Isoprene is a reactive hydrocarbon emitted to the atmosphere in large quantities by terrestrial vegetation. Isoprene emissions vary widely among plant species and as a function of meteorology, and they have a significant impact on tropospheric chemistry. Emissions can be estimated with empirical models, but this approach is subject to large uncertainties. In this talk, I will show how satellite and eddy covariance measurements can be used to constrain the empirical isoprene emission model MEGAN (Model of Emissions of Gases and Aerosols from Nature). Model biases can be reduced by optimizing the vegetation-specific standard emission rate with satellite-based constraints, while diurnal variability can be improved by optimizing the model's temperature dependence with eddy covariance measurements. Uncertainties in the observational constraints are a significant challenge for this work. Large regionally varying discrepancies exist between different satellite-based emissions estimates, which can potentially be attributed to biases in chemistry-transport models. The impact of these biases is sensitive to satellite overpass time, making direct comparison between different satellite-based emissions estimates difficult. Further optimization of empirical models such as MEGAN is dependent on reducing the uncertainties in these satellite constraints, as well as increasing the availability of ground-based eddy covariance measurements.