Crop leaf water potential estimation method based on hyperspectral reflectivity

Abstract

The invention relates to a crop leaf water potential estimation method based on hyperspectral reflectivity. The crop leaf water potential estimation method comprises the following steps: measuring the spectral reflectivity and leaf water potential of a leaf under a preset wave band; determining a candidate vegetation index based on the spectral reflectance; screening a target vegetation index from the candidate vegetation indexes based on the relationship between the candidate vegetation indexes and the leaf water potential; and constructing a target prediction model based on the leaf water potential and the target vegetation index, and predicting the leaf water potential of the leaf based on the target prediction model. According to the method, the leaf water potential measurement efficiency is greatly improved, a feasible technical scheme is provided for crop water and fertilizer deficiency diagnosis, and the method has important scientific significance and practical value for improving automation and intellectualization of water and fertilizer management.

Description

Technical Field

[0001] The present invention belongs to the technical field of agricultural intelligent irrigation, and specifically relates to a method for estimating water potential of crop leaves based on hyperspectral reflectance. Background Art

[0002] Drought is the main natural disaster threatening crop production, and its impact on agronomy is the sum of all other natural disasters. Due to climate change and human activities, the global spatial and temporal pattern of precipitation has become more variable. The frequency, duration and intensity of drought events in the future are expected to increase. The drying of the atmosphere will lead to a vapor pressure deficit and an increase in irrigation water demand, further exacerbating the contradiction between food security and water shortage. Drought affects physiological processes by affecting the water status of crops. Crop leaves are the most sensitive tissues under drought stress and directly reflect the degree of drought in plants. The parameters that characterize the water status of leaves mainly include leaf moisture content, leaf relative moisture content and leaf water potential.

[0003] Leaf water potential is an important indicator that reflects the response of plants to water stress. It is usually used as the main controlling factor of stomatal conductance under stress. On the one hand, it directly affects the opening and closing of leaf stomata, and on the other hand, it directly affects the hydraulic transmission efficiency of crops. Both have a synergistic effect on the photosynthetic physiological process of leaves. Leaf water potential measurement plays an important role in agricultural production and the assessment of crop physiological status. At present, the traditional method is to cut the leaves, put them in a moist self-sealing bag to reduce water loss, and use a water potential meter to measure as quickly as possible. Water potential meters generally have two categories: plant water potential pressure chambers and dew point water potential meters, but both require destructive leaf samples for measurement. Crop water potential measurement in field environments is difficult to meet the needs of continuous high-throughput monitoring. Therefore, it is very necessary to develop a non-destructive, convenient and efficient method for determining crop water potential. Summary of the invention

[0004] In order to solve the problems in the prior art, the present invention provides a crop leaf water potential estimation method based on hyperspectral reflectance, characterized in that it includes: measuring the spectral reflectance and leaf water potential of leaves in a predetermined band; determining a candidate vegetation index based on the spectral reflectance; based on the relationship between the candidate vegetation index and the leaf water potential, screening a target vegetation index from the candidate vegetation index; and constructing a target prediction model based on the leaf water potential and the target vegetation index, and predicting the leaf water potential of the leaves based on the target prediction model.

[0005] Furthermore, a target prediction model is constructed based on leaf water potential and a target vegetation index, and predicting the leaf water potential of leaves based on the target prediction model includes: constructing a target prediction model based on leaf water potential and a target vegetation index, selecting an optimal prediction model from the target prediction model, and predicting the leaf water potential of leaves based on the optimal prediction model.

[0006] Furthermore, screening the optimal prediction model from the target prediction models includes selecting a target prediction model that meets predetermined conditions from the target prediction models as the optimal prediction model, and the predetermined conditions are related to the relationship between the predicted water potential value of the target prediction model and the actually measured water potential value.

[0007] Further, the relationship between the predicted water potential value and the actually measured water potential value includes at least one of a coefficient of determination, a mean absolute error, a root mean square error, and a normalized root mean square error.

[0008] Furthermore, the normalized root mean square error of the optimal prediction model is less than 20%.

[0009] Furthermore, the predetermined wavelength band is 350~2500nm.

[0010] Further, based on the relationship between the candidate vegetation index and the leaf water potential, screening the target vegetation index from the candidate vegetation index includes: screening the vegetation index whose correlation coefficient between the candidate vegetation index and the leaf water potential is greater than a predetermined value from the candidate vegetation index as the target vegetation index.

[0011] Further, the predetermined value is 0.8.

[0012] Further, the spectral reflectance of the leaf in a predetermined wavelength band is first measured, and then the leaf water potential is measured.

[0013] Furthermore, the target prediction model is constructed by using the multivariate stepwise regression method or the partial least squares regression method.

[0014] In summary, the present invention has the following advantages: the leaf water potential value is inverted according to the spectral reflectance, the theoretical basis is clear, and the operability is strong. The high spectral reflectance estimation model of leaf water potential is determined by the preliminary basic measurement, and the spectrometer can be used to realize large-scale, non-destructive and rapid leaf water potential measurement during the growth period. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flow chart of the crop leaf water potential estimation method based on hyperspectral reflectance according to the present invention.

[0016] Figure 2 This is the correlation coefficient graph between vegetation index and leaves of each variety.

[0017] Figure 3(a) to Figure 3(e) are the measured values ​​and predicted values ​​of the PLSR (partial least squares) model.

[0018] Figures 4(a) to 4(e) show the measured and predicted values ​​of the multivariate linear stepwise regression model. DETAILED DESCRIPTION

[0019] In the following description of examples or embodiments of the present invention, reference will be made to the accompanying drawings, in which specific examples or embodiments that may be implemented are shown by way of illustration, and in which the same reference numerals and symbols may be used to designate the same or similar components, even if they are shown in different drawings. In addition, in the following description of examples or embodiments of the present invention, when it is determined that a detailed description of well-known functions and components incorporated herein may make the subject matter in some embodiments of the present invention less clear, the description will be omitted. Terms such as "including", "having", "comprising", "consisting of", "comprising", and "forming" used herein are generally intended to allow for the addition of other components unless these terms are used with the term "only". As used herein, the singular form is intended to include the plural form unless the context clearly indicates otherwise.

[0020] Reference Figure 1 , showing a flow chart of a crop leaf water potential estimation method based on hyperspectral reflectance according to the present invention.

[0021] like Figure 1 As shown, the method includes step S101: measuring the spectral reflectance and leaf water potential of the leaf in a predetermined band.

[0022] In an embodiment, 60-90 fully expanded and mature leaves of the crop can be collected before dawn, and the cut ends at the base of the leaves are immersed in air-free pure water and stored in a dark place to ensure that the plants are in a state of near saturated turgor pressure.

[0023] In the embodiment, the leaf samples can be brought back to the room and the ambient temperature can be controlled at 26°C to ensure the stability and consistency of the test conditions. The hyperspectral reflectance and leaf water potential are measured at different intervals (the first 15 minutes, the measurement is performed every 5 minutes, and 2-3 leaves are measured each time, and then the measurement is performed at intervals of 10 minutes, 30 minutes, 1 hour and 2 hours respectively).

[0024] In an embodiment, the hyperspectral reflectance may be measured first and then the leaf water potential may be measured.

[0025] In the embodiment, the spectral reflectance of the blade is measured using a portable ground object spectrometer equipped with a blade clip, and the effective band of the ground object spectrometer is 350~2500nm. Therefore, the wavelength range of the spectral reflectance of the present invention can be 350~2500nm. The spectral sampling can be 10 spectra as a sampling spectrum, that is, 10 spectra are recorded each time, and the average value is used as the spectral reflectance of the sample point. Each blade is measured three times, and the average is taken as the spectral reflectance of the blade. A whiteboard is used for calibration before each reflectance measurement.

[0026] In the embodiment, the leaf water potential can be measured by using a plant water potential pressure chamber, where leaves are placed in the leaf chamber, and the cutout is left outside the leaf chamber through a rubber plug for observation. The leaf chamber is slowly pressurized, and the negative value of the initial pressure when water drops drip out of the cutout is the leaf water potential value.

[0027] The method includes step S103: determining a candidate vegetation index based on spectral reflectance.

[0028] In the present invention, the candidate vegetation index may be various vegetation indices, such as the normalized difference vegetation index, but the present invention is not limited thereto, and may also be other vegetation indices.

[0029] For example, the Normalized Difference Vegetation Index ( NDVI ) can be determined according to the following formula: , in, R i , R j are the spectral reflectances at bands i and j respectively.

[0030] The method comprises step S105: based on the relationship between the candidate vegetation index and the leaf water potential, selecting the target vegetation index from the candidate vegetation index.

[0031] In an embodiment, a vegetation index whose correlation coefficient between the candidate vegetation index and the leaf water potential is greater than a predetermined value may be selected as the target vegetation index. The predetermined value may be 0.8, but the present invention is not limited thereto.

[0032] The method includes step S107: constructing a target prediction model based on leaf water potential and a target vegetation index, and predicting the leaf water potential of leaves based on the target prediction model.

[0033] In an embodiment, the selected target vegetation index can be used as an independent variable and the leaf water potential can be used as a dependent variable to construct a target prediction model. The target prediction model can be constructed based on a partial least squares method or a multivariate linear regression method, but the present invention is not limited thereto.

[0034] Optionally, in order to further improve the accuracy of the target prediction model, a target prediction model that meets predetermined conditions may be selected from the target prediction models as an optimal prediction model, and the leaf water potential of the leaf may be predicted by the optimal prediction model.

[0035] In this case, in order to construct the target prediction model and select the optimal prediction model from the target prediction model, the data corresponding to the leaf water potential and the target vegetation index can be sorted in ascending order according to the leaf water potential value, and the data can be divided into two groups according to the odd and even numbers of the serial numbers. The first group is used to construct multiple target prediction models, and the second group can be used to verify multiple target prediction models, that is, to select the optimal prediction model from multiple target prediction models.

[0036] After obtaining multiple target prediction models through the first set of data, the multiple target prediction models can be screened through the second set of data to obtain the optimal prediction model.

[0037] The predetermined condition may be related to the relationship between the predicted water potential value of the target prediction model and the actually measured water potential value. The relationship between the predicted water potential value of the target prediction model and the actually measured water potential value may, for example, include a coefficient of determination (R 2 ), mean absolute error (MAE), root mean square error (RMSE), and normalized root mean square error (NRMSE), where: , (1) , (2) , (3) , (4) Where: O i and P i Respectively i The actual measured water potential value and the predicted water potential value; O max and O min Respectively represent the maximum and minimum values ​​of the actually measured water potential; n Represents the number of observations.

[0038] As described above, the second set of data can be input into the target prediction model obtained according to the first set of data to screen the optimal prediction model. The second set of data can be predicted according to the target prediction model, and the optimal prediction model can be screened based on the relationship between the predicted water potential value and the actual measured water potential value in the second set of data.

[0039] You can select R from the target prediction model 2 The target prediction model whose RMSE, MAE and NRMSE are respectively smaller than the second threshold, the third threshold and the fourth threshold is taken as the optimal prediction model, wherein the first threshold, the second threshold, the third threshold and the fourth threshold can be determined according to actual conditions, and the present invention does not specifically limit this.

[0040] For example, you can select R 2 The model with the largest RMSE, MAE and NRMSE is selected as the optimal prediction model.

[0041] In particular, in order to ensure the prediction accuracy of the model, the NRMSE of the optimal prediction model is required to be less than 20%. When the NRMSE is greater than 30%, it is not suitable as the optimal prediction model.

[0042] Therefore, according to the present invention, a method for estimating the water potential of crop leaves based on hyperspectral reflectance is provided. The method measures the hyperspectral reflectance and corresponding water potential of a series of leaves with different moisture contents based on the different spectral reflectance characteristics of leaves with different moisture contents, calculates and screens the vegetation index highly correlated with the leaf water potential, uses the screened vegetation index as an independent variable and the leaf water potential as a dependent variable, constructs a model of leaf water potential and vegetation index, and makes predictions based on the model. Furthermore, a model with an NRMSE of less than 20% can be selected as the optimal model.

[0043] Compared with the traditional leaf water potential measurement method, the traditional method requires cutting the leaves and putting them into a moist ziplock bag to reduce water loss. The water potential meter is used for measurement as quickly as possible. The water potential meter generally has two categories: plant water potential pressure chamber and dew point water potential meter. However, the traditional method requires destructive leaf sample measurement. In the field environment, crop water potential measurement is difficult to meet the continuous high-throughput monitoring requirements. The present invention overcomes the problems of destructive sampling of crop leaf water potential measurement and provides a non-destructive, convenient and efficient method for determining crop water potential. It greatly improves the efficiency of crop leaf water potential measurement and provides a feasible technical solution for crop water and fertilizer deficiency diagnosis.

[0044] In order to further describe the method of the present invention, the present invention is illustrated by taking the test base of China Institute of Water Resources and Hydropower Research in Daxing District, Beijing as an example.

[0045] The test was conducted at the China Institute of Water Resources and Hydropower Research Test Base in Daxing District, Beijing. The geographical location is 39°39′ north latitude and 116°15′ east longitude. The climate is a typical semi-arid continental monsoon climate with an average annual rainfall of 540 mm. The base has an automatic weather station, which can obtain relevant meteorological data. The data is automatically recorded every 30 minutes and can be downloaded regularly. According to the international soil texture classification standard, the 0-100 cm depth of the site is loam, and the average field holding capacity and bulk density are 30.58% and 1.41 g / cm respectively. 3 .

[0046] 70 leaves of three varieties of field corn and two varieties of soybean were selected at the National Water-Saving Irrigation Beijing Engineering Technology Research Center Experimental Base in Weishanzhuang Town, Daxing District, Beijing. Each target leaf was sampled before dawn, and the cut was placed in airless pure water to ensure that the plant was in a state close to saturated turgor pressure. The selected samples were all fully expanded and mature leaves.

[0047] The leaves were brought back to the laboratory, and the experimental environment temperature was controlled at 26°C to ensure the stability and consistency of the test conditions. The retrieved corn leaves were neatly placed on a drying table. After the leaves were retrieved, the leaf spectral reflectance and leaf water potential were measured alternately. For the first 15 minutes, the measurements were made every 2 minutes, and 2-3 leaves of each variety were measured each time. After that, the measurements were carried out at intervals of 5 minutes, 10 minutes, 30 minutes, 1 hour, and 2 hours.

[0048] The leaf spectral reflectance was measured using a portable ground feature spectrometer (FieldSpec4 Wide-Res, produced by ASD, USA) equipped with a leaf clip. The effective band of the ground feature spectrometer is 350~2500nm. The spectrum sampling is based on 10 spectra as a sampling spectrum, that is, 10 spectra are recorded each time, and the average value is used as the canopy spectral reflectance of the sample point. Each leaf was measured three times, and the average was taken as the leaf spectral reflectance. A white board was used for correction before each reflectance measurement. After measuring the leaf spectral reflectance, the leaf water potential was measured again. A plant water potential pressure chamber was used to put the leaf into the leaf chamber, and the incision was left outside the leaf chamber through a rubber plug for observation. The leaf chamber was slowly pressurized, and the negative value of the initial pressure when water droplets dripped out of the incision was the leaf water potential value.

[0049] like Figure 2 As shown in the figure, 17 published vegetation indices were correlated with leaf water potential of different varieties of corn and soybean to screen out spectral vegetation indices that are sensitive to leaf water potential. For corn, the sensitive vegetation indices screened out for the Epoch 168 variety include: SR (1300,1450), SEVEN , SRWI , I AM , Dn (1530,525), MSI , SEVEN 1640 and SEVEN 2130; the screening results of Jingke 968 varieties are the same; the vegetation index of Zhengdan 958 varieties is: SR (1300,1450), I AM , Dn (1530,525), MSI , SEVEN 1640 and SEVEN 2130; For soybean, the screening results of the vegetation index of Zhonghuang 13 varieties are: YES , SR (1300,1450), SEVEN , SRWI , R(610,560) / ND(810,610),VARI; Vegetation index selected by Qihuang 34: YES , SR (1300,1450), SEVEN , SRWI , YES / NDVI , the selected vegetation indices were used to establish the input variables of PLSR and multivariate linear stepwise regression models.

[0050] PLSR and multivariate linear stepwise regression analysis: The selected spectral vegetation indices were used as model inputs to construct PLSR models and multivariate linear stepwise regression models for each crop and variety. The construction and verification results are shown in Tables 1 and 2 and Figures 3(a) to 3(e) and 4(a) to 4(e), where Figures 3(a) to 3(e) and 4(a) to 4(e) represent Jiyuan 168, Zhengdan 958, Jingke 968, Zhonghuang 13, and Qihuang 34, respectively.

[0051] Table 1 Partial least squares (PLSR) model results

[0052] Table 2 Results of multivariate linear stepwise regression model

[0053] According to the results in Tables 1 and 2 and Figures 3(a) to 3(e) and 4(a) to 4(e), the prediction effect of the PLSR model in soybean is better, and the R 2 All above 0.63, NRMSE The prediction accuracy of the multivariate linear stepwise regression model in corn was higher, and the R 2 More than 0.88, NRMSE It is lower than 10.57%, which shows that the established model has good prediction effect and strong stability.

[0054] In summary, the natural drying method was used to alternately measure the leaf hyperspectral reflectance and leaf water potential of typical crops such as corn and soybean, and a model for leaf water potential inversion using hyperspectral reflectance was constructed, which achieved non-destructive high-pass measurement of leaf water potential during the growth period. This method greatly improves the efficiency of leaf water potential measurement, provides a feasible technical solution for crop water and fertilizer deficiency diagnosis, and has important scientific significance and practical value for improving the automation and intelligence of water and fertilizer management.

[0055] Although the specific implementation of the present invention is described in detail in conjunction with the drawings, it should not be understood as limiting the scope of protection of this patent. Within the scope described in the claims, various modifications and variations that can be made by those skilled in the art without creative work still fall within the scope of protection of this patent.

Claims

1. A method for estimating crop leaf water potential based on hyperspectral reflectance, characterized in that: include: Determine the spectral reflectance and leaf water potential of leaves in a predetermined band; determining a candidate vegetation index based on the spectral reflectance; Based on the relationship between the candidate vegetation index and the leaf water potential, screening a target vegetation index from the candidate vegetation index; as well as A target prediction model is constructed based on the leaf water potential and the target vegetation index, and the leaf water potential of the leaf is predicted based on the target prediction model.

2. The method for estimating crop leaf water potential based on hyperspectral reflectance according to claim 1, characterized in that: Constructing a target prediction model based on the leaf water potential and the target vegetation index, and predicting the leaf water potential of the leaves based on the target prediction model includes: constructing a target prediction model based on the leaf water potential and the target vegetation index, selecting an optimal prediction model from the target prediction models, and predicting the leaf water potential of the leaves based on the optimal prediction model.

3. The method for estimating crop leaf water potential based on hyperspectral reflectance according to claim 2, characterized in that: Screening the optimal prediction model from the target prediction model includes selecting a target prediction model that meets a predetermined condition from the target prediction model as the optimal prediction model, and the predetermined condition is related to the relationship between the predicted water potential value of the target prediction model and the actually measured water potential value.

4. The method for estimating crop leaf water potential based on hyperspectral reflectance according to claim 3, characterized in that: The relationship between the predicted water potential value and the actually measured water potential value includes at least one of a coefficient of determination, a mean absolute error, a root mean square error, and a normalized root mean square error.

5. The method for estimating crop leaf water potential based on hyperspectral reflectance according to claim 4, characterized in that: The normalized root mean square error of the optimal prediction model is less than 20%.

6. The method for estimating crop leaf water potential based on hyperspectral reflectance according to claim 1, characterized in that: The predetermined wavelength range is 350-2500 nm.

7. The method for estimating crop leaf water potential based on hyperspectral reflectance according to claim 1, characterized in that: Based on the relationship between the candidate vegetation index and the leaf water potential, screening the target vegetation index from the candidate vegetation index includes: screening the vegetation index from the candidate vegetation index whose correlation coefficient between the candidate vegetation index and the leaf water potential is greater than a predetermined value as the target vegetation index.

8. The method for estimating crop leaf water potential based on hyperspectral reflectance according to claim 7, characterized in that: The predetermined value is 0.

8.

9. The method for estimating crop leaf water potential based on hyperspectral reflectance according to claim 1, characterized in that: First, the spectral reflectance of the leaves in a predetermined band is measured, and then the leaf water potential is measured.

10. The method for estimating crop leaf water potential based on hyperspectral reflectance according to any one of claims 1 to 9, characterized in that: The target prediction model is constructed by using a multivariate stepwise regression method or a partial least squares regression method.

Images