Optimal band characterization in reformation of hyperspectral indices for species diversity estimation
Introduction
Tropical forests represents the most diverse ecosystems due to its repository of biological diversity (Foody, 2003; Pavan Kumar, Sajjad, Rehman and Jain, 2018; Thomas et al., 2004; Whitmore, 1990; Malhi et al., 2021a). Those support different life forms living in the tropics by providing natural resources and suitable habitat conditions (G S Kiran and Malhi, 2011; Legendre et al., 2009). Different aspects of diversity such as position, dimension and species can be utilized to examine these forest stand (Ozdemir and Karnieli, 2011; St-Louis et al., 2009; Wilson, 2015) that can aid in the implementation of proper planning and conservation strategies (Bettinger and Tang, 2015; G Sandhya Kiran and Mudaliar, 2012; Pandey et al., 2019).
Alpha diversity is utilized for determining diversity in species of a community that basically characterizes the biodiversity of a forest area (Magurran, 1988). Various Diversity indices (DIs), such as Shannon Weiner DI, Margalef DI, McIntosh DI, Brillouin DI are the common measures for quantifying the alpha diversity. The traditional means of measuring these diversity indices require extensive field surveys. These are found to be time, resource and cost intensive (Rocchini et al., 2010; Skidmore et al., 2015), as well as proficiency of resource is required (Warren et al., 2014). These surveys require the influence of assessor (Rocchini et al., 2016) and are restricted to a small scale only (Lengyel et al., 2008). Remote sensing holds a great potential and aids in circumventing these limitations of field surveys (Lopes et al., 2017; Malhi et al., 2020). Several attempts were made based on the use of multispectral remote sensing data with the high spatial resolution to assess alpha diversity using Landsat (Madonsela et al., 2017), LISS IV (Pavan Kumar et al., 2015), Sentinel (Pavan Kumar et al., 2018) and Quickbird (Fricker et al., 2015). The limitation of these attempts were in getting the low spectral resolution output that restricts the detection of the subtle changes.
To resolve the issue of low spectral resolution, attempts were made using hyperspectral narrow bands and hyperspectral vegetation indices in the range between400–2500 nm, derived from either spaceborne, airborne or field measured instruments that would help in identifying specific characteristics of plants (Anand et al., 2020; Thenkabail et al., 2013). These include pigment contents, leaf area index, biomass, crops stress, management properties (e.g., nitrogen application, tillage), and other biochemical properties (e.g., lignin, cellulose, plant residue) (Thenkabail and Lyon, 2016; Ullah et al., 2012; Srviastava et al., 2020). Spectral signatures derived from hyperspectral reflectance values represent the unique characteristics of plant or features (Singh et al., 2020). These are carried out by the selection of suitable bands from hyperspectral data, consisting of more than 200 bands (S. Wang and Chang, 2007). One can select a different number of variable bands to generate suitable vegetation indices using ratios to perform plant characterization. Further, band selection of hyperspectral data helps in eliminating redundant information at the band level (Bajwa et al., 2004; Malhi et al., 2021b). This allows removing Hughes phenomenon which is associated with hyperspectral data (Hughes, 1968). There are several data mining methods that are used in past for handling data dimensionality problem (Bajwa et al., 2004; Thenkabail et al., 2004; Thenkabail et al., 2004b; Thenkabail et al., 2011), such as Principal Compoenent Analysis (PCA) and regression trees. To model the relationship between scalar and exploratory variables such as hyperspectral indices, wavelength and partial least square, linear regression are utilized and their plots were drawn. These approaches not only help in reducing the data redundancy and dimensionality but also help in finding the best fit from these variables. Several other methods are utilized to process and analyze hyperspectral datasets for unmixing, continuum removal, finding the derivatives using neural networks, hierarchy endmember mixture analysis, PCA, Segmented PCA and more (Pandey et al., 2014; Thenkabail and Lyon, 2016). The recent development in band selection technique also includes Weighted Kernal Regularization (WKR) (Sun, Yang, Peng and Du, 2019), correntropy-based sparse spectral clustering (CSSC) (Sun, Peng, Yang and Du, 2019), fast and latent low-rank subspace clustering (FLLRSC) (Sun et al., 2020) and unmixing models like non-negative matrix factorization (NMF) (Pauca et al., 2006; J. Peng, Zhou, Sun, Du and Xia, 2020).
In past decades, several studies demonstrated that a relatively small number of specific narrow bands captured the optimal information which is essential for the assessment, quantification and discrimination of species (Chan and Paelinckx, 2008; Thenkabail et al., 2014; Thenkabail et al., 2011). The importance of selected band and its characterization are essential in the species discrimination and assessment. Since, different plant species give a varied response to different wavelengths (Cavender-Bares et al., 2017; Jetz et al., 2016) variations within optimum blue, red, green and Near-Infrared regions (NIR) hyperspectral bands of Electromagnetic Radiation (EMR) can aid in measuring species diversity (Anand et al., 2018; Y. Peng et al., 2019). Earth Observation (EO) hyperspectral data has provided a significant enhancement and better opportunity in assessing biological diversity using spectral channels than multispectral data sets (Anand et al., 2020; Jha et al., 2019; R. Lucas, Bunting, Paterson and Chisholm, 2008). EO hyperspectral data namely Hyperion can provide new information on alpha diversity, primarily due to contiguous bands. Due to continuous narrow spectral channels with a bandwidth of 10 nm ranging from 0.357 to 2.576 micromter spectral range, Hyperion data has the ability to provide unique spectral reflectance of different ground features (Carlson et al., 2007; Clark et al., 2005; K. L. Lucas and Carter, 2008; Malhi et al., 2020a, Malhi et al., 2020b; Thenkabail and Lyon, 2016). These differences in the wavelengths are mainly attributed to the variations in the pigment types and their amount. As indicated about the spectral capabilities of hyperspectral data, those indices are capable of differentiating plant species from adjacent land covers (Gitelson et al., 2002, Gitelson et al., 2002; A. Huete et al., 2002; Rouse et al., 1974; Sims and Gamon, 2003).
In purview of the above, the present study aims to: (i) propose a methodology for identifying optimum bands within blue, red, green and NIR regions of EMR for estimating species DIs namely Shannon Weiner DI, Margalef DI, McIntosh DI and Brillouin DI (2) develop modified hyperspectral indices using identified bands for DIs measurements (3) create a database of species diversity inventory of Shoolpaneshwar Wildlife Sanctuary (SWS), India after validation.
Section snippets
Study area
This research was conducted in a very dense and highly diverse forests of SWS of Narmada District in the Gujarat State, India (Fig. 1). The geographical location of the study area is 21° 03′ North - 73° 05′ East and to 21° 59′ North - 74° 10’ East. The sanctuary was established in the year 1989 and covers 607.71 km2 of geographical area. The forests in the sanctuary are remnants of some of the finest forests. This forest is mixed and contains two water reservoirs namely Sardar Sarovar and
Estimation of species diversity
Estimation of species diversity was carried out by calculating DIs which help in quantifying the biological variability. This will aid in juxtaposing biological entities made of diverse components both at spatial and temporal scale. Measures of four different DIs; Shannon Weiner DI, Margalef DI, McIntosh DI and Brillouin DI from the quadrat sampling gave the idea of the species distribution pattern of the study area.
For fifteen sampling plots, the range for Shannon Weiner DI was between 0.54
Conclusions
Spatial information for the identification of species diversity is essential for the timely execution and conservation of healthy ecosystem and biodiversity due to its importance in understanding the sustainable development. The unprecedented advancements in the technology have led to invention of sensors for enhanced extraction of information from the gathered data. Hyperspectral Remote Sensing which is available in large continuous narrow wavebands is identified as a prominent tool for
Author contributions
Conceptualization, R.K.M.M. and P.K.S.; methodology, R.KM.M., A.A, P.S.; software, A.A., R.K.M.M.; validation, R.K.M.M., A.A., A.N.M. and G.S.K.; formal analysis, A.A., R.K.M.M.; investigation, P.K.S., R.K.M.M., A.A.; resources, P.K.S., G.S.K.; data curation, P.K.S.R.K.M.M.; writing—original draft preparation, R.K.M.M. and A.A.; writing—review and editing, P.K.S, G.P.P., R.K.M.M., G.S.K., A.N.M., A.A., P.S.; visualization, RK.M.M., A.A.; supervision, P.K.S., G.P.P., R.K.M.M.; project
Funding
The present research was funded by “Department of Science and Technology and Science and Engineering Research Board, PDF/2017/002620”.
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.
Acknowledgements
R.K.M M extends thanks to the Department of Science and Technology and Science and Engineering Research Board for funding PDF/2017/002620. Authors thank Ministry of Environment, Forest and Climate Change and Gujarat Forest Department for providing necessary permission to execute field sampling in SWS. Authors are also thankful to local forest officials for providing required help during field sampling.
References (117)
- et al.
Comparing prediction power and stability of broadband and hyperspectral vegetation indices for estimation of green leaf area index and canopy chlorophyll density
Rem. Sens. Environ.
(2001) - et al.
Evaluation of Random Forest and Adaboost tree-based ensemble classification and spectral band selection for ecotope mapping using airborne hyperspectral imagery
Rem. Sens. Environ.
(2008) - et al.
Hyperspectral discrimination of tropical rain forest tree species at leaf to crown scales
Rem. Sens. Environ.
(2005) Calculating the vegetation index faster
Rem. Sens. Environ.
(1990)- et al.
Assessing crop residue cover using shortwave infrared reflectance
Rem. Sens. Environ.
(2004) Caution is needed when applying Margalef diversity index
Ecol. Indicat.
(2010)- et al.
Use of a green channel in remote sensing of global vegetation from EOS-MODIS
Rem. Sens. Environ.
(1996) - et al.
Remote sensing of chlorophyll concentration in higher plant leaves
Adv. Space Res.
(1998) - et al.
Hyperspectral vegetation indices and novel algorithms for predicting green LAI of crop canopies: modeling and validation in the context of precision agriculture
Rem. Sens. Environ.
(2004) - et al.
Overview of the radiometric and biophysical performance of the MODIS vegetation indices
Rem. Sens. Environ.
(2002)