Wheat foliar pathogens inhabit crop canopies that are subjected to substantial spatio-temporalvariations in temperature. Given the pronounced diversity in thermal individual responses within alocal Zymoseptoria tritici population, these environmental heterogeneities may lead to phenotypicselection in the field. Signatures of response to thermal selection over the course of annual Septoriatritici blotch (STB) epidemics have been previously identified. However, the consequences onpopulation dynamics of the interplay between interindividual phenotypic variation and thermal canopyheterogeneity have not yet been investigated. We address this issue using a three-step approach. First,we quantified the extent of environmental (measurement of temperature distribution within wheatcanopies) and phenotypic (standardised thermal phenotyping of Z. tritici) variations encountered by orin natural pathogen populations. Second, we dissected the processes underlying spatio-temporalchanges in the phenotypic composition of populations facing different types and extent ofmicroclimatic heterogeneities by performing in planta polycyclic selection experiments (growthchamber and field experiments). Third, we explored in silico the way in which phenotypic diversityaffect population adaptive dynamics (quantitative assessments of population vulnerability andresilience to thermal changes). Our results show how environmental signals, interindividualphenotypic variations and ecological processes have affected Z. tritici population dynamics in ourexperiments. In particular, three major findings emerged from this investigation: (i) the occurrence andthe epidemiological consequences of short-term selection driven by seasonal temperature variationsover an annual epidemic; (ii) the critical importance of spatio-temporal thermal heterogeneity in wheatcanopies in the maintenance of phenotypic diversity within Z. tritici populations through the presenceof thermal refugia; (iii) the quantitative impacts of oversimplifications currently adopted in diseaseprediction models. By shedding new light on population adaptive potential to environmentalvariations, these insights would help to improve predictions of the eco-evolutionary responses ofpopulations to changing climate.