Reconstruction of within-field spatio-temporal variability from fixed and on-board proximal sensors: application to simulated leaf area data in viticulture
Abstract
Purpose This study investigates the reconstruction of spatio-temporal variability in vineyards by combining data from fixed and mobile sensors: fixed sensors monitor temporal changes on a few specific plants, while mobile sensors collect spatial data at specific times. It focuses on the Leaf Area Index (LAI) as a key indicator of crop development that can be measured on almost any plant at any time. Methods The methodology involves the simulation of LAI data from April to October, taking into account phenological, environmental and plant variability, as well as sensor errors. Data fusion methods, spatio-temporal inverse distance weighting (STIDW), spatio-temporal kriging (STK) and generalized additive models (GAM), were evaluated to reconstruct LAI dynamics from available sensor data. Results Although these three methods all gave satisfactory results with reconstruction error not higher than $$10\%$$ , STK and GAM performed better than STIDW in reconstructing spatial structure. However, GAM showed flexibility in modelling non-linear relationships and outperformed other models in the case of a more heterogeneous plot. Different sampling scenarios showed that increasing the number of spatial maps and fixed sensors improved the reconstruction by about $$5\%$$ , with spatial maps having a greater impact. Conclusion The spatio-temporal variability of the vineyard can be reconstructed with a limited number of fixed sensors and monthly acquisitions with mobile sensors at the beginning of the season, for example 4 monthly acquisitions from April to July and from 3 to 8 fixed sensors per hectare depending on the complexity of the plot. This study highlights the benefits of data fusion to improve precision farming applications.