Satellites like NASA’s OCO-2 collect important information about Earth’s surface, such as solar-induced chlorophyll fluorescence (SIF), which helps researchers study plant health and carbon uptake. However, these measurements are often taken over large and irregular areas, making it difficult to understand fine-scale patterns—especially in complex regions like cities.

We present a new method that uses Bayesian statistics to estimate high-resolution maps from these coarse satellite observations. Our approach treats the satellite data as averages over a fine-scale grid and uses a flexible model that accounts for spatial patterns and land cover. By drawing connections between two popular types of spatial models—discrete Spatially Autoregressive (SAR) and continuous Matérn models—we create a system that is both accurate and computationally efficient.

We apply this method to SIF data collected over Boston, showing that it can produce detailed maps that reveal differences across neighborhoods. To test our method’s accuracy, we simulate satellite-like measurements from high-resolution albedo data (a measure of surface reflectivity) and show that our model successfully recovers the original fine-scale patterns. This work offers a new way to downscale satellite data while providing reliable estimates of uncertainty, helping researchers make better use of Earth observation data in cities and other heterogeneous landscapes.