yield nitrogen application model group and the economically optimal nitrogen application model group.

[0117] Method 1: The average contribution of each soil physicochemical property data and each climate characteristic data is used to form a contribution threshold. Data that is less than the contribution threshold in each soil physicochemical property data and each climate characteristic data are removed, and the remaining data constitute each target data.

[0118] Method 2: Based on the soil physicochemical properties data and climate characteristic data, sort them from largest to smallest contribution, remove the data after the preset percentile in sequence, and the remaining data constitute the target data.

[0119] Furthermore, the network to be trained in step D can be any one of the following: XGBoost model, Random Forest model, Gradient Boosting Tree model, or Deep Neural Network model. In specific implementations of this invention, the Random Forest (RF) regression model is preferred, but GBDT, XGBoost, or deep learning models can also be used. However, the Random Forest regression model has higher interpretability and more stable variance properties.

[0120] The parameter settings for building the random forest regression model are shown in Table 1 below.

[0121] Table 1

[0122]

[0123] Step E. Based on the highest yield nitrogen application model group and the economically optimal nitrogen application model group, calculate and determine the nitrogen application amount for each location to be tested in the target area with respect to the highest yield target and the nitrogen application amount with respect to the economically optimal target.

[0124] In practical applications, step E above involves obtaining preset soil physicochemical property data and preset climate characteristic data for each test location in the target test area. Then, each nitrogen application model in the highest yield nitrogen application model group is applied to obtain the highest yield nitrogen application amount for each nitrogen application model corresponding to the test location, and the average value is calculated to form the nitrogen application amount for the test location with respect to the highest yield target. Thus, the nitrogen application amount for each test location in the target test area with respect to the highest yield target is obtained.

[0125] Simultaneously, for each detection location in the target area, preset soil physicochemical property data and preset climate characteristic data are obtained for each detection location. Then, each nitrogen application model in the economically optimal nitrogen application model group is applied to obtain the economically optimal nitrogen application amount for each nitrogen application model corresponding to the detection location, and the average value is calculated to form the nitrogen application amount of the detection location with respect to the economically optimal target. Thus, the nitrogen application amount of each detection location in the target area with respect to the economically optimal target is obtained.

[0126] Based on the implementation of step E above, the nitrogen application models in the highest yield nitrogen application model group are further designed and applied. After obtaining the highest yield nitrogen application amount corresponding to each nitrogen application model at the location to be detected, the highest yield nitrogen application amount is sorted from smallest to largest, and the highest yield nitrogen application amount ranked at the 2.5% position is obtained to form the lower limit of the confidence interval. The highest yield nitrogen application amount ranked at the 97.5% position is obtained to form the upper limit of the confidence interval, thus forming the confidence interval corresponding to the highest yield nitrogen application.

[0127] Similarly, by applying each nitrogen application model in the economically optimal nitrogen application model group, after obtaining the economically optimal nitrogen application amount corresponding to each nitrogen application model for the location to be detected, the process also includes sorting each economically optimal nitrogen application amount from smallest to largest, obtaining the economically optimal nitrogen application amount ranked at the 2.5% position to form the lower limit of the confidence interval, obtaining the economically optimal nitrogen application amount ranked at the 97.5% position to form the upper limit of the confidence interval, and thus forming the confidence interval corresponding to the economically optimal nitrogen application.

[0128] In practical applications, the design method of this invention is verified and illustrated using the nitrogen fertilizer gradient test data of single-season rice in the Yangtze River Delta as an example.

[0129] First, data from a nitrogen fertilizer gradient experiment on rice in the Yangtze River Delta region from 2006 to 2010 were obtained, comprising 917 plot experiments with four nitrogen application levels: 0, 50%MN, MN, and 150%MN, where MN represents the medium nitrogen application rate set by local agricultural experts. Soil properties (pH, SOC, TN, TP, Texture, CEC) and climatic factors (Tmp, Pre) were collected concurrently. Missing values ​​were filled using IDW interpolation, and outliers were removed using the box plot IQR method. After quality control, valid data from 162 experimental sites were retained.

[0130] Next, the lmer() package of R language lme4 was used to fit the mixed effects. Through parameter estimation, the fixed effects estimates (global average) were obtained in this embodiment as shown in Table 2:

[0131] Table 2

[0132]

[0133] The output of the random effects section shows:

[0134] Site random intercept b 0s Standard deviation = 0.31;

[0135] Site random slope b 1s Standard deviation = 0.009;

[0136] This indicates significant differences in baseline yield and nitrogen sensitivity among different sites, supporting a modeling strategy that uses mixed effects rather than a single fixed parameter.

[0137] The explanatory power metric for the model is:

[0138] condition = 0.72 (total explanatory power of fixed effects + random effects);

[0139] Marginal = 0.41 (explained by fixed effects only).

[0140] These results further illustrate that the regional nitrogen application-yield relationship exhibits significant site heterogeneity in space, and the fixed-effects model alone cannot explain all the variation. The mixed-effects model of this invention can obtain a more reasonable curve shape and more accurate parameter estimates.

[0141] The fixed coefficients and random effects determined in this way form the mathematical basis for subsequent solutions to the maximum yield nitrogen application rate YHN and the economically optimal nitrogen application rate ONR.

[0142] Then, based on the model parameters, the maximum nitrogen application rate YHN and the economically optimal nitrogen application rate ONR for each test point are calculated.

[0143] This embodiment sets the rice price at 2051.48 CNY / t based on the "Compilation of National Agricultural Product Cost and Benefit Data" from 2006 to 2010. -1 The cost of nitrogen fertilizer is 4.255 CNY·kg N. -1 The highest yield nitrogen application rate YHN and the economically optimal nitrogen application rate ONR were obtained for 162 test sites, as follows: Figure 2 As shown in Table 3, the distribution statistics for these 162 stations are as follows:

[0144] Table 3

[0145]

[0146] Statistical analysis shows that the economically optimal nitrogen application rate ONR is significantly lower than the nitrogen application rate YHN with the highest yield on average. This indicates that although increasing nitrogen application rate can lead to increased yield, the yield gains from this increase cannot offset the cost of nitrogen fertilizer. Therefore, it is considered a "non-optimal input" in an economic sense.

[0147] Furthermore, using the economically optimal nitrogen application rate (ONR) of the 162 experimental sites obtained in the previous steps as supervision labels, and using the corresponding site variables such as pH, SOC, TN, Texture, CEC, Tmp, Pre, Year, and Province as feature inputs, a random forest model (ntree=1000, mtry=5) was constructed and trained 500 times using Monte Carlo iterations. Each training iteration outputs the predicted value of ONR. Finally, the mean and standard deviation of the 500 predictions were statistically analyzed. The model performance of this embodiment on the validation set (accounting for 20% of the sample) is shown in Table 4.

[0148] Table 3

[0149] index numerical values R² (coefficient of determination) 0.68 <![CDATA[RMSE(kg N ha -1 )]]> 23.5

[0150] The probability distribution of ONR predictions was obtained through 500 Monte Carlo repetitions, and the 95% confidence interval (CI) was calculated as shown in Table 4.

[0151] Table 4

[0152]

[0153] After completing 500 Monte Carlo repetitions of training of the random forest model, this embodiment extracts and statistically analyzes the IncMSE (Increase in Mean Squared Error) weights for each training iteration. For each input variable, the "increase in prediction error after variable permutation" is calculated in each of the 500 repetitions, and the average IncMSE value of the 500 iterations is used as the final importance index of the variable, as shown in Table 5.

[0154] Table 5

[0155]

[0156] The ranking of variable importance (IncMSE) shows that the six most critical factors affecting the economically optimal nitrogen application rate (ONR) are: pH, Sand, province, Tmp, baseline yield, and C / N ratio. Figure 3 As shown.

[0157] Based on the above variable importance results, this invention can provide an "interpretive layer" in the fertilization recommendation output, so that fertilization decisions can not only provide recommended values, but also provide the scientific basis for forming the recommended values, thereby improving the understanding and trust of agricultural management departments, agricultural consultants and farmers in the process of applying this technology.

[0158] In this embodiment, data on the distribution of single-season rice planting in the Yangtze River Delta region in 2022 was obtained from the National Ecosystem Science Data Center. The spatial resolution of this data is 10m. Considering the resolution of the climate raster data, this embodiment resamples the 10m data to a 1km resolution and spatially matches the soil attribute data (pH, SOC, TN, Texture, CEC) of each grid point with the climate attribute data (Tmp, Pre) of the same period to form a feature database covering a total of 189,000 grid points in the region.

[0159] After inputting the above grid features into the random forest model trained by S4, this embodiment obtains the spatial prediction results of the economically optimal nitrogen application rate (ONR) at a resolution of 1km in the Yangtze River Delta region in 2022. The statistics are shown in Table 6 below.

[0160] Table 6

[0161]

[0162] The results indicate that there are significant differences in the optimal nitrogen application requirements among different fields within the same region, and the model output can be used to construct field-level fertilization recommendation maps.

[0163] Furthermore, in this embodiment, the rice growing season Tmp and Pre data for 2021–2100 under the GFDL-ESM4 model output from the IPCC SSP climate paths (SSP119, SSP245, SSP585) provided by the National Tibetan Plateau Data Center were used. The above-mentioned climate characteristics were input into the aforementioned economically optimal nitrogen application rate prediction random forest model to predict the economically optimal nitrogen application rate (ONR) for future years. The statistical results are shown in Table 7 below.

[0164] Table 7

[0165]

[0166] The results of this example demonstrate that, within the framework of this invention, the economically optimal nitrogen application rate (ONR) can be spatially and temporally predicted under different future climate pathways, such as... Figure 4 As shown, its uncertainty range can be quantified, thus providing an information basis for agricultural management departments to plan future nitrogen fertilizer input strategies.

[0167] The rice nitrogen fertilizer application prediction method designed in the above technical solution first constructs rice experimental data samples based on multi-point nitrogen fertilizer gradient test data. Then, it constructs a quadratic mixed-effect yield response model including fixed-effect coefficients and random-effect terms, and fits it to obtain a yield-nitrogen application rate response model. Then, based on the yield-nitrogen application rate response model and the economic objective function, it solves for the highest yield nitrogen application rate and the economically optimal nitrogen application rate for each experimental point. Then, through Monte Carlo resampling, it trains to obtain the highest yield nitrogen application model group and the economically optimal nitrogen application model group. Finally, for each location in the target test area, it determines the nitrogen application rate for the highest yield target and the nitrogen application rate for the economically optimal target, thereby constructing a high-resolution regional optimal nitrogen application map. The design scheme of this invention combines the interpretability of empirical models with the generalization ability of machine learning, which can significantly improve the decision-making accuracy of farmland nitrogen fertilizer management recommendations, and provide quantitative, regional, and interannually variable economically optimal nitrogen application guidance for rice production. It has significant agricultural production application value and promotion potential.

[0168] In the specific design, considering both the highest yield nitrogen application rate and the economically optimal nitrogen application rate, and combining machine learning to extrapolate the model, the nitrogen application recommendation can be refined to the field level, rather than the regional average, significantly improving the accuracy of nitrogen application and increasing the precision of field-level fertilization. Furthermore, for the economically optimal nitrogen application rate, the "economic fertilization theory" is used as a constraint, and the model retains the decomposition of fixed effects and random effects, giving the model a clear agronomic meaning rather than being a pure statistical black box, thus achieving an "interpretable + generalizable" fertilization decision-making model.

[0169] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A method for predicting nitrogen fertilizer application rate in rice, characterized in that, Includes the following steps: Step A. Based on the analysis time of each test site, collect field yield data of nitrogen fertilizer gradient at multiple points, and simultaneously collect preset soil physicochemical property data and preset climate characteristic data of each test site to construct data samples for each rice experiment, and then proceed to Step B; Step B. Construct a quadratic mixed-effect yield response model that includes fixed-effects coefficients and random-effects terms, and fit the yield-nitrogen application rate response model based on the data samples from each rice experiment, and then proceed to step C; Step C. Based on the yield-nitrogen application rate response model, solve for the highest yield nitrogen application rate corresponding to each test point, and construct the economic objective function to solve for the economically optimal nitrogen application rate corresponding to each test point, and then proceed to step D; Step D. Based on each test site, Monte Carlo resampling is used. According to the soil physicochemical properties data and climate characteristics data of the test sites, the highest yield nitrogen application rate and the economically optimal nitrogen application rate of the test sites are used as target values. The network to be trained is trained separately to obtain the highest yield nitrogen application model group and the economically optimal nitrogen application model group. Then proceed to step E. Step E. Based on the highest yield nitrogen application model group and the economically optimal nitrogen application model group, calculate and determine the nitrogen application amount for each location to be tested in the target area with respect to the highest yield target and the nitrogen application amount with respect to the economically optimal target.

2. The method for predicting nitrogen fertilizer application rate in rice according to claim 1, characterized in that: Step A includes the following steps A1 to A2; Step A1. Based on a preset number of test points, and each test point includes rice test areas with preset nitrogen application rates, for each analysis time, the nitrogen application rate and rice yield of each rice test area in each test point corresponding to the analysis time are statistically analyzed, and preset soil physicochemical property data and preset climate characteristic data of each test point corresponding to the analysis time are collected, and then proceed to step A2. Step A2. Based on each rice experimental area and each analysis time in each experimental site, construct rice experimental data samples using the nitrogen application rate, rice yield, and preset soil physicochemical properties and climate characteristics data of the corresponding analysis time for each rice experimental area, and then obtain each rice experimental data sample.

3. The method for predicting nitrogen fertilizer application rate in rice according to claim 2, characterized in that: Step A also includes the following steps A3 to A5, where step A3 is performed after step A2 is completed. Step A3. For each rice experiment data sample, if there are missing values, use the inverse distance weighted interpolation method to fill in the missing values, update the rice experiment data sample, and then proceed to step A4; Step A4. Based on the data samples from each rice experiment, determine the normal value range for each target data type, including nitrogen application rate, rice yield, preset soil physicochemical properties data, and preset climate characteristic data, and then proceed to step A5. Step A5. For each rice experiment data sample, determine whether there are any outliers in the data values ​​of each target data type corresponding to the rice experiment data sample that are outside the corresponding normal value range. If so, delete the rice experiment data sample; otherwise, do not perform any further processing.

4. The method for predicting nitrogen fertilizer application rate in rice according to claim 3, characterized in that: In step A4, based on each rice experiment data sample, for each target data type of nitrogen application rate, rice yield, preset soil physicochemical property data, and preset climate characteristic data, the following steps A4-1 to A4-2 are executed; Step A4-1. Sort the data values ​​of the target data type in ascending order, and obtain the data value that is in the 25th position of the sort. And to obtain the data value that is in the 75th position of the sorted list. Then press Calculate the interquartile range Then proceed to step A4-2; Step A4-2. Calculation The result constitutes the lower limit of the normal numerical range corresponding to the target data type, and the calculation The result constitutes the upper limit of the normal numerical range corresponding to the target data type, and thus constitutes the normal numerical range corresponding to the target data type.

5. The method for predicting nitrogen fertilizer application rate in rice according to claim 2, characterized in that: Step B includes the following steps B1 to B2; Step B1. Construct a quadratic mixed-effects output response model including fixed-effects coefficients and random-effects terms as follows: ； in, , Indicates the number of test sites. , Indicates the first The number of rice experimental areas in each experimental site Indicates the first The first test site The rice experimental area corresponds to the first Rice yield at the time of analysis Indicates the first The first test site The rice experimental area corresponds to the first Nitrogen application rate at each analysis time point , , All are fixed effects coefficients to be trained. Indicates baseline rice yield. This represents the slope of the linear term of the nitrogen fertilizer effect. This represents the quadratic effect of the linear effect of nitrogen fertilizer. , , All represent random effects to be trained. Indicates the first Random intercepts at each test point Follows a normal distribution , Indicates the random effects to be trained variance Indicates the first The random slope of each test point Follows a normal distribution , Indicates the random effects to be trained variance Indicates the first The test point corresponds to the first The random intercept of each analysis time period, Follows a normal distribution , Indicates the random effects to be trained variance Indicates the first The first test site The rice experimental area corresponds to the first The analysis time follows a normal distribution. random error, express The variance; then proceed to step B2; Step B2. Based on the data samples from each rice experiment, use the R language... In the bag The function is fitted to the quadratic mixed-effects output response model using the restricted maximum likelihood estimation method to obtain the fixed effects coefficients. , , The value of the random effect at each test site , The value of the random effect at each experimental point corresponding to each analysis time. The value of, and the variance of random effects , , , The value of is obtained, that is, the yield-nitrogen application rate response model.

6. The method for predicting nitrogen fertilizer application rate in rice according to claim 5, characterized in that: Step C includes the following steps C1 to C2; Step C1. Eliminate yield-nitrogen application rate response models for the test sites. and The expected rice yield model is as follows, where, Indicates the first Nitrogen application rate at each test site Indicates the first Rice yield at each experimental site express Expected value; ； Execute the expected rice yield model with respect to nitrogen application rate The first derivative of , and set the derivative to zero: ； Determine the maximum nitrogen application rate as follows: ； Then, based on step B, the yield-nitrogen application rate response model is obtained. and The value is determined for each test point, based on the corresponding value of the test point. The nitrogen application rate corresponding to the highest yield at the test site was calculated. , Indicates the first The highest yield nitrogen application rate corresponding to each test point is obtained, and then the highest yield nitrogen application rate corresponding to each test point is obtained; then proceed to step C2; Step C2. Construct the economic benefit function for the test sites. As follows, among which, This indicates the unit price of rice. This indicates the unit price of nitrogen fertilizer; ； Execute the economic return function Regarding nitrogen application rate The first derivative of , and set the derivative to zero: ； Will Substituting into the above formula, the update is as follows: ； Determine the economically optimal nitrogen application rate as follows: ； Then, based on step B, the yield-nitrogen application rate response model is obtained. and The value is determined for each test point, based on the corresponding value of the test point. The economically optimal nitrogen application rate corresponding to the test site was calculated. , Indicates the first The economically optimal nitrogen application rate corresponding to each test site is determined, and then the economically optimal nitrogen application rate corresponding to each test site is obtained.

7. The method for predicting nitrogen fertilizer application rate in rice according to claim 1, characterized in that: In step D, the nitrogen application rate with the highest yield and the economically optimal nitrogen application rate are used as target values, respectively, and the total number of Monte Carlo resampling times is preset. ,initialization Perform steps D1 to D4 to obtain the nitrogen application model group corresponding to the target amount, namely the highest yield nitrogen application model group and the economically optimal nitrogen application model group. Step D1. For each test point, calculate the average value of the preset soil physicochemical properties data corresponding to each analysis time, and construct the average value of each soil physicochemical property corresponding to the test point. At the same time, calculate the average value of the preset climate characteristic data corresponding to each analysis time, and construct the average value of each climate characteristic corresponding to the test point. Then proceed to step D2. Step D2. Based on each test point, use the Bootstrap resampling method to sample test points with replacement, and construct the first... Sub-sample subset, and proceed to step D3; Step D3. Based on the first The subsample, with the mean values ​​of soil physicochemical properties and climate characteristics of each experimental point as input and the target quantity of each experimental point as output, is used to train the network to be trained, resulting in the trained network, which constitutes the first subsample corresponding to the target quantity. First, select a nitrogen application model, and then proceed to step D4; Step D4. Judgment Is it equal to That is, obtaining the target quantity corresponding to One nitrogen application model constitutes the nitrogen application model group corresponding to the target amount; otherwise, for... The value is incremented by 1 and the process returns to step D1.

8. The method for predicting nitrogen fertilizer application rate in rice according to claim 7, characterized in that: In step D, using the highest-yield nitrogen application rate and the economically optimal nitrogen application rate as target values, the nitrogen application model group corresponding to the target values ​​is obtained by executing steps D1 to D3 each time. After implementing the nitrogen application model, the increase in mean square error for each soil physicochemical property and each climate characteristic was calculated, and then... After executing steps D1 to D3, the corresponding data for each soil physicochemical property and each climate characteristic are obtained. The mean square error was increased by each time; then, based on the mean square error increase, the data were sorted from largest to smallest, and the data on each soil physicochemical property and each climate characteristic were sorted to represent the contribution of each soil physicochemical property and each climate characteristic to the output of the nitrogen application model group corresponding to the corresponding target quantity from largest to smallest. Then, determine each target data in any of the following ways, and then execute step D. Based on each test point, resample through Monte Carlo. According to each target data of the test point, use the highest yield nitrogen application rate and the economically optimal nitrogen application rate of the test point as the target values, and train the network to be trained respectively to obtain the highest yield nitrogen application model group and the economically optimal nitrogen application model group. Method 1: The average contribution of each soil physicochemical property data and each climate characteristic data is used to form a contribution threshold. Data that is less than the contribution threshold in each soil physicochemical property data and each climate characteristic data are removed, and the remaining data constitute each target data. Method 2: Based on the soil physicochemical properties data and climate characteristic data, sort them from largest to smallest contribution, remove the data after the preset percentile in sequence, and the remaining data constitute the target data.

9. The method for predicting nitrogen fertilizer application rate in rice according to claim 1, characterized in that: Step E includes obtaining preset soil physicochemical property data and preset climate characteristic data for each test location in the target test area, applying each nitrogen application model in the highest yield nitrogen application model group to obtain the highest yield nitrogen application amount for each nitrogen application model corresponding to the test location, and calculating the average value to form the nitrogen application amount for the test location with respect to the highest yield target, thereby obtaining the nitrogen application amount for each test location in the target test area with respect to the highest yield target; Simultaneously, for each detection location in the target area, preset soil physicochemical property data and preset climate characteristic data are obtained for each detection location. Then, each nitrogen application model in the economically optimal nitrogen application model group is applied to obtain the economically optimal nitrogen application amount for each nitrogen application model corresponding to the detection location, and the average value is calculated to form the nitrogen application amount of the detection location with respect to the economically optimal target. Thus, the nitrogen application amount of each detection location in the target area with respect to the economically optimal target is obtained.

10. The method for predicting nitrogen fertilizer application rate in rice according to claim 9, characterized in that: In step E, after applying each nitrogen application model in the highest yield nitrogen application model group to obtain the highest yield nitrogen application amount corresponding to each nitrogen application model at the location to be detected, the method further includes sorting each highest yield nitrogen application amount from smallest to largest, obtaining the highest yield nitrogen application amount ranked at the 2.5% position to form the lower limit of the confidence interval, obtaining the highest yield nitrogen application amount ranked at the 97.5% position to form the upper limit of the confidence interval, and thus forming the confidence interval corresponding to the highest yield nitrogen application. Similarly, by applying each nitrogen application model in the economically optimal nitrogen application model group, after obtaining the economically optimal nitrogen application amount corresponding to each nitrogen application model for the location to be detected, the process also includes sorting each economically optimal nitrogen application amount from smallest to largest, obtaining the economically optimal nitrogen application amount ranked at the 2.5% position to form the lower limit of the confidence interval, obtaining the economically optimal nitrogen application amount ranked at the 97.5% position to form the upper limit of the confidence interval, and thus forming the confidence interval corresponding to the economically optimal nitrogen application.

Images