Search

This paper proposes a method to support sugarcane yield forecast using vegetation spectral indices, principal component analysis and historic yield data. The study area is located in the State of São Paulo, Brazil, and is divided into 11 production plots (108.75 ha), where sugarcane of the RB85 5536 variety is cultivated on red latossol (oxissol‐type) soil and flat topography. The data employed in the study include radiometrically and geometrically corrected enhanced thermatic mapper Plus (ETM+)/Landsat‐7 and ASTER/Terra images, acquired in June and April 2001, respectively, and historic harvest data measured in 2000 and 2001. The method comprises several steps: (a) enhancement of specific spectral responses of vegetation constituents; (b) reduction of spectral dimensions with prioritization of information and weighing of parameters related to foliar area; the data processed through these steps are reduced to a single image (the synthesis image), from which the mean DN (digital number) per cultivated area is calculated; (c) the image DNs are subsequently transformed into ton of stalk per hectare (t ha-1) through normalization, which requires knowledge of the previous year's yield for the cultivated production plots under analysis. Yield estimates using the method showed greater precision in comparison to the ubiquitous visual methods employed by the sugarcane agro‐industry in Brazil. Using factual productivity data of the year 2000 harvest only, the method achieved estimate errors varying between 2.57% and 5.65%, compared with 9.06% expected by the sugar factory; whereas using data from the year 2001 harvest, error margins were remarkably lower, around 1%. [ABSTRACT FROM AUTHOR]