Keywords

1 Introduction

Since the 1970s, remote sensing technology has been used in large area crop yield prediction. Precision agriculture has become an important modern agricultural production mode to improve the utilization rate of agricultural resources, reduce the cost of agricultural production and improve the agricultural ecological environment [1]. As an important grain crop in China, the safe production of wheat is an important guarantee for the sustainable development of our country. Quickly and timely understanding of the distribution of wheat, generalization, growth status and yield dynamics, to facilitate government decision-makers and producers to take various regulatory measures to achieve the purpose of increasing production, income and efficiency [2]. Estimation of crop yield has gone through the sampling survey, statistical analysis, statistical model and meteorological satellite remote sensing data estimation stage [3]. China’s yield estimation by remote sensing are summarized as follows: Spectral information - index - growth information - yield, spectrum - moisture and nitrogen - yield, spectral information - index - growth information - growth model – yield [4]. Many experts and scholars have studied the NDVI index based on the spectral reflectance of different crops, He et al. [5] research shows that the growth trend in the 3–5 month NDVI index, but the amplification of index was smaller, the production precision is highest in the May NDVI index; others believe that the current work of remote sensing is mainly by using single fertility period or some fertility period of spectral vegetation index and yield for regression analysis to estimate yield, such as NDVI index in China [6,7,8,9].

In this study, ASD spectroscopy was used to obtain reflectance data of crops during the whole growth period, calculate NDVI index and yield inversion. The reliability of the instrument is verified indirectly. At the same time, it lays a foundation for large-scale production forecasting.

2 Materials and Methods

2.1 Overview of the Study Area

The experimental area is located in Changping District, Beijing National Precision Agriculture Research and Demonstration Base, located at latitude 40° 11′, longitude 116° 27′, altitude 36 m, the base of fertile land, flat terrain, soil type for the soil base, covers an area of about 2500 acres, climate type is a typical warm temperate semi-humid continental monsoon climate, the winter wheat crop is the base in late September early October planted (Fig. 1).

Fig. 1.
figure 1

National precision agriculture research and demonstration base

Full size image

2.2 Research Design

There are two varieties of wheat used in this experiment, A and B, respectively, cross planting. There were 16 plots in the test field, and each plot was divided into three sub-districts for three replicates, each with an area of 48 m2 (6 m Ã— 8 m) and a line spacing of 15 cm. The experimental field also conducted nitrogen treatment (N1, N2, N3, N4) and water treatment (excessive, normal and rainfed). This experiment is to obtain the reflectance of crops, and the function of dealing with the corresponding NDVI index, the final function model using the measured output and the NDVI index to establish the estimation. Therefore, we should try to reduce the impact of irrelevant factors, so that the experiment from the normal water area selected experimental plots, such as the block 3, 6, 7, 10, 11, 15 six plots. Among them, using the data of the three plots 3, 6, 10 and 11 to establish the function model of different band length and reflectance. The experimental data of 12 cells were selected from 18 experimental communities to establish the functional model of the corresponding NDVI index in the measured area. The experimental data of the other six communities were used to validate the model (Fig. 2).

Fig. 2.
figure 2

Sketch map of experimental plots

Full size image

2.3 Instrument Introduction

The following is a portable instrument for measuring crop growth parameters (Before using the instrument, the integrating sphere is used to perform spectral calibration on the instrument to ensure that the instrument can be used normally, and the basic parameters of the instrument are as follows). The instrument has two channels. By searching the relevant research status at home and abroad, the final choice is 650 nm and 810 nm as the channel wavelength of the instrument. The experiment date is from March 28 to May 17; the test time is 6:30 to 18:30, every half an hour to conduct a test. Before the experiment, the equipment should be set up and checked first, the instrument’s sensor is facing the sun and level, and the sensor probe is 0.5 m from the canopy of the rice. Each collection area needs to collect 5 points of data, take the average as the spectral reflection of the sample point (Fig. 3, Table 1).

Fig. 3.
figure 3

Instrument for measuring crop growth parameters

Full size image
Table 1. Hand-held portable crop growth parameters speedometer parameters
Full size table

2.4 Determination Method

2.4.1 Spectral Data

The winter wheat canopy spectra were measured using a FieldSpec Pro FR2500 back-hanging field hyperspectral radiometer manufactured by American Analytical Spectral Device (ASD), the band value is 350–2500 nm. The canopy spectra were measured from 6 a.m. to 16 p.m., with clear weather, no wind or very small wind speed. When it was measured, the sensor probe is vertical downward, the vertical height is about 1 m from the top of the canopy, and 10 sampling spectra are recorded at each observation point, and the average value is taken as the reflection value of the observation point. In the process of measurement, the standard white board correction should be carried out before and after the observation of each group of targets (the standard white board reflectivity is 1, so the target spectrum is dimensionless relative reflectivity).

2.4.2 Normalized Difference Vegetation Index

NDVI: Refers to the increase in the near infrared range of green leaves scattering and red band range of chlorophyll absorption differences, generally used to detect vegetation growth status, vegetation coverage and eliminate partial radiation errors. In terms of the normalized difference vegetation index, the sensitivity will decrease when the LAI value is very high (LAI is the leaf area index), that is, when the vegetation is luxuriant. Calculation formula:

$$ {\text{NDVI}} = \frac{{\rho_{NIR} - \rho_{RED} }}{{\rho_{NIR} + \rho_{RED} }} = \frac{NIR - R}{NIR + R} $$
(1)

The range of NDVI value is −1 ~ 1, a negative value indicates that the ground cover for cloud, water, snow, high reflection of visible light; 0 said that the rocks or bare soil, NIR and R are approximately equal; positive, said there is vegetation cover and increases with increasing coverage; 0.2 ~ 0.8 is the general scope of green vegetation area [10].

2.4.3 Yield

This experiment was carried out on the basis of the experimental data(yield, asd)in 2014. The maximum measured output is 549.42 (kg/mu), the minimum is 223.03 (kg/mu), and the average value is 378.05 (kg/mu).Then, the NDVI index and the measured output are used to establish the model, and the NDVI index is tested to calculate the predicted output, and the corresponding RMSE and NRMSE are calculated.

2.5 Model Accuracy Evaluation

2.5.1 Validation Techniques

In this paper, we use 2/3 of the experimental data to establish the model, another third of the experimental data were validation for the model just created. In this paper, a total of 6 plots and 18 plots were selected for the experiment. The experimental data of 12 plots were randomly selected to establish the function model. The experimental data of 6 other plots were used to validate the model.

2.5.2 Coefficient of Determination

The coefficient of determination (coefficient of - determination, R) was used as the evaluation of the relationship between remote sensing estimation of biomass and biomass measurement standard. Calculation formula:

$$ {\text{R}}^{2} = \frac{{(\sum {xy - \frac{{\sum {x\sum y } }}{n}} )^{2} }}{{(\sum {x^{2} } - \frac{{(\sum x )^{2} }}{n})(\sum {y^{2} - \frac{{(\sum y )^{2} }}{n}} )}} $$
(2)

x, y stands for sample value, n stands for sample number; n stands for the number of samples. The higher the R2, the better the correlation between the variables involved in the evaluation, the higher the reference value; The smaller the R2, the lower the correlation, the smaller the reference value [11].

2.5.3 Root Mean Square Error

Root mean square error was used as a standard for evaluating the estimated biomass and measured biomass errors.

Calculation formula:

$$ RMSE = \sqrt {\frac{{\sum\limits_{i = 1}^{n} {(y_{i} - \mathop y\limits^{ \wedge }_{i} )^{2} } }}{n}} $$
(3)
$$ {\text{NRMSE}} = {\text{RMSE}}/{\text{X}} $$
(4)

\( y_{i} \) stands for measured value, \( \mathop y\limits^{ \wedge }_{i} \) stands for analog value, n stands for the number of samples, X stands for mean of the measured value. RMSE is very sensitive to larger errors. The smaller the RMSE, the higher the accuracy of the estimation results. RMSE has increased the penalty for single data (squared penalty), which makes the evaluation even more rigorous [12].

2.6 Statistical Analysis

In this study, determination coefficient (R2) and root-mean-square error were regarded as indicators to interpret and quantify the relationship between NDVI and vegetation index. Generally speaking, the higher R2, the smaller RMSE, and the better ability of the estimated model between the NDVI and vegetation index, otherwise, the estimated ability is poor.

3 Results and Analysis

3.1 The Relationship Between the Length and Reflectivity of Different Plots of Winter Wheat in Different Bands

Refer to the literature at home and abroad indicated that the relationship between the length and reflectivity of Winter Wheat in different bands is linear regression model. From the measured data, the interception wavelength is between 640–660 nm and 800–820 nm. The function models of different band length and reflectivity of each block are obtained as follows (Tables 2, 3):

Table 2. The relationship between band and reflectivity in 640–660 nm
Full size table
Table 3. The relationship between band and reflectivity in 800–810 nm
Full size table

3.2 A Functional Model for Measuring NDVI Index of Yield and Whole Growth Period Was Established

This study is based on the average reflectance of different bands of crops obtained in each plot obtained in 2014. The corresponding NDVI index is obtained by formula (1), following Table 4.

Table 4. NDVI index of each experimental plot
Full size table

The measured output and NDVI data of 12 plots were obtained and analyzed statistically, obtained the function model of measuring yield and NDVI index, as follows. The relationship between measured yield and NDVI index is y = 45.504e2.6504x, The R2 is 0.78 by using formula (1–2), the RMSE is 40.795(kg/mu) by using formula (1–3), the NRMSE is 10.79%. The above data show that the model has certain reliability (Fig. 4).

Fig. 4.
figure 4

NDVI index and measured yield model for full growth period

Full size image

3.3 Verification of Function Model

Using the NDVI index of the other six communities that did not participate in the model building are respectively used in the function model obtained in the previous section. Compared with measured output, calculated the RMSE is 49.297 (kg/mu) by using formula (1–3), the NRMSE is 13.04%. The above data show that the model has certain reliability.

4 Conclusion

  1. 1.

    When the band length is 640–660 nm, the fitting model of band length and crop reflectivity is higher, and the influence of nitrogen treatment is smaller. When the band length is 800–810 nm, the fitting degree of normal nitrogen and nitrogen treatment is higher than that of semi nitrogen and nitrogen free treatment.

  2. 2.

    The estimation model established by using yield and NDVI index measured the decision coefficient R value is 0.78, the root mean square error of the value of RMSE is 40.795 (kg/mu), mean square error NRMSE value is 10.79%.

  3. 3.

    When model validation is performed, the root mean square error RMSE is 49.297 (kg/mu), and the standard root mean square error is 13.04%.