DESC0024957

Solar panels get dirty and less efficient even in regions with frequent rain. This issue, known as soiling, causes significant revenue losses for solar farms. When a new solar farm is being planned (the development phase), accurate energy generation forecasts are critical to the financial viability of the proposed asset. Soiling losses are a key input to energy generation forecasts but estimations from current providers have a high degree of uncertainty. We currently use a comprehensive approach to address soiling for sites that are already built (operational sites) that employs a predictive soiling model informed by data analytics and a proprietary method utilizing a digital microscope to image particles on the panel surface. Using the vast and novel dataset from the current solar farms we are analyzing, we plan to improve our predictive soiling model such that it will be accurate for development-phase sites (where data analytics and the microscopy images are not available). The overall approach to model improvement involves three steps. First, we will improve predictions on how precipitation events reduce or enhance soiling losses. Next, we will utilize a novel machine learning approach to quantify the impact of highly-localized emission sources. Finally, we will model how pollen and fungal growth contribute to soiling losses that are tightly adhered to the panel surface. Each of these three research tasks will be integrated into the current predictive soiling model in Phase I and Phase II. Soiling is a major contributor to financial risk for solar financiers, slowing the adoption of solar. The resulting soiling model will estimate soiling to within 25%, reducing the financial risk of building utility and commercial scale solar assets and accelerating the energy transition. When the economic attractiveness of solar is improved, our energy infrastructure, energy affordability, air quality, and climate will all benefit.