China’s Terrestrial UNVI Multidimensional Dataset (2018–2021)

Zhao, H. Q.¹  Liu, X. Q.¹  Zhang, L. F. ^2*  Chen, J. H. ²  Fu, H. C.¹  Ma, K.¹

1. College of Geoscience and Surveying Engineering, China University of Mining and Technology (Beijing), Beijing 100083, China;

2. Aerospace Information Research Institute, Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences, Beijing 100101, China

Abstract: Vegetation index data products are widely used in the inversion of physical and chemical parameters of vegetation, land cover classification and change studies. Traditional vegetation index products, such as AVHRR-NDVI, are sensitive to soil background changes and have the problem of high value saturation, which easily causes a decrease in sensitivity to vegetation detection. China’s terrestrial UNVI multidimensional dataset (2018–2021) was developed based on the MODIS surface reflectance product MOD09GA using the Universal Pattern Decomposition algorithm UPDM (Universal Pattern Decomposition Method), which takes 16 days as the synthesis cycle. The practices show that the UNVI has more advantages in reflecting the change in vegetation cover and quantitative inversion of vegetation physical and chemical parameters compared with traditional NDVI products. The synthesis algorithm takes the number of days without cloud data in the synthesis period N as the judgment condition and uses the angle normalized synthesis method, the maximum synthesis method in a limited perspective, the direct calculation method and the maximum synthesis method MVC as the main synthesis algorithm to calculate the UNVI. Thus, the 2018–2021 China terrestrial UNVI products with a time resolution of 16 d and a spatial resolution of approximately 463 m were synthesized. The dataset includes the UNVI products of China terrestrial vegetation indices in 23 time periods with 16 d intervals from 2018 to 2021. UNVI vegetation index products can provide more comprehensive and convenient long-time series vegetation index data products for scholars participating in research on global change and human activities.

Keywords: UNVI; MODIS; vegetation index; BRDF-C; long time series

DOI: https://doi.org/10.3974/geodp.2022.04.16

CSTR: https://cstr.escience.org.cn/CSTR:20146.14.2022.04.16

Dataset Availability Statement:

The dataset supporting this paper was published and is accessible through the Digital Journal of Global Change Data Repository at: https://doi.org/10.3974/geodb.2022.12.01.V1 or https://cstr.escience.org.cn/CSTR:20146.11.2022.12.01.V1.

1 Introduction

Vegetation occupies a large proportion of the land surface and is the most intuitive information in remote sensing images. Vegetation is an important part of the geographical environment. It adapts to landform, climate, hydrology, soil and biological conditions and is controlled by many factors, so it has great dependence on and sensitivity to the geographical environment^[1]. Therefore, vegetation coverage and exuberance can reflect the changes and rules of geographical conditions in a certain area to a certain extent.

Satellite remote sensing technology has gradually become a powerful means to acquire environmental information because of its advantages, such as a large detection range, fast data acquisition speed, short period, less restriction by ground and large amount of information. According to the spectral reflection and absorption characteristics of vegetation, a series of indicators sensitive to changes in surface vegetation and able to effectively reflect vegetation cover and biomass are obtained by using visible and near-infrared band data of multispectral remote sensing through analysis and operation and band combination, which are called the vegetation index^[2]. It can be used to reflect vegetation coverage, vegetation vitality, biomass and other vegetation growth conditions qualitatively and quantitatively^[3].

The Universal Normalized Vegetation Index (UNVI) is a new type of ground cover analysis method proposed by Fujiwara et al.^[4] in 1996: the Pattern Decomposition Method (PDM), which is suitable for analyzing multispectral satellite data. Various Vegetation Indices (VI) derived from the PDM are based on the idea that the reflectance of a given surface is a linear overlay of several standard land cover types so that multispectral data can be used for the quantitative inversion of vegetation indices. However, the VI algorithm derived from the PDM needs to calculate the complex coefficient matrix, which is very inconvenient for users to use. Therefore, to popularize these PDM-based VIs in a more user-friendly way, Zhang et al.^[5] emphasized the concept of universality of VIs. As a result, the Universal Pattern Decomposition algorithm (UPDM) is generated to establish a generalized transformation matrix based on different sensors. The algorithm divides the ground into four standard coverage types: soil, water, vegetation, and vegetation between yellow and green. At the same time, Zhang conducted accurate ground tests for different sensors^[6] and obtained multiple conversion matrices including MODIS, ETM+, GLI and other sensors for the synthesis of UNVI. Therefore, PDM has universality, and VI can be calculated only with a ready-made transformation matrix. According to the UPDM algorithm, Zhang proposed the universal normalized vegetation index UNVI with many advantages^[7].

As a vegetation index independent of the characteristics of sensors, the general normalized vegetation index UNVI can better meet the needs of long-term change research based on multisensor data. This dataset uses the multiday observation synthesis method to complete the production of a 1:1,000,000 UNVI multidimensional dataset for land areas of China from 2018 to 2021.

2 Metadata of the Dataset

The metadata of China’s terrestrial UNVI multidimensional dataset (2018–2021)^[8] are summarized in Table 1. It includes the dataset full name, short name, authors, year of the dataset, temporal resolution, spatial resolution, data format, data size, data files, data publisher, and data sharing policy, etc.

Table 1  Metadata summary of the China’s terrestrial UNVI multidimensional dataset (2018-2021)

3 Methods

Due to the influence of the atmosphere, clouds and other factors, it is difficult to obtain high-quality vegetation index products from the observation data of a single day, so the industry usually adopts the multiday observation synthesis method to obtain higher-quality vegetation index products^[10,11]. Vegetation index synthesis refers to selecting the vegetation index that can best represent the actual vegetation conditions on the surface by using the appropriate vegetation index synthesis algorithm within the appropriate synthesis period and then synthesizing a vegetation index raster image with minimal influence on atmospheric conditions, cloud conditions, observational geometry, geometric accuracy, etc.^[12]. Selecting a suitable vegetation index synthesis algorithm according to the above factors has become the key to producing better quality vegetation index products.

For NASA’s official MODIS Vegetation Index product, the VI algorithm filters the data based on quality, cloud cover, and observational geometry. However, as MODIS is a push-sweep sensor, the size of pixels increases with increasing observation zenith angle, and the spatial resolution change caused by this influence can be up to four times^[13,14], which will cause a large BRDF characteristic error. Therefore, CV-MVC and MVC synthesis methods are used instead of BRDF synthesis methods.

Zhang et al.^[15] further upgraded the synthesis algorithm of MODIS vegetation index products by adding the discrimination of cloud cover within 16 days and designed the synthesis process of the UNVI vegetation index in a more refined way. We added not only the BRDF algorithm for anisotropic surfaces but also the Constrained View angle-Maximum Value Composite (CV-MVC), Vegetation Index Computation (VI) and Maximum Value Composite (MVC). They are used as standby algorithms when the angle normalization synthesis method is not applicable to cope with the harsh condition that the BRDF algorithm requires at least 5 d of high cleanliness pixels. The purpose of selecting the appropriate synthesis method is to reduce the influence of adverse atmospheric conditions, cloudy conditions and adverse geometric conditions and to ensure the spatiotemporal consistency of the synthesized vegetation index products.

3.1 Principles of the UNVI Algorithm

The general normalized vegetation index UNVI is a full-spectrum vegetation index proposed by Zhang based on the general pattern decomposition algorithm UPDM. It assumes that the spectrum of any ground feature on the surface is a linear combination of the spectra of several standard ground features, namely, soil, water, vegetation and yellow leaves between green leaves and dead leaves^[7]. The expression formula of UNVI is as follows:

                                                                               (1)

where i is the band number, R(i) is the reflectance of ground objects in band i, and Pw, Pv and Ps represent the normalized reflectance of three standard ground objects (water, vegetation and soil, respectively) in the spectral range of band i. Cw, Cv and Cs represent the UPDM coefficients of standard water bodies, vegetation, soil and yellow leaves, respectively^[7].

For some studies, only three components of UPDM are sufficient, and approximately 95.5% of the spectral reflectance information of land cover can be converted into three decomposition coefficients and decomposed into three standard models, with an error of approximately 4.2% per degree of freedom^[16]. However, other studies may require more detailed analyses of vegetation change. Therefore, Zhang^[17] added a yellow leaf coefficient as a supplementary spectral model, namely, the four-parameter UPDM. The four standard ground features include soil, water bodies, vegetation and yellow leaves. The expression formula of the improved UNVI is as follows:

                                                             (2)

where P4 represents the normalized reflectance of the newly added standard ground object yellow leaves in the spectral range of band i and C4 represents the UPDM coefficient of standard yellow leaves.

To simplify the use of UPDM, Zhang et al.^[18] deduced a simple coefficient matrix M for different satellite sensors and optimized the calculation of the UNVI coefficient matrix. The simplified matrix calculation formula is as follows:

                                                                                                                                 (3)

where, R=[R₁, R₂,…, R_n]^T (T represents matrix transpose) is the column vector of the reflectivity observed values of the original remote sensing data, M=[M_w, M_v, M_s, M₄]^T is a 4×n matrix, n represents the number of bands, and the subscript of M has the same meaning as in Equation (2). C=[C_w, C_v, C_s, C₄]^T is the column vector of UPDM coefficients. For different sensors, the band selected to calculate UNVI and the value of coefficient matrix M are different. The M matrix corresponding to the MODIS sensor used in this design is:

                 (4)

According to the above four coefficients, UNVI can be expressed as:

                                                                                                       (5)

where, the UPDM coefficient can be calculated by Equation (3), a is the standard soil model coefficient, and its value is a=0.1. This value is obtained by setting the UNVI value of withered yellow leaves as 0 and that of lush vegetation as 1. The denominator  represents the sum of the total reflectance. Since the reflectance information of soil and dead vegetation is mixed in pixels, the approximate reflectance information of vegetation is obtained by subtracting the UPDM coefficients of corrected soil and withered or dead vegetation from the UPDM coefficients of vegetation in the molecule. In areas with higher vegetation density, the molecular values are higher because the soil and yellow leaves are covered by healthy vegetation. Meanwhile, in areas with sparse vegetation, higher  and  values will be generated due to the scattering of light by soil and yellow leaves, leading to a decrease in molecular values. Therefore, UNVI is more sensitive to a wider range of vegetation dynamics than traditional VIs^[18,19].

3.2 Principles of Composition Algorithm Selection

As the selection of the synthesis algorithm is determined by many factors, such as the specific application purpose, sensor characteristics, and atmospheric parameters of the corresponding region, after the completion of screening MOD09GA data, the appropriate synthesis algorithm is selected according to the cloudless data days N of each group of data.

In the production process of this dataset, although the angle normalization method BRDF-C can effectively remove the influence of changing observation geometry, which represents the most advanced vegetation index synthesis method at present, it is easily affected by the time resolution of the dataset, the limitation of the number of effective observations and the interference of thin clouds. Therefore, only when the number of days N ≥5 without cloud data was the angle normalization synthesis method BRDF-C selected to fit all the cloud-free observation data within the synthesis period to the equivalent reflectance value of the substar under the irradiation condition band by band and pixel by pixel, and then the synthesized value was obtained by the UNVI calculation formula and recorded in UNVI_DATA. At the same time, Q = N × 10 + 1 is calculated and recorded in QC_BAND (quality control band). In particular, when the synthesis value of UNVI is not within the interval [0.3–UNVI_MVC, UNVI_MVC +0.05] (UNVI_MVC represents the UNVI value synthesized by the maximum synthesis method) or the reflectance of the fitting substellar point is negative, the angle normalization synthesis method is abandoned. CV-MVC was used to synthesize UNVI instead.

At the same time, when there is a small amount of cloud-free data, the BRDF-C method will obtain the wrong reflectivity fitting value and further obtain the wrong UNVI value. Therefore, when the number of days of cloud-free data is 1 < N < 5, the UNVI value of all high-quality cloud-free data within the synthesis cycle will be calculated first. Then, CV-MVC uses the maximum value synthesis method in a limited view angle to select the maximum value from these several high-quality data as the final UNVI synthesis value of the pixel and record it in UNVI_DATA. Meanwhile, Q = N×10 + 2 is calculated and recorded in QC_BAND. When the number of days without cloud data N =1, the direct calculation of UNVI with the high-quality original data of this day is more accurate than the angle normalization method. Therefore, the UNVI composite value was obtained by the direct calculation method and recorded in UNVI_DATA. Meanwhile, Q = N×10 + 3 was calculated and recorded in QC_BAND. When the number of cloud-free data days N =0, there are no high-quality cloud-free data in the synthesis cycle. Compared with the other three algorithms, the MVC of the maximum synthesis method can remove atmospheric influences, including residual clouds, and its composite value is closer to the observed data of the substar. Therefore, the maximum value synthesis method is used to calculate the UNVI value of all 16 d data within this synthesis period, and the maximum value is selected as the UNVI value of this synthesis period and recorded in UNVI_DATA. Meanwhile, Q = 4 is calculated and recorded in QC_BAND. In this process, to avoid synthesis failure caused by low reflectance values or no effective reflectance data, data containing –28,672 were removed, and the UNVI of all invalid values was set to –999 to represent invalid data. Then, MVC synthesis was carried out.

3.3 Data Processing

In 2018, for example, this dataset uses MODIS/terra surface reflectance daily L2G Global 1 km and 500 m SIN Grid V006 products with time ranges from January 1, 2018, to December 31, 2018, as raw data[1]. UNVI products are produced using 7 reflectance band data from 01 to 07, sensor zenith angle data, sensor azimuth, solar azimuth and quality control band data of the data product.

Read classified and stored MOD09GA data. The 2018-01-01 data were set as the initial synthetic data, and the 365 days of MOD09GA data in 2018 were divided into groups every16 days in chronological order. The data from days 337 to 349 in 2018 were less than

16days (13 days), and the data from days 337 to 349 were selected as one group. Then, the MOD09GA data are read in sequence in groups. Check whether there are invalid values and negative reflectance values in the reflectance band of each set of input MOD09GA data. If there are invalid values, the data of this day will be eliminated. If the pixel reflectance is negative (in the range [–100, 0]), change this value to 1.

Determine data cloud cover. Read the quality control band data (state_1 km_1.img) in a block, convert the data into binary, and take the first three digits as the quality parameter of the pixel. 000 or 011 indicates high-quality data without clouds. In other cases, the data have a cloud and are of low quality. Finally, the number of days each pixel contains cloud-free data within this synthesis cycle is counted as N.

According to the advantages and disadvantages of the exponential synthesis model and method of BRDF, CV-MVC, VI and MVC as well as their applicable conditions, and according to the days N of cloud-free data within each synthesis cycle, the UNVI synthesis method of each pixel within the synthesis cycle is determined. The specific synthesis process is as follows:

Figure 1  UNVI product synthesis flow chart

After obtaining UNVI data through the above four synthesis algorithms, UNVI products of 23 synthesis cycles in China’s land area in 2018 were obtained through mosaic, coordinate transformation and data clipping.

Finally, MARS software was used to store the UNVI synthesis data in multidimensional data format, and the UNVI products of 23 synthesis cycles were stored in multidimensional data format. Finally, the UNVI dataset product of China’s land vegetation index in 2018 was obtained. The dataset was an MDD file stored in TSB format. Its file contains the 2018 UNVI composite data and its header file.

4 Data Results and Validation

4.1 Data Composition

The UNVI dataset of China’s land vegetation index is stored in the MDD multidimensional data format, including 23 groups of files, and each group of files contains a UNVI product with a synthesis period of 16 days. The composition of the dataset is shown in Table 2.

Table 2  Composition of the UNVI multidimensional dataset (2018-2021) (taking 2018 as an example)

4.2 Data Products

The UNVI product covers all the land areas of China, including the UNVI data of 23 time phases from January 1 to December 31, 2018 to 2021 in a 16-day synthesis cycle. Each phase has a band for storing UNVI data, and the effective range is [–2, 2]. The product of this dataset is the coordinate system GCS_WGS_1984, the tape number is 49 N, and the spatial resolution is 463 m.

4.3 Data Validation

The spatial and temporal changes in UNVI datasets in China during 2018–2021 were analyzed.

(1) Time change analysis

Taking the UNVI data in 2019 as an example, the time variation in the average UNVI in China was analyzed. To eliminate the influence of outliers on the average value of UNVI, only effective pixels with values of [–2, 2] in the region are selected for average calculation. However, this operation still cannot exclude the cloud cover area, so the average calculated value of UNVI will be smaller than the real value. The greater the cloud cover area is, the greater the degree of underestimation, such as the 001–016 days of 2019.

As shown in Figure 2, the average value of UNVI in one year presents the characteristics of a maximum value in summer and an obvious peak value in summer and a minimum value and an insignificant peak value in winter, which is consistent with the change trend of vegetation in one year.

(2) Spatial change analysis

Taking UNVI data in 2019 as an example, UNVI products on days 001–016 (Figure 3a, 4a), 097–112 (Figure 3b, 4b), 177–192 (Figure 3c, 4c) and 273–288 (Figure 3d, 4d) of each year are selected for spatial change analysis.

Figure 3 and Figure 4 show that the UNVI in China varies in different seasons. The spatial distribution of UNVI was as follows: (1) in winter (days 001–016), the UNVI in northern China was generally smaller, while that in southern China was larger; (2) in spring (days 097–112), the UNVI in northern and southern regions increased gradually, but the UNVI in northern Hebei and further north had no significant change compared with that in winter; (3) in summer (days 177–192), both the northern and southern regions reached the period of relatively large UNVI; (4) in autumn (days 273–288), the spatial distribution of UNVI was similar to that in summer, and the value decreased slightly. However, southern regions are more susceptible to cloud cover than northern regions, as shown in Figure 4a. Generally, its spatial distribution features are large in the south and small in the north, which agrees with the growth characteristics of vegetation in China.

5 Discussion and Conclusion

This dataset is based on the MOD09GA product and traditional vegetation index synthesis algorithm, and the UNVI vegetation index synthesis algorithm designed by Zhang’s team is used to produce China land vegetation index UNVI products. In the synthesis process of the vegetation index, due to the influence of atmospheric conditions, cloud interference and other factors, it is difficult to produce high-quality vegetation index products with single-day observation data. Therefore, this design adopts a multiday observation synthesis method to produce higher-quality UNVI products.

In this dataset, MARS software can be used to analyze and study the phenological changes of vegetation, quantitative inversion products of physical and chemical parameters of vegetation can be generated, and the phenological changes of the whole year can be studied^[15]. Compared with traditional vegetation indices (such as NDVI, EVI, etc.), this dataset has more advantages in reflecting vegetation cover changes and quantitative inversion of physical and chemical parameters^[20,21]. It can be used for qualitative and quantitative inversion of physical and chemical parameters of vegetation or classification of vegetation cover types, as well as study of annual phenological changes^[22–24]. In addition, some scholars^[20] have found that UNVI has considerable potential in drought monitoring. The above results show that UNVI is superior to or equal to other mainstream vegetation indices in retrieving the physicochemical parameters of vegetation, vegetation cover change and vegetation classification and has better application prospects. Therefore, this dataset can provide more comprehensive and convenient long time series data products of the vegetation index for scholars engaged in global change research.

In summary, compared with traditional vegetation index products, this dataset has more obvious advantages and has more universal applicability to different sensors. Although this dataset has many advantages, it can be further deepened and expanded in terms of the algorithm. For example, the UNVI synthesis algorithm based on this paper may result in a discrepancy between the UNVI synthesis result and the real value because the sensor data used in the cloud detection result are incorrect or when the number of days N of high-quality cloud-free data is less. Although the above two cases rarely occur, the results show that the algorithm still has room for improvement and progress.

Author Contributions

Zhang, L. F. and Zhao, H. Q. designed the algorithms of dataset. Zhao, H. Q., Liu, X. Q., Chen, J. H., Fu, H. C. and Ma, K. contributed to the data processing and analysis. Liu, X. Q., and Zhao, H. Q. wrote the data paper.

Conflicts of Interest

The authors declare no conflicts of interest.

References

[1]      Wang, Q., Yang, Y. P., Huang, J. Z., et al. Environmental Remote Sensing [M]. Beijing: Science Press, 2004: 366.

[2]      Han, A. H. Study on monitoring method of forest biomass and carton storage based on remote sensing [D]. Beijing: Beijing Forestry University, 2009.

[3]      Tian, Q. J., Min, X. J. Advances in study on vegetation indices [J]. Advances in Earth Sciences, 1998(4): 10-16.

[4]      Noboru, F., Akiko, O., Motomasa, D. Pattern decomposition method for hyper-multispectral satellite data analysis [P]. SPIE Asia-Pacific Remote Sensing, 2001.

[5]      Zhang, L. F., Mitsushita, Y., Furumi, S., et al. Universality of modified pattern decomposition method for satellite sensors [R]. Asia GIS Conference, Wuhan, 2003.

[6]      Zhang, L. F. The universal pattern decomposition method and the vegetation index based on the UPDM [D]. Wuhan: Wuhan University, 2005.

[7]      Zhang, L. F., Zhang, L. P., Muramatsu, K., et al. Universal pattern decomposition method based on hyper spectral satellite remote sensing data [J]. Information Science of Wuhan University, 2005(3): 264-268.

[8]      Zhao, H. Q., Liu, X. Q., Zhang, L. F., et al. China’s terrestrial UNVI multidimensional dataset (2018-2021) [J/DB/OL]. Digital Journal of Global Change Data Repository, 2022. https://doi.org/10.3974/geodb.2022.

12.01.V1. https://cstr.escience.org.cn/CSTR:20146.11.2022.12.01.V1.

[9]      GCdataPR Editorial Office. GCdataPR data sharing policy [OL]. https://doi.org/10.3974/dp.policy.2014.05 (Updated 2017).

[10]   De Wasseige, C., Vancutsem, C., Defourny, P. Sensitivity analysis of compositing strategies: modelling and experimental investigations. Vegetation 2000 [C]. Lake Maggiore, Italy. 2000: 267-274.

[11]   Holben, B. N. Characterization of maximum value composites from temporal AVHRR data [J]. International Journal of Remote Sensing, 1986, 7(11): 1417-1434.

[12]   Lovell, J. L., Graetz, R. D., King, E. A. Compositing AVHRR data for the Australian continent: seeking best practice [J]. Canadian Journal of Remote Sensing, 2003, 29(6): 770-782.

[13]   Duchenmin, B., Maisongrande, P. Normalisation of directional effects in 10-day global syntheses derived from VEGETATION/SPOT: I. Investigation of concepts based on simulation [J]. Remote Sensing of Environment, 2002, 81(1): 90-100.

[14]   Long, X., Li, J., Liu, Q. H. Review on VI compositing algorithm [J]. Remote Sensing Technology and Application, 2013, 28(6): 969-977.

[15]   Zhang, L. F., Zhong, T., Liu, H. L., et al. UNVI multidimensional dataset of 2017 China’s terrestrial at 1:1 000000 scale [J]. Journal of Remote Sensing, 24(11): 1293-1298.

[16]   Daigo, M., Ono, A., Fujiwara, N., et al. Pattern decomposition method for hyper-multi-spectral data analysis [J]. International Journal of Remote Sensing, 2004, 25(6): 1153-1166.

[17]   Zhang, L. F., Furumi, S., Muramatsu, K., et al. Sensor-independent analysis method for hyper-multispectral data based on the pattern decomposition method [J]. International Journal of Remote Sensing, 2006, 27(21): 4899-4910.

[18]   Zhang, L. F., Qiao, N., Baig, M. H. A., et al. Monitoring vegetation dynamics using the universal normalized vegetation index (UNVI): an optimized vegetation index-VIUPD [J]. Remote Sensing Letters, 2019, 10(7): 629-638.

[19]   Zhang, L. F., Furumi, S., Muramatsu, K., et al. A new vegetation index based on the universal pattern decomposition method [J]. International Journal of Remote Sensing, 2007, 28(1): 107-124.

[20]   Jiao, W. Z., Zhang, L. F., Chang, Q., et al. Evaluating an enhanced vegetation condition index (VCI) based on VIUPD for drought monitoring in the continental United States [J]. Remote Sensing, 2016, 8(3): 224.

[21]   Du, H. S., Jiang, H. L., Zhang, L. F., et al. Evaluation of spectral scale effects in estimation of vegetation leaf area index using spectral indices methods [J]. Chinese Geographical Science, 2016, 26(6): 731-744.

[22]   Wang, S. H., Yang, D., Li, Z., et al. A global sensitivity analysis of commonly used satellite-derived vegetation indices for homogeneous canopies based on model simulation and random forest learning [J]. Remote Sensing, 2019, 11(21): 2547.

[23]   Liu, H. L., Zhang, F. Z., Zhang, L. F., et al. UNVI-based time series for vegetation discrimination using separability analysis and random forest classification [J]. Remote Sensing, 2020, 12(3): 529.

[24]   Jiang, H. L., Zhang, L. F., Yang, H., et al. Research on spectral scale effect in the estimation of vegetation leaf chlorophyll content [J]. Spectroscopy and Spectral Analysis, 2016, 36(1): 169-176.