|Myers, David - UNIVERSITY OF MISSOURI|
|Miles, Randall - UNIVERSITY OF MISSOURI|
|Grunwald, Sabine - UNIVERSITY OF FLORIDA|
Submitted to: International Conference on Precision Agriculture Abstracts & Proceedings
Publication Type: Abstract Only
Publication Acceptance Date: May 2, 2008
Publication Date: July 20, 2008
Citation: Myers, D.B., Miles, R.J., Kitchen, N.R., Sudduth, K.A., Sadler, E.J., Grunwald, S. 2008. Multi-Sensor Estimation of Claypan Soil Profile Properties [abstract]. In: 9th International Conference on Precision Agriculture Conference Abstracts. 9th International Conference on Precision Agriculture, July 20-23, 2008, Denver, Colorado. p. 200. Technical Abstract: Large quantities of data are needed to solve land use and soil management problems, yet lab analysis of soil data is costly and time consuming. Soil property sensors on mobile platforms have the capability to estimate soil properties at many more locations than reference lab measurements. The fusion of multiple sensor signals may be synergistic - estimating soil properties more accurately than single sensor systems. This hypothesis is examined in the Central Claypan Region of Northeast Missouri for an emulated penetrometer sensor with capability to measure cone-index (CI), bulk apparent electrical conductivity (ECp), and visible to near-infrared diffuse reflectance spectra (VNIR-DRS). A multi-sensor penetrometer was emulated by combining in-situ measurements of ECp and CI from an existing conductivity penetrometer with ex-situ VNIR-DRS measurements on die-pressed moist soil cores. Sites in four agricultural fields (n=75) were measured and sampled in May-June 2007. Soil properties of interest were measured from 2.5 cm lengths within cores and split into independent calibration and validation datasets (16 sites each). Measurements included soil texture by hydrometer, 1:1 (water:soil) suspension pH, and organic carbon (OC). Modeling was performed by partial least squares regression (PLSR). Reflectance, CI, and ECp were normalized to unit variance. Results demonstrate the potential of VNIR-DRS for in-situ estimation of soil properties in claypan soils; however, fusion of ECp and CI data did not improve the estimation of the target properties. Relative to the target variables, CI and ECp had larger variability than many VNIR bands. Good correlation of these wavelengths to soil properties and the overall information content of the megavariate VNIR signal overwhelmed any information added by CI and ECp. Between-site noise in ECp and CI is large in these fields, probably due to differences in field soil moisture and management. Several features of VNIR spectra informed the PLSR models. Water and X-OH absorption features at 1410, 1940, and 2200 nm were most important for clay estimation. These bands influenced pH models but were secondary to a broad reflectance peak at 800 nm (640 to 970 nm). Soil pH was negatively correlated to reflectance in the visible-red part of this peak. Bright chroma mottles of 2.5 and 5 YR (reddish) hues coincide with maximum profile acidity at the claypan boundary and may be informing this relationship. Organic carbon estimates were influenced by a broad range of mostly visible wavelengths (425 to 1000 nm). These were negatively correlated with OC suggesting soil melanization as the predictive factor. Sensor fusion did not improve soil property estimation in this instance. This was due to the large information content in VNIR-DRS spectra which highlights the potential of this method. Penetrometer based VNIR-DRS sensors should prove to be a valuable resource for filling the spatial gaps left by traditional sampling and reference lab measurements.