Take any low-carbon energy scenario, and you will find an increasing and potentially considerable amount of solar photovoltaic panels in it. But to what extent can we actually deploy this technology?
Technical factors, like solar radiation and buildings geometry, provide potential upper bounds, while past and expected economic trends of solar technical change influence how close we get to those upper bounds. One question remains: are we representing all the relevant factors in projecting solar deployment rates?
A team of engineers at Stanford just published on Joule interesting insights which provide a novel way to answer to this question. Focusing on the United States, they train a deep learning model to recognize and measure residential photovoltaic panels from satellite imagery. The result is a large database mapping density and extent of current solar installations all over the US. Then, they correlate this information with economic and environmental factors.
The authors quantify several enabling and limiting factors of solar deployment. For example, solar deployment peaks at a given population density, activates only above a certain solar radiation level, is hindered despite its profitability in areas with low or inequitably distributed income, and saturates in high-income areas. These and other factors are then included in a comprehensive predictive model of solar deployment.
We are likely to see an increasing amount of applications of these novel techniques to mine new data and insights on how we consume and produce energy. And these efforts constitute precious opportunities for integrated assessment modellers to improve and validate the realism of their scenarios.