PubAg

Main content area

Soil characterization across catenas via advanced proximal sensors

Author:
Duda, Bogdan M., Weindorf, David C., Chakraborty, Somsubhra, Li, Bin, Man, Titus, Paulette, Laura, Deb, Shovik
Source:
Geoderma 2017 v.298 pp. 78-91
ISSN:
0016-7061
Subject:
X-ray fluorescence spectroscopy, carbon, catenas, computer software, data collection, geographic information systems, land management, landscapes, markets, models, nitrogen content, particle size, physicochemical properties, prediction, reflectance spectroscopy, soil chemical properties, soil organic matter, soil physical properties, soil resources, soil sampling, topography, Eastern European region, Romania, Texas
Abstract:
As countries of Eastern Europe look to advance their agricultural markets through large scale agronomic production, high resolution mapping of soil resources will be essential. Portable X-ray fluorescence (PXRF) spectrometry and diffuse reflectance spectroscopy (DRS) are non-invasive, proximal sensing techniques which provide quantitative data germane to physicochemical soil properties in seconds. While these techniques have been widely used to characterize individual soil samples, sample sets, or variability across individual fields, less work has been done at the catena scale (even less so in Eastern Europe), where variability due to topographic differences substantively affects a wide number of soil properties. The present study was conducted on three catenas of the Transylvanian Plain, Romania, each with 100 sampling points randomly established in ArcGIS. Laboratory analysis (particle size analysis, total carbon, total nitrogen, soil organic matter) was conducted at Texas Tech University, USA. Following Savitzky–Golay first derivative transformation, DRS spectra were used to predict soil physicochemical parameters of interest via support vector regression. The whole dataset was randomly divided into a 70% training (n=210) and 30% test set (n=90). Across all catenas, a combined PXRF+DRS approach showed better parameter prediction relative to either sensor independently as evidenced by higher R2, lower RMSE, higher RPD, and higher RPIQ values. For each parameter, the 100 points per catena were used as input data to develop a PXRF+DRS predictive model, and the output data from each model was kriged using ArcGIS 10.3.1. Spatial analysis strongly reflected management and landscape dynamics across the catenas. Combined proximal sensor approaches show considerable advantages over traditional laboratory approaches, allowing for high sample throughput, greater analytical density, and less expensive data, with minimal fall off in data quality. The combined PXRF+DRS approach showed excellent potential for providing the data needed to support optimized soil resource mapping and land management decisions in Eastern Europe or worldwide.
Agid:
5654108