Isoprene is the most abundant non-methane hydrocarbon in Earth's atmosphere. It is emitted naturally by plants, primarily in tropical regions. As a highly reactive compound, isoprene has a large impact on tropospheric chemistry. Uncertainties in isoprene emissions estimates are therefore a serious obstacle for atmospheric chemistry models. Isoprene emissions can be estimated using the Model of Emissions of Gases and Aerosols from Nature (MEGAN). In MEGAN, emissions are represented by a location-dependent standard emission rate which is scaled up or down by time-dependent changes in meteorology and canopy environment. The standard rates and scaling algorithms are highly uncertain, in part because they are empirical parameterizations based on limited data. In this talk, I will show how data assimilation can be used to improve the parameterizations in MEGAN by taking advantage of top-down isoprene emissions estimates. By optimizing the standard emission rates as well as the scaling of emissions with temperature, I will demonstrate how we can reduce biases and improve the seasonal cycle of emissions in MEGAN. A case study for the North African savannahs will be presented, along with a discussion of some of the challenges associated with this approach.