Characterization of a nations forest resources are important for effective planning and management of it. Characterization of forests at each level have their own profound applications. Forest/Non-forest and greenness maps are used to assess climatic impacts of forests (Prijith et al., 2021; Yu et al., 2020), loss of biodiversity (Mahanand et al., 2021), assessment of forest loss (Hansen, 2013; Rahman & Sumantyo, 2010) and forest degradation (Gao et al., 2020; Joshi et al., 2015). Forest type maps are used for efficient utilization and sustainable management of forests resources (Hościło & Lewandowska, 2019; Zhu & Liu, 2014) and forest conservation activities (Fedrigo et al., 2018). Species level maps are used for conservation and management activities (Behera et al., 2021; Ferreira et al., 2019),estimating economic output from forest (Naidoo et al., 2012; Peerbhay et al., 2013), biodiversity monitoring (Jha et al., 2019), invasive species detection and control (Boschetti et al., 2007).
Remote sensing techniques and technologies have evolved significantly over the past couple of decades. We are now able to observe forests not just in terms of their reflectance spectrum but also in terms of their texture (Ferreira et al., 2019), phenology (Araya et al., 2018; Srinet et al., 2020) and structure (Neuenschwander & Pitts, 2019; Potapov et al., 2021; Singhal et al., 2021) and from varied platforms starting from Satellite based to Airborne (UAVs and Airplanes) to Terrestrial Sensors.
Large scale (global (Martone et al., 2018; Shimada et al., 2014), Pan-European (Pekkarinen et al., 2009),China (Qin et al., 2015)) forest/non-forest maps have been made by use of readily available remote sensing datasets. While some forest types at country level have been mapped using readily available satellite datasets like mangroves of China (Chen et al., 2017) and evergreen forests of Brazil (Sheldon et al., 2012) etc. A lot of research has demonstrated mapping of forest types at smaller scales using advanced remote sensing methods (Fedrigo et al., 2018; Hościło & Lewandowska, 2019; Suzanne Mariëlle Marselis et al., 2018; Pasquarella et al., 2018; Zhu & Liu, 2014). While species level maps at smaller scales been generated using Airbourne Hyperspectral Data (Behera et al., 2021; Boschetti et al., 2007; Jha et al., 2019; Naidoo et al., 2012; Peerbhay et al., 2013; Pinheiro et al., 2016; Wietecha et al., 2019),Airborne LiDAR Data (Blomley et al., 2017; Kukkonen et al., 2019; Shi et al., 2018; Suratno et al., 2009), textures from high resolution satellite imagery (Ferreira et al., 2019), Terrestrial Laser Scanners (Terryn et al., 2020) and by fusion of multiple satellite datasets (Clark, 2020; Grabska et al., 2020; Hościło & Lewandowska, 2019; Wolter & Townsend, 2011). But for a national effort we need a robust, scalable and economically viable group of datasets and techniques.
Earlier efforts in this field have been largely based on Multi spectral datasets alone and some element of phenology was incorporated by the use of multi temporal datasets (Reddy et al., 2015; Roy et al., 2015). (van Leeuwen et al., 2021) has discussed limitations to species identification using spectral methods. He has used 3D spectroscopic landscape simulations to quantify loss of classification accuracy (of species present in the landscape) with the spatial resolution and with species richness of the landscape. A lot of the newer studies have demonstrated fairly, the potential of structural attributes to produce regional species level maps using Airbourne LiDAR data and for mapping tree species richness using satellite based LiDAR data (Suzanne M. Marselis et al., 2020; Suzanne Mariëlle Marselis et al., 2019). With increase in ease of access to and processing of a lot multispectral dataset, scientists are now easily able to extract or track the vegetative phenology (Araya et al., 2018; Xu et al., 2021) of the target vegetation. With 4 passes of MODIS and VIIRS available globally daily, changes to phenology can be observed with cloud cover being the one of its major limitations. These sensors have products at much coarser resolution (250m) and hence are much more suited for forest type mapping (Srinet et al., 2020) than for species level analysis. With Sentinel 2A and 2B pair repetivity of 5 days can be achieved for a particular target with spatial resolution of 10m which can potentially capture some species level information.
Reflectance spectra, phenology and structure are 3 aspects of forests which can be readily observed from the space. In this study we try to assess how our current capacity to observe the 3 observable aspects of forests are useful for characterising forests at community level. This study aims at mapping tree communities of forests with best possible accuracy possible with current readily available remote sensing datasets.