Grapevine Leaf Area Index Estimation with Machine Learning and Unmanned Aerial Vehicle Information

Abstract:

Leaf Area Index (LAI) is an important structural property of vegetation canopy and is one of the basic quantities driving the algorithms used in biogeochemical, ecological, and meteorological applications. However, the methods to obtain accurate LAI, even at a field scale, are still challenging. Although direct methods can provide an accurate estimation of LAI, these are time-consuming and labor-intensive efforts that are inappropriate when the research scale is regional or global. On the other hand, indirect methods, such as remote sensing based methods, show promise in avoiding those drawbacks and also provide spatial results. For this research, UAV based information (optical and thermal images and point cloud information) and intensive field measurements from previous efforts on LAI by the USDA ARS Grape Remote Sensing Atmospheric Profile and Evapotranspiration eXperiment (GRAPEX) project are used in machine learning (ML) models to generate accurate LAI estimation at the field scale. These LAI modeling efforts and analyses, using such as Random Forest and Support Vector Machine (SVM) ML approaches for multiple vineyards across California (part of the GRAPEX project), are presented, along with a detailed discussion of the procedure followed for LAI modeling, ML input generation and selection, and performance of each of the models.