Regional crop yield estimation method and system using particle filter data assimilation

Abstract

The application provides a regional crop yield estimation method and system based on particle filtering data assimilation, which comprises the following steps: running each particle at a fixed time interval during the critical growth period of crops until each particle runs to the mature period of crops, and obtaining the first LAI output by each particle; updating the initial weight of each particle according to the posterior probability density function, the updated first LAI of each particle and the likelihood function, and obtaining the updated weight of each particle; resampling each particle in the initial particle set and updating the initial particle set under the condition that the particle divergence is less than a preset value; and obtaining the estimated value of the crop yield according to the crop yield and the maximum weight of each particle under the condition that each particle runs to the mature period of crops. The application can improve the sequential sampling efficiency of particles, reduce the resampling frequency, maintain the diversity of particle input parameters, and improve the speed and accuracy of crop yield prediction.

Description

Technical Field

[0001] This invention relates to the field of agricultural remote sensing technology, and in particular to a method and system for estimating regional crop yields using particle filter data assimilation. Background Technology

[0002] Traditional methods for estimating crop yields mainly include statistical surveys, crop model-based forecasting, and agricultural meteorological forecasting. These methods, due to their inherent limitations, struggle to achieve high-precision regional crop yield estimation. In contrast, estimation methods based on satellite remote sensing technology, with their spatial continuity and temporal dynamics, possess unique advantages in regional crop yield estimation. Furthermore, combining remote sensing technology with crop growth models based on mechanisms such as photosynthesis, respiration, transpiration, and nutrient production can achieve high-precision regional yield estimation. Data assimilation methods, combining the advantages of point-based crop growth models with area-based remote sensing observations, have become a hot topic in quantitative agricultural remote sensing research both domestically and internationally in recent years.

[0003] However, existing assimilation yield estimation methods are generally based on the Gaussian assumption of crop growth model state vector and remote sensing observation vector and the weak nonlinear assumption of crop growth model. However, the inconsistency between these two assumptions and the actual situation introduces errors into the assimilation yield estimation, which cannot meet the requirements of high-precision crop yield estimation at the regional scale. Summary of the Invention

[0004] The present invention provides a method and system for estimating regional crop yield using particle filter data assimilation. This method addresses the problems in existing technologies where the Gaussian assumptions based on crop growth model state vectors and remote sensing observation vectors, along with the weak nonlinear assumptions of crop growth models, lead to errors in yield estimation due to inconsistencies with actual conditions. Furthermore, while the standard particle filter algorithm, which is well-suited for solving nonlinear and non-Gaussian problems, suffers from severe particle degradation when applied to the assimilation of remote sensing and crop growth models, resulting in unsatisfactory assimilation speed and accuracy. The method provided in this invention overcomes this problem of the standard particle filter algorithm, improving both assimilation speed and accuracy.

[0005] This invention provides a regional crop yield estimation method using particle filter data assimilation, comprising:

[0006] During the critical growth period of the crop, each particle in the initial particle set is run at fixed time intervals until each particle reaches the crop maturity stage, and the first leaf area index (LAI) output by each particle is obtained. Each particle in the initial particle set is obtained based on the crop growth model and the posterior sample set of the key parameters of the crop growth model. The posterior sample set is obtained by calibrating the key parameters of the crop growth model. The key parameters are the parameters in the crop growth model that are related to crop yield in the region.

[0007] Based on the posterior probability density function of the simulated LAI error for each particle, the updated first LAI, and the likelihood function, the initial weights of each particle are updated to obtain the updated weights of each particle. The updated first LAI is obtained by updating each first LAI according to the proposed transition probability density function of each first LAI. The likelihood function is a probability density function of Gaussian distribution. The mean of the likelihood function is the mean of all second LAIs in the first set, and the variance of the likelihood function is the variance of all second LAIs. The first set is a set composed of the second LAIs of crops in the region. The second LAI is obtained by adding Gaussian noise to the third LAI during the critical growth period of the crop. The third LAI is the LAI of the crop during the critical growth period obtained from the target product.

[0008] If the particle divergence is less than a preset value, each particle in the initial particle set is resampled to update the initial particle set. The preset value is determined according to a preset ratio of the number of particles, and the particle divergence is determined according to the update weight of each particle.

[0009] When each particle reaches the crop maturity stage, an estimated value of the crop yield is obtained based on the crop yield output by each particle and the final weight of each particle. The final weight is determined based on the updated weight corresponding to each particle at the crop maturity stage.

[0010] According to a method for estimating regional crop yield using particle filter data assimilation provided by the present invention, the method for obtaining the initial particle set includes:

[0011] Based on the crop yield in the region and the third LAI of the crop in the region, the key parameters in the crop growth model are calibrated to obtain the posterior sample set of the key parameters.

[0012] The initial particle set is obtained based on the posterior sample set and the crop growth model. Each particle in the initial particle set is obtained by inputting a sample into the crop growth model. The sample is obtained by randomly sampling from the posterior sample set.

[0013] According to a particle filter data assimilation method for regional crop yield estimation provided by the present invention, the key parameters in the crop growth model are calibrated based on the crop yield within the region and the third LAI of the crop within the region to obtain a posterior sample set of the key parameters, including:

[0014] Based on the crop yield and the third LAI, the key parameters are calibrated using the Markov chain Monte Carlo algorithm to obtain the posterior sample set.

[0015] According to the regional crop yield estimation method for particle filter data assimilation provided by the present invention, the method for obtaining the posterior probability density function of the LAI error of each particle simulation includes:

[0016] Run each particle to obtain the posterior probability density function of the simulated LAI error for each particle.

[0017] According to a method for estimating regional crop yield using particle filter data assimilation provided by the present invention, the method for obtaining the proposed transition probability density function of each first LAI includes:

[0018] The first set and the second set are input into the ensemble Kalman filter algorithm to obtain the proposed transition probability density function of each first LAI, and the second set is the set of first LAIs output by each particle.

[0019] According to a regional crop yield estimation method using particle filter data assimilation provided by the present invention, when each particle reaches the crop maturity stage, the step of obtaining an estimated value of the crop yield based on the crop yield output by each particle and the final weight of each particle includes:

[0020] The crop yield output by each particle is weighted and summed according to the final weight of each particle to obtain an estimated value of the crop yield.

[0021] The present invention also provides a regional crop yield estimation system using particle filter data assimilation, comprising: a data acquisition module, a weight update module, a particle update module, and a yield estimation module;

[0022] The data acquisition module is used to run each particle in the initial particle set at fixed time intervals during the critical growth period of the crop until each particle reaches the crop maturity period, and to acquire the first leaf area index (LAI) output by each particle. Each particle in the initial particle set is obtained based on the crop growth model and the posterior sample set of the key parameters of the crop growth model. The posterior sample set is obtained by calibrating the key parameters of the crop growth model. The key parameters are parameters in the crop growth model that are related to crop yield in the region.

[0023] The weight update module is used to update the initial weight of each particle based on the posterior probability density function of the simulated LAI error of each particle, the updated first LAI, and the likelihood function, to obtain the updated weight of each particle. The updated first LAI is obtained by updating each first LAI according to the proposed transition probability density function of each first LAI. The likelihood function is a probability density function of Gaussian distribution. The mean of the likelihood function is the mean of all second LAIs in the first set, and the variance of the likelihood function is the variance of all second LAIs. The first set is a set composed of the second LAIs of crops in the region. The second LAI is obtained by adding Gaussian noise to the third LAI during the critical growth period of the crop. The third LAI is the LAI of the crop during the critical growth period obtained from the target product.

[0024] The particle update module is used to resample each particle in the initial particle set and update the initial particle set when the particle divergence is less than a preset value. The preset value is determined according to a preset ratio of the number of particles, and the particle divergence is determined according to the update weight of each particle.

[0025] The yield estimation module is used to obtain an estimated value of the crop yield based on the crop yield output by each particle and the final weight of each particle when each particle reaches the crop maturity stage. The final weight is determined based on the updated weight corresponding to each particle at the crop maturity stage.

[0026] The present invention also provides an electronic device, including a processor and a memory storing a computer program, wherein the processor executes the program to implement a regional crop yield estimation method for particle filter data assimilation as described above.

[0027] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the regional crop yield estimation method as described above using particle filter data assimilation.

[0028] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements a regional crop yield estimation method for particle filter data assimilation as described above.

[0029] The regional crop yield estimation method and system provided by the present invention using particle filtering data assimilation updates the LAI output of each particle by introducing the proposed transition probability density function of the LAI output of each particle, and uses the updated LAI to update the initial weights of the particles. This improves the sequential sampling efficiency of particles, reduces the number of particle resampling, maintains the diversity of particle input parameters, and improves the speed and accuracy of crop yield prediction. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0031] Figure 1 This is one of the flowcharts of the regional crop yield estimation method for particle filter data assimilation provided by the present invention;

[0032] Figure 2 This is the second flowchart of the regional crop yield estimation method for particle filter data assimilation provided by the present invention.

[0033] Figure 3 This is a schematic diagram of the posterior probability density distribution of the LAI error simulated by the model provided by this invention;

[0034] Figure 4 This is a schematic diagram of the structure of the regional crop yield estimation system for particle filter data assimilation provided by the present invention;

[0035] Figure 5 This is a schematic diagram of the physical structure of the electronic device provided by the present invention. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0037] Existing techniques commonly used for assimilation of remote sensing and crop growth models, such as strongly constrained four-dimensional variational and ensemble Kalman filtering, are generally based on Gaussian assumptions regarding the crop growth model's state vector and remote sensing observation vector, as well as weak nonlinear assumptions about the crop growth model. This contradicts reality and leads to reduced prediction accuracy. Particle filtering algorithms, based on the idea of ​​Monte Carlo simulation, use a set of particles to simulate the probability density function of the assimilated state vector, eliminating the need for Gaussian and linear assumptions. However, directly applying particle filtering algorithms to the assimilation of remote sensing and crop growth models results in severe particle degradation, i.e., excessively large differences in particle weights and reduced particle divergence. This necessitates extensive resampling operations, causing the algorithm to run too slowly. Furthermore, the diversity of the particle input parameter set is lost, further reducing the accuracy of the Monte Carlo simulation.

[0038] The particle filtering data assimilation method for regional crop yield estimation provided by this invention can solve nonlinear and non-Gaussian assimilation problems. It calibrates crop growth models, such as the WOFOST model, based on the measured crop yield within the region and the LAI (third LAI) data provided in the target product GLASS 250_m LAI, obtaining the posterior sample set of key parameters and the posterior probability density function of the model's simulated LAI error. It samples from the posterior sample set of key parameters and inputs it into the WOFOST model to obtain the initial particle set and initial particle weights. It then inputs meteorological data to drive the WOFOST model and uses a particle filtering algorithm to assimilate the GLASS 250_m LAI data within the crop growth period. The proposed distribution of particle sampling is the posterior distribution of the state vector obtained after ensemble Kalman filtering. Finally, it performs a weighted summation of the estimated crop yield values ​​for each particle based on the final particle weights to obtain the estimated crop yield within the region.

[0039] The regional crop yield estimation method based on particle filter data assimilation provided by this invention can overcome the problems of low sampling efficiency and severe particle degradation when the standard particle filter algorithm is applied to remote sensing observation and crop growth model data assimilation, thereby improving the accuracy and speed of regional crop yield estimation. The specific implementation is as follows:

[0040] Figure 1 This is one of the flowcharts illustrating the regional crop yield estimation method using particle filter data assimilation provided by the present invention, such as... Figure 1 As shown, the method includes:

[0041] Step 110: During the critical growth period of the crop, run each particle in the initial particle set at fixed time intervals until each particle reaches the crop maturity period, and obtain the first leaf area index (LAI) output by each particle. Each particle in the initial particle set is obtained based on the crop growth model and the posterior sample set of the key parameters of the crop growth model. The posterior sample set is obtained by calibrating the key parameters of the crop growth model. The key parameters are the parameters in the crop growth model that are related to the crop yield in the region.

[0042] Step 120: Update the initial weights of each particle based on the posterior probability density function of the simulated LAI error for each particle, the updated first LAI, and the likelihood function, to obtain the updated weights of each particle. The updated first LAI is obtained by updating each first LAI based on the proposed transition probability density function of each first LAI. The likelihood function is a Gaussian probability density function, the mean of the likelihood function is the mean of all second LAIs in the first set, and the variance of the likelihood function is the variance of all second LAIs. The first set is a set of second LAIs of crops within the region. The second LAI is obtained by adding Gaussian noise to the third LAI within the critical growth period of the crop. The third LAI is the LAI of the crop within the critical growth period obtained from the target product.

[0043] Step 130: If the particle divergence is less than a preset value, resample each particle in the initial particle set and update the initial particle set. The preset value is determined according to a preset ratio of the number of particles, and the particle divergence is determined according to the update weight of each particle.

[0044] Step 140: When each particle reaches the crop maturity stage, an estimated value of the crop yield is obtained based on the crop yield output by each particle and the final weight of each particle. The final weight is determined based on the updated weight corresponding to each particle at the crop maturity stage.

[0045] It should be noted that the above method can be implemented by computer equipment.

[0046] Optionally, the critical growth period of the crop can specifically include various stages of the crop in the region from emergence to maturity (i.e., crop maturity). The critical growth period will vary for different crops. For example, the critical growth period of winter wheat can specifically include: emergence, greening, jointing, heading, flowering, and maturity.

[0047] This fixed time interval can be set freely as needed, for example, the fixed time interval can be set to daily, every three days, every five days, etc.

[0048] Each particle in the initial particle set is obtained by inputting a posterior sample set into a crop growth model such as the WOFOST model. Specifically, the posterior sample set can be obtained by calibrating the key parameters in the WOFOST model. These key parameters can be parameters in the WOFOST model related to crop yield (TWSO) in the region, such as the effective accumulated temperature from seedling emergence to flowering (TSUM1), the number of days a leaf survives at 35°C (SPAN), and the maximum CO2 assimilation rate per leaf (AMAXTB).

[0049] Assuming the fixed time interval is set to 1 day, each particle in the initial particle set is run day by day during the critical growth period of the crop until each particle has run to the maturity period of the crop, and the leaf area index (LAI) output by each particle run day by day is obtained (i.e. the first LAI).

[0050] The posterior probability density function of the LAI error simulated by each particle can be specifically obtained by plotting the first LAI output by each particle as a histogram and fitting the histogram with a suitable probability density function (e.g., Gaussian distribution).

[0051] Each updated first LAI can be specifically obtained by updating each first LAI according to the proposed transition probability density function of the first LAI output by each particle.

[0052] The likelihood function can be specifically a Gaussian likelihood function with the mean of all second LAIs in the first set as the mean and the variance of all second LAIs as the variance. The first set can be specifically a set of second LAIs of crops in the region. The second LAI can be specifically obtained by adding Gaussian noise (Gaussian perturbation) to the third LAI during the critical growth period of the crop. The third LAI can be specifically the LAI of the crop during the critical growth period of the crop obtained from the target product (GLASS 250_m LAI).

[0053] Based on the posterior probability density function of the simulated LAI error for each particle, the updated first LAI, and the likelihood function, the initial weights of each particle are updated to obtain the updated weights of each particle, i.e., the updated weights.

[0054] The particle divergence is calculated based on the updated weight of each particle, and then compared with a preset value. This preset value can be determined based on a preset proportion of the number of particles in the initial particle set, or more specifically, it can be greater than or equal to half the number of particles.

[0055] When the particle divergence is less than the preset value, each particle in the initial particle set is resampled to update the initial particle set. This resampling can specifically be done by resampling the particles using simple random resampling. The interval corresponding to the particle in [0, 1] is assigned according to the update weight of each particle. The length of the interval is proportional to the size of the update weight of the particle. N random numbers are generated that follow a uniform distribution in [0, 1], where N is the number of particles in the initial particle set. The particles corresponding to the interval to which each random number belongs are copied and added to the new particle set. Finally, the newly generated particle set replaces the initial particle set, and an initial weight is set for each particle in the newly generated particle set.

[0056] When each particle reaches the crop maturity stage, an estimated crop yield for the region is obtained based on the crop yield output by each particle and the final weight of each particle. This final weight is determined by the updated weight corresponding to each particle when it reaches the crop maturity stage.

[0057] The regional crop yield estimation method for particle filter data assimilation provided by this invention updates the LAI of each particle output by introducing the proposed transition probability density function of the LAI of each particle output, and uses the updated LAI to update the initial weight of the particle. This improves the sequential sampling efficiency of particles, reduces the number of particle resampling, maintains the diversity of particle input parameters, and improves the speed and accuracy of crop yield prediction.

[0058] Furthermore, in one embodiment, the method for obtaining the initial particle set may specifically include:

[0059] Based on the crop yield in the region and the third LAI of the crop in the region, the key parameters in the crop growth model are calibrated to obtain the posterior sample set of the key parameters.

[0060] The initial particle set is obtained based on the posterior sample set and the crop growth model. Each particle in the initial particle set is obtained by inputting a sample into the crop growth model. The sample is obtained by randomly sampling from the posterior sample set.

[0061] Further, in one embodiment, the step of calibrating the key parameters in the crop growth model based on the crop yield within the region and the third LAI of the crop within the region to obtain the posterior sample set of the key parameters may specifically include:

[0062] Based on the crop yield and the third LAI, the key parameters are calibrated using the Markov chain Monte Carlo algorithm to obtain the posterior sample set.

[0063] Optionally, based on crop yield within the region and the third LAI of crops within the region (such as the LAI observations obtained from GLASS 250_m LAI), the key parameters of the WOFOST model are calibrated using the Markov chain Monte Carlo (MCMC) algorithm to obtain the posterior sample set.

[0064] Specifically: using the measured yield per unit area as crop yield and GLASS 250_m LAI data during the critical growth period of crops as observations, the MCMC method is used to estimate the posterior probability distribution of key parameters in the WOFOST model within the region, and obtain the posterior sample set of key parameters.

[0065] Samples are obtained by sampling from the posterior sample set of the key parameter, and each sample is input into the WOFOST model to obtain each particle corresponding to the initial particle set.

[0066] Furthermore, in one embodiment, the method for obtaining the posterior probability density function of the simulated LAI error for each particle may specifically include:

[0067] Run each particle to obtain the posterior probability density function of the simulated LAI error for each particle.

[0068] Optionally, the posterior sample set is sampled to obtain a sampled sample, which is then input into the WOFOST model to obtain individual particles. Each particle is then run to obtain the posterior probability density function of the simulated LAI error for each particle. Specifically:

[0069] By plotting the histogram of the simulated LAI error of the WOFOST model under parameter adjustment after parameter posterior of the WOFOST model after calibration by the MCMC method, and selecting an appropriate probability density function for fitting based on the shape of the histogram, the posterior probability density function is obtained.

[0070] Furthermore, in one embodiment, the method for obtaining the proposed transition probability density function of each first LAI may specifically include:

[0071] The first set and the second set are input into the ensemble Kalman filter algorithm to obtain the proposed transition probability density function of each first LAI, and the second set is the set of first LAIs output by each particle.

[0072] Optionally, the first set and the second set are input into a ensemble Kalman filter algorithm to obtain the proposed transition probability density function for each first LAI, wherein the second set can be specifically the set of first LAIs output by each particle.

[0073] For example, running each particle daily, when there is a remote sensing observation (i.e., there is a third LAI obtained from GLASS250_m LAI), a Gaussian perturbation is added to the third LAI of GLASS 250_m LAI to generate a remote sensing observation set, i.e., the first set;

[0074] The set of remote sensing observations and the set of particle state variables LAI (i.e., the second set) are input into the ensemble Kalman filter algorithm to obtain the proposed transition probability density function of the first LAI, and the first LAI output by each particle is sampled and updated according to the probability density function.

[0075] Furthermore, in one embodiment, when each particle reaches the crop maturity stage, obtaining an estimated crop yield based on the crop yield output by each particle and the final weight of each particle may specifically include:

[0076] The crop yield output by each particle is weighted and summed according to the final weight of each particle to obtain an estimated value of the crop yield.

[0077] Optionally, the crop yield output by each particle at the crop maturity stage can be weighted and summed according to the final weight of each particle, and the resulting yield is the estimated value of the crop yield.

[0078] For example, Figure 2 This is the second flowchart illustrating the regional crop yield estimation method using particle filter data assimilation provided by this invention. Figure 2 As shown:

[0079] Province X was selected as the study area, and the third LAI of GLASS250_m LAI in the study area from March to May 2021 was selected.

[0080] S1, based on the third LAI of crop yield and critical growth period in the region, the WOFOST model is calibrated using the MCMC method to obtain the posterior sample set of key parameters in the WOFOST model;

[0081] Step S1 specifically involves using measured yield per unit area (crop yield) and GLASS 250_mLAI data during the critical growth period of the crop as observations. The MCMC method is used to estimate the posterior probability distribution of WOFOST model parameters within the study area, obtaining the posterior sample set of key parameters. These key parameters are those in the WOFOST model that are sensitive to crop yield TWSO, such as TSUM1, SPAN, and AMAXTB.

[0082] S2, sample from the posterior sample set and input into the WOFOST model to obtain the initial particle set and the posterior probability density function of the WOFOST model to simulate the LAI error, and set the initial weights of the particles;

[0083] Step S2 specifically involves plotting a histogram of the uncertainty (error of LAI value) of the simulated model under parameter adjustment after MCMC calibration of the WOFOST model, and selecting an appropriate probability density function (posterior probability density function) for fitting based on the shape of the histogram. Here, a Gaussian distribution is chosen, such as... Figure 3 As shown. The initial weights of all particles are set to 0. N is the number of particles in the initial particle set.

[0084] S3 runs each particle in the particle set every day. When there is a remote sensing observation (i.e., the third LAI), a Gaussian perturbation is added to the third LAI to generate a remote sensing observation set.

[0085] S4. Input the remote sensing observation set (first set) and the set of particle state variables LAI (second set) into the ensemble Kalman filter algorithm to obtain the proposed transition probability density function of the LAI of each particle, and sample and update the first LAI output by each particle according to the proposed transition probability density function.

[0086] Step S4 specifically involves inputting the set of remote sensing observations and the set of particle state variables (LAIs) into a set Kalman filter algorithm to obtain the posterior distribution of the LAI for each particle, and then extracting a sample from the posterior distribution to update the first LAI of the particle output.

[0087] S5, run each particle to obtain the posterior probability density function of the simulated LAI error, and calculate the updated weight of each particle based on the posterior probability density function, the likelihood function and the proposed transition probability density function of each first LAI.

[0088] Step S5 specifically involves constructing a Gaussian likelihood function with the mean of the LAI in the remote sensing observation set as the mean and the variance of the LAI in the remote sensing observation set as the variance. Based on the WOFOST model corresponding to the first LAI after particle update, the posterior probability density function value, likelihood function value, and proposed transition probability density function value of the LAI error are simulated, and the particle update weight is calculated according to the following formula:

[0089]

[0090] In the formula, The updated weight for the i-th particle; z i Let f(d|z) be the updated LAI value for the i-th particle, d be the LAI observation value provided by the GLASS 250_mLAI product, and f(d|z) be the likelihood function;R (·) represents the posterior probability density function of the LAI error simulated by WOFOST; q(z) represents the proposed transition probability density function; w i Let be the initial weight of the i-th particle.

[0091] S6. Calculate the particle divergence based on the updated particle weights obtained in S5. If the particle divergence value is less than half the number of particles, resample the particle set.

[0092] Step S6 specifically involves calculating the particle divergence using the following formula:

[0093]

[0094] Where Neff represents particle divergence.

[0095] If the particle divergence is less than half the number of particles in the set, the particles in the initial particle set are resampled. Specifically, simple random resampling is used to resample the particles. First, the interval corresponding to the particle in [0, 1] is assigned according to the updated weight of each particle. The length of the interval is proportional to the size of the updated weight of the particle. Second, N random numbers that follow a uniform distribution in [0, 1] are generated, where N is the number of particles in the initial particle set. The particles corresponding to the interval to which each random number belongs are copied and added to the new particle set. Finally, the newly generated particle set replaces the initial particle set, and the weight of each particle is set to...

[0096] S7. Repeat steps S3 to S6 until all particles' corresponding models reach the crop maturity stage. Output all yield results and final particle weights from the model run on the last day.

[0097] S8, based on the final particle weights of S7, performs a weighted summation of the yield results of each particle, and the resulting yield is the estimated yield of crops in the region, i.e., the estimated value of crop yield.

[0098] The regional crop yield estimation method based on particle filter data assimilation provided by this invention improves the sequential sampling efficiency of particles by introducing the posterior distribution of LAI obtained by the ensemble Kalman filter algorithm as the proposal distribution and sampling from the proposal distribution, greatly reducing the number of resampling times, maintaining the diversity of the particle input parameter set, and improving the prediction speed and accuracy.

[0099] The regional crop yield estimation system based on particle filter data assimilation provided by this invention is described below. The regional crop yield estimation system based on particle filter data assimilation described below can be referred to in correspondence with the regional crop yield estimation method based on particle filter data assimilation described above.

[0100] Figure 4This is a schematic diagram of the regional crop yield estimation system for particle filter data assimilation provided by the present invention, as shown in the figure. Figure 4 As shown, it includes:

[0101] The module includes a data acquisition module 410, a weight update module 411, a particle update module 412, and a yield estimation module 413.

[0102] The data acquisition module 410 is used to run each particle in the initial particle set at fixed time intervals during the critical growth period of the crop until each particle reaches the maturity period of the crop, and to acquire the first leaf area index (LAI) output by each particle. Each particle in the initial particle set is obtained based on the crop growth model and the posterior sample set of the key parameters of the crop growth model. The posterior sample set is obtained by calibrating the key parameters of the crop growth model. The key parameters are parameters in the crop growth model that are related to the crop yield in the region.

[0103] The weight update module 411 is used to update the initial weight of each particle according to the posterior probability density function of the simulated LAI error of each particle, the updated first LAI, and the likelihood function, to obtain the updated weight of each particle. The updated first LAI is obtained by updating each first LAI according to the proposed transition probability density function of each first LAI. The likelihood function is a probability density function of Gaussian distribution. The mean of the likelihood function is the mean of all second LAIs in the first set, and the variance of the likelihood function is the variance of all second LAIs. The first set is a set of second LAIs of crops in the region. The second LAI is obtained by adding Gaussian noise to the third LAI during the critical growth period of the crop. The third LAI is the LAI of the crop during the critical growth period of the crop obtained from the target product.

[0104] The particle update module 412 is used to resample each particle in the initial particle set and update the initial particle set when the particle divergence is less than a preset value. The preset value is determined according to a preset ratio of the number of particles, and the particle divergence is determined according to the update weight of each particle.

[0105] The yield estimation module 413 is used to obtain an estimated value of the crop yield based on the crop yield output by each particle and the final weight of each particle when each particle reaches the crop maturity stage. The final weight is determined based on the updated weight corresponding to each particle at the crop maturity stage.

[0106] The regional crop yield estimation system for particle filtering data assimilation provided by this invention updates the LAI of each particle output by introducing the proposed transition probability density function of the LAI of each particle output, and uses the updated LAI to update the initial weights of the particles. This improves the sequential sampling efficiency of particles, reduces the number of particle resampling, maintains the diversity of particle input parameters, and improves the speed and accuracy of crop yield prediction.

[0107] Figure 5 This is a schematic diagram of the physical structure of an electronic device provided by the present invention, such as... Figure 5 As shown, the electronic device may include a processor 510, a communication interface 511, a memory 512, and a bus 513, wherein the processor 510, the communication interface 511, and the memory 512 communicate with each other via the bus 513. The processor 510 can call logical instructions in the memory 512 to execute the following methods:

[0108] During the critical growth period of the crop, each particle in the initial particle set is run at fixed time intervals until each particle reaches the crop maturity stage, and the first leaf area index (LAI) output by each particle is obtained. Each particle in the initial particle set is obtained based on the crop growth model and the posterior sample set of the key parameters of the crop growth model. The posterior sample set is obtained by calibrating the key parameters of the crop growth model. The key parameters are the parameters in the crop growth model that are related to crop yield in the region.

[0109] Based on the posterior probability density function of the simulated LAI error for each particle, the updated first LAI, and the likelihood function, the initial weights of each particle are updated to obtain the updated weights of each particle. The updated first LAI is obtained by updating each first LAI according to the proposed transition probability density function of each first LAI. The likelihood function is a probability density function of Gaussian distribution. The mean of the likelihood function is the mean of all second LAIs in the first set, and the variance of the likelihood function is the variance of all second LAIs. The first set is a set composed of the second LAIs of crops in the region. The second LAI is obtained by adding Gaussian noise to the third LAI during the critical growth period of the crop. The third LAI is the LAI of the crop during the critical growth period obtained from the target product.

[0110] If the particle divergence is less than a preset value, each particle in the initial particle set is resampled to update the initial particle set. The preset value is determined according to a preset ratio of the number of particles, and the particle divergence is determined according to the update weight of each particle.

[0111] When each particle reaches the crop maturity stage, an estimated value of the crop yield is obtained based on the crop yield output by each particle and the final weight of each particle. The final weight is determined based on the updated weight corresponding to each particle at the crop maturity stage.

[0112] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer power supply (which may be a personal computer, server, or network power supply, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0113] Furthermore, this invention discloses a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, and when these instructions are executed by a computer, the computer can execute the regional crop yield estimation method for particle filter data assimilation provided in the above-described method embodiments, for example including:

[0114] During the critical growth period of the crop, each particle in the initial particle set is run at fixed time intervals until each particle reaches the crop maturity stage, and the first leaf area index (LAI) output by each particle is obtained. Each particle in the initial particle set is obtained based on the crop growth model and the posterior sample set of the key parameters of the crop growth model. The posterior sample set is obtained by calibrating the key parameters of the crop growth model. The key parameters are the parameters in the crop growth model that are related to crop yield in the region.

[0115] Based on the posterior probability density function of the simulated LAI error for each particle, the updated first LAI, and the likelihood function, the initial weights of each particle are updated to obtain the updated weights of each particle. The updated first LAI is obtained by updating each first LAI according to the proposed transition probability density function of each first LAI. The likelihood function is a probability density function of Gaussian distribution. The mean of the likelihood function is the mean of all second LAIs in the first set, and the variance of the likelihood function is the variance of all second LAIs. The first set is a set composed of the second LAIs of crops in the region. The second LAI is obtained by adding Gaussian noise to the third LAI during the critical growth period of the crop. The third LAI is the LAI of the crop during the critical growth period obtained from the target product.

[0116] If the particle divergence is less than a preset value, each particle in the initial particle set is resampled to update the initial particle set. The preset value is determined according to a preset ratio of the number of particles, and the particle divergence is determined according to the update weight of each particle.

[0117] When each particle reaches the crop maturity stage, an estimated value of the crop yield is obtained based on the crop yield output by each particle and the final weight of each particle. The final weight is determined based on the updated weight corresponding to each particle at the crop maturity stage.

[0118] On the other hand, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the regional crop yield estimation method for particle filter data assimilation provided in the above embodiments, including, for example:

[0119] During the critical growth period of the crop, each particle in the initial particle set is run at fixed time intervals until each particle reaches the crop maturity stage, and the first leaf area index (LAI) output by each particle is obtained. Each particle in the initial particle set is obtained based on the crop growth model and the posterior sample set of the key parameters of the crop growth model. The posterior sample set is obtained by calibrating the key parameters of the crop growth model. The key parameters are the parameters in the crop growth model that are related to crop yield in the region.

[0120] Based on the posterior probability density function of the simulated LAI error for each particle, the updated first LAI, and the likelihood function, the initial weights of each particle are updated to obtain the updated weights of each particle. The updated first LAI is obtained by updating each first LAI according to the proposed transition probability density function of each first LAI. The likelihood function is a probability density function of Gaussian distribution. The mean of the likelihood function is the mean of all second LAIs in the first set, and the variance of the likelihood function is the variance of all second LAIs. The first set is a set composed of the second LAIs of crops in the region. The second LAI is obtained by adding Gaussian noise to the third LAI during the critical growth period of the crop. The third LAI is the LAI of the crop during the critical growth period obtained from the target product.

[0121] If the particle divergence is less than a preset value, each particle in the initial particle set is resampled to update the initial particle set. The preset value is determined according to a preset ratio of the number of particles, and the particle divergence is determined according to the update weight of each particle.

[0122] When each particle reaches the crop maturity stage, an estimated value of the crop yield is obtained based on the crop yield output by each particle and the final weight of each particle. The final weight is determined based on the updated weight corresponding to each particle at the crop maturity stage.

[0123] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0124] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer power supply (which may be a personal computer, server, or network power supply, etc.) to execute the methods described in various embodiments or some parts of the embodiments.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for estimating regional crop yield using particle filter data assimilation, characterized in that, include: During the critical growth period of the crop, each particle in the initial particle set is run at fixed time intervals until each particle reaches the crop maturity stage, and the first leaf area index (LAI) output by each particle is obtained. Each particle in the initial particle set is obtained based on the crop growth model and the posterior sample set of the key parameters of the crop growth model. The posterior sample set is obtained by calibrating the key parameters of the crop growth model. The key parameters are the parameters in the crop growth model that are related to crop yield in the region. Based on the posterior probability density function of the simulated LAI error for each particle, the updated first LAI, and the likelihood function, the initial weights of each particle are updated to obtain the updated weights of each particle. The updated first LAI is obtained by updating each first LAI according to the proposed transition probability density function of each first LAI. The likelihood function is a probability density function of Gaussian distribution. The mean of the likelihood function is the mean of all second LAIs in the first set, and the variance of the likelihood function is the variance of all second LAIs. The first set is a set composed of the second LAIs of crops in the region. The second LAI is obtained by adding Gaussian noise to the third LAI during the critical growth period of the crop. The third LAI is the LAI of the crop during the critical growth period obtained from the target product. The method for obtaining the proposed transition probability density function of each first LAI includes: inputting the first set and the second set into a ensemble Kalman filter algorithm to obtain the proposed transition probability density function of each first LAI, wherein the second set is the set composed of the first LAIs output by each particle; the formula for calculating the update weight of each particle is: ； in, For the first Update weights for each particle. For the updated version LAI value of each particle, LAI observations provided for the GLASS 250_mLAI product. Let be the likelihood function. The posterior probability density function of the LAI error is simulated using WOFOST. To propose a transition probability density function, For the first The initial weights of each particle; If the particle divergence is less than a preset value, each particle in the initial particle set is resampled to update the initial particle set. The preset value is determined based on a preset proportion of the number of particles, and the particle divergence is determined based on the update weight of each particle. The formula for calculating the particle divergence is: ； in, Represents particle divergence, The number of particles in the initial particle set; When each particle reaches the crop maturity stage, an estimated value of the crop yield is obtained based on the crop yield output by each particle and the final weight of each particle. The final weight is determined based on the updated weight corresponding to each particle at the crop maturity stage.

2. The regional crop yield estimation method based on particle filter data assimilation according to claim 1, characterized in that, The method for obtaining the initial particle set includes: Based on the crop yield in the region and the third LAI of the crop in the region, the key parameters in the crop growth model are calibrated to obtain the posterior sample set of the key parameters. The initial particle set is obtained based on the posterior sample set and the crop growth model. Each particle in the initial particle set is obtained by inputting a sample into the crop growth model. The sample is obtained by randomly sampling from the posterior sample set.

3. The regional crop yield estimation method based on particle filter data assimilation according to claim 2, characterized in that, The key parameters in the crop growth model are calibrated based on the crop yield within the region and the third LAI of the crops within the region to obtain a posterior sample set of the key parameters, including: Based on the crop yield and the third LAI, the key parameters are calibrated using the Markov chain Monte Carlo algorithm to obtain the posterior sample set.

4. The regional crop yield estimation method based on particle filter data assimilation according to claim 1, characterized in that, The method for obtaining the posterior probability density function of the LAI error for each particle simulation includes: Run each particle to obtain the posterior probability density function of the simulated LAI error for each particle.

5. The regional crop yield estimation method based on particle filter data assimilation according to any one of claims 1-4, characterized in that, When each particle reaches the crop maturity stage, obtaining an estimated crop yield based on the crop yield output by each particle and the final weight of each particle includes: The crop yield output by each particle is weighted and summed according to the final weight of each particle to obtain an estimated value of the crop yield.

6. A regional crop yield estimation system using particle filter data assimilation, characterized in that, include: The module includes a data acquisition module, a weight update module, a particle update module, and a yield estimation module. The data acquisition module is used to run each particle in the initial particle set at fixed time intervals during the critical growth period of the crop until each particle reaches the crop maturity period, and to acquire the first leaf area index (LAI) output by each particle. Each particle in the initial particle set is obtained based on the crop growth model and the posterior sample set of the key parameters of the crop growth model. The posterior sample set is obtained by calibrating the key parameters of the crop growth model. The key parameters are parameters in the crop growth model that are related to crop yield in the region. The weight update module is used to update the initial weight of each particle based on the posterior probability density function of the simulated LAI error of each particle, the updated first LAI, and the likelihood function, to obtain the updated weight of each particle. The updated first LAI is obtained by updating each first LAI according to the proposed transition probability density function of each first LAI. The likelihood function is a probability density function of Gaussian distribution. The mean of the likelihood function is the mean of all second LAIs in the first set, and the variance of the likelihood function is the variance of all second LAIs. The first set is a set composed of the second LAIs of crops in the region. The second LAI is obtained by adding Gaussian noise to the third LAI during the critical growth period of the crop. The third LAI is the LAI of the crop during the critical growth period obtained from the target product. The method for obtaining the proposed transition probability density function of each first LAI includes: inputting the first set and the second set into a ensemble Kalman filter algorithm to obtain the proposed transition probability density function of each first LAI, wherein the second set is the set composed of the first LAIs output by each particle; the formula for calculating the update weight of each particle is: ； in, For the first Update weights for each particle. For the updated version LAI value of each particle, LAI observations provided for the GLASS 250_mLAI product. Let be the likelihood function. The posterior probability density function of the LAI error is simulated using WOFOST. To propose a transition probability density function, For the first The initial weights of each particle; The particle update module is used to resample each particle in the initial particle set and update the initial particle set when the particle divergence is less than a preset value. The preset value is determined based on a preset proportion of the number of particles, and the particle divergence is determined based on the update weight of each particle. The formula for calculating the particle divergence is: ； in, Represents particle divergence, The number of particles in the initial particle set; The yield estimation module is used to obtain an estimated value of the crop yield based on the crop yield output by each particle and the final weight of each particle when each particle reaches the crop maturity stage. The final weight is determined based on the updated weight corresponding to each particle at the crop maturity stage.

7. An electronic device comprising a processor and a memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the regional crop yield estimation method based on particle filter data assimilation as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the regional crop yield estimation method based on particle filter data assimilation as described in any one of claims 1 to 5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the regional crop yield estimation method for particle filter data assimilation as described in any one of claims 1 to 5.

Images