Wheat yield remote sensing prediction method combining phenological parameters

Abstract

The invention discloses a wheat yield remote sensing prediction method combining phenological parameters, which comprises the following steps: S1, performing field observation in a key growth period of winter wheat, synchronously collecting canopy hyperspectral reflectivity data, leaf area index (LAI) and SPAD value of each observation sample point, and recording wheat grain yield of the corresponding sample point; s2, performing noise reduction preprocessing on the acquired canopy hyperspectral data, and extracting sensitive spectral parameters by combining principal component analysis (PCA) and correlation analysis methods; s3, taking the sensitive spectrum parameters, LAI and SPAD values as independent variables, taking the wheat grain yield as a dependent variable, and adopting partial least squares regression (PLSR) to construct a multivariable yield prediction model; and S4, performing wheat yield prediction on an independent test sample or regional scale remote sensing data by using the trained model. The method overcomes the defect that only yield sensitive spectrum parameters are used for predicting the effect, and accurate estimation of the model is achieved.

Description

Technical Field

[0001] This invention relates to the field of agricultural remote sensing and crop yield prediction technology, and in particular to a method for remote sensing prediction of wheat yield based on a combination of climatic parameters. Background Technology

[0002] Crop yield is a core indicator for measuring agricultural production levels. Accurate and timely crop yield forecasting is crucial for food security, agricultural management decisions, and optimal resource allocation. Wheat, as one of my country's major food crops, has long relied on manual field sampling and laboratory analysis for yield forecasting, such as calculating theoretical yield based on the number of effective spikes per unit area, the number of grains per spike, and the thousand-grain weight. While these methods offer high accuracy, they suffer from low efficiency, high cost, and significant disruption, making them unsuitable for the large-scale, dynamic, and real-time monitoring demands of modern agriculture.

[0003] In recent years, with the rapid development of remote sensing technology, retrieving crop growth status using ground object spectrometers or satellite remote sensing data has become a research hotspot. Numerous studies have shown a strong correlation between vegetation indices (such as NDVI, EVI, and SR) and crop biomass and final yield, and constructing empirical or semi-empirical regression models based on remote sensing data from key growth stages has become the mainstream approach. However, these models generally face the following problems:

[0004] The indirect nature of the spectrum-yield relationship: Wheat yield is not a direct response variable of canopy spectrum, but rather the result of long-term accumulation of processes such as photosynthesis, dry matter accumulation, and nutrient translocation throughout the entire growth period. Relying solely on spectral information from one or more growth stages is insufficient to fully reflect the dynamic process of yield formation, leading to poor extrapolation ability and inadequate stability of the model.

[0005] Band redundancy and overfitting risks: Hyperspectral data has extremely high band resolution and high correlation between adjacent bands. Directly using multi-band modeling can easily lead to model complexity, poor interpretability, and no significant improvement in prediction accuracy.

[0006] Lack of physiological mechanism support: Traditional remote sensing models are mostly statistical empirical models, which ignore the influence of the internal physiological processes of crop growth, making it difficult to reveal the biological basis of yield formation and limiting the universality and transferability of the models.

[0007] Therefore, there is an urgent need for a comprehensive prediction method that integrates remote sensing spectral information with key crop physiological parameters to improve the accuracy and reliability of remote sensing estimation of wheat yield. Summary of the Invention

[0008] The present invention aims to solve the aforementioned problems by providing a remote sensing prediction method for wheat yield based on combined climatological parameters.

[0009] To solve the above-mentioned technical problems, the technical solution provided by the present invention is as follows: a remote sensing prediction method for wheat yield based on combined phenological parameters, comprising the following steps: S1, conducting field observations during the key growth period of winter wheat, simultaneously collecting canopy hyperspectral reflectance data, leaf area index (LAI), and SPAD values ​​at each observation point, and recording the wheat grain yield at the corresponding point; S2, performing noise reduction preprocessing on the collected canopy hyperspectral data, and extracting sensitive spectral parameters by combining principal component analysis (PCA) and correlation analysis methods; S3, using the aforementioned sensitive spectral parameters, LAI, and SPAD values ​​as independent variables, and wheat grain yield as the dependent variable, constructing a multivariate yield prediction model using partial least squares regression (PLSR); S4, using the trained model to predict wheat yield on independent test samples or regional-scale remote sensing data.

[0010] As an improvement, the key growth period includes the jointing stage, the flag-raising stage, the flowering stage, and the grain-filling stage.

[0011] As an improvement, the canopy hyperspectral reflectance data were collected by a portable ground object spectrometer under clear, cloudless, and windless weather conditions between 10:00 AM and 2:00 PM. The probe was positioned 1 meter vertically above the canopy with a field of view of 25°. Whiteboard calibration was performed every 20 minutes. At least 10 spectral curves were collected for each observation point, and the average value was taken as the final spectral value for that point.

[0012] As an improvement, the noise reduction preprocessing uses the Savitzky-Golay polynomial smoothing method to filter the original spectral curve and eliminate the influence of random noise.

[0013] As an improvement, the method for extracting the sensitive spectral parameters includes: (a) performing principal component analysis (PCA) on the full-band spectral data in the range of 350–2500 nm, and selecting principal components that explain a total variance of ≥85%; (b) identifying bands that significantly contribute to spectral variation based on the factor loadings of each principal component; (c) calculating the correlation coefficients between each vegetation index (including one or more of NDVI, RVI, SR, and MCARI) and measured yield, and screening the index with the highest correlation; (d) removing redundant bands or indices with absolute values ​​of correlation coefficients greater than 0.85 with other variables, and retaining spectral features with strong independence and high representativeness as sensitive spectral parameters.

[0014] As an improvement, 5–10 representative winter wheat plants were selected at each observation point, and the length and width of each leaf were measured. The leaf area index per unit surface area was calculated using the formula LAI = Σ(length × width × 0.83) / ground area.

[0015] As an improvement, the method for determining the SPAD value is as follows: using a portable chlorophyll meter (SPAD-502), six wheat plants are randomly selected within the observation point, and the SPAD reading at the middle position of their flag leaves is measured. The average of the six measurements is taken as the canopy SPAD value of the observation point.

[0016] As an improvement, before harvest, the number of effective ears was counted in a 0.5m×0.5m quadrat. The ears of wheat were cut off, brought back to the laboratory, dried and threshed, and the thousand-grain weight and average number of grains per ear were measured. The yield per unit area was calculated according to the formula "theoretical yield = number of ears per mu × number of grains per ear × thousand-grain weight". This yield data was used as the measured yield data for model building and verification.

[0017] The advantages of this invention compared to the prior art are:

[0018] 1. Significantly Improved Prediction Accuracy: Compared to traditional models that only use spectral data, this invention enhances the model's ability to biologically represent crop growth processes by introducing two key physiological parameters, LAI and SPAD, significantly improving the R-value of yield prediction across multiple growth stages (especially the grain-filling stage). 2 (Increases by 0.15–0.30), reducing RMSE by more than 20%.

[0019] 2. Enhance the model's mechanistic rationality and stability: LAI reflects the population's light energy interception capacity, and SPAD indicates nitrogen nutrition and photosynthetic potential. Both are closely related to dry matter accumulation and grain yield, making the model more physiologically grounded and reducing the uncertainty of empirical modeling.

[0020] 3. Reduce band redundancy and overfitting risks: Input variables are optimized through PCA and correlation screening, and PLSR is used to handle multivariate collinearity problems, achieving a simple and efficient model.

[0021] 4. Non-destructive and repeatable monitoring: All parameters can be quickly acquired on-site using portable devices without destructive sampling, supporting dynamic tracking and real-time yield prediction throughout the entire growth period. Attached Figure Description

[0022] Figure 1 This is a flowchart of the steps of a remote sensing prediction method for wheat yield based on combined climatic parameters according to the present invention.

[0023] Figure 2 This is a map showing the distribution of experimental plots for a remote sensing prediction method for wheat yield based on composite climatic parameters according to the present invention.

[0024] Figure 3 This is an average spectral curve of wheat canopy at each growth stage of a remote sensing prediction method for wheat yield based on composite climatic parameters according to the present invention.

[0025] Figure 4This is a reference spectral index and its calculation formula used in the remote sensing prediction method for wheat yield based on composite climatic parameters of the present invention.

[0026] Figure 5 This is a principal component analysis result diagram of a remote sensing prediction method for wheat yield based on composite climatic parameters according to the present invention.

[0027] Figure 6 This refers to the correlation coefficient between the spectral index and wheat yield of a remote sensing prediction method for wheat yield based on composite climatic parameters according to the present invention.

[0028] Figure 7 This invention relates to a wheat yield remote sensing prediction method based on combined climate parameters, specifically a wheat yield prediction model constructed using only spectral parameters.

[0029] Figure 8 This invention relates to a wheat yield remote sensing prediction method based on combined climate parameters, specifically a wheat yield prediction model constructed using only spectral parameters.

[0030] Figure 9 This invention relates to the accuracy of a wheat yield prediction model based on the combined phenological parameters and spectral parameters in a remote sensing prediction method for wheat yield using combined phenological parameters. Detailed Implementation

[0031] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "inner", "outer", "center", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation structure and operation. Therefore, they should not be construed as limiting this invention.

[0032] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "provided with," "installed," "connected," "linked," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0033] The present invention will now be described in further detail with reference to the accompanying drawings.

[0034] Combined with appendix Figure 1 A remote sensing prediction method for wheat yield based on combined climatological parameters includes the following steps:

[0035] Step 1: Field Experiment Design and Phenological Parameter Collection

[0036] Field observation plots were established in typical winter wheat growing areas, preferably located in major producing areas such as the Huang-Huai-Hai Plain, to record geographical location, climate conditions and soil background information;

[0037] The key growth stages of wheat were selected for observation, including the jointing stage, the flag-raising stage, the flowering stage, and the grain-filling stage.

[0038] Multiple representative observation points were set up within the sample plots during each growth period to simultaneously collect the following phenological parameters:

[0039] Leaf Area Index (LAI): Using the leaf area coefficient method, 5–10 representative plants were selected at each sampling point, and the length and width of each leaf were measured according to the formula.

[0040] LAI = ∑(length × width × 0.83) / ground area

[0041] Calculate the leaf area per unit surface area;

[0042] Canopy chlorophyll content (SPAD value): Using a portable chlorophyll meter (such as SPAD-502), the SPAD readings at the middle position of the flag leaf of 6 wheat plants were randomly measured at the observation point, and the average value was taken as the canopy chlorophyll index at that point.

[0043] Actual yield data (GY): Before harvest, the number of effective ears was counted in a 0.5m × 0.5m quadrat. Ears were cut, brought back to the laboratory, dried, threshed, and the thousand-grain weight and number of grains per ear were measured according to the formula.

[0044] Theoretical yield = number of ears per mu × number of grains per ear × weight of 1000 grains

[0045] Calculate the yield for each sample point for model training and validation.

[0046] Step 2: Acquisition and Preprocessing of Canopy Hyperspectral Data

[0047] Wheat canopy reflectance was measured using a portable ground object spectrometer (such as ASD FieldSpec3) under clear, cloudless, and windless weather conditions (10:00 AM–2:00 PM).

[0048] The probe is positioned 1 meter vertically above the canopy, with a field of view of 25°. Whiteboard calibration is performed every 20 minutes.

[0049] At least 10 spectral curves were collected for each sampling point, and the average value was taken as the canopy spectral reflectance of that point.

[0050] The original spectral data was denoised by using the Savitzky-Golay smoothing method to eliminate random noise and obtain smoothed spectral curves.

[0051] Commonly used vegetation indices, such as NDVI, RVI, SR, and MCARI, are extracted, covering sensitive band regions such as green peak, red edge, and near-infrared.

[0052] Step 3: Extraction of sensitive spectral parameters

[0053] Principal component analysis (PCA) was performed on the full-band (350–2500 nm) spectral data to extract principal components (PCs) that could explain ≥85% of the total variance;

[0054] Analyze the factor loadings of each principal component to identify the bands that contribute the most to spectral variation;

[0055] By combining the correlation analysis between spectral indices and yield, the characteristic bands and vegetation indices with the highest correlation with yield were selected as sensitive spectral parameters.

[0056] Redundant variables with high correlation (|r|>0.85) were removed, while spectral features with strong representativeness and high independence were retained.

[0057] Step 4: Building a Multivariate Yield Prediction Model

[0058] Two comparative models were constructed to verify the superiority of the present invention:

[0059] Model A (Control Model): A yield prediction model is established using multiple linear regression (MLR) with only sensitive spectral parameters (single band or vegetation index) as independent variables.

[0060] Model B (the model of this invention): Using sensitive spectral parameters, LAI and SPAD values ​​as independent variables, a multivariate prediction model is constructed using partial least squares regression (PLSR), making full use of the covariance between spectral and physiological parameters to extract latent variables for modeling;

[0061] The dataset was divided into a training set (70%) and a test set (30%), which were used for model training and independent validation, respectively.

[0062] Through the coefficient of determination (R) 2 The performance of a model is evaluated using metrics such as root mean square error (RMSE) and relative analysis error (RPD).

[0063] Step 5: Model Application and Spatial Prediction of Yield

[0064] The trained PLSR model is applied to regional-scale remote sensing images (such as Sentinel-2, Landsat, or UAV hyperspectral data), and combined with the inverted LAI and SPAD spatial distribution maps to generate a spatial prediction map of wheat yield.

[0065] This enables remote sensing estimation of wheat yield from small plots to regional scales, supporting precision agricultural management and policy formulation.

[0066] Specific embodiments of the present invention are as follows:

[0067] Example: Winter wheat yield forecast in Qihe County, Shandong Province

[0068] Test location: Qihe County, Dezhou City, Shandong Province (36.48°N, 116.47°E), a warm temperate semi-humid monsoon climate with an average annual temperature of 14.3℃ and an average annual precipitation of 645.9mm;

[0069] Trial period: October 2023 – June 2024;

[0070] Variety: Jimai 22;

[0071] Observation frequency: once each during the jointing stage, flag-raising stage, flowering stage, and grain-filling stage;

[0072] Instruments: ASD FieldSpec3 spectrometer, SPAD-502 chlorophyll meter;

[0073] (1) Collection of phenological parameters and yield data during the wheat growth period (jointing stage, flag-raising stage, flowering stage, grain-filling stage).

[0074] Combined with appendix Figure 2 The field experimental plot is located in Qihe County, Shandong Province, China (36.48°, 116.47°), north of the Yellow River. This region has a semi-humid continental climate and is situated on the alluvial plain of the lower Yellow River. The parent material of the soil is loess. The terrain is flat, with a ground elevation of approximately +27 to +30 meters and a slope of approximately 0.05%. It has a warm temperate semi-humid monsoon climate, with an average annual temperature of 14.3℃ and an average annual precipitation of 645.9 mm. Rainfall is concentrated from late June to late September, and the average frost-free period is approximately 210 days.

[0075] Leaf area index (LAI): After collecting canopy spectra at different growth stages, 5-10 representative wheat plants were selected and converted into LAI per unit ground area using the leaf area coefficient method (length * width * 0.83).

[0076] Canopy chlorophyll content: In this study, SPAD value was used as the actual measurement index of chlorophyll content. SPAD was measured using a portable chlorophyll meter. SPAD was measured in the middle of the flag leaf of 6 wheat plants randomly in the canopy spectral measurement area, and the average value was taken as the canopy SPAD value of the sample.

[0077] Wheat grain yield determination (GY): Before winter wheat harvest, a 0.5m × 0.5m area is selected at the sampling point to determine the actual number of plants and the total number of effective ears within the sampling area; wheat ears are cut off and brought back to the laboratory for drying and threshing to obtain the average number of grains per ear; 1000 dry wheat grains are randomly selected, weighed 3 times, and the average value is taken; theoretical yield = number of ears per mu × number of grains per ear × weight of 1000 grains to obtain the final yield of winter wheat.

[0078] (2) Collection of canopy spectral data during wheat growth period

[0079] Wheat canopy spectra at various growth stages were measured using an ASD FieldSpec3 portable ground spectrometer. All canopy spectral measurements were recorded on clear, windless days between 10:00 AM and 2:00 PM, with whiteboard calibration performed every 20 minutes during data collection. The sensor probe was positioned 1 m perpendicular to the top of the canopy, with a field of view of 25°. Representative plants were selected, and the average of 10 measured spectral curves from each plot was taken as the canopy spectral reflectance value. After acquiring the wheat canopy spectral curves, noise reduction was performed using a moving window fitting polynomial smoothing method (Savitzky-Golay smoothing) to obtain the original spectral reflectance. See Appendix for details. Figure 3 .

[0080] Spectral indices, based on combinations of two or more spectral bands, carry more vegetation physiological information than single bands and exhibit better sensitivity than peaks, troughs, and single bands. Therefore, using spectral indices to build physiological information prediction models can, to some extent, eliminate the band overfitting problem caused by using too many bands, making statistical analysis results more accurate and convincing. In index calculations using band combinations, discrete spectra from specific regions are typically combined. Most indices utilize three regions describing typical reflectance patterns of green vegetation: green, red, and near-infrared bands.

[0081] (3) Extraction of sensitive spectral parameters

[0082] Principal component analysis (PCA) is used to recombine numerous correlated indicators in the spectrum into a new set of independent composite indicators to replace the original indicators, thereby extracting key bands. Components (PCs) with a high explanatory power of the total variance and bands with high factor loadings are considered the most representative indicators of the canopy spectrum. The retained principal components are selected based on their explanatory power of the total variance. When a single component contains multiple bands, the correlation coefficient is used to determine whether the variables are redundant; among highly correlated variables, only one is selected, and the rest are discarded. This is combined with other methods to further refine the analysis. Figure 5 .

[0083] The correlation coefficient method was used to calculate the correlation between spectral indices and wheat yield, and the correlation coefficients between wheat canopy spectral indices and wheat yield at each growth stage were obtained, as shown in the attached figure. Figure 6 .

[0084] (4) Model building

[0085] To compare the superiority of the method proposed in this patent, a wheat yield prediction model using only remote sensing data and a wheat yield prediction model using a combination of phenological parameters will be constructed. The former uses simple regression to construct the model, with the single feature band extracted by PCA and the highest correlation spectral index extracted by the correlation coefficient method as independent variables to construct the regression model for wheat yield. The latter is a multivariate model based on partial least squares regression, with phenological parameters added to the independent variables to construct the prediction model.

[0086] Comparison Appendix Figure 7 Appendix Figure 8 and attached Figure 9 It can be seen that combining spectral parameters with wheat phenological parameters at different growth stages can improve the prediction effect of wheat yield, especially compared with the wheat yield prediction model built using only spectral parameters, which shows a significant improvement in accuracy. Therefore, when constructing a remote sensing model for wheat yield, it is advisable to consider using phenological parameters as supplementary data to spectral parameters and input them into the model as independent variables.

[0087] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structural designs, lids, and embodiments without departing from the spirit of the present invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for remote sensing prediction of wheat yield based on composite climatological parameters, characterized in that, Includes the following steps: S1. Conduct field observations during the critical growth period of winter wheat, and simultaneously collect canopy hyperspectral reflectance data, leaf area index (LAI) and SPAD values ​​at each observation point, and record the wheat grain yield at the corresponding point. S2. The collected canopy hyperspectral data are preprocessed for noise reduction, and sensitive spectral parameters are extracted by combining principal component analysis (PCA) and correlation analysis. S3. Using the aforementioned sensitive spectral parameters, LAI and SPAD values ​​as independent variables and wheat grain yield as the dependent variable, a multivariate yield prediction model is constructed using partial least squares regression (PLSR). S4. Use the trained model to predict wheat yield from independent test samples or regional-scale remote sensing data.

2. The method for remote sensing prediction of wheat yield based on composite climatological parameters according to claim 1, characterized in that, The critical growth periods include the jointing stage, the flag-raising stage, the flowering stage, and the grain-filling stage.

3. The method for remote sensing prediction of wheat yield based on composite climatological parameters according to claim 1, characterized in that, The canopy hyperspectral reflectance data were collected by a portable ground object spectrometer under clear, cloudless, and windless weather conditions between 10:00 AM and 2:00 PM. The probe was positioned 1 meter vertically above the canopy with a field of view of 25°. Whiteboard calibration was performed every 20 minutes. At least 10 spectral curves were collected for each observation point, and the average value was taken as the final spectral value for that point.

4. The method for remote sensing prediction of wheat yield based on composite climatological parameters according to claim 1, characterized in that, The noise reduction preprocessing uses the Savitzky-Golay polynomial smoothing method to filter the original spectral curve and eliminate the influence of random noise.

5. The method for remote sensing prediction of wheat yield based on composite climatic parameters according to claim 1, characterized in that, The method for extracting the sensitive spectral parameters includes: (a) Principal component analysis (PCA) was performed on the full-band spectral data in the range of 350–2500 nm, and principal components that explained ≥85% of the total variance were selected; (b) Identify the bands that contribute significantly to spectral variation based on the factor loadings of each principal component; (c) Calculate the correlation coefficient between each vegetation index (including one or more of NDVI, RVI, SR, and MCARI) and the measured yield, and screen the index with the highest correlation. (d) Redundant bands or indices with an absolute value of correlation coefficient with other variables greater than 0.85 are removed, and spectral features with strong independence and high representativeness are retained as sensitive spectral parameters.

6. The method for remote sensing prediction of wheat yield based on composite climatological parameters according to claim 5, characterized in that, At each observation point, select 5–10 representative winter wheat plants, measure the length and width of each leaf, and calculate the leaf area index per unit surface area using the formula LAI = Σ(length × width × 0.83) / ground area.

7. The method for remote sensing prediction of wheat yield based on composite climatological parameters according to claim 1, characterized in that, The method for determining the SPAD value is as follows: using a portable chlorophyll meter (SPAD-502), six wheat plants are randomly selected within the observation point, and the SPAD reading at the middle position of their flag leaves is measured. The average of the six measurements is taken as the canopy SPAD value of the observation point.

8. The method for remote sensing prediction of wheat yield based on composite climatological parameters according to claim 7, characterized in that, Before harvest, the number of effective ears was counted in a 0.5m×0.5m quadrat. The ears of wheat were cut off, brought back to the laboratory, dried and threshed, and the thousand-grain weight and average number of grains per ear were measured. The yield per unit area was calculated according to the formula "theoretical yield = number of ears per mu × number of grains per ear × thousand-grain weight", and used as the measured yield data for model building and verification.

Images