Quantifying the effect of landscape composition on disease dynamics remains challenging because it depends on many factors: disease epidemiological traits, climatic effect, cultural practices used to control diseases and landscape features. With usual ecology modelling, it is difficult to quantify and disentangle these factors. The objective of our study was to separate landscape effect from all these other factors. To this end, we have developed a dynamic disease model that integrates disease development, climate and fungicides effects. This model was applied to the case of banana leaf streak disease (BLSD) caused by Pseudocersospora fijiensis in Martinique. This disease is one of the main biotic constraint of banana production all over the world. Each process of our model was calibrated on a dataset including 83 plots producing Cavendish banana located all over Martinique. For each plot, between 2015 and 2019 the stage of evolution of the disease (SED, represents the dynamic of fungal infection on young leaves), the types of fungicide treatment applied, and the Piche evaporation were measured weekly. The model was used in two steps. First, we ran the model for each week based on measures of previous week. Then, we established a GLM of the residues of the model as a response to the weeks after the beginning of the epidemic, the week of the year, and the Piche evaporation. This GLM aimed at taking account the effects of i) the increase of the disease pressure over the island (constantly growing since its first detection in 2011), ii) the seasonality of the disease, and iii) the inoculum potential, respectively. We then subtracted the prediction of this GLM to the simulated SED, leading to a corrected predicted SED (SEDc). We hypothesized that SEDc to be related to the landscape effect on the disease, development and sporulation because all other factors were extracted. Finally, we correlated the SEDc with landscape composition variables calculated in buffers around