A method and system for forecasting winter wheat meteorological yield

Abstract

The present invention provides a winter wheat meteorological yield prediction method and system, the method comprising: performing band calculation on winter wheat MODIS remote sensing data based on a preset time series, obtaining MODIS remote sensing information characteristics of a winter wheat target study area; dividing time periods according to a preset time series, collecting environmental data and management data of the winter wheat target study area, and inputting them into a WOFOST model, obtaining WOFOST simulation quantity characteristics of the winter wheat growth period in different time periods; inputting MODIS remote sensing data and WOFOST simulation quantity characteristics into a trained winter wheat meteorological yield prediction model, and obtaining a winter wheat meteorological yield prediction result. The present invention adds the simulation quantity from jointing to heading period of the WOFOST crop model, constructs a yield prediction model integrating WOFOST and LSTM, and improves the yield prediction effect.

Description

Technical Field

[0001] The present invention relates to the field of agricultural remote sensing technology, and in particular to a method and system for forecasting meteorological yield of winter wheat. Background Art

[0002] In recent years, the issue of wheat yield has received widespread attention from the society. Effectively grasping the information on wheat growth process and making yield forecasts are of great significance for adjusting the wheat planting structure and making agricultural decisions by relevant departments.

[0003] Remote sensing data has the advantages of high resolution, low cost, and wide regional coverage, and is used in yield prediction research. Many studies resample remote sensing data to a monthly scale and use different types of remote sensing data to predict crop yields. In addition, deep learning plays an important role in artificial intelligence. By setting up complex networks, it has a high feature analysis capability, fast speed, high accuracy, and obvious advantages in processing multi-dimensional features.

[0004] With the continuous development of remote sensing data and deep learning, researchers at home and abroad have combined the two to predict crop yields. Many studies have incorporated remote sensing data in the form of different crop growth periods, different growth period combinations, and histograms into deep learning yield prediction models. However, the effect of deep learning depends on feature samples, which are easily affected by the growth environment of crops in different periods, resulting in inaccurate prediction of crop yields. Therefore, there is an urgent need for a winter wheat meteorological yield prediction method and system to solve the above problems. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides a method and system for forecasting winter wheat meteorological yield.

[0006] The present invention provides a winter wheat meteorological yield prediction method, comprising:

[0007] Perform band calculation on the winter wheat MODIS remote sensing data based on the preset time series to obtain the MODIS remote sensing information characteristics of the winter wheat target study area;

[0008] Dividing time periods according to the preset time series, collecting environmental data and management data of the winter wheat target research area, and inputting them into the WOFOST model to obtain WOFOST simulation quantity characteristics of the winter wheat growth period in different time periods;

[0009] The MODIS remote sensing information features and the WOFOST simulation features are input into a trained winter wheat meteorological yield prediction model to obtain the yield prediction result of the winter wheat target study area, wherein the trained winter wheat meteorological yield prediction model is obtained by training a long short-term memory neural network with sample MODIS remote sensing information features, sample WOFOST simulation features and sample meteorological yield features.

[0010] According to a winter wheat meteorological yield prediction method provided by the present invention, the trained winter wheat meteorological yield prediction model is obtained by training through the following steps:

[0011] Perform band calculation on MODIS remote sensing data based on a preset time series to obtain a normalized vegetation index and a normalized vegetation moisture index, draw a frequency histogram, obtain the frequency value of the frequency histogram, and construct sample MODIS remote sensing information features;

[0012] Divide the time periods according to the preset time series, extract the WOFOST simulation quantities of each ear differentiation period of winter wheat from jointing to heading period corresponding to each time period, and construct sample WOFOST simulation quantity characteristics, wherein the sample WOFOST simulation quantity characteristics include crop growth process, leaf area index, dry matter mass and physiological action characteristics;

[0013] Based on the meteorological production data, the sample meteorological production characteristics are constructed;

[0014] The sample MODIS remote sensing information features, the sample WOFOST simulation features and the sample meteorological yield features are input into the winter wheat meteorological yield prediction model for training to obtain a trained winter wheat meteorological yield prediction model.

[0015] According to a winter wheat meteorological yield prediction method provided by the present invention, constructing sample meteorological yield characteristics according to meteorological yield data includes:

[0016] According to the 5-year moving average method, the actual yield sample data of winter wheat was detrended to obtain the trend yield;

[0017] According to the difference between the actual yield sample data and the trend yield, meteorological yield data is obtained to construct sample meteorological yield characteristics.

[0018] According to a winter wheat meteorological yield prediction method provided by the present invention, before dividing the time periods according to the preset time series, collecting the environmental data and management data of the winter wheat target research area, and inputting them into the WOFOST model, and obtaining the WOFOST simulation quantity characteristics of the winter wheat growth period in different time periods, the method also includes: localizing and calibrating the WOFOST model.

[0019] According to a winter wheat meteorological yield prediction method provided by the present invention, the long short-term memory neural network includes 1 input layer, 5 hidden layers, 1 fully connected layer and 1 output layer, wherein the input layer standardizes the data through centralization and converts the data into a standard normal distribution.

[0020] According to a winter wheat meteorological yield prediction method provided by the present invention, before performing band calculation on the winter wheat MODIS remote sensing data based on a preset time series, the method further includes:

[0021] The winter wheat MODIS remote sensing data based on the preset time series were mosaicked, projected and clipped.

[0022] According to a winter wheat meteorological yield prediction method provided by the present invention, the band operation is performed on MODIS remote sensing data based on a preset time series to obtain a normalized vegetation index and a normalized vegetation moisture index, including:

[0023] According to the pixel values ​​of near-infrared band 1 and red light band in MOD09A1, the normalized vegetation index is obtained, and the formula is:

[0024] NDVI = (NIR1-R) / (NIR1+R);

[0025] According to the pixel values ​​of near infrared band 1 and near red band 2 in MOD09A1, the normalized vegetation moisture index is obtained, and the formula is:

[0026] NDWI=(NIR1-NIR2) / (NIR1+NIR2);

[0027] Among them, NDVI represents the normalized difference vegetation index, NDWI represents the normalized difference vegetation water index; R represents the pixel value of the red light band, corresponding to the original effective value of the first band of MOD09A1; NIR1 represents the pixel value of the near infrared band 1, corresponding to the original effective value of the second band of MOD09A1; NIR2 represents the pixel value of the near infrared band 2, corresponding to the original effective value of the fifth band of MOD09A1.

[0028] The present invention also provides a winter wheat meteorological yield prediction system, comprising:

[0029] The remote sensing information feature acquisition module is used to perform band calculation on the winter wheat MODIS remote sensing data based on the preset time series to obtain the MODIS remote sensing information features of the winter wheat target research area;

[0030] A simulation quantity characteristic acquisition module is used to divide the time period according to the preset time series, collect the environmental data and management data of the winter wheat target research area, and input them into the WOFOST model to obtain the WOFOST simulation quantity characteristics of the winter wheat growth period in different time periods;

[0031] The meteorological yield prediction module is used to input the MODIS remote sensing information characteristics and the WOFOST simulation characteristics into the trained winter wheat meteorological yield prediction model to obtain the yield prediction result of the winter wheat target study area, wherein the trained winter wheat meteorological yield prediction model is obtained by training the long short-term memory neural network with sample MODIS remote sensing information characteristics, sample WOFOST simulation characteristics and sample meteorological yield characteristics.

[0032] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the above-mentioned winter wheat meteorological yield prediction methods are implemented.

[0033] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above-mentioned winter wheat meteorological yield prediction methods are implemented.

[0034] The winter wheat meteorological yield prediction method and system provided by the present invention perform band calculation on winter wheat MODIS remote sensing data to obtain MODIS remote sensing information characteristics of a winter wheat target study area, use a WOFOST model to obtain WOFOST simulation quantity characteristics of the winter wheat growth period in different time periods, and then input the MODIS remote sensing data and the WOFOST simulation quantity characteristics into a trained winter wheat meteorological yield prediction model to obtain a meteorological yield prediction result of the winter wheat target study area. Compared with the traditional winter wheat yield prediction method, by drawing on the idea of ​​data assimilation, on the basis of LSTM deep learning based on MODIS remote sensing information, the simulation quantity of the WOFOST crop model from jointing to heading period is added, a yield prediction model integrating WOFOST and LSTM is constructed, and the yield prediction effect is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0036] Figure 1A schematic flow chart of the winter wheat meteorological yield prediction method provided by the present invention;

[0037] Figure 2 A schematic diagram of prediction results using a combination of crop growth process, total dry matter weight, and physiological action simulation as characteristic variables in the winter wheat meteorological yield prediction model provided by the present invention;

[0038] Figure 3 A schematic diagram showing the comparison of the results of the winter wheat meteorological yield prediction model provided by the present invention in different temperature ranges with the model without adding WOFOST simulation;

[0039] Figure 4 A schematic diagram of the LSTM structure in the winter wheat meteorological yield prediction model provided by the present invention;

[0040] Figure 5 A schematic diagram of the structure of the winter wheat meteorological yield prediction system provided by the present invention;

[0041] Figure 6 This is a schematic structural diagram of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] Figure 1 The schematic diagram of the flow chart of the winter wheat meteorological yield prediction method provided by the present invention is as follows: Figure 1 As shown, the present invention provides a winter wheat meteorological yield prediction method, comprising:

[0044] Step 101, performing band calculation on winter wheat MODIS remote sensing data based on a preset time series to obtain MODIS remote sensing information characteristics of a winter wheat target research area;

[0045] Step 102, dividing the time periods according to the preset time series, collecting the environmental data and management data of the winter wheat target research area, and inputting them into the WOFOST model to obtain the WOFOST simulation quantity characteristics of the winter wheat growth period in different time periods;

[0046] Step 103, input the MODIS remote sensing information features and the WOFOST simulation features into the trained winter wheat meteorological yield prediction model to obtain the yield prediction result of the winter wheat target study area, wherein the trained winter wheat meteorological yield prediction model is obtained by training a long short-term memory neural network with sample MODIS remote sensing information features, sample WOFOST simulation features and sample meteorological yield features.

[0047] In the present invention, in step 101, the remote sensing data of the Moderate-resolution Imaging Spectroradiometer (MODIS) of winter wheat at different growth stages are obtained, and the band operation is performed on the MODIS remote sensing data of winter wheat in the order of time series to obtain the required Normalized Difference Vegetation Index (NDVI) and Normalized Difference Water Index (NDWI), and a frequency histogram is drawn to obtain the MODIS remote sensing information characteristics of the winter wheat target study area. Optionally, the order of the time series is based on the growth stage of winter wheat from jointing to heading, and the MODIS remote sensing data synthesized in 8 days are arranged in order of time.

[0048] Furthermore, in step 102, the preset time series is divided into a plurality of time periods, each time period corresponds to the differentiation period of each ear of winter wheat, and the crop, soil, meteorological, and management data are used as input data of the World Food Studies (WOFOST) crop model to obtain the WOFOST simulation characteristics of the winter wheat growth period in different time periods.

[0049] Among them, crop data includes the selected simulated crop varieties, soil data includes soil type and related soil hydrological characteristics, meteorological data includes meteorological data format, available sites, rainfall, temperature and other data; management data includes the start time of crop simulation, the end time of crop simulation, and the maximum duration of the crop cycle.

[0050] Furthermore, in step 103, the image histogram frequency value of the MODIS remote sensing information feature and the WOFOST multi-type analog quantity feature are combined and input into the winter wheat meteorological yield prediction model as feature variables, and the trained winter wheat meteorological yield prediction model is used to perform prediction to obtain the yield prediction result of the winter wheat target study area. Among them, the winter wheat meteorological yield prediction model is a long short-term memory neural network (LSTM) model.

[0051] The winter wheat meteorological yield prediction method provided by the present invention performs band calculation on winter wheat MODIS remote sensing data to obtain MODIS remote sensing information characteristics of a winter wheat target study area, uses a WOFOST model to obtain WOFOST simulation quantity characteristics of the winter wheat growth period in different time periods, and then inputs the MODIS remote sensing information characteristics and the WOFOST simulation quantity characteristics into a trained winter wheat meteorological yield prediction model to obtain a meteorological yield prediction result of the winter wheat target study area. Compared with the traditional winter wheat yield prediction method, by drawing on the idea of ​​data assimilation, on the basis of LSTM deep learning based on MODIS remote sensing information, the simulation quantity of the WOFOST crop model from jointing to heading period is added, a yield prediction model integrating WOFOST and LSTM is constructed, and the yield prediction effect is improved.

[0052] Based on the above embodiment, the trained winter wheat meteorological yield prediction model is obtained by training through the following steps:

[0053] Step 201, performing band operation on MODIS remote sensing data based on a preset time series to obtain a normalized vegetation index and a normalized vegetation moisture index, and drawing a frequency histogram to obtain frequency values ​​of the frequency histogram to construct sample MODIS remote sensing information features;

[0054] Step 202, dividing the time periods according to the preset time series, extracting the WOFOST simulation quantities of each ear differentiation stage of winter wheat from jointing to heading period corresponding to each time period, and constructing sample WOFOST simulation quantity characteristics, wherein the sample WOFOST simulation quantity characteristics include crop growth process, leaf area index, dry matter mass and physiological action characteristics;

[0055] Step 203, constructing sample meteorological production characteristics according to meteorological production data;

[0056] Step 204: input the sample MODIS remote sensing information features, the sample WOFOST simulation quantity features and the sample meteorological yield features into a winter wheat meteorological yield prediction model for training, so as to obtain a trained winter wheat meteorological yield prediction model.

[0057] In the present invention, in step 201, MODIS remote sensing data synthesized for 8 days from jointing to heading in 2004-2017 is obtained based on the MOD09A1 product with a resolution of 500m, and the product is obtained from LAADS Web. Band operation is performed on the MODIS remote sensing data synthesized for 8 days from jointing to heading of winter wheat to obtain the normalized vegetation index and the normalized vegetation moisture index. The specific method is as follows:

[0058] According to the pixel values ​​of near-infrared band 1 and red light band in MOD09A1, the normalized vegetation index is obtained, and the formula is expressed as follows:

[0059] NDVI = (NIR1-R) / (NIR1+R);

[0060] According to the pixel values ​​of near infrared band 1 and near red band 2 in MOD09A1, the normalized vegetation moisture index is obtained, and the formula is:

[0061] NDWI=(NIR1-NIR2) / (NIR1+NIR2);

[0062] Among them, NDVI represents the normalized difference vegetation index, NDWI represents the normalized vegetation water index; MOD09A1 represents the 8-day synthetic data product with a resolution of 500 m for bands 1-7, which is used to estimate the surface spectral reflectance of the study area; R represents the pixel value of the red band, corresponding to the original effective value of the first band of MOD09A1; NIR1 represents the pixel value of the near infrared band 1, corresponding to the original effective value of the second band of MOD09A1; NIR2 represents the pixel value of the near infrared band 2, corresponding to the original effective value of the fifth band of MOD09A1.

[0063] Furthermore, in order to solve the problems of inconsistent shapes and ranges of winter wheat planting areas in different counties, the pixel values ​​of remote sensing images of each county are counted by frequency histogram. Statistics show that NDVI is between 0 and 1, and NDWI is between -0.25 and 0.25. The present invention divides the NDVI of the 7 time periods from jointing to heading into 5 intervals at an interval of 0.2, and divides the NDWI into 5 intervals at an interval of 0.1. The frequency values ​​of each interval generated by each image histogram are selected as feature values, totaling 70 feature variables, to complete the construction of MODIS remote sensing information sample features.

[0064] In the present invention, in step 202, in order to match the MODIS remote sensing data synthesized in 8 days, the data obtained in step S1 from jointing to heading stage is divided into 7 time periods in units of 8 days, each time period corresponds to each ear differentiation stage, and the WOFOST simulation quantities of each ear differentiation stage from jointing to heading stage are extracted, which are mainly divided into four aspects: crop growth process, leaf area index, dry matter weight, and physiological effect, thereby completing the construction of the sample characteristics of the WOFOST simulation information.

[0065] Among them, the average jointing to heading period of winter wheat in Henan Province, which is prone to late frost from March 14 to May 8, was selected, and the jointing to heading period was divided into 7 time periods with 8 days as a unit. Each time period corresponds to each ear differentiation period, as shown in Table 1:

[0066] Table 1

[0067] Time period date Ear differentiation stage 1 3 / 14-3 / 21 Early floret differentiation stage 2 3 / 22-3 / 29 Late floret differentiation 3 3 / 30-4 / 6 Pistil and pistil differentiation stage 4 4 / 7-4 / 14 Early medicinal period 5 4 / 15-4 / 22 Late medicinal period 6 4 / 23-4 / 30 Tetrad period 7 5 / 1-5 / 8 Heading and flowering period

[0068] In the present invention, in step 203, the yield of crops is formed under the combined influence of various natural and non-natural factors. Generally, the actual yield of crops is decomposed into trend yield, meteorological yield and random yield. Due to the difference in meteorological conditions between years, the fluctuation of crop yield is caused, and the corresponding yield component is called meteorological yield. Meteorological yield data of multiple years from 2004 to 2017 are selected to construct sample meteorological yield characteristics.

[0069] In the present invention, in step 204, the frequency values ​​of the image histogram obtained in step 201 are respectively used with the crop growth process, leaf area index, dry matter weight, and physiological effect-related variables of the seven ear differentiation stages obtained in step 202 as characteristic variables, and the meteorological yield obtained in step 203 is used as the target variable to construct an LSTM meteorological yield prediction model, and compared with the results using only remote sensing information.

[0070] The model uses 2004-2014 as training data and 2015-2017 as validation data. 2 , RMSE and MAE are used as the evaluation indicators of the prediction accuracy of winter wheat meteorological yield in Henan Province: 1) R2 is used as the standard for evaluating the correlation between the model prediction value and the measured value, R 2 The closer it is to 1, the better the interpretation of the independent variable in the model is for the dependent variable; 2) RMSE is used to evaluate the clustering of samples. The closer RMSE is to 0, the higher the clustering of samples; 3) MAE is used as a standard for evaluating the error between the predicted value and the measured value. The closer MAE is to 0, the higher the accuracy of the estimation result. The specific formula is as follows:

[0071]

[0072]

[0073]

[0074] Among them, X i is the measured value, Y i is the predicted value, is the measured average value, is the predicted mean, and N is the number of observed samples.

[0075] The WOFOST analog quantity division and accuracy comparison are shown in Appendix 2:

[0076] Table 2

[0077]

[0078] According to Table 2, the model constructed by combining DVS, total dry matter weight, and physiological action simulation has the highest accuracy. 2 The R of the model constructed using DVS was 0.707, RMSE was 765.452 kg / ha, and MAE was 588.977 kg / ha. 2 The R value of the model reconstructed using total dry matter was 0.758, the RMSE was 702.565 kg / ha, and the MAE was 537.788 kg / ha. 2 The R value of the model constructed by combining the physiological action simulations was 0.754, the RMSE was 743.953 kg / ha, and the MAE was 553.239 kg / ha. 2 It reached 0.762, RMSE reached 706.099 kg / ha, and MAE reached 529.092 kg / ha.

[0079] Furthermore, a combination of multiple types of simulated quantities consisting of crop growth process, total dry matter weight, and physiological action simulated quantities with high accuracy was selected as feature variables, combined with the frequency value of the image histogram, and meteorological yield as the target variable to construct an LSTM meteorological yield prediction model, and compared it with the model using only remote sensing information. The model used 2004-2014 as training data and 2015-2017 as validation data. R 2 , RMSE and MAE were used to evaluate the accuracy.

[0080] Figure 2 This is a schematic diagram of the prediction results of the winter wheat meteorological yield prediction model provided by the present invention using a combination of multiple types of simulation quantities consisting of crop growth process, total dry matter weight, and physiological action simulation quantities as characteristic variables. Figure 2 As shown in (a), the model R of multiple types of analog quantity combination 2reached 0.773, RMSE reached 709.278 kg / ha, and MAE reached 531.324 kg / ha, compared with Figure 2 (b) Only RS is used as the characteristic variable, and a combination of multiple types of simulations consisting of crop growth process, total dry matter weight, and physiological action simulations is used as the characteristic variable R 2 Both improved, RMSE and MAE decreased, and thus the prediction accuracy of the winter wheat meteorological yield prediction model was also improved.

[0081] Furthermore, the model constructed based on the combination of the above-mentioned high-precision multi-type simulation quantities was used to analyze the ability of winter wheat yield prediction under different minimum temperature ranges and compared with the results using only remote sensing data. 2 , RMSE and MAE for accuracy assessment. The specific methods are as follows:

[0082] The LSTM yield model constructed by extracting the frequency value of the fusion of multi-type simulation combination and image histogram has a total of 300 validation samples from 2015 to 2017. Taking the average minimum temperature from jointing to heading in Henan Province as an indicator, according to the winter wheat frost damage grade index in the "Meteorological Industry Standard of the People's Republic of China: Crop Frost Damage Grade (QX / T 88-2008)" and considering the balance of the distribution of validation samples, the minimum temperature is divided into three intervals: less than -1℃, (-1℃, 0℃] and (0℃, 2℃]. Among them, there are 120 validation samples in the interval of less than -1℃, 94 validation samples in the interval of (-1℃, 0℃], and 86 validation samples in the interval of (0℃, 2℃]. The accuracy of yield prediction of winter wheat meteorological yield prediction model under the influence of late frost is evaluated based on the validation samples in different intervals, and compared with the results without adding WOFOST simulation.

[0083] Figure 3 This is a schematic diagram comparing the results of the winter wheat meteorological yield prediction model provided by the present invention in different temperature ranges with those without adding WOFOST simulation. Figure 3 As shown in the figure, the model with WOFOST simulation quantity has an R 2 The model R at (-1℃, 0℃] 2 The model R at (0℃, 2℃] reached 0.717, RMSE reached 767.957kg / ha, and MAE reached 554.500kg / ha; 2 The model reached 0.722, RMSE reached 750.645 kg / ha, and MAE reached 619.295 kg / ha. The R 2The results show that the yield prediction accuracy of the model is higher and the sensitivity is stronger in the range of -1°C. The meteorological yield prediction model for winter wheat of the present invention can effectively monitor the yield of winter wheat under the influence of late frost.

[0084] On the basis of the above embodiment, the sample meteorological yield characteristics are constructed according to the meteorological yield data, including:

[0085] According to the 5-year moving average method, the actual yield sample data of winter wheat was detrended to obtain the trend yield;

[0086] According to the difference between the actual yield sample data and the trend yield, meteorological yield data is obtained to construct sample meteorological yield characteristics.

[0087] Specifically, the actual yield of crops includes meteorological yield, trend yield and random yield. Meteorological yield refers to the yield affected by changes in weather and climate, which has the characteristics of pulsation; trend yield refers to the yield caused by social development and scientific and technological innovation; random yield refers to the yield change affected by random factors such as earthquakes and social changes. The present invention uses the 5a sliding average method to detrend the actual yield sample data to obtain the trend yield, and obtains the meteorological yield according to the yield calculation formula. The specific formula is as follows:

[0088] y=y t +y c +e;

[0089] Among them, y is the actual yield, y t is the trend output, y c is the meteorological output, and e is the random output.

[0090] Since random yield has no regular pattern, it is ignored in the present invention and is not included in the equation. Thus, the meteorological yield data can be obtained by calculating the difference between the actual yield sample data and the trend yield.

[0091] On the basis of the above embodiment, before dividing the time periods according to the preset time series, collecting the environmental data and management data of the winter wheat target research area, and inputting them into the WOFOST model, and obtaining the WOFOST simulation quantity characteristics of the winter wheat growth period in different time periods, the method also includes: localizing and calibrating the WOFOST model.

[0092] In the present invention, the crop, soil, meteorological and management data from 2004 to 2017 in the winter wheat study area were collected as the input data of the WOFOST model, and the meteorological data from 2004 to 2017 were obtained from the NASA website, including the minimum temperature, total radiation, maximum temperature, wind speed, vapor pressure and precipitation. In order to achieve a better simulation effect, the model was localized and calibrated to complete the simulation of the WOFOST model.

[0093] Among them, the parameters of TSUMEM, TSUM1 and TSUM2 in the WOFOST model were calculated based on the multi-year development period data and temperature data of each station and adjusted after model calibration; LAIEM and SPAN affect the initial growth and growth process of LAI, and then affect the light interception rate. The final localized parameter values ​​of TSUMEM, TSUM1, TSUM2, LAIEM and SPAN were determined by consulting the literature and calibrating the model, and were 120.5 (℃﹒d-1), 1480.3 (℃﹒d-1), 704.2 (℃﹒d-1), 0.131 and 35.0 (d), respectively.

[0094] By localizing and calibrating the WOFOST model, the model can be more adapted to the local climate and environment, thereby better achieving the effect of dynamically simulating the crop growth and development process.

[0095] Based on the above embodiment, the long short-term memory neural network includes 1 input layer, 5 hidden layers, 1 fully connected layer and 1 output layer, wherein the input layer standardizes the data through centralization and converts the data into a standard normal distribution.

[0096] Figure 4 This is a schematic diagram of the LSTM structure in the winter wheat meteorological yield prediction model provided by the present invention. Figure 4 As shown, in the present invention, LSTM is selected to establish a winter wheat meteorological yield prediction model. Its structure includes: 1 input layer, 5 hidden layers, 1 fully connected layer and 1 output layer. The input layer is mainly composed of remote sensing information and WOFOST simulation information. Since the input data is large and complex, it needs to be standardized. The data is converted into a standard normal distribution in a centralized way to avoid the influence of outliers on the model. The specific formula is as follows:

[0097]

[0098] Among them, μ represents the mean of the training samples and s represents the standard deviation.

[0099] The input layer is followed by a hidden layer, which can learn the time series information from the jointing to the heading period of winter wheat. Dropout is added to the hidden layer to remove the network units according to the ratio to avoid overfitting; finally, the learned information features are combined and output to the 1-dimensional space through the fully connected layer to obtain the prediction results of the output layer; in order to improve the convergence speed of the model, the stochastic gradient descent (SGD) method is used to optimize the model. After multiple experiments and tuning, the final hyperparameters of the model are set as follows: the number of samples selected for one training is 50, the ratio of removing neural network units is 0.2, the learning rate is 0.01, and the number of single training iterations for all batches in forward and backward propagation is 700.

[0100] Through the LSTM model, the input remote sensing information features and WOFOST simulation quantity features are processed and analyzed, which speeds up the processing of multidimensional features. The model is optimized according to the stochastic gradient descent method, which can improve the convergence speed of the winter wheat meteorological yield prediction model, thereby effectively improving the prediction accuracy of the winter wheat meteorological yield prediction model.

[0101] On the basis of the above embodiment, before performing band calculation on the winter wheat MODIS remote sensing data based on the preset time series, the method further includes:

[0102] The winter wheat MODIS remote sensing data based on the preset time series were mosaicked, projected and clipped.

[0103] In the present invention, the winter wheat MODIS remote sensing data is mosaicked, more than two scenes of images are seamlessly spliced ​​to form a complete target area, and the transparency and feathering radius of the remote sensing data image are changed to make the mosaicked area of ​​the image better integrated;

[0104] Project winter wheat MODIS remote sensing data, customize projection settings according to remote sensing data requirements, or convert it into required projection information according to application requirements;

[0105] The winter wheat MODIS remote sensing data were cropped to retain the images of the winter wheat target study area and ensure that the cropped part was rich in information.

[0106] By performing the above preprocessing on the winter wheat MODIS remote sensing data, the geometric distortion and information error of the remote sensing image caused by factors such as attitude changes, altitude, speed, and atmospheric interference during the remote sensing imaging process can be corrected.

[0107] Figure 5 The schematic diagram of the structure of the winter wheat meteorological yield prediction system provided by the present invention is as follows: Figure 5As shown, the present invention provides a winter wheat meteorological yield prediction system, including a remote sensing information feature acquisition module 501, an analog quantity feature acquisition module 502 and a meteorological yield prediction module 503, wherein the remote sensing information feature acquisition module 501 is used to perform band calculation on the winter wheat MODIS remote sensing data based on a preset time series, and obtain the MODIS remote sensing information features of the winter wheat target study area; the analog quantity feature acquisition module 502 is used to divide the time period according to the preset time series, collect the environmental data and management data of the winter wheat target study area, and input them into the WOFOST model to obtain the WOFOST analog quantity features of the winter wheat growth period in different time periods; the meteorological yield prediction module 503 is used to input the MODIS remote sensing information features and the WOFOST analog quantity features into a trained winter wheat meteorological yield prediction model, and obtain the meteorological yield prediction result of the winter wheat target study area, wherein the trained winter wheat meteorological yield prediction model is obtained by training a long short-term memory neural network with sample MODIS remote sensing information features, sample WOFOST analog quantity features and sample meteorological yield features.

[0108] The winter wheat meteorological yield prediction system provided by the present invention performs band calculation on winter wheat MODIS remote sensing data to obtain MODIS remote sensing information characteristics of a winter wheat target study area, uses a WOFOST model to obtain WOFOST simulation quantity characteristics of the winter wheat growth period in different time periods, and then inputs the MODIS remote sensing data and the WOFOST simulation quantity characteristics into a trained winter wheat meteorological yield prediction model to obtain a meteorological yield prediction result of the winter wheat target study area. Compared with a traditional winter wheat yield prediction method, by drawing on the idea of ​​data assimilation, on the basis of LSTM deep learning based on MODIS remote sensing information, a simulation quantity from jointing to heading period of a WOFOST crop model is added, a yield prediction model integrating WOFOST and LSTM is constructed, and a yield prediction effect is improved.

[0109] In the present invention, the remote sensing information feature acquisition module 501 can process the historical reserved data to obtain the remote sensing information feature variables, and can also obtain the remote sensing information feature variables of the study area in real time online.

[0110] Furthermore, the analog quantity feature acquisition module 502 can output a single analog quantity to the meteorological yield prediction module 503 , or a combination of multiple types of analog quantities with higher precision to the meteorological yield prediction module 503 .

[0111] Furthermore, the meteorological yield prediction system of the present invention can also be applied to crops such as rice, corn and soybean.

[0112] The system provided by the present invention is used to execute the above-mentioned method embodiments. Please refer to the above-mentioned embodiments for the specific processes and detailed contents, which will not be repeated here.

[0113] Figure 6 A schematic diagram of the structure of an electronic device provided by the present invention, such as Figure 6 As shown, the electronic device may include: a processor (processor) 610, a communication interface (Communications Interface) 620, a memory (memory) 630 and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 can call the logic instructions in the memory 630 to execute the winter wheat meteorological yield prediction method, which includes: performing band operations on the winter wheat MODIS remote sensing data based on a preset time series to obtain the MODIS remote sensing information characteristics of the winter wheat target study area; dividing the time period according to the preset time series, collecting the environmental data and management data of the winter wheat target study area, and inputting them into the WOFOST model to obtain the WOFOST simulation quantity characteristics of the winter wheat growth period in different time periods; inputting the MODIS remote sensing information characteristics and the WOFOST simulation quantity characteristics into the trained winter wheat meteorological yield prediction model to obtain the yield prediction result of the winter wheat target study area, wherein the trained winter wheat meteorological yield prediction model is obtained by training a long short-term memory neural network based on sample MODIS remote sensing information characteristics, sample WOFOST simulation quantity characteristics and sample meteorological yield characteristics.

[0114] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0115] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the winter wheat meteorological yield prediction method provided by the above methods, and the method includes: performing band calculations on winter wheat MODIS remote sensing data based on a preset time series to obtain MODIS remote sensing information characteristics of a target study area of ​​winter wheat; dividing time periods according to the preset time series, collecting environmental data and management data of the target study area of ​​winter wheat, and inputting them into a WOFOST model to obtain WOFOST simulation quantity characteristics of the winter wheat growth period in different time periods; inputting the MODIS remote sensing information characteristics and the WOFOST simulation quantity characteristics into a trained winter wheat meteorological yield prediction model to obtain a yield prediction result of the target study area of ​​winter wheat, wherein the trained winter wheat meteorological yield prediction model is obtained by training a long short-term memory neural network based on sample MODIS remote sensing information characteristics, sample WOFOST simulation quantity characteristics and sample meteorological yield characteristics.

[0116] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the processor executes the above-mentioned winter wheat meteorological yield prediction method, the method comprising: performing band operations on winter wheat MODIS remote sensing data based on a preset time series to obtain MODIS remote sensing information characteristics of a target study area of ​​winter wheat; dividing time periods according to the preset time series, collecting environmental data and management data of the target study area of ​​winter wheat, and inputting them into a WOFOST model to obtain WOFOST simulation quantity characteristics of the winter wheat growth period in different time periods; inputting the MODIS remote sensing information characteristics and the WOFOST simulation quantity characteristics into a trained winter wheat meteorological yield prediction model to obtain a yield prediction result of the target study area of ​​winter wheat, wherein the trained winter wheat meteorological yield prediction model is obtained by training a long short-term memory neural network based on sample MODIS remote sensing information characteristics, sample WOFOST simulation quantity characteristics and sample meteorological yield characteristics.

[0117] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0118] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for forecasting winter wheat meteorological yield, characterized in that: include: Perform band calculation on the winter wheat MODIS remote sensing data based on the preset time series to obtain the MODIS remote sensing information characteristics of the winter wheat target study area; Dividing time periods according to the preset time series, collecting environmental data and management data of the winter wheat target research area, and inputting them into the WOFOST model to obtain WOFOST simulation quantity characteristics of the winter wheat growth period in different time periods; Inputting the MODIS remote sensing information features and the WOFOST simulation features into a trained winter wheat meteorological yield prediction model to obtain a meteorological yield prediction result for the winter wheat target study area, wherein the trained winter wheat meteorological yield prediction model is obtained by training a long short-term memory neural network based on sample MODIS remote sensing information features, sample WOFOST simulation features and sample meteorological yield features; The trained winter wheat meteorological yield prediction model is obtained by training in the following steps: Perform band calculation on MODIS remote sensing data based on a preset time series to obtain a normalized vegetation index and a normalized vegetation moisture index, draw a frequency histogram, obtain the frequency value of the frequency histogram, and construct sample MODIS remote sensing information features; Divide the time periods according to the preset time series, extract the WOFOST simulation quantities of each ear differentiation period of winter wheat from jointing to heading period corresponding to each time period, and construct sample WOFOST simulation quantity characteristics, wherein the sample WOFOST simulation quantity characteristics include crop growth process, leaf area index, dry matter mass and physiological action characteristics; Based on the meteorological production data, the sample meteorological production characteristics are constructed; Inputting the sample MODIS remote sensing information features, the sample WOFOST simulation quantity features and the sample meteorological yield features into a long short-term memory neural network for training to obtain a trained winter wheat meteorological yield prediction model; The sample meteorological yield characteristics are constructed based on the meteorological yield data, including: According to the 5-year moving average method, the actual yield sample data of winter wheat was detrended to obtain the trend yield; According to the difference between the actual yield sample data and the trend yield, meteorological yield data is obtained to construct sample meteorological yield characteristics.

2. The winter wheat meteorological yield prediction method according to claim 1, characterized in that: Before dividing the time periods according to the preset time series, collecting the environmental data and management data of the winter wheat target research area, and inputting them into the WOFOST model, and obtaining the WOFOST simulation quantity characteristics of the winter wheat growth period in different time periods, the method also includes: localizing and calibrating the WOFOST model.

3. The winter wheat meteorological yield prediction method according to claim 1, characterized in that: The long short-term memory neural network includes 1 input layer, 5 hidden layers, 1 fully connected layer and 1 output layer, wherein the input layer standardizes the data through centralization and converts the data into a standard normal distribution.

4. The winter wheat meteorological yield prediction method according to claim 1, characterized in that: Before performing band calculation on the winter wheat MODIS remote sensing data based on the preset time series, the method further includes: The winter wheat MODIS remote sensing data based on the preset time series were mosaicked, projected and clipped.

5. The winter wheat meteorological yield prediction method according to claim 1, characterized in that: The band operation is performed on the MODIS remote sensing data based on the preset time series to obtain the normalized vegetation index and the normalized vegetation moisture index, including: According to the pixel values ​​of near infrared band 1 and red light band in MOD09A1, the normalized vegetation index is obtained, and the formula is: NDVI = (NIR1-R) / (NIR1+R); According to the pixel values ​​of near infrared band 1 and near red band 2 in MOD09A1, the normalized vegetation moisture index is obtained, and the formula is: NDWI=(NIR1-NIR2) / (NIR1+NIR2); Among them, NDVI represents the normalized difference vegetation index, NDWI represents the normalized difference vegetation water index; R represents the pixel value of the red light band, corresponding to the original effective value of the first band of MOD09A1; NIR1 represents the pixel value of the near infrared band 1, corresponding to the original effective value of the second band of MOD09A1; NIR2 represents the pixel value of the near infrared band 2, corresponding to the original effective value of the fifth band of MOD09A1.

6. A winter wheat meteorological yield prediction system, characterized in that: include: The remote sensing information feature acquisition module is used to perform band calculation on the winter wheat MODIS remote sensing data based on the preset time series to obtain the MODIS remote sensing information features of the winter wheat target research area; A simulation quantity characteristic acquisition module is used to divide the time period according to the preset time series, collect the environmental data and management data of the winter wheat target research area, and input them into the WOFOST model to obtain the WOFOST simulation quantity characteristics of the winter wheat growth period in different time periods; A meteorological yield prediction module is used to input the MODIS remote sensing information features and the WOFOST simulation features into a trained winter wheat meteorological yield prediction model to obtain a meteorological yield prediction result for the winter wheat target study area, wherein the trained winter wheat meteorological yield prediction model is obtained by training a long short-term memory neural network based on sample MODIS remote sensing information features, sample WOFOST simulation features and sample meteorological yield features; The trained winter wheat meteorological yield prediction model is obtained by training in the following steps: Perform band calculation on MODIS remote sensing data based on a preset time series to obtain a normalized vegetation index and a normalized vegetation moisture index, draw a frequency histogram, obtain the frequency value of the frequency histogram, and construct sample MODIS remote sensing information features; Divide the time periods according to the preset time series, extract the WOFOST simulation quantities of each ear differentiation period of winter wheat from jointing to heading period corresponding to each time period, and construct sample WOFOST simulation quantity characteristics, wherein the sample WOFOST simulation quantity characteristics include crop growth process, leaf area index, dry matter mass and physiological action characteristics; Based on the meteorological production data, the sample meteorological production characteristics are constructed; Inputting the sample MODIS remote sensing information features, the sample WOFOST simulation quantity features and the sample meteorological yield features into a long short-term memory neural network for training to obtain a trained winter wheat meteorological yield prediction model; The sample meteorological yield characteristics are constructed based on the meteorological yield data, including: According to the 5-year moving average method, the actual yield sample data of winter wheat was detrended to obtain the trend yield; According to the difference between the actual yield sample data and the trend yield, meteorological yield data is obtained to construct sample meteorological yield characteristics.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the winter wheat meteorological yield prediction method according to any one of claims 1 to 5 are implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the winter wheat meteorological yield prediction method according to any one of claims 1 to 5 are implemented.

Images