Photosynthesis-irradiance (PI) relationships are important for phytoplankton ecology and quantifying carbon fixation rates in the environment. However, the parameters of PI relationships are typically unknown across space and time. Here we use machine learning, satellite remote-sensing, and a global database of in-situ PI relationships to build models that predict PI parameters globally as a function of satellite-observed variables. Using only surface light, temperature, and chlorophyll, we achieve an R² of 76% for predicting photosynthesis rates at saturating light (𝑃^B_max) and an R² of 58% for predicting the light saturation parameter (𝐸_k). Predictability is maximized when averaging covariates over monthly timescales, indicating that environmental history and community turnover are important for predicting PI relationships. These results will help improve satellite-based primary production models, quantify emergent timescales in photosynthetic communities, and provide global estimates of photosynthetic parameters that can be used in plankton ecology and biogeochemistry models.