Broad-scale cruciferous weed patch classification in winter wheat using QuickBird imagery for in-season site-specific control

This article explores the potential use of multi-spectral high-spatial resolution QuickBird imagery to detect cruciferous weed patches in winter wheat fields. In the present study, research was conducted on six individual naturally infested fields (field-scale study: field area ranging between 3 and 52 ha) and on a QuickBird-segmented winter wheat image (broad-scale study: area covering approximately 263 winter wheat fields, approximately 2 656 ha) located in the province of Cordoba (southern Spain). To evaluate the feasibility of mapping cruciferous weed patches in both the field-scale and broad-scale studies, two supervised classification methods were used: the Maximum likelihood classifier (MLC) and vegetation indices. Then, the best classification methods were selected to develop in-season site-specific cruciferous weed patch treatment maps. The analysis showed that cruciferous weed patches were accurately discriminated in both field-scale and broad-scale scenarios. Thus, considering the broad-scale study, classification accuracies of 91.3 and 89.45 % were obtained using the MLC and blue/green (B/G) vegetation indices, respectively. The site-specific treatment maps obtained from the best classifiers indicated that there is a great potential for reducing herbicide use through in-season, cruciferous weed patch site-specific control on both a field-scale and broad-scale. For example, it can be determined that by applying site-specific treatment maps on a broad-scale, herbicide savings of 61.31 % for the no-treatment areas and 13.02 % for the low-dose herbicide areas were obtained.This work was partially financed by the Spanish Ministry of Science and Innovation through projects AGL2008-04670-CO3-03 and AGL2011-30442-CO2-01 (FEDER).Peer Reviewe