Análise espaçotemporal da sigatoka amarela da bananeira utilizando sensoriamento remoto e geoestatística

• Autor(a) principal: Rodrigues, Julia Dal Poggetto
• Data de Publicação: 2017
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFLA
• Texto Completo: http://repositorio.ufla.br/jspui/handle/1/15623
• topic: Banana - Doenças e pragas - Distribuição espacial; Banana - Doenças e pragas - Análise de séries temporais; Sigatoka amarela; Geologia - Métodos estatísticos; Sensoriamento remoto; Bananas - Diseases and pests - Spatial distribution; Bananas - Diseases and pests - Time-series analysis; Yellow sigatoka; Geostatistics; Remote sensing; Engenharia Agrícola; Fitopatologia

Yellow Sigatoka leaf spot, caused by Pseudocercospora musae (Mycosphaerella musicola), is among the diseases that most affect banana crop. The disease progress occurs in both time and space, and the Remote Sensing sciences and the space-time Geostatistics provide support for the analysis of this dynamic process. Therefore, the objective of this study was to perform the spatiotemporal prediction of yellow Sigatoka at different times, to calculate and evaluate vegetation indices derived from the Enhanced Thematic Mapper Plus (ETM +) sensor, from the Landsat 7 satellite, to infer about the disease occurrence, to evaluate the existence of spatial and temporal correlation between data obtained by the sensor and data in situ, and finally to analyze the spectral signature of the plant affected by the disease. The first study used images from the Landsat 7 satellite, ETM + sensor, with atmospheric correction method Dark Object Substraction 1 (DOS1) and Second Simulation of Sattelite Signal in the Solar Spectrum (6S). Normalized Difference Vegetation Index (NDVI), Green Standard Difference Vegetation Index (GNDVI) and Normalized Difference Water Index (NDWI) were calculated. In the spectral signature related to the months of September and October, the mid-infrared region allowed the characterization of the disease in the plant. The NDVI and GNDVI indices showed differences in the atmospheric correction and in relation to the disease the NDWI presented better result. In the second study, Gneiting's separable spatiotemporal, Double Exponential, and non-separable covariance models were tested with the Weight Least Squares (WLS), Restricted Maximum Likelihood (REML) and Likelihood Pairwise methods. The Gneiting‟s non-separable model, WLS adjustment method, in which the trend was modeled, allowed to reduce as uncertainties of spatial and temporal prediction of the disease, as well as to characterize the pattern of monocycle temporality of the disease.