The United Nations’ Paris Agreement forces nations and industries to transition from conventional hydrocarbon-based energy sources to more sustainable and less-emitting alternatives such as geothermal energy. Geothermal doublets are used to provide a source of green heat to greenhouses and residential buildings. In industry, reservoir simulators are used to make predictions about the energy production and thermal lifetime of such a geothermal project; however they can be computational expensive. Upscaling approximates the expensive fine-grid simulation and reduces the necessary computational power, but is by definition not mathematically exact. This thesis tests a newly derived thermal dispersion-based upscaling method, referred to as Taylor-based upscaling, and compares it to conventional arithmetic upscaling and a reference fine-grid simulation. The upscaling methods are applied on reservoir descriptions from the Margretheholm-1A and the HON-GT-01 reservoirs, and simulations are done using Delft Advanced Research Terra Simulator (DARTS). The reservoirs are modelled as 2D layer-cake models in the form of a geothermal doublet with a fine longitudinal resolution, 1000 grid blocks between injector and producer, to minimize numerical dispersion. The results are presented as 2D-temperature profiles and breakthrough-curves. The deviations are quantified in the form of L2-norm calculations and by comparing the amount of days it takes for each upscaling method to reach a 1◦C or 5◦C drop in production temperature. The results show that arithmetic upscaling underestimates the spreading of the cold-temperature front leading to later breakthrough, whereas Taylor-based upscaling overestimates this spreading. For the Margretheholm-1A reservoir Taylor-based upscaling mimics the fine-grid reference simulation better compared to arithmetic upscaling, whereas arithmetic averaging performs better for the HON-GT-01 reservoir. Increasing the vertical thermal conductivity (kz ) leads to Taylor-based upscaling coming closer to the fine-grid reference simulation; however different reservoir descriptions require a different kz adjustment. In conclusion, the Taylor-based upscaling method does not outperform conventional arithmetic upscaling for all reservoir descriptions. Also, no universal rule was found for increasing the kz -value to improve the Taylor-based upscaling method.