In Central America, coffee is grown in agroforestry systems. Since 2012, coffee leaf rust, caused by the fungus Hemileia vastatrix, has produced major epidemics in this region. To prevent future epidemics, the European Union through its PROCAGICA program (Programa Centroamericano de Gestión Integral de la Roya del Café) promotes the creation of an early warning system based on weather monitoring.To build models to forecast the disease we must first identify which microclimatic variables are responsible for rust development and then be able to estimate these variables under different agroforestry systems as a function of the data provided by weather stations, established at full sun. From a trial set up in Costa Rica where disease and weather data were monitored, we deduced, without a priori [1], that the different disease development stages (see figure) were the result of complex combinations of microclimatic variables acting at diffe-rent periods (times and durations). Then, to estimate the effect of agroforestry systems on these microclimatic variables, a second trial was conducted in Costa Rica within an altitudinal gradient. In each site, meteorological stations were set up in a full sun reference plot and coffee plots with different shade trees. Using boosted regression tree method, we found that microclimate under shading depends mainly on full sun weather with nonlinear relationship, hour, shade tree species, orientation, canopy openness and plot slope in this order.