CN-121998233-A - Rice nitrogenous fertilizer application amount prediction method

CN121998233ACN 121998233 ACN121998233 ACN 121998233ACN-121998233-A

Abstract

The invention relates to a rice nitrogen fertilizer application amount prediction method, which comprises the steps of firstly constructing each rice test data sample based on multipoint nitrogen fertilizer gradient test data, then constructing a secondary mixed effect yield response model containing a fixed effect coefficient and a random effect item, fitting to obtain a yield-nitrogen application amount response model, then solving the highest yield nitrogen application amount and the economic optimal nitrogen application amount of each test point based on the yield-nitrogen application amount response model and an economic objective function, obtaining a highest yield nitrogen application model group and an economic optimal nitrogen application model group through Monte Carlo resampling by training, and finally determining the nitrogen application amount related to the highest yield target and the nitrogen application amount related to the economic optimal target aiming at each position of a target to be tested, thereby forming a high resolution area optimal nitrogen application map.

Inventors

ZHAO XU
CAI SIYUAN
Sun Jiabei

Assignees

中国科学院南京土壤研究所

Dates

Publication Date: 20260508
Application Date: 20251209

Claims (10)

1. The rice nitrogen fertilizer application amount prediction method is characterized by comprising the following steps of: Step A, collecting multi-point nitrogen fertilizer gradient field yield data based on each analysis time of each test point, synchronously collecting preset soil physicochemical property data and preset climate characteristic data of each test point, constructing each rice test data sample, and then entering step B; Step B, constructing a secondary mixed effect yield response model containing a fixed effect coefficient and a random effect item, fitting to obtain a yield-nitrogen application quantity response model based on each rice test data sample, and then entering the step C; Step C, based on the yield-nitrogen application quantity response model, solving the highest yield nitrogen application quantity corresponding to each test point, constructing an economic objective function, solving the economic optimal nitrogen application quantity corresponding to each test point, and then entering the step D; Step D, resampling by Monte Carlo based on each test point, taking the highest yield nitrogen application amount and the economic optimal nitrogen application amount of the test point as target amounts respectively according to the soil physicochemical property data and the climate characteristic data of the test point, training the test points respectively aiming at a network to be trained to obtain a highest yield nitrogen application model group and an economic optimal nitrogen application model group, and then entering the step E; And E, calculating and determining the nitrogen application quantity of each to-be-detected position in the target to-be-detected area relative to the target with the highest yield and the nitrogen application quantity relative to the economic optimal target based on the highest yield nitrogen application model group and the economic optimal nitrogen application model group respectively.
2. The method for predicting the nitrogen fertilizer application amount of rice according to claim 1, wherein the step A comprises the following steps A1 to A2; step A1, based on a preset number of test points, each test point respectively comprises rice test areas with preset different nitrogen application levels, and according to each analysis time, the nitrogen application amount and the rice yield of each rice test area in each test point respectively corresponding to the analysis time are counted, and preset soil physicochemical property data and weather characteristic data of each test point respectively corresponding to the analysis time are collected, and then the step A2 is carried out; And A2, constructing a rice test data sample based on each rice test area in each test point and each analysis time, and presetting each soil physicochemical property data and each climate characteristic data of the test point according to the nitrogen application amount, the rice yield and the analysis time corresponding to the rice test area, so as to obtain each rice test data sample.
3. The method for predicting the nitrogen fertilizer application amount of rice as recited in claim 2, wherein the step A further comprises the steps A3 to A5, and the step A3 is performed after the step A2 is performed; step A3, respectively aiming at each rice test data sample, if a missing value exists, carrying out deficiency supplementing by adopting an inverse distance weighted interpolation method, updating the rice test data sample, and then entering the step A4; Step A4, determining a normal numerical range under the target data type according to each rice test data sample, and respectively aiming at the nitrogen application amount, the rice yield, the preset soil physicochemical property data and each target data type of preset climate characteristic data, and then entering step A5; And A5, respectively aiming at each rice test data sample, judging whether an abnormal value corresponding to a normal numerical range exists in the data value of each target data type corresponding to the rice test data sample, if so, deleting the rice test data sample, otherwise, not performing any further processing.
4. The method for predicting the nitrogen fertilizer application amount of rice in claim 3, wherein in the step A4, based on each rice test data sample, the following steps A4-1 to A4-2 are executed for each target data type of nitrogen application amount, rice yield, preset soil physicochemical property data and preset climate characteristic data respectively; step A4-1, sorting the data values of the target data type according to the order from small to large, and obtaining the data value with the sorting at 25% of the bits And obtaining a data value in which the ordering is located at 75% of the bits Then press down Calculating to obtain the quartile range Then enter step A4-2; Step A4-2. Calculating As a result of (a) forming the lower limit of the normal numerical range corresponding to the target data type, and calculating As a result, the upper limit of the normal value range corresponding to the target data type is formed, and the normal value range corresponding to the target data type is further formed.
5. The method for predicting the nitrogen fertilizer application amount of rice according to claim 2, wherein the step B comprises the following steps B1 to B2; step B1, constructing a secondary mixed effect yield response model comprising a fixed effect coefficient and a random effect term, wherein the secondary mixed effect yield response model comprises the following steps: ; Wherein, the , The number of test points is indicated, , Represent the first The number of rice test areas in each test point, Represent the first In the test points The rice test area corresponds to The rice yield at each analysis time was determined, Represent the first In the test points The rice test area corresponds to The nitrogen application amount at each analysis time was determined, 、、 Are all fixed effect coefficients to be trained, The baseline rice yield is indicated, Represents the slope of the primary term of the linear effect of the nitrogen fertilizer, The secondary term effect of the linear effect of the nitrogen fertilizer is represented, 、、 All represent the random effect to be trained, Represent the first The random intercept of the individual test points, Obeys normal distribution , Representing random effects to be trained Is a function of the variance of (a), Represent the first The random slope of the individual test points is, Obeys normal distribution , Representing random effects to be trained Is a function of the variance of (a), Represent the first The test points correspond to The random intercept of each analysis time, Obeys normal distribution , Representing random effects to be trained Is a function of the variance of (a), Represent the first In the test points The rice test area corresponds to Obeying the normal too distribution of the individual analysis times Is used for the random error of (a), Representation of Then go to step B2; step B2, based on each rice test data sample, using R language In bags The function is fitted to the secondary mixed effect yield response model through a limiting maximum likelihood estimation method to obtain a fixed effect coefficient 、、 Random effect of each test point 、 Random effect of each test point for each analysis time And random effect variance 、、、 And obtaining a yield-nitrogen application response model.
6. The method for predicting the nitrogen fertilizer application amount of rice as recited in claim 5, wherein the step C comprises the following steps C1 to C2; Step C1, rejecting response models of test points on yield-nitrogen application amount And (3) with The expected model of the rice yield after that is as follows, among others, Represent the first The nitrogen application amount of each test point is equal to that of the test point, Represent the first The rice yield at each test point was measured, Representation of Is a desired value of (2); ; executing Rice yield expected model with respect to Nitrogen application amount And let the derivative be zero: ; Determination of the maximum yield of nitrogen application The following are provided: ; and then fitting according to the step B to obtain a yield-nitrogen application quantity response model And (3) with For each test point, according to the corresponding value of the test point Calculating to obtain the highest yield nitrogen application amount corresponding to the test point , Represent the first C2, obtaining the highest-yield nitrogen application amount corresponding to each test point, and then entering step C2; step C2, constructing an economic benefit function of the test points As follows, among others, The unit price of the rice is represented, Indicating the nitrogen fertilizer unit price; ; Performing economic benefit functions Regarding the nitrogen application amount And let the derivative be zero: ; Will be Substituting the above, the update is as follows: ; Determining an economically optimal nitrogen application The following are provided: ; and then fitting according to the step B to obtain a yield-nitrogen application quantity response model And (3) with For each test point, according to the corresponding value of the test point Calculating to obtain the economic optimal nitrogen application amount corresponding to the test point , Represent the first And the economic optimal nitrogen application amount corresponding to each test point is further obtained.
7. The method of claim 1, wherein in step D, the maximum yield nitrogen application amount and the economic optimum nitrogen application amount are used as target amounts respectively, and the total number of times of resampling is performed according to a preset Monte Carlo Initializing Step D1 to step D4 are executed to obtain a nitrogen application model group corresponding to the target quantity, namely a nitrogen application model group with the highest yield and an economic optimal nitrogen application model group; D1, respectively aiming at each test point, solving the average value of the preset soil physicochemical property data of each test point corresponding to each analysis time to form the average value of each soil physicochemical property corresponding to the test point, simultaneously solving the average value of the preset climate characteristic data of each test point corresponding to each analysis time to form the average value of each climate characteristic corresponding to the test point, and then entering the step D2; step D2, based on each test point, the test points are extracted in a replaced mode through a Bootstrap resampling mode, and the first test point is constructed A sub-sample subset and enter step D3; Step D3 based on the first The secondary sample subset takes the average value of the physicochemical properties of each soil and the average value of the climatic characteristics of each test point as input, the target quantity of each test point as output, trains the network to be trained to obtain a trained network, and forms the first corresponding target quantity Performing nitrogen application models, and then entering a step D4; step D4, judging Whether or not to be equal to If yes, obtain the corresponding target amount Forming a nitrogen application model group corresponding to the target quantity by using the nitrogen application models, otherwise, aiming at The value of (1) is updated by adding 1 and returns to step D1.
8. The method for predicting nitrogen fertilizer applying amount of paddy rice according to claim 7, wherein in the step D, the highest yield nitrogen applying amount and the economic optimal nitrogen applying amount are used as target amounts respectively, and in the process of obtaining the nitrogen applying model group corresponding to the target amounts, each time the steps D1 to D3 are executed, the first corresponding to the target amounts is obtained After each nitrogen application model, calculating and obtaining the mean square error increment corresponding to each soil physicochemical property data and each climate characteristic data respectively, thereby completing After the execution of the sub-steps D1 to D3, obtaining physical and chemical property data of each soil and weather characteristic data which correspond to each other respectively Sequencing the physicochemical property data and the climate characteristic data of each soil according to the sequence, and sequentially representing the contribution degree of the physicochemical property data and the climate characteristic data of each soil to the output quantity of the nitrogen application model group corresponding to the corresponding target quantity from large to small; Determining each target data according to any one of the following modes, and then executing a step D, resampling by Monte Carlo based on each test point, and respectively taking the highest yield nitrogen application amount and the economic optimal nitrogen application amount of the test point as target amounts according to each target data of the test point, and respectively training aiming at a network to be trained to obtain a highest yield nitrogen application model group and an economic optimal nitrogen application model group; mode 1, forming a contribution threshold by using the average value of the contribution of the physicochemical property data of each soil and the contribution of the climate characteristic data, removing all data smaller than the contribution threshold in the physicochemical property data of each soil and the climate characteristic data, and forming all target data by the rest of all data; And 2, sorting the soil physicochemical property data and the climate characteristic data according to the contribution degree from large to small, sequentially removing each data after the preset percentile, and forming each target data by the rest of each data.
9. The method for predicting the nitrogen fertilizer application amount of rice according to claim 1, wherein the step E comprises the steps of obtaining preset physical and chemical property data of each soil and preset climate characteristic data of each position to be detected for each position to be detected in a target area, applying each nitrogen application model in a nitrogen application model group with the highest yield, obtaining the highest yield nitrogen application amount of each nitrogen application model corresponding to each position to be detected, calculating an average value, forming the nitrogen application amount of the position to be detected with respect to the highest yield target, and further obtaining the nitrogen application amount of each position to be detected in the target area with respect to the highest yield target; Meanwhile, aiming at each to-be-detected position in the target to-be-detected area, preset soil physicochemical property data and weather characteristic data of each to-be-detected position are obtained, each nitrogen application model in the economic optimal nitrogen application model group is applied, the economic optimal nitrogen application amount of each to-be-detected position corresponding to each nitrogen application model is obtained, the average value is obtained, the nitrogen application amount of the to-be-detected position relative to the economic optimal target is formed, and the nitrogen application amount of each to-be-detected position relative to the economic optimal target in the target to-be-detected area is obtained.
10. The method for predicting the nitrogen fertilizer application amount of paddy rice according to claim 9, wherein each nitrogen application model in the highest-yield nitrogen application model group is applied in the step E, and after the highest-yield nitrogen application amount of each nitrogen application model corresponding to the position to be detected is obtained, the method further comprises sorting the highest-yield nitrogen application amount from small to large, and obtaining the highest-yield nitrogen application amount of which the sorting is positioned at the 2.5% position, so as to form a confidence interval lower limit, and obtaining the highest-yield nitrogen application amount of which the sorting is positioned at the 97.5% position, so as to form a confidence interval upper limit, so as to form a confidence interval corresponding to the highest-yield nitrogen application; After the economic optimal nitrogen application amounts of the economic optimal nitrogen application models in the economic optimal nitrogen application model group are obtained, sequencing from small to large for the economic optimal nitrogen application amounts of the economic optimal nitrogen application models respectively, and obtaining the economic optimal nitrogen application amounts of the economic optimal nitrogen application models in the 2.5% position to form a confidence interval lower limit, obtaining the economic optimal nitrogen application amounts of the economic optimal nitrogen application models in the 97.5% position to form a confidence interval upper limit, and further forming a confidence interval corresponding to the economic optimal nitrogen application.

Description

Rice nitrogenous fertilizer application amount prediction method Technical Field The invention relates to a rice nitrogen fertilizer application amount prediction method, and belongs to the technical fields of farmland nitrogen fertilizer management, crop models and accurate agricultural big data modeling. Background At present, the farmland nitrogen fertilizer application amount depends on experience recommended values or a linear/secondary response model based on a single-point test, and the recommendation lacks space-time extrapolation capability and cannot reflect climate-soil-variety interaction difference, so that nitrogen application under-application or over-application is caused, and further the nitrogen fertilizer utilization rate (NUE) and economic return are reduced. Although the existing machine learning method can perform spatial interpolation prediction, the existing machine learning method often lacks a clear economic objective function, and the yield benefit-nitrogenous fertilizer cost is difficult to be used as a direct optimization index. Therefore, there is an urgent need for a coupled prediction technique with an economical optimal nitrogen application target as a core, which combines interpretability (empirical secondary response) and generalization capability (machine model). Disclosure of Invention The technical problem to be solved by the invention is to provide the rice nitrogen fertilizer application amount prediction method, which can effectively improve the nitrogen application utilization efficiency, the economic benefits of farmers and the nitrogen fertilizer resource utilization efficiency, and realize the engineering floor and policy popularization of the optimization of the rice nitrogen application under the accurate agricultural condition. The invention designs a rice nitrogen fertilizer application amount prediction method, which aims to solve the technical problems and adopts the following technical scheme that: Step A, collecting multi-point nitrogen fertilizer gradient field yield data based on each analysis time of each test point, synchronously collecting preset soil physicochemical property data and preset climate characteristic data of each test point, constructing each rice test data sample, and then entering step B; Step B, constructing a secondary mixed effect yield response model containing a fixed effect coefficient and a random effect item, fitting to obtain a yield-nitrogen application quantity response model based on each rice test data sample, and then entering the step C; Step C, based on the yield-nitrogen application quantity response model, solving the highest yield nitrogen application quantity corresponding to each test point, constructing an economic objective function, solving the economic optimal nitrogen application quantity corresponding to each test point, and then entering the step D; Step D, resampling by Monte Carlo based on each test point, taking the highest yield nitrogen application amount and the economic optimal nitrogen application amount of the test point as target amounts respectively according to the soil physicochemical property data and the climate characteristic data of the test point, training the test points respectively aiming at a network to be trained to obtain a highest yield nitrogen application model group and an economic optimal nitrogen application model group, and then entering the step E; And E, calculating and determining the nitrogen application quantity of each to-be-detected position in the target to-be-detected area relative to the target with the highest yield and the nitrogen application quantity relative to the economic optimal target based on the highest yield nitrogen application model group and the economic optimal nitrogen application model group respectively. As a preferable technical scheme of the invention, the step A comprises the following steps A1 to A2; step A1, based on a preset number of test points, each test point respectively comprises rice test areas with preset different nitrogen application levels, and according to each analysis time, the nitrogen application amount and the rice yield of each rice test area in each test point respectively corresponding to the analysis time are counted, and preset soil physicochemical property data and weather characteristic data of each test point respectively corresponding to the analysis time are collected, and then the step A2 is carried out; And A2, constructing a rice test data sample based on each rice test area in each test point and each analysis time, and presetting each soil physicochemical property data and each climate characteristic data of the test point according to the nitrogen application amount, the rice yield and the analysis time corresponding to the rice test area, so as to obtain each rice test data sample. The invention has the preferable technical scheme that the step A also comprises the following steps A3 to A5, and the step A3 is