A method for predicting rice yield using an integrated spectral index

Abstract

The present application relates to the technical field of rice yield prediction, and particularly relates to a method for predicting rice yield by integral spectral index. The specific method comprises the following steps: obtaining relative growth length days at a sampling time point to determine a phonological stage corresponding to the sampling time point; obtaining canopy reflectance spectral data at different sampling time points to determine spectral index values corresponding to the sampling time points; and according to the determined phonological stage and spectral index values corresponding to the sampling time points, obtaining comprehensive spectral index values corresponding to continuous phonological stages, and substituting the comprehensive spectral index values into a corresponding yield prediction model to obtain a rice predicted yield. The present application uses NDWI to establish a yield prediction model to predict rice yield by integral spectral index, and on the basis of ensuring the accuracy of the prediction result, dynamically predicts the grain yield at the growth and development stage of rice, and solves the problem of inaccurate predicted yield result of the yield prediction model at a specific growth stage.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of rice yield prediction, and particularly relates to a method for predicting rice yield by integral spectral index. BACKGROUND

[0002] With the continuous progress of agricultural science and technology, real-time monitoring of rice growth conditions and yield estimation have become an important part of modern agricultural management. Traditional rice yield estimation methods, such as manual field investigation and quadrat determination, can provide relatively accurate data, but the process is time-consuming and labor-intensive, and it is difficult to achieve large-area and real-time monitoring. Therefore, it is particularly important to develop a fast and quantitative method for characterizing rice growth conditions and yield estimation.

[0003] Traditional yield estimation mainly relies on the manual method of destructive sampling, which is extremely laborious and tedious and cannot be widely applied to farm practices. In contrast, remote sensing is a non-destructive method for quickly and accurately capturing crop growth characteristics in the field environment. Remote sensing technology has been successfully applied to evaluate phenology, biomass, photosynthetic traits, and yield. However, existing monitoring models of yield traits are basically empirical models that mainly establish a linear or nonlinear relationship between yield and spectral indices calculated by multiple spectral sensitive bands related to yield traits in the visible-near infrared region. However, although the monitoring model can be applied to each growth stage, the model usually cannot explain the impact of the yield formation process on yield. Yield formation is a cumulative process of crop growth and development, so the yield prediction model for a specific growth stage cannot well evaluate the final yield of crops.

[0004] Therefore, the present application provides a method for predicting rice yield by integral spectral index, which can accurately predict the grain yield of rice growth and development stages under different irrigation systems. SUMMARY

[0005] In order to solve the problem of inaccurate yield prediction results of the existing yield prediction model for a specific growth stage, the purpose of the present application is to provide a method for predicting rice yield by integral spectral index.

[0006] To achieve the above-mentioned purpose, the technical scheme of the present application is as follows.

[0007] The present application provides a method for predicting rice yield by integral spectral index, comprising the following steps:

[0008] Obtain the relative length of day at the sampling time point to determine the corresponding phenological stage of the sampling time point; the phenological stage includes the booting stage, the flowering stage, the grain filling stage and the mature stage;

[0009] Obtain the canopy reflectance spectral data at different sampling time points to determine the spectral index value of the corresponding sampling time point;

[0010] According to the determined phenological stage and spectral index value corresponding to the sampling time point, a comprehensive spectral index value corresponding to a continuous phenological stage is obtained, and is substituted into a corresponding yield prediction model to obtain a rice yield prediction.

[0011] The yield prediction model is a yield prediction model corresponding to a corresponding phenological stage with a spectral index or a comprehensive spectral index as an abscissa and a relative yield as an ordinate.

[0012] Preferably, the rice yield prediction is a rice yield prediction under different irrigation systems.

[0013] Preferably, the corresponding yield prediction model includes a yield prediction model corresponding to the booting stage, a yield prediction model corresponding to the booting stage-flowering stage, a yield prediction model corresponding to the booting stage-grain filling stage, and a yield prediction model corresponding to the booting stage-mature stage.

[0014] Further preferably, the yield prediction model corresponding to the booting stage is y=-0.68+40.10x-238.01x 2 , R 2 =0.57; the yield prediction model corresponding to the booting stage-flowering stage is y=-0.31+162.89x-5058.92x 2 , R 2 =0.41; the yield prediction model corresponding to the booting stage-grain filling stage is y=-1.44+196.25x-3878.75x 2 , R 2 =0.73; and the yield prediction model corresponding to the booting stage-mature stage is y=1.95+15.13x-239.50x 2 , R 2 =0.79; wherein x is a spectral index or a comprehensive spectral index; y is a relative yield; and R 2 is a determination coefficient.

[0015] Further preferably, the spectral index is a normalized difference water index.

[0016] The calculation formula of the spectral index is:

[0017] NDWI=(R 860 -R 1640 ) / (R 860 +R 1640 ); wherein NDWI is a normalized difference water index; and R is canopy reflectance spectrum data of a corresponding waveband. For example, R 860 is canopy reflectance spectrum data of a waveband of 860 nm; and R 1640 is canopy reflectance spectrum data of a waveband of 1640 nm.

[0018] Further preferably, the formula for calculating the comprehensive spectral index is:

[0019] SI inte =∑SI i ×t i ; wherein, SI inte is the comprehensive spectral index; SI i is the difference between the instantaneous spectral indices of two adjacent sampling time points; ∑SI i is the sum of SI i ; t i is the continuous phenological stage experienced by the two adjacent sampling time points. i represents the sampling time point; the two adjacent sampling time points are represented as i and i-1, respectively.

[0020] Preferably, the continuous phenological stage includes the period from the booting stage to the flowering stage, the period from the booting stage to the grain filling stage, and the period from the booting stage to the maturity stage.

[0021] Preferably, the specific method for obtaining the relative growth length day of the sampling time point is:

[0022] The accumulated growth length day of the sampling time point and the total growth length day of the entire growth period are obtained to calculate the relative growth length day of the sampling time point.

[0023] Further preferably, the formula for calculating the relative growth length day is:

[0024] RGDD = AGDD / TGDD; wherein, RGDD is the relative growth length day, d℃; AGDD is the accumulated growth length day, d℃; TGDD is the total growth length day of the entire growth period, d℃.

[0025] Further preferably, the formula for calculating the accumulated growth length day is:

[0026]

[0027] wherein, AGDD is the accumulated growth length day; N is the number of growth days from the beginning of the growth of the rice to the sampling time point; T max is the maximum temperature per day; T min is the minimum temperature per day; T base is the minimum temperature at which the rice begins to have physiological activities.

[0028] wherein, the accumulated growth length day is the sum of the accumulated average of the maximum temperature per day and the minimum temperature per day from the first day to the Nth day of the growth of the plant; at this time, N is the number of growth days from the beginning of the growth of the rice to the sampling time point. The basic growth temperature is T base .

[0029] The total length of the day of the entire growth period is the sum of the daily maximum temperature plus the average of the daily minimum temperature minus the cumulative sum of the base growth temperature between the entire growth period.

[0030] Advantages of the present application:

[0031] 1、The present application uses cumulative phenological stages to improve the monitoring ability of rice yield under different irrigation systems, and when multiple phenological stages are included in the spectral model, the NDWI parameter has significant advantages. Based on this, the present application proposes a yield prediction model using NDWI, and uses the yield prediction model to predict the rice yield by integral spectral index, which ensures the accuracy of the prediction result and dynamically predicts the grain yield of the rice growth and development stage.

[0032] 2、The method of the present application will help to quickly evaluate the yield potential of different water management modes in rice plants in a larger area, and also help to guide farmers to establish potential yield targets in water-sensitive stages in the future main phenological stages to achieve precise irrigation. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 A calculation diagram of the comprehensive spectral index taking the reflectance value of the spectrum at 1970 band as an example.

[0034] Figure 2 A result diagram of the Pearson correlation coefficient between the 14 spectral indices and the relative yield.

[0035] Figure 3 A result diagram of the correlation between the spectral index of different phenological stages and the relative yield. Among them, A represents the correlation between the spectral index of the booting stage and the relative yield; B represents the correlation between the spectral index of the flowering stage and the relative yield; C represents the correlation between the spectral index of the grain filling stage and the relative yield; D represents the correlation between the spectral index of the maturity stage and the relative yield.

[0036] Figure 4 A result diagram of the correlation between the comprehensive spectral index or the cumulative comprehensive spectral index of the continuous phenological stage and the relative yield. Among them, A' represents the correlation between the comprehensive spectral index of the booting stage to the flowering stage and the relative yield; B' represents the correlation between the comprehensive spectral index of the flowering stage to the grain filling stage and the relative yield; C' represents the correlation between the comprehensive spectral index of the grain filling stage to the maturity stage and the relative yield; D' represents the correlation between the cumulative comprehensive spectral index of the booting stage to the grain filling stage and the relative yield; E' represents the correlation between the cumulative comprehensive spectral index of the booting stage to the maturity stage and the relative yield; F' represents the correlation between the cumulative comprehensive spectral index of the flowering stage to the maturity stage and the relative yield.

[0037] Figure 5In the figure, (a) is the quantitative relationship diagram of NDII-based spectral index at the booting stage and relative yield; (b) is the quantitative relationship diagram of NDWI-based spectral index at the booting stage and relative yield; (c) is the quantitative relationship diagram of MSI-based spectral index at the booting stage and relative yield; (d) is the quantitative relationship diagram of SRWI-based spectral index at the booting stage and relative yield; (e) is the quantitative relationship diagram of NDII-based spectral index at the flowering stage and relative yield; (f) is the quantitative relationship diagram of NDWI-based spectral index at the flowering stage and relative yield; (g) is the quantitative relationship diagram of MSI-based spectral index at the flowering stage and relative yield; (h) is the quantitative relationship diagram of SRWI-based spectral index at the flowering stage and relative yield; (i) is the quantitative relationship diagram of NDII-based spectral index at the grain filling stage and relative yield; (j) is the quantitative relationship diagram of NDWI-based spectral index at the grain filling stage and relative yield; (k) is the quantitative relationship diagram of MSI-based spectral index at the grain filling stage and relative yield; (l) is the quantitative relationship diagram of SRWI-based spectral index at the grain filling stage and relative yield; (m) is the quantitative relationship diagram of NDII-based spectral index at the maturity stage and relative yield; (m) is the quantitative relationship diagram of NDWI-based spectral index at the maturity stage and relative yield; (o) is the quantitative relationship diagram of MSI-based spectral index at the maturity stage and relative yield; (p) is the quantitative relationship diagram of SRWI-based spectral index at the maturity stage and relative yield.

[0038] Figure 6 In the figure, (a) is the quantitative relationship diagram of NDII-based comprehensive spectral index at the booting stage-flowering stage and relative yield; (b) is the quantitative relationship diagram of NDWI-based comprehensive spectral index at the booting stage-flowering stage and relative yield; (c) is the quantitative relationship diagram of MSI-based comprehensive spectral index at the booting stage-flowering stage and relative yield; (d) is the quantitative relationship diagram of SRWI-based comprehensive spectral index at the booting stage-flowering stage and relative yield; (e) is the quantitative relationship diagram of NDII-based comprehensive spectral index at the flowering stage-grain filling stage and relative yield; (f) is the quantitative relationship diagram of NDWI-based comprehensive spectral index at the flowering stage-grain filling stage and relative yield; (g) is the quantitative relationship diagram of MSI-based comprehensive spectral index at the flowering stage-grain filling stage and relative yield; (h) is the quantitative relationship diagram of SRWI-based comprehensive spectral index at the flowering stage-grain filling stage and relative yield; (i) is the quantitative relationship diagram of NDII-based comprehensive spectral index at the grain filling stage-maturity stage and relative yield; (j) is the quantitative relationship diagram of NDWI-based comprehensive spectral index at the grain filling stage-maturity stage and relative yield; (k) is the quantitative relationship diagram of MSI-based comprehensive spectral index at the grain filling stage-maturity stage and relative yield; (l) is the quantitative relationship diagram of SRWI-based comprehensive spectral index at the grain filling stage-maturity stage and relative yield.

[0039] Figure 7 Fig. 1 is a diagram of quantitative relationships between the spectral indices and the relative yield, wherein (a) is a diagram of quantitative relationships between the NDII-based cumulative integrated spectral index at the booting stage to the grain-filling stage and the relative yield; (b) is a diagram of quantitative relationships between the NDWI-based cumulative integrated spectral index at the booting stage to the grain-filling stage and the relative yield; (c) is a diagram of quantitative relationships between the MSI-based cumulative integrated spectral index at the booting stage to the grain-filling stage and the relative yield; (d) is a diagram of quantitative relationships between the SRWI-based cumulative integrated spectral index at the booting stage to the grain-filling stage and the relative yield; (e) is a diagram of quantitative relationships between the NDII-based cumulative integrated spectral index at the booting stage to the maturity stage and the relative yield; (f) is a diagram of quantitative relationships between the NDWI-based cumulative integrated spectral index at the booting stage to the maturity stage and the relative yield; (g) is a diagram of quantitative relationships between the MSI-based cumulative integrated spectral index at the booting stage to the maturity stage and the relative yield; (h) is a diagram of quantitative relationships between the SRWI-based cumulative integrated spectral index at the booting stage to the maturity stage and the relative yield; (i) is a diagram of quantitative relationships between the NDII-based cumulative integrated spectral index at the flowering stage to the maturity stage and the relative yield; (j) is a diagram of quantitative relationships between the NDWI-based cumulative integrated spectral index at the flowering stage to the maturity stage and the relative yield; (k) is a diagram of quantitative relationships between the MSI-based cumulative integrated spectral index at the flowering stage to the maturity stage and the relative yield; (l) is a diagram of quantitative relationships between the SRWI-based cumulative integrated spectral index at the flowering stage to the maturity stage and the relative yield.

[0040] Figure 8 Fig. 2 is a diagram of results of verifying relationships between the spectral indices and the relative yield by NRMSE, wherein (a) is a diagram of results of verifying relationships between the spectral indices and the relative yield by NRMSE; (b) is a diagram of results of verifying relationships between the integrated spectral indices and the relative yield by NRMSE; (c) is a diagram of results of verifying relationships between the cumulative integrated spectral indices and the relative yield by NRMSE. NRMSE is a normalized root mean square error.

[0041] Figure 9 Fig. 3 is a diagram of results of determining coefficients and NRMSE for predicting the relative yield according to the spectral indices, the integrated spectral indices and the cumulative integrated spectral indices, wherein (a) is a diagram of results of determining coefficients; (b) is a diagram of results of NRMSE.

[0042] Figure 10 Fig. 4 is a flowchart of predicting the yield of rice by using the yield prediction model established by the NDWI according to an embodiment of the present application. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical scheme and advantages of the present application more clearly understood, the present application will be further described in detail below with reference to the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0044] Reflectance spectral characteristics are determined by population traits, biochemical composition and morphological characteristics of plants. These traits are closely related to environmental temperature due to the changes in chlorophyll content and water content at different environmental temperatures. In addition, when using spectral practice, the ecological environment around the crops will be different, resulting in significant differences in the prediction ability of grain yield at different growth stages. Previous studies have shown that temperature-related phenological information can improve the spectral monitoring performance of wheat plants in different nitrogen nutrient states from booting to grain filling stages. Therefore, temperature-related phenological information will be considered as a valuable growth and development factor to improve the spectral estimation ability of crop growth process. However, the current research on phenological information is mainly used to divide different phenological stages of crops. In contrast, there are fewer studies on combining phenological information and spectral indicators to construct new indicators of rice plants. The present application aims to integrate phenological information into the crop growth state monitoring model, in order to effectively accumulate multiple instantaneous growth information at different specific growth stages, and to reduce the model error caused by the change of environmental factors during measurement. Based on this, it is necessary to combine phenological information and spectral parameters to construct new indicators to estimate the grain yield at the whole growth stage.

[0045] In order to more accurately monitor and estimate rice yield, scientists are constantly exploring new vegetation indices. At present, there are 14 types of spectral indices known to be well correlated with biomass, chlorophyll content and soil water status during crop growth. The present application eliminates unimportant indicators from 14 different types of spectral indices using the Pearson correlation coefficient, so as to facilitate the use of spectral indices closely related to yield to establish relevant yield prediction models.

[0046] The present application aims to effectively improve the accuracy of the results of integral spectral index in predicting rice yield by introducing phenological stages. Subsequent experiments have also proved that the cumulative phenological stage has a positive effect on improving the monitoring ability of rice yield under different irrigation. At the same time, when multiple phenological stages are incorporated into the spectral model, the NDWI parameter has a significant advantage. Therefore, we propose an application scheme using NDWI, trying to dynamically predict the grain yield at the growth and development stage of rice, as shown in Figure 10 The method for predicting rice yield by integral spectral index of the present application will be further described below with specific embodiments.

[0047] The methods described in the following examples are conventional unless otherwise specified. The reagents and materials described in the following examples are commercially available unless otherwise specified. Spectral scanning equipment, ASD Field Spec4, Boulder, CO, USA. In the following examples, the phenological stages include the booting stage, the anthesis stage, the filling stage and the maturity stage. The booting stage is abbreviated as B; the anthesis stage is abbreviated as A; the filling stage is abbreviated as F; and the maturity stage is abbreviated as M. The relative growth length day is abbreviated as RGDD; and the accumulated growth length day is abbreviated as AGDD.

[0048] 1. Sampling and measurement.

[0049] In order to obtain the canopy reflectance spectral data of rice at different phenological stages, the spectral scanning equipment was used to collect the canopy reflectance spectral data at the booting stage, the anthesis stage, the filling stage and the maturity stage of rice, respectively. The wave band amplitude range of the spectral scanning equipment was 350 nm-2500 nm.

[0050] All measurements were performed between 10:00 and 14:00 under clear and cloudless weather conditions. After 10 measurements of each sample, the average value was calculated as the final data, and the reference plate was corrected every 15 min. During each observation period, representative plants of all treatments were monitored. In all experiments, manual sampling was used to sample from three randomly selected 2 m 2 areas, and the yield of rice was accurately determined. The grains in the ear were taken out, and the grain yield was adjusted to the standard state with a moisture content of 13.5%.

[0051] 1.1. Relative yield.

[0052] The calculation formula of the relative yield is Y=Y i / Y max ; wherein Y is the relative yield; Y i is the yield value of different irrigation treatments; and Y max is the highest yield value in all irrigation treatments.

[0053] Different irrigation treatments are based on different water management modes, such as aerobic rice, flood irrigation, dry-wet alternate irrigation system, etc. The soil water potential of aerobic rice can reach-30 KPa. Flood irrigation, i.e. conventional flooding irrigation. Dry-wet alternate irrigation is to irrigate when the soil water potential reaches-15 KPa.

[0054] 1.2. Spectral index.

[0055] In order to study the growth status of rice, the canopy reflectance spectral data of different growth stages of rice collected are used as original spectral data. The original spectral data collected are preprocessed in View Spec Pro 6.0 software. According to the research results of predecessors, 14 different categories of spectral indices which are better correlated with the biomass of rice, the content of chlorophyll and the soil moisture condition in the growth process of crops are selected. The 14 categories of spectral indices and the calculation methods are shown in Table 1.

[0056] Table 1 List of spectral indices

[0057]

[0058]

[0059] Note: R represents the canopy reflectance spectral data of the corresponding waveband. For example, R 860 represents the canopy reflectance spectral data of the 860 nm waveband; R 1640 represents the canopy reflectance spectral data of the 1640 nm waveband. Through these canopy reflectance spectral data, the 14 categories of spectral indices of rice at different growth stages are calculated by using the calculation methods of the corresponding spectral indices. These categories of spectral indices are selected according to the research results of predecessors, and they have better correlation with the biomass of rice, the content of chlorophyll and the soil moisture condition.

[0060] 1.3, Relative growth length.

[0061] The growth length, referred to as GDD, is the cumulative effective accumulated temperature value experienced by a plant to complete a certain growth stage under actual environmental conditions. The calculation formula of the growth length is as follows:

[0062] Wherein, GDD is the growth length; T max is the highest temperature of the day; T min is the lowest temperature of the day; T base is the minimum temperature at which rice begins physiological activity, that is, the basic growth temperature, and in this embodiment, T base = 10℃.

[0063] The accumulated growth length, referred to as AGDD, is a quantitative index for evaluating the temperature accumulation experienced by crops within a certain time range to assess the growth conditions and growth progress of crops. The calculation formula of the accumulated growth length is as follows: Wherein, AGDD is the accumulated growth length; N is the growth days from the beginning of rice growth to the sampling time point; T max is the highest temperature of the day; T min is the lowest temperature of the day; T base is the minimum temperature at which rice begins physiological activity, that is, the basic growth temperature, and in this embodiment, T base=10℃.

[0064] Total growth days (TGDD) for the entire growing season is the sum of AGDD accumulated from the beginning to the end of the crop's growth period.

[0065] Relative growth days (RGDD) are the ratio of the accumulated growth days of a phenological period to the total growth days of the entire growing season. The formula for calculating relative growth days is:

[0066] RGDD = AGDD / TGDD; where RGDD is the relative growth degree-days (d℃); AGDD is the cumulative growth degree-days (d℃); and TGDD is the total growth degree-days (d℃) for the entire growth period.

[0067] 1.4 Comprehensive Spectral Index.

[0068] Comprehensive Spectral Index (SI) inte The spectral index value, denoted as SI, is obtained by weighted averaging of spectral index values ​​from different phenological stages. The SI value is calculated using the methods described in Table 1 for the corresponding spectral index categories. Table 1 lists 14 categories of spectral indices, and their calculation processes are similar to those in Table 1. Figure 1 Therefore, taking the reflectance value at the 1970 spectral band as an example, we will explain the calculation process of the comprehensive spectral index corresponding to the reflectance value at the 1970 spectral band. The calculation diagram is shown below. Figure 1 . Figure 1 In the diagram, the vertical projection of two sampling time points on the curve onto the horizontal axis is calculated, and the area of ​​the vertical projection between the curve corresponding to the two sampling time points and the horizontal axis is equal to SI. inte That is, SI inte It is obtained by integrating the SI during a certain growth stage. The formula for calculating the comprehensive spectral index is: SI inte =ΣSI i ×t i Among them, SI inte SI is the comprehensive spectral index. i It is the difference between the instantaneous spectral indices at two adjacent sampling time points; ∑SI i It is SI i The sum; t i It represents the continuous phenological stages experienced by two adjacent sampling time points. i represents the sampling time point; two adjacent sampling time points are represented as i and i-1, respectively.

[0069] To improve the correlation between the comprehensive spectral index and the above relative yield, the three-phenological-stage comprehensive spectral index, from the booting stage to the flowering stage, the flowering stage to the filling stage, and the filling stage to the mature stage, is combined into three continuous phenological stages, the cumulative comprehensive spectral index from the booting stage to the filling stage, the booting stage to the mature stage, and the flowering stage to the mature stage. Among them, the booting stage to the flowering stage is represented as booting stage-flowering stage; the flowering stage to the filling stage is represented as flowering stage-filling stage; the filling stage to the mature stage is represented as filling stage-mature stage; the booting stage to the filling stage is represented as booting stage-filling stage; the booting stage to the mature stage is represented as booting stage-mature stage; and the flowering stage to the mature stage is represented as flowering stage-mature stage.

[0070] 2. Data analysis and model validation.

[0071] The Pearson correlation coefficient, abbreviated as PCC, helps to understand the relationship between the features and the response variable and measures the linear correlation between variables. The Pearson correlation coefficient is used in the embodiments of the present application to eliminate unimportant indicators in the spectral index data. Indicators closely related to yield are used to establish a model to improve the performance of the model. The leave-one-out cross-validation, abbreviated as LOOCV, is performed on the regression function to determine the coefficient R 2 , the root mean square error RMSE and the normalized root mean square error NRMSE to test the reliability of the model.

[0072] R 2 is the coefficient of determination, which is used to quantify the degree of fitting of the model to the data. R 2 closer to 1, the better the model fits, and the stronger the explanatory power of the independent variable to the dependent variable. R 2 The calculation formula is:

[0073] wherein, x i is the measured value of SI; is the average value of the measured SI; is the predicted value of SI predicted by the model; and n is the sample size.

[0074] RMSE is the root mean square error, which is used to measure the standard deviation of the difference between the predicted value and the true value of the model. It quantifies the average error amplitude of the predicted value. The smaller the RMSE, the more accurate the prediction of the model. The calculation formula of RMSE is: wherein, wherein, x i is the measured value of SI; is the predicted value of SI predicted by the model; and n is the sample size.

[0075] NRMSE, or Normalized Root Mean Square Error, is normalized by dividing the RMSE by the standard deviation of the true values. It provides a measure of error relative to data variation, making it easier to compare model performance across different datasets. A smaller NRMSE value indicates higher predictive accuracy. NRMSE is helpful for comparisons between datasets with different scales because it eliminates differences in error magnitude caused by varying data ranges. The formula for calculating NRMSE is:

[0076] Where RMSE is the root mean square error; S is the measured average of the relative yield.

[0077] 3. Results and Analysis.

[0078] 3.1 Feature index screening.

[0079] Based on experimental data from four phenological stages, a systematic analysis was conducted on the 14 spectral indices from step 1.2 to identify suitable spectral index values ​​for monitoring rice yield. The Pearson correlation coefficients between the 14 spectral indices and relative yield are shown below. Figure 2 As shown. Figure 2 The graph shows the Pearson correlation coefficients between the 14 spectral indices and relative yields. * indicates a significant correlation at the 0.05 level; ** indicates a significant correlation at the 0.01 level.

[0080] Depend on Figure 2 The results show that the spectral indices NDII, NDWI, MSI and SRWI have the best correlation with relative yield, with correlation coefficients of -0.57, -0.55, 0.55 and -0.52, respectively, and P<0.01.

[0081] 3.2 Correlation between characteristic indicators of different phenological stages and yield.

[0082] The absolute values ​​of the correlation between spectral indices of individual phenological stages and relative yield at maturity under different irrigation regimes, such as... Figure 3 As shown. By Figure 3 The results showed that the spectral indices for monitoring rice yield were NDII, NDWI, MSI, and SRWI. These four spectral indices showed correlations of 0.60–0.74 with relative yield at the booting, flowering, grain-filling, and maturity stages. At each phenological stage, these four spectral indices demonstrated good correlations with relative yield, with p < 0.05.

[0083] The absolute value of the correlation between the composite spectral index of continuous phenological periods or the cumulative composite spectral index and relative yield, such as... Figure 4As shown. Continuous phenological periods, such as booting stage to flowering stage, flowering stage to grain-filling stage, and grain-filling stage to maturity stage. Relative yield is based on the relative yield at maturity. Figure 4 The results showed that the correlation between the comprehensive spectral index and relative yield reached 0.47–0.84 during the booting-flowering, flowering-grain-filling, and grain-filling-maturity stages. Among them, the NDII and SRWI during the booting-flowering stage, and the MSI and SRWI during the flowering-grain-filling stage, did not reach the 0.05 significance level.

[0084] The cumulative comprehensive spectral index showed correlation coefficients of 0.61–0.89 with relative yield during the booting-filling, booting-maturity, and flowering-maturity stages. The correlation between the cumulative comprehensive spectral index and relative yield was generally good, with p < 0.05.

[0085] In summary, under different irrigation regimes, the spectral index, composite spectral index, and cumulative composite spectral index of each phenological stage all showed good correlation between NDWI and relative yield.

[0086] 3.3 Estimating rice yield using spectral indices.

[0087] Table 2. Quantitative Relationship between Spectral Indices and Relative Yields at Different Phenological Stages

[0088]

[0089]

[0090] This invention establishes a quantitative relationship between spectral indices and relative yield at different phenological stages based on different irrigation regimes. The quantitative relationship between spectral indices and relative yield at the booting, flowering, grain-filling, and maturity stages is shown below. Figure 5 .according to Figure 5 As shown in Table 2, relative yield is expressed as a quadratic function of spectral indices at different phenological stages. The coefficients of determination between spectral indices and relative yield range from 0.43 to 0.59, with NDWI showing better predictive ability for relative yield, averaging R0.05. 2 =0.54.

[0091] 3.4. Estimating rice yield using comprehensive spectral indices.

[0092] Table 3. Quantitative Relationship between Comprehensive Spectral Index and Relative Yield in Different Continuous Phenological Stages

[0093]

[0094]

[0095] Under different irrigation regimes, the relative yield in different continuous phenological stages is also expressed as a quadratic function of the integrated spectral index, and the quantitative relationship between the integrated spectral index and the relative yield in the stages of booting to flowering, flowering to grain filling, and grain filling to maturity is as follows: Figure 6 According to the results in Table 3 and Figure 6 , the estimation performance of the integrated spectral index on the relative yield in the stages of booting to flowering, flowering to grain filling, and grain filling to maturity is compared, and is 0.26-0.43, 0.31-0.73, and 0.64-0.79, respectively. It is observed that the estimation performance of the integrated spectral index is not as good as that of the spectral index in the stage of booting to flowering, but is significantly improved in the stage of grain filling to maturity.

[0096] In order to improve the estimation performance of the integrated spectral index on the relative yield under different irrigation regimes, the integrated spectral index in the above continuous phenological stages is combined into three continuous phenological stages, i.e., the stages of booting to grain filling, booting to maturity, and flowering to maturity, and the quantitative relationship between the cumulative integrated spectral index and the relative yield in the stages of booting to grain filling, booting to maturity, and flowering to maturity is as follows: Figure 7 .

[0097] Table 4 Simulation relationship between the cumulative integrated spectral index and the relative yield in different continuous phenological stages

[0098]

[0099]

[0100] According to the results in Table 4 and Figure 7 , the determination coefficient of the cumulative integrated spectral index on the relative yield in each continuous phenological stage reaches 0.48-0.86. The estimation performance of the cumulative integrated spectral index on the relative yield in the stage of booting to maturity is generally good. Meanwhile, compared with the spectral index and the integrated spectral index in a single phenological stage, the cumulative integrated spectral index has better estimation performance, as shown in Fig. (a). Figure 9 However, no matter the spectral index, the integrated spectral index, or the cumulative integrated spectral index, the determination coefficient of NDWI on the relative yield in each phenological stage is good. Figure 9 Fig. is a result diagram of the determination coefficient and NRMSE of the relative yield predicted according to the spectral index, the integrated spectral index, and the cumulative integrated spectral index. In the figure, (a) is a result diagram of the determination coefficient, and (b) is a result diagram of NRMSE.

[0101] 3.5. Verification of the quantitative relationship.

[0102] The leave-one-out cross validation is performed on the spectral index, the comprehensive spectral index and the accumulated comprehensive spectral index to evaluate the regression relationship of the estimated relative yield under different irrigation systems, and the result graph of the relationship between the spectral index, the comprehensive spectral index and the accumulated comprehensive spectral index and the relative yield is verified by NRMSE, as shown in Figure 8 . According to the verification result of Figure 8 , it is proved that the quantitative relationship between the spectral index, the comprehensive spectral index and the accumulated comprehensive spectral index and the relative yield is the best in the middle and late growth period, that is, the booting stage to the maturity stage. Meanwhile, compared with the spectral index and the comprehensive spectral index of a single phenological stage, the accumulated comprehensive spectral index has a lower NRMSE, as shown in the (b) graph of Figure 9 . Therefore, when estimating the relative yield in the maturity stage, the stability of the accumulated comprehensive spectral index estimation model is better than that of the spectral index and the comprehensive spectral index.

[0103] According to the above results, it is shown that the continuous phenological stage has a positive effect on improving the monitoring ability of different irrigation rice yield. Meanwhile, when multiple phenological stages are included in the spectral model, the NDWI parameter has a significant advantage. Therefore, the yield prediction model is established by using the NDWI, and the spectral rice yield is predicted by using the yield prediction model. The yield prediction model attempts to dynamically predict the grain yield in the growth and development stage of rice on the basis of ensuring the accuracy of the prediction result, as shown in Figure 10 . Figure 10 The flow chart of the yield prediction model established by using the NDWI for predicting the rice yield according to the embodiment of the present application is shown in the figure. The * indicates a significant correlation at the 0.05 level. The ** indicates a significant correlation at the 0.01 level.

[0104] The method for predicting the rice yield by using the integral spectral index includes the following steps:

[0105] Step 1: The accumulated growth length day at the sampling time point and the total growth length day in the whole growth period are obtained to calculate the relative growth length day at the sampling time point and determine the corresponding phenological stage of the sampling time point; the phenological stage includes the booting stage, the flowering stage, the grain filling stage and the maturity stage. The calculation formula of the relative growth length day is: RGDD=AGDD / TGDD; wherein, RGDD is the relative growth length day, d℃; AGDD is the accumulated growth length day, d℃; TGDD is the total growth length day in the whole growth period, d℃. Through step 1, the sampling time point can be accurately divided into the booting stage, the flowering stage, the grain filling stage and the maturity stage of the rice plant.

[0106] Step 2, the canopy reflectance spectrum data at different sampling time points are collected by using a spectral scanning device, and the spectral index corresponding to the sampling time points is obtained according to the spectral index calculation formula of the normalized difference water index.

[0107] NDWI = (R 860 -R 1640 ) / (R 860 +R 1640 ); wherein, NDWI is the normalized difference water index; R is the canopy reflectance spectrum data corresponding to the wave band. For example, R 860 is the canopy reflectance spectrum data at the wave band of 860 nm; R 1640 is the canopy reflectance spectrum data at the wave band of 1640 nm.

[0108] According to the determined phenological stage and spectral index value corresponding to the sampling time point, the comprehensive spectral index corresponding to the continuous phenological stage is calculated by using the comprehensive spectral index calculation formula; the comprehensive spectral index calculation formula is: SI inte =∑SI i ×tx; wherein, SI inte is the comprehensive spectral index; SI i is the difference value of the instantaneous spectral index of the adjacent two sampling time points; ∑SI i is the sum of SI i ; t i is the continuous phenological stage experienced by the adjacent two sampling time points. i represents the sampling time point; the adjacent two sampling time points are represented as i and i-1 respectively.

[0109] Step 3, according to the phenological stage determined in step 1, the spectral index or comprehensive spectral index corresponding to the phenological stage is input into the corresponding yield prediction model to calculate the predicted yield of rice. For example, Figure 10 , the specific method is:

[0110] According to the phenological stage determined in step 1, it is judged whether it is the flowering period, when the output result is no, the spectral index corresponding to the booting stage is input into the yield prediction model of the booting stage to calculate the predicted yield.

[0111] It is judged whether it is the flowering period, when the output result is yes, it is further judged whether it is the grain filling period, when the output result is no, the comprehensive spectral index corresponding to the booting stage-flowering period is input into the yield prediction model of the booting stage-flowering period to calculate the predicted yield.

[0112] It is judged whether it is the grain filling period, when the output result is yes, it is further judged whether it is the mature period, when the output result is no, the comprehensive spectral index corresponding to the booting stage-grain filling period is input into the yield prediction model of the booting stage-grain filling period to calculate the predicted yield.

[0113] When the output result is yes, the comprehensive spectral index corresponding to the booting stage-mature stage is input into the yield prediction model of the booting stage-mature stage to calculate the predicted yield.

[0114] The yield prediction model of different phenological stages is shown in Table 5.

[0115] Table 5 Quantitative relationship between comprehensive spectral index and relative yield at different phenological stages

[0116] Phenological stage Relationship between the integrated spectral index and the relative yield Heading stage y = -0.68 + 40.10x - 238.01x 2 , R 2 = 0.57 * ]]> From heading stage to flowering stage <![CDATA[y=-0.31+162.89x-5058.92x 2 ,R 2 =0.41]]> From heading stage to milking stage y = -1.44 + 196.25x - 3878.75x 2 , R 2 = 0.73 ** ]]> From heading stage to maturity stage y = 1.95 + 15.13x - 239.50x 2 , R 2 = 0.79 ** ]]>

[0117] Note: The comprehensive spectral index is taken as the horizontal coordinate, denoted as x, and the relative yield is taken as the vertical coordinate, denoted as y; the cumulative comprehensive spectral index is the cumulative comprehensive spectral index based on NDWI.

[0118] The method for predicting the yield of rice provided by the embodiment of the present application will help to quickly evaluate the yield potential of different water management modes in rice plants in a larger area, such as aerobic rice, flood irrigation and alternating dry-wet irrigation systems. In addition, the method can guide farmers to establish potential yield targets in water-sensitive stages in the future main phenological stages to achieve precise irrigation. These established relationships show that the yield can be quantitatively evaluated by the cumulative comprehensive spectral index. Of course, more cases need to be verified in the future to establish a stable and universal model.

[0119] The preferred embodiments of the present application are merely used for the purpose of illustration, and are not used to limit the present application, and any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting rice yield using an integral spectral index, characterized in that, Includes the following steps: The relative growth degree-days at the sampling time points are obtained to determine the phenological stage corresponding to the sampling time points; the phenological stages include the heading stage, flowering stage, grain-filling stage, and maturity stage; Acquire canopy reflectance spectral data at different sampling time points to determine the spectral index value at the corresponding sampling time point; the spectral index is the normalized differential water index. Based on the phenological stage and spectral index value corresponding to the determined sampling time point, the comprehensive spectral index value corresponding to the continuous phenological stage is obtained, and then substituted into the corresponding yield prediction model to obtain the predicted rice yield. The yield prediction model is a yield prediction model corresponding to the phenological stage with the spectral index or comprehensive spectral index as the horizontal axis and the relative yield as the vertical axis. The formula for calculating the comprehensive spectral index is as follows: ； Among them, SI inte SI is the comprehensive spectral index. i It is the difference between the instantaneous spectral indices at two adjacent sampling time points; It is SI i The sum; t i It refers to the continuous phenological stages experienced at two adjacent sampling time points.

2. The method for predicting rice yield using the integrated spectral index according to claim 1, characterized in that, The predicted rice yield is the predicted rice yield under different irrigation systems.

3. The method for predicting rice yield using the integral spectral index according to claim 1, characterized in that, The corresponding yield prediction models include yield prediction models for the booting stage, yield prediction models for the booting to flowering stage, yield prediction models for the booting to grain-filling stage, and yield prediction models for the booting to maturity stage.

4. The method for predicting rice yield using the integral spectral index according to claim 3, characterized in that, The yield prediction model for the booting stage is y = -0.68 + 40.10x - 238.01x 2 R 2 =0.57; The yield prediction model for the booting to flowering stage is y = -0.31 + 162.89x' - 5058.92x'. 2 R 2 =0.41; The yield prediction model for the booting stage to the grain-filling stage is y = -1.44 + 196.25x' - 3878.75x'. 2 R 2 =0.73; The yield prediction model for the booting stage to maturity stage is y = 1.95 + 15.13x' - 239.50x'. 2 R 2 =0.79; Where x is the spectral index; x' is the comprehensive spectral index; y is the relative yield; R 2 The coefficient of determination.

5. The method for predicting rice yield using the integral spectral index according to claim 4, characterized in that, The formula for calculating the spectral index is as follows: FIRE=(R 860 -R 1640 ) / (R 860 +R 1640 )4 Wherein, NDWI is the normalized differential water index; R is the canopy reflectance spectral data for the corresponding band.

6. The method for predicting rice yield using the integrated spectral index according to claim 1, characterized in that, The continuous phenological stages include the booting stage to flowering stage, the booting stage to grain-filling stage, and the booting stage to maturity stage.

7. The method for predicting rice yield using the integrated spectral index according to claim 1, characterized in that, The specific method for obtaining the relative growth days at the sampling time point is as follows: The cumulative growth days at the sampling time point and the total growth days throughout the entire growth period are obtained to calculate the relative growth days at the sampling time point.

8. The method for predicting rice yield using the integral spectral index according to claim 7, characterized in that, The formula for calculating the relative growth days is: ； Wherein, RGDD is the relative growth degree days, d℃; AGDD is the cumulative growth degree days, d℃; and TGDD is the total growth degree days, d℃ for the entire growth period.

9. The method for predicting rice yield using the integral spectral index according to claim 8, characterized in that, The formula for calculating accumulated growth days is: ； Where AGDD represents the accumulated growth days; N represents the number of growth days from the start of rice growth to the sampling time point; T max The highest temperature of the day; T min The lowest temperature of the day; T base This is the lowest temperature at which rice begins its physiological activities.

Images