Thematic resolution of mapped data (i.e. number of classes considered) influence landscape structure representations and thus our ability to explain biodiversity patterns. We assessed the effet of thematic resolution to explain bird species patterns in agricultural-forest landscapes of a LTER site in south-western France. We also compared an unsupervised classification (k-means clustering) from an image of Normalized Difference Vegetation Index reflecting the plant productivity and a supervised classification based on a land cover map. We built several maps with 2, 4, 5, 6 and 7 classes for each classification method. Based on 573 point counts and these maps, we built generalized additive models with binomial distribution for 36 species. We compared these models using the Akaike Information Criterion and the percentage of explained deviance. Species distributions were explained in a similar manner by the unsupervised and supervised classification methods (averaged %D² = 0.15 in both cases). The gain of explanatory performance was the higher between the thematic resolution of 2 and 4 classes for both classification methods (averaged %D² from 0.12 to 0.17 for the unsupervised classification and from 0.11 to 0.15 for the supervised classification). We suggest that NDVI images with unsupervised classification can replace land cover maps which are often long to obtain, to explain bird community patterns in agricultural landscapes like ours.