Artificial Neural Network Model to Estimate Vine Water Status Using a Multispectral Camera aboard an UAV
Tomás Poblete,* Samuel Ortega-Farias, Miguel A. Moreno,
and Matthew Bardeen
*Universidad de Talca, Citra-Utalca, Avenida Lircay s/n, Casilla 410, Talca, Chile (firstname.lastname@example.org)
A field experiment was carried out to implement an artificial neural network (ANN) model for estimating the spatial variability of vine water potential of a drip-irrigated Carménère vineyard located in the Pencahue Valley, Maule Region, Chile (35°25´ LS; 71°44l LW; 90 m asl). For this study, a helicopter-based unmanned aerial vehicle (UAV) was equipped with a multispectral camera at very high resolution (6 cm × 6 cm), while a pressure chamber was used to measure the midday stem water potential (MSWP) at the time of the UAV overpass, near solar noon. A multilayer perceptron ANN type was used to develop a model, using as input nodes the spectral information from five wavelengths (550, 570, 670, 700, and 800 nm) and as output node, the MSWP. As a reference, correlations between MSWP and several spectral indices such as NDVI, MSR, GNDVI, PRI, and TCARI/OSAVI were developed in this study. There were significant linear correlations between MSWP and spectral indices NDVI, GNDVI, and MSR, with r2 ranging between 0.32 and 0.35. The best correlation was observed between MSWP and MSR. The ANN model could predict MSWP with an r2 value of 0.87 and a root mean square error of 0.12 MPa. Results demonstrated that multispectral cameras placed on an UAV could provide a good tool to evaluate the intra-vineyard spatial variability of vine water status.
Funding Support: Chilean National Science Foundation (CONICYT)