Insights for improving bacterial blight management in coffee field using spatial big data and machine learning

Abstract

Pseudomonas syringae pv. garcae, the causal agent of coffee disease bacterial blight, causes losses in nurseries and coffee fields. In this work, the objective was to evaluate integrated bacterial blight management in a coffee (Coffea arabica L.) field based on disease spatial pattern and ecological variables. The coffee field was composed by 85 georeferenced sample points, containing 5 plants representing a georeferenced point, being the spatial support of the experiment. Disease intensity classes were predicted in the field by machine learning algorithms fitted to big data on surface reflectance and spectral indices derived from digital image processing of Landsat-8 OLI/TIRS, as well as morphometric and hydrological attributes determined by geocomputation algorithms. Geostatistical modeling was used to characterize the spatial pattern and map the disease to gain epidemiological knowledge and precisely manage bacterial blight. Random forest algorithm enabled to detect the importance of relief morphometry associated with bacterial blight spatial progress in the coffee field, mainly according to the altitude and flow line curvature of the terrain. Probabilistic information on disease spatial pattern, modeled considering external trend effects of the topography variation, can be useful information for disease spatial prediction and integrated management based on georeferenced disease sampling. Multiple environmental variables may be carefully considered to evaluate mechanisms of interactions of bacterial blight with coffee plants and the physical environment.