Improving leaf area index estimation accuracy of wheat by involving leaf chlorophyll content information
Introduction
Leaf area index (LAI) is defined as the ratio of total one-sided leaf area to the ground area (Chen and Black, 1992), which is a crucial indicator for characterizing the land surface vegetation states. LAI affects many biological and physical vegetation processes, such as photosynthesis and respiration (Chen and Cihlar, 1996), and has been widely used in the hydrological, crop yield and ecological models (Zhang et al., 2020, Xia et al., 2021) due to its ability to describe mass (e.g., water and carbon) and energy (e.g., radiation and heat) exchange between biosphere and atmosphere (Yan et al., 2019). Therefore, accurate LAI estimation is of great importance for a variety of earth systems, agriculture and ecological studies.
Remote sensing provides a faster and cost-effective method for LAI estimation over large areas. The traditional LAI estimation method using remote sensing data builds empirical statistical models based on sensitive band reflectances, vegetation indices (VIs), or spectral transform values (Chen et al., 2020). Empirical methods are easily affected by the vegetation types, experimental locations, and sampling times. Therefore, empirical models show limited capabilities when applied to large-scale and multitype vegetation areas. In contrast, the physical-based methods have no such limitation because they consider various vegetation biophysical and biochemical parameters, and soil reflectances. However, the radiation transfer model (RTM) used in physical-based method requires many input parameters for accurate simulations. Therefore, the inversion of RTM is very difficult, and hybrid LAI estimation models have been widely used for LAI estimation (Sinha et al., 2020). By combining a physical model and machine learning algorithm, hybrid methods have advantages of both empirical and physical inversion algorithms.
The VIs are essential inputs for both statistical and hybrid models. For example, the normalized difference vegetation index (NDVI) is widely used for LAI estimation (Hasegawa et al., 2010, Sinha et al., 2020). However, the Red-NIR band combinations easily suffer from saturation at moderate-to-dense canopies, which usually cause underestimation for moderate-to-high LAI values (LAI > 3) (Delegido et al., 2013). The reason is that NIR band reflectances increase rapidly due to scattering of light in intercellular volumes of leaf mesophylls (Dorigo et al., 2007), while red band reflectances exhibit fewer variations since they become saturated with high chlorophyll content. To improve the sensitivity to moderate-to-high LAI values, various new VIs have been developed in the past few decades. One solution is to attenuate the reflectance contrast between the red and NIR bands, such as modified simple ratio index (MSR) (Chen 1996). Although this method is effective for multispectral bands (e.g. Landsat-8) (Dong et al., 2020), it still shows a limited capability for reducing the underestimation phenomenon. During the past 20 years, the development of hyperspectral imaging sensors has provided plenty of narrow bands from visible to NIR wavelength (Delegido et al., 2013). The red-edge (RE) wavelength is discovered to be influenced by multiple scattering between leaf layers, and is strongly affected by LAI. Moreover, because the RE bands are sensitive to vegetation conditions, several studies have used them to formulate VIs and achieved more accurate LAI estimation in the moderate-to-high dense canopy regions (Brown et al., 2019). Therefore, a number of RE-based VIs, such as the red edge normalized difference vegetation (NDVIRE) and red edge chlorophyll index (CIRE), have been widely used for LAI estimation (Sibanda et al., 2019).
Although, the RE-based VIs have contributed to LAI estimation accuracy improvement, some concerns still need to be solved (George et al., 2018). Although the RE-based VIs developed using hyperspectral reflectances possess great potential, those narrow RE band-based VIs can hardly applied to large region because of the small amount of hyperspectral data. Fortunately, several spaceborne sensors that involve RE bands have been designed and launched, such as Rapid-Eye, WorldView-2, Sentinel-2 (S2), and Chinese GF-6. Among those sensors, the S2 Multispectral Instrument (MSI) provides the most detailed RE bands. Thus, it is of great significance to discuss the potential of improving LAI estimation accuracy by using S2 data. Other than VIs, another important but often ignored parameter is leaf chlorophyll content (LCC). It also has a great influence on RE band reflectance (Xie et al., 2018). The increase of LCC not only causes strong absorption in the red spectral wavelengths, but also leads to a shift in the RE band toward longer wavelengths (Herrmann et al., 2011). In recent years, some studies have attempted to reduce the effect of LCC change by excluding LCC-sensitive bands in LAI estimation (Sun et al., 2020). However, since it is impossible to fully separate the influence of these two parameters on reflectance, the effects of LCC changes are still nonnegligible for LAI estimation. Another disadvantage of the above method is that the canopy information provided by RS data is not fully used. Although LCC-sensitive bands are not recommended for LAI-sensitive VI modifications, they can be combined with LAI-sensitive VIs to increase the information on canopy status for LAI estimation. According to plant physiology studies (Dordas and Sioulas, 2008), photosynthesis which is directly influenced by LCC is one of the crucial factors for plant growth (including LAI enlargement). Those studies have indicated that LCC is also capable of characterizing the vegetation growth status and related to LAI. However, there is no obvious linear relationship between these two parameters. Fortunately, the development of machine learning algorithm provides an opportunity to solve complex nonlinear relationships between LAI and LCC. Therefore, based on the powerful nonlinear expression ability of machine learning algorithm, it is possible to adopt LCC information in characterizing the current state of LAI.
As discussed above, this study aims to test the hypothesis of improving LAI estimation accuracy of wheat by involving LCC information. To achieve this objective, two issues need to be resolved. Firstly, the basic LAI estimation model should be established using LAI-sensitive VIs. Secondly, since LCC information can be involved in various ways, the best LCC introducing form needs to be determined for LAI estimation accuracy improvement.
Section snippets
Materials and methods
First, the PROSAIL (PROSPECT and Scattering by Arbitrarily Inclined Leaves (SAIL)) model was used to simulate the S2 multispectral reflectances and the corresponding LAI and LCC values (Fig. 1). Global sensitivity analysis was applied to identify LAI-sensitive bands and LCC-sensitive bands for VIs construction. To select the best VIs for LAI and LCC estimation, several VIs were calculated by different band combinations. LAI and LCC estimation models were established using the random forest
EFAST sensitivity analysis
The Si and STi values representing the contributions of LAI and LCC to the reflectances are presented in Fig. 5. These two indices show the similar results regarding the sensitivity of each band. The reflectances of green and RE1 bands are more sensitive to the LCC, while the reflectance of RE2, RE3, NIR1, and NIR2 bands are significantly sensitive to LAI. The sum of Si values of LAI and LCC in the red band is 0.915, which indicates 91.5% of the simulated red band canopy reflectance can be
Discussion
Accurate LAI estimation based on medium-resolution multispectral data is always a challenging task in quantitative remote sensing. Many studies have confirmed that saturation occurs at moderate-to-high LAI region when using traditional combined Red-NIR VIs (Sinha et al., 2020). To solve this problem, RE bands that are more sensitive to LAI are widely applied to replace red band. However, some studies have indicated that the RE bands are also influenced by LCC, which is a factor that cannot be
Conclusions
This study proposed a novel approach to improve LAI estimation accuracy by involving LCC information in the model. The main conclusions are as follows:
- (1)
In the basic LAI estimation model, the use of RE2-based VIs and red-based VIs can achieve better LAI estimations for the entire growth period. In addition, the RE2-based VIs can achieve higher LAI estimation accuracies when LAI > 3, but they also caused overestimation when LAI ≤ 3.
- (2)
The addition of pure LCC can improve LAI estimation accuracy, but
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This research was funded by the National Key R&D Program of China (2020YFE0200700), the National Natural Science Foundation of China (42171318), the Major Special Project - the China High-Resolution Earth Observation System (30-Y30F06-9003-20/22), and the Tang Scholar Program (K. Jia is a Tang Scholar of Beijing Normal University).
References (37)
- et al.
Crop nitrogen monitoring: Recent progress and principal developments in the context of imaging spectroscopy missions
Remote Sens. Environ.
(2020) - et al.
Sensitivity of spectral reflectance to variation in live fuel moisture content at leaf and canopy level
Remote Sens. Environ.
(2004) - et al.
Comparing prediction power and stability of broadband and hyperspectral vegetation indices for estimation of green leaf area index and canopy chlorophyll density
Remote Sens. Environ.
(2001) - et al.
Retrieving leaf area index of boreal conifer forests using Landsat TM images
Remote Sens. Environ.
(1996) - et al.
Remote estimation of crop and grass chlorophyll and nitrogen content using red-edge bands on sentinel-2 and -3
Int. J. Appl. Earth Obs. Geoinf.
(2013) - et al.
Evaluation of the MERIS terrestrial chlorophyll index (MTCI)
Adv. Space Res.
(2007) - et al.
A red-edge spectral index for remote sensing estimation of green LAI over agroecosystems
Eur. J. Agron.
(2013) - et al.
Estimating crop biomass using leaf area index derived from Landsat 8 and sentinel-2 data
ISPRS J. Photogramm. Remote Sens.
(2020) - et al.
Safflower yield, chlorophyll content, photosynthesis, and water use efficiency response to nitrogen fertilization under rainfed conditions
Ind. Crops Prod.
(2008) - et al.
A review on reflective remote sensing and data assimilation techniques for enhanced agroecosystem modeling
Int. J. Appl. Earth Obs. Geoinf.
(2007)
Sentinel-2: EAS’s optical high-resolution mission for GMES operational services
Remote Sens. Environ.
Relationships between leaf chlorophyll content and spectral reflectance and algorithms for non-destructive chlorophyll assessment in higher plant leaves
J. Plant Physiol.
Improving the estimation of leaf area index by using remotely sensed NDVI with BRDF signatures
Remote Sens. Environ.
LAI assessment of wheat and potato crops by venμs and Sentinel-2 bands
Remote Sens. Environ.
Retrieving LAI, chlorophyll and nitrogen contents in sugar beet crops from multi-angular optical remote sensing: Comparison of vegetation indices and PROSAIL inversion for field phenotyping
Field Crops Research
Fractional Vegetation Cover Estimation Algorithm for Chinese GF-1 Wide Field View Data
Remote Sens. Environ.
Development of a two-band enhanced vegetation index without a blue band
Remote Sens. Environ.
Optimization of soil-adjusted vegetation indices
Remote Sens. Environ.
Cited by (13)
A chlorophyll-constrained semi-empirical model for estimating leaf area index using a red-edge vegetation index
2024, Computers and Electronics in AgricultureInsights from field phenotyping improve satellite remote sensing based in-season estimation of winter wheat growth and phenology
2023, Remote Sensing of EnvironmentImproving leaf chlorophyll content estimation through constrained PROSAIL model from airborne hyperspectral and LiDAR data
2022, International Journal of Applied Earth Observation and GeoinformationCitation Excerpt :But broad spectral bands of multispectral images have limited ability to capture spectral characteristics, which are inadequate to quantify accurate LCC (Blackburn 2007). Subsequently, the development of hyperspectral images provides more narrow spectral bands and records more rich hyperspectral reflectance information (Chen et al. 2022; Dalponte et al. 2012). Recent studies have shown that hyperspectral images generally perform better than multispectral images for LCC estimation (Lu et al. 2019; Navarro-Cerrillo et al. 2014).