Isoprene is a reactive hydrocarbon emitted in large quantities by terrestrial vegetation. Emission rates vary among plant species and are sensitive to environmental conditions. As a precursor to air quality and climate pollutants including ozone and carbon monoxide, isoprene has a large impact on global atmospheric chemistry and climate. However, these impacts are difficult to predict due to uncertainties in isoprene emission estimates. In this talk, I will tackle these uncertainties from two different perspectives. First, I will show that the widely-used isoprene emission model MEGAN sometimes fails to capture the observed variability of the isoprene emission temperature response across a diverse range of ecosystems. I will then show how these errors can be corrected by re-parameterizing the model with observational constraints in a Bayesian data assimilation framework. The re-parameterized model shows increased temperature sensitivity at several field sites, which has important implications for atmospheric composition in a warming climate. Afterwards, I will use the GEOS-Chem global 3D chemical transport model to probe the complex interactions between biogenic isoprene and biomass burning emissions in the Amazon rainforest during the exceptionally large August 2019 fires. I will show that modelled isoprene in the Amazon is highly sensitive to the choice of biomass burning emission inventory used in the simulations due to the strong chemical coupling between atmospheric isoprene and NOx from fires. This coupling is a potentially significant source of error in satellite-based "top-down" isoprene emission estimates which rely on atmospheric chemistry models to estimate emissions from concentration measurements. These model results are also compared with newly available ground-based isoprene retrievals from a Fourier Transform Infrared (FTIR) spectrometer at Porto Velho, Brazil, to help assess the accuracy of the different biomass burning emission inventories.