Skip to main content
SLU publication database (SLUpub)
Research article - Peer-reviewed, 2010

Increased sample point density in farm soil mapping by local calibration of visible and near infrared prediction models

Wetterlind, Johanna; Stenberg, Bo; Söderström, Mats

Abstract

For use as decision support for variable rate applications in precision agriculture, the commonly used sample point density of one sample per hectare is often not enough. However, increasing the sampling density using laboratory analyses is too expensive for farmers to implement. It is therefore important to find methods for rationalisation. To this end, farm-scale visible and near infrared reflection (vis-NIR) calibrations were established on two farms in southern Sweden (Hacksta and Sjbstorp) for soil texture, soil organic matter, total N, pH and plant-available P. K and Mg. By keeping the laboratory analyses to a minimum to be used for vis-NIR calibrations and only collecting vis-NIR spectra from the vast majority of the samples, the sampling density could be increased without significantly increasing the cost. In this study 25 samples were used in the calibrations. Six different calibration sample selection methods were compared, selected from three different datasets originating from a larger context aiming at covering soil variations. Using only 25 calibration samples resulted in good predictions for clay at both farms, r(2) values of 0.81 and 0.89 and RMSEP values of 3.6 and 3.9%. Sand, soil organic matter and total nitrogen were well predicted at Hacksta (r(2) = 0.87, 0.90 and 0.89 and RMSEP = 3.0, 0.28 and 0.018% respectively) but 25 samples proved to be too few at the geologically divided farm Sjostorp. For predicting pH and plant-available P. K and Mg, more than 25 calibration samples were needed at both farms, although with 75% of all reference samples (92 and 94 at Hacksta and Sjostorp respectively) in the calibration these parameters also showed potential for building useful NIR calibrations (RPD values between 2.3 and 2.8 except for the predictions for pH at one of the farms resulting in an RPD value of 1.6). However, predictions for silt content were less reliable and the small number of calibration samples was not the limiting factor in this case. The promising results are encouraging for further development of cost-effective high resolution farm soil maps using NIR spectroscopy. (C) 2010 Elsevier By. All rights reserved.

Keywords

Diffuse reflectance spectroscopy; Plant-available P, K and Mg; Soil pH; Precision agriculture; Clay; Soil organic matter

Published in

Geoderma
2010, Volume: 156, number: 3-4, pages: 152-160
Publisher: ELSEVIER SCIENCE BV