Modelling mixed scenarios of canopy and soil spectral reflectance to improve SOC prediction

Three and four times as much carbon is stored in Earth's soil as organic matter, about 60-80%, compared to what is found in the atmosphere and terrestrial plants. A quantifiable fraction of soil organic matter is soil organic carbon (SOC), which is a key property of soil quality. Since the advent of optical remote sensing technologies and especially with the development of soil and imaging spectroscopy, empirical statistical approaches have often been employed to link the spectral signatures with soil properties. Common approaches are indirect modelling through the multivariate statistical algorithms and machine learning algorithms, where correlation processes and nonlinear relationships between variables are taken. An alternative, to these methods, is forward radiative transfer modelling (RTM) or physical modelling that predicts the spectral reflectance of soils in the solar domain (0.4–2.5 μm) in different scenarios. This approach nowadays is mostly used for modeling wet soils and potentially inferring soil moisture content from soil reflectance, but not used for retrieving other properties such as mixed vegetation content and soil properties such as organic carbon content. In this research supported by WORLDSOILS project, we aim to couple a multilayer radiative transfer model of soil reflectance (MARMIT) to soil-leaf-canopy model (SLC-1D RTM) based on LUCAS soil spectral library (SSL) to simulate reflectances of mixed soil-vegetation scenarios as function of water, vegetation and SOC content. In the integrated model called MARMIT-SLC, changes in the spectral reflectance of the soil surface are considered that include the occurrence of soil moisture, dryness, in addition to the effects of early green crops and dry crop residues. This development may improve the coverage and accuracy of SOC predictions based on remote sensing data. For this, upscaling simulations over a large spatial scale of landscapes are performed. Preliminary results show that the accuracy of SOC predictions obtained from the laboratory's VNIR-SWIR spectra based on LUCAS 2009 soil datasets have increased. Although the RTM approach has been developed systematically to validate the suitability of the improved soil algorithm for global soil mapping, there are challenges in model evaluation and validation of results due to the lack of ground data availability.

Keywords: soil spectroscopy, RTM, MARMIT model, SLC model, Leaf area index, fractional vegetation cover, SOC.