Electrical imaging of soil water availability to grapevine: a benchmark experiment of several machine-learning techniques. - Archive ouverte HAL Access content directly
Journal Articles Precision Agriculture Year : 2016

Electrical imaging of soil water availability to grapevine: a benchmark experiment of several machine-learning techniques.

Abstract

Electrical resistivity (ER) can be used to assess soil water in the field. This study investigated the possibility of extending the use of ER to measure plant available soil water variables, i.e. available soil water (ASW), total transpirable SW (TTSW), and fraction of transpirable SW (FTSW) using a pedotransfer approach. In a vineyard, 224 electrical resistivity tomography (ERT) transects and 672 time domain reflectometry (TDR) soil water profiles were acquired over 2 years. Soil physical–chemical properties were measured on 73 soil samples from eight different sites. To estimate the amount of soil water available to plants, grapevine (Vitis vinifera L.) water status was monitored by means of leaf water potentials. A benchmark experiment was carried out to compare four machine-learning techniques: multivariate adaptive regression splines (MARS), k-nearest neighbours (KNN), random forest (RF), and gradient boosting machine (GBM). Model interpretation led to a deeper understanding of the relationships between electrical resistivity and soil properties when predicting soil water availability for the plant. The models assessed had good predictive performance and were therefore used to map ASW, TTSW and FTSW in the vineyard. ER coupled to machine-learning algorithms was shown to be a good proxy for quantification and visualisation of plant available soil water with low disturbance.
No file

Dates and versions

hal-01403963 , version 1 (28-11-2016)

Identifiers

Cite

Luca Brillante, Benjamin Bois, Olivier Mathieu, Jean Lévêque. Electrical imaging of soil water availability to grapevine: a benchmark experiment of several machine-learning techniques.. Precision Agriculture, 2016, 17 (6), pp.637-658. ⟨10.1007/s11119-016-9441-1⟩. ⟨hal-01403963⟩
92 View
0 Download

Altmetric

Share

Gmail Facebook X LinkedIn More