An automatic classifier based on a discriminant analysis (DA) was used to classify eight classes in relation to different stages of rice fields during the flooding season. This methodology is characterized by the fact that, once the training phase has been carried out, training areas are not required to perform new classifications. If the images have been radiometrically corrected in a consistent way, the classifier can be used in a retrospective mode using past images. For this study, the training phase was conducted with data taken in October 2006 and January 2007 while the automatic classifier was applied to a total of 10 Landsat-5 Thematic Mapper (TM) images from the 2004-05 and 2006-07 seasons. An average level of accuracy of 93.4% (range 89.7-98.7%) demonstrates the capability of the method to obtain high-quality and quasi-instantaneous classifications and to carry out retrospective studies even when training areas are not available for past dates. Two examples of how the method can be used are included in this article: (i) a study of the temporal evolution of flooding covers by period and (ii) the use of vector enrichment as a thematic updating tool for the cadastre. An additional objective of the study was to analyse the importance of the different bands to ascertain the suitability of alternative sensors with spectral configurations other than those provided by Landsat. This analysis demonstrates that the absence of shortwave infrared (SWIR) bands results in a decrease of almost nine percentage points in the accuracy levels of the classification while the blue band can be excluded with minimal impact on the results. © 2011 Taylor & Francis.