Premium
Constructing a layered electrical conductivity model using k nearest‐neighbour predictions and a combination of two proximal sensors
Author(s) -
Piikki K.,
Wetterlind J.,
Söderström M.,
Stenberg B.
Publication year - 2014
Publication title -
european journal of soil science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.244
H-Index - 111
eISSN - 1365-2389
pISSN - 1351-0754
DOI - 10.1111/ejss.12201
Subject(s) - emi , resolution (logic) , electrical resistivity and conductivity , arable land , standard deviation , conductivity , dipole , analytical chemistry (journal) , computational physics , mathematics , physics , electromagnetic interference , statistics , chemistry , artificial intelligence , computer science , telecommunications , agriculture , geography , quantum mechanics , archaeology , chromatography
Summary A strategy to produce layered high‐resolution apparent electrical conductivity ( EC a ) models of agricultural fields by distance‐weighted k nearest‐neighbour prediction ( k NN ) was tested at three farms. Electromagnetic induction ( EMI ) measurements were combined with measurements made with a dipole probe. Depth‐layer‐specific EC a values from the probe measurements were interpolated in the attribute space of the EMI measurements with the distance‐weighted k NN method. This analysis resulted in high‐resolution EC a maps for depth intervals of 0–0.2 and 0.4–0.6 or 0.4–0.8 m. The EC a values measured with the dipole probe ranged between 6.1 and 40.2 mS m −1 , and at two of the three farms investigated it was possible to create EC a maps at two depths with mean absolute errors of 1.1–3.8 mS m −1 . At the third farm the predictions were less accurate. Combining data from two fundamentally different sensors of EC a for k NN predictions was deemed to be an efficient way to produce 3‐D information on arable soil. However, it seems to be essential that the dipole probe and the EMI measurements are made under similar conditions.