The increase in computational power and richness of Earth system data has allowed new methods for simulating natural processes with higher precision and accuracy than previously imagined. Older methods to increase skill of computer model simulations include parameter inference, where the parameters of a forward simulation model are optimized to better represent reality and allow the model to capture dynamics seen in the observed data. However, these methods are limited by our physical understanding of the underlying system, making it impossible to capture certain dynamics when the model is under-represented. Machine learning methods have emerged as a potential tool to bypass the limitations of our physical understanding, and they can create simulations with much higher skill than previous methods. This work investigates and compares the skill of photosynthesis simulations from various model formulations including those with optimized parameters and those from machine learning.