Soil adjusted vegetation water content retrievals in grasslands

Soil contamination of canopy reflectance over grasslands can cause errors in empirical vegetation water content (VWC) retrievals using the NDII (Normalized Difference Infrared Index, [0.86-1.64]/[0.86+1.64]). Minimization of soil contamination by NDII relies on the existence of a quasi straight soil line and quasi straight VWC isolines (lines of equal VWC) in the 1.64-0.86 µm reflectance space. Further the VWC isolines are expected to meet at the origin of the 1.64-0.86 µm reflectance space. Considering soil moisture as the primary determinant of soil reflectance variation at a given location, this study investigates the effect of soil moisture on the nature of soil lines and VWC isolines under grassland conditions. Reflectance simulations from coupled soil-leaf-canopy reflectance models under grassland conditions show that soil lines and VWC isolines are expected to be curved and may not converge at the origin. This behaviour is attributed to disproportionate soil moisture related absorption processes operating at 1.64 µm and 0.86 µm. A new technique that accounts for these inconsistencies in NDII assumptions is proposed for VWC retrievals. The technique consists of using separate regression relationships between VWC and a Soil Adjusted NDII (SANDII) based on the volumetric soil moisture category of the background. SANDII, based on the idea borrowed from the Soil Adjusted Vegetation Index (SAVI) is an origin shifted transformation of NDII. The optimum origin that reduces VWC retrieval errors is shown to be soil moisture category specific. The proposed technique requires categorical soil moisture information in order to decide which regression relationship to apply for VWC retrievals. Climatology, meteorological models or microwave observations are expected to be reliable resources for such categorical soil moisture information. Evaluations of the proposed technique using simulated reflectances showed that absolute errors in VWC retrievals were reduced by an average 20% as compared to the traditional NDII regression method. Such improvements are expected to be significant for fire-risk applications. Finally supporting evidence for the need of an origin translated NDII is provided using data collected over pastures during the Soil Moisture Experiment 2003 (SMEX03) field campaign.