Forecast systems for coffee diseases

Abstract

Developing techniques to notify coffee growers about the ideal timing for applying fungicides, both chemical and biological, is crucial to increase the financial and environmental sustainability of coffee growing. The aim of this study was to develop a warning system for the incidence of coffee rust (Hemileia vastatrix) and phoma leaf spot (Phoma spp), using multiple linear regression models (MLRM) based on meteorological variables. Validate these models in several municipalities and then regionalize this platform for the coffee growing regions of south and cerrado of Minas Gerais state, using geostatistics to obtain surface maps with Digital Elevation Model (DEM). The models developed by Pinto et al. (2002) and Silva (2018) were validated in five municipalities: Carmo do Rio Claro (CRC), Monte Santo de Minas (MSM), Nova Resende (NR), Rio Paranaíba (RP), and Serra do Salitre (SS), for rust and phoma leaf spot, respectively. Experiments were conducted in a randomized block design with five treatments and five replications. The experimental plot consisted of six rows with 20 central plants in the useful area. Evaluations of rust and phoma leaf spot were carried out every two weeks. MLRM were adjusted for five municipalities based on data collected from automatic weather stations. Meteorological variables were lagged relative to the disease assessment dates. After adjusting 730 models, four were selected, two for each disease. Subsequently, they were validated in 15 municipalities, i.e., an additional 10 besides the initial five. Additionally, apart from the initial 15 properties, one property in each municipality, 35 other properties were randomly included to validate the models, near those locations. Models with meteorological variables collected 15-30 days prior to rust incidence (DAI) in CRC and NR showed promise for rust alert, while models with meteorological variables collected 07-15 and 15 DAI, adjusted for CRC and MSM respectively, showed promise for phoma leaf spot. These four models were the best for alerting about incidence at the control level with 15 DAI for rust and with 07 and 15 DAI for phoma leaf spot. The alert is issued when the 5% control level is reached for each disease, with four consecutive alerts in two rust alert models and three consecutive alerts in one phoma leaf spot alert model. The models were validated in different areas from those where they were previously adjusted and regionalized, using universal kriging with moderate spatial resolution of 1 km based on DEM to obtain surface maps of 10 meteorological variables for the southern and cerrado regions of Minas Gerais. The time frame from December 2020 to May 2021 was used to comply the rust warning systems developed in chapter 1 and by Pinto et al. (2002). Universal kriging was used as an external trend modeling of altitude, being a robust method for interpolating surfaces of meteorological variables. The interpolated data were used to generate phytosanitary alert maps, showing the favorability of weather conditions for the occurrence of rust.