Evaluation of the use of UAV-derived Vegetation Indices and Environmental Variables for Grapevine Water Status Monitoring based on Machine Learning Algorithms and SHAP Analysis

Thumbnail Image
Open Access Location
Journal Title
Journal ISSN
Volume Title
(c) The author/s (CC BY)
Monitoring and management of grapevine water status (GWS) over the critical period between flowering and veraison plays a significant role in producing grapes of premium quality. The spatial and temporal variation of GWS, is conventionally characterized by using a pressure bomb. This is laborious and time-consuming, which limits the number of samples. Although unmanned aerial vehicles (UAVs) can provide mapping across the entire vineyards efficiently, most commercial UAV-based multispectral sensors do not contain a shortwave infrared band which makes the monitoring of GWS problematic. As GWS is an integrated response to vegetation characteristics, temporal trends, weather conditions, and soil/terrain data, it is assumed that these ancillary variables have the potential to enhance the capability of UAVs for GWS monitoring. The goal of this study is to explore whether and which of these ancillary variables may improve the accuracy of GWS estimation using UAV, and provide insights into the contribution that each selected variable contributes to the variation in GWS. A UAV was flown over two Pinot Noir vineyards in New Zealand to generate aerial images with 4.3 cm resolution, and 18 vegetation indices (VIs) were computed for every sampled grapevine. The strongly correlated VIs were used as the core input for later GWS modeling. Ancillary data included soil/terrain, weather, and temporal variables. Slope and elevation were extracted from a digital elevation model, and apparent electrical conductivity (ECa) was obtained from an EM38 survey. A local weather station provided continuous air temperature, humidity, rainfall, wind speed, and irradiance data, which were computed as variables at weekly and daily intervals. Day of the year (DOY) was used to represent the temporal trend along the growing season. Three machine learning algorithms (elastic net, random forest regression, and support vector regression) were used to regress the predictors against stem water potential (Ψstem), measured by a pressure bomb and used as a proxy for GWS. Shapley Additive exPlanations (SHAP) analysis (a statistical tool that weighs the importance of each variable in a model) was used to assess the relationship between selected variables and Ψstem. The results show that Transformed Chlorophyll Absorption Reflectance Index (TCARI) and Excess Green Index (ExG) are the best correlated VIs, but their correlation with Ψstem is poor (rTCARI = 0.6; rExG = 0.58). The coefficient of determination (R2) of the TCARI-based model increased from 0.35 to 0.7 when DOY and elevation were included as ancillary inputs. R2 of the ExG-based model increased from 0.3 to 0.74 when DOY, elevation, slope, ECa, and daily average windspeed, were included as ancillary inputs. Support vector regression was the best model to describe the relationship between Ψstem and selected predictors. This study has provided proof of the concept of developing GWS estimation models that potentially enhance the monitoring capacities of UAVs for GWS, as well as provide individual GWS mapping at the vineyard scale. This may enable growers to improve irrigation management, leading to controlled vegetative growth and optimized berry quality.
2022, 34 pp. ? - ? (12)