4.1. Summary of remote sensing technique
There are various kinds of RS practices founded on their three-dimensional, spectral, chronological as well as radiometric resolution and platforms on where they are set-up (shown in Fig. 1). Based on type of use, property to be quantified, as well as correctness of results, suitable approach is selected.
Rapid as well as extensive uses of these applications are due to developments in sensors requirements. Sensors set-up on satellite platforms have upgraded from panchromatic towards multispectral as well as upcoming hyperspectral i.e., EnMAP, HyspIRI as well as PRISMA. Owning to availability of such hyperspectral sensors, essential information regarding soil’s condition, SOC estimation can be obtained via RS applications. Additionally, to meet present as well as upcoming demands for soil monitoring essential datum for correct up-to-date soil maps can also be obtained via RS practice. RS practices have merits that these are environmentally sound practices to obtain data related to soil properties, provide data of those sites/ places that are inaccessible, provide concise information and lower the chances of laborious soil sampling work (Angelopoulou, Tziolas et al. 2019).
RS practices have issues that they possess little signal to noise proportion (Minu, Shetty et al. 2016), little spectral resolution (Gomez, Rossel et al. 2008), as well as undergo geometric and atmospheric manipulations (Jakob, Zimmermann et al. 2017). There exists another problem i.e., scale effects, for example, variations take place when retrieval models as well as algorithms are derived at small and large levels (Wu and Li 2009). Additionally, external parameters like soil wetness, structure, raggedness, vegetation greatly affect correct quantitative estimation via RS practices (Wu and Li 2009). Table 4 summarized some merits as well as demerits of RS platforms for SOC monitoring (Angelopoulou, Tziolas et al. 2019).
Multivariate statistical practices find applications for model calibration employing PLSR, and pre-processing practices changes in each study. However, for generating prediction models for soil characteristics, there is significant interest in connection with machine learning approaches having power to outperform PLSR as shown in Fig. 2 (Angelopoulou, Tziolas et al. 2019).