Remote Sensing of Surface Carbon and Water Contents using Bare Soil Imagery.
Javed Iqbal, Purdue University, Department of Agronomy, 915 W. State Street, West Lafayette, IN 47907, Phillip Owens, Purdue Univ., West Lafayette, IN 47907, and Jeffery L. Willers, USDA-ARS, 810 HIGHWAY 12 EAST, MISSISSIPPI STATE, MS 39762.
Knowledge of spatial soil diversity and landscape dynamics is fundamental to understanding of the biogeochemical cycle of soil carbon and water content. Remote sensing data are useful for quantifying land-based measurements like soil texture, carbon and water content. Mapped surface soil properties acquired through remote sensing can be used as inputs for global biogeochemical models. The objective of this study was to explore the relationship between bare soil reflectance and surface soil texture (sand, silt, and clay), organic matter, and soil moisture. High spatial (2 m) and spectral resolution (414-920 nm) hyperspectral /multispectral aerial imageries were collected over the Mississippi Delta and Mississippi blackland prairie regions. Major soil types included Commerce (fine-silty, mixed, superactive, nonacid, thermic Fluvaquentic Endoaquepts), Robinsonville (coarse-loamy, mixed, superactive, nonacid, thermic Typic Udifluvents), and Convent (coarse-silty, mixed, superactive, nonacid, thermic Fluvaquentic Endoaquepts) and Brooksville (Fine, smectitic, thermic Aquic Hapluderts). Over three hundred surface soil samples were collected within these study areas. ArcView® GIS was used to generate sampling locations for random, transect, and target soil sampling methods. Each soil sample represented a composite of six sub-samples collected within a two meter square area. These samples sites were selected to represent the range of aspect, slope, elevation, and parent materials within each study area. To reduce the dimensionality of the hyperspectral data set, PCA analysis was applied. The selected bands were used in generating statistical relationships between spectral reflectance and surface soil properties data. Stepwise (backward & forward) and partial least square statistical methods were used to generate surface maps of soil texture, organic matter, and surface soil moisture. Multivariate analysis revealed that the near infrared band (950 nm) is the best predictor of percent clay (R2 = 0.683) and silt (R2 = 0.634), while a combination of red (650 nm) and green (550 nm) bands was the best predictor of organic matter. Surface soil moisture dynamic was highly spatially correlated with soil texture maps. Once these relationships were established, the ERDAS® Imagine Spatial Module was used to generate surface maps for percent clay, percent silt and percent organic matter. These final products not only could be used for management purposes but also to quantify the spatial patterns and temporal dynamics of natural resources as impacted by climate change.