Premium
Multisensor On‐The‐Go Mapping of Soil Organic Carbon Content
Author(s) -
Knadel Maria,
Thomsen Anton,
Greve Mogens H.
Publication year - 2011
Publication title -
soil science society of america journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.836
H-Index - 168
eISSN - 1435-0661
pISSN - 0361-5995
DOI - 10.2136/sssaj2010.0452
Subject(s) - calibration , sampling (signal processing) , mean squared error , remote sensing , kriging , environmental science , soil carbon , spatial variability , digital soil mapping , soil science , range (aeronautics) , correlation coefficient , mathematics , computer science , soil map , statistics , materials science , soil water , geology , filter (signal processing) , composite material , computer vision
Detailed information on field‐scale variability of soil organic C (SOC) is essential for improved C management. Conventional sampling methods are costly because of large spatial variability and the high sampling density required. To reduce costs, automated in situ methods are needed. We compared mapping SOC using a mobile sensor platform (MSP) and conventional grid sampling on a highly variable agricultural field in Denmark. Sixty‐four samples collected on a 25‐m grid were used to generate a reference map of SOC distribution using kriging. Mobile sensory data (visible–near infrared spectra, electrical conductivity [EC], and temperature) obtained with a MSP were used to create a map of predicted C. To predict SOC, a calibration model was developed based on 15 representative samples. The best calibration model using a second Savitzky–Golay derivative on spectral data with EC as auxiliary data resulted in values as follows: root mean square error of prediction = 5.94; R 2 = 0.84; and ratio of standard error of prediction to SD [RPD] = 2.3. This study showed that the quality of those maps can be improved and spatial sampling intensities can be reduced by incorporating auxiliary data as a source of secondary information. An increased RPD value (2.3) was obtained for the sensor fusion measurements in comparison with those obtained using spectral data only (RPD = 1.9). The map based on MSP measurements detected more of the local SOC variation. High values for the error of prediction may have originated from the large SOC range (1.44–42.9%), the small number of calibration samples, and a sampling strategy that was not optimal. We concluded that more samples should be used when mapping highly variable fields.