Understanding particulate matter pollution and photovoltaic power output using data and models

Abstract

Particulate matter (PM) both in the atmosphere and deposited on solar photo-voltaic (PV) panels reduce PV energy generation efficiency. Atmospheric PM near the surface, particularly those with an aerodynamic diameter ≤ 2.5 µm (PM2.5), have well-documented deleterious impacts on human health. It is imperative to develop models to accurately simulate PM impacts on PV efficiency in order to enlighten relevant policy-making aimed at reducing these impacts. It is important to develop models to accurately capture magnitudes and variabilities of ground-level PM2.5 concentrations in order to inform follow-on epidemiological studies. In this thesis I use a combination of satellite observations, in situ measurements, process-driven models, and data-driven models to understand PM pollution and PV power output with a focus on their links from global to regional scales.

In Chapter 2, I integrate the GEOS-Chem global 3-D model of atmospheric composition, equipped with online radiative transfer calculations, with PVLIB-Python which is a solar PV performance model, to quantify PV efficiency losses due to atmospheric and deposited PM. I calculate three PV efficiencies: 1) real PV efficiency considering atmospheric and deposited PM; 2) hypothetical PV efficiency considering atmospheric PM only; and 3) hypothetical PV efficiency assuming no PM. By comparing these PV efficiencies, I find that regions with low PV efficiency are typically associated with high PM-induced PV efficiency losses, and that the losses due to deposited PM far exceed those due to atmospheric PM, with the maximum magnitude of the former almost eight times that of the latter. Desert regions including the Sahara, Arabian-Peninsula, Central Asia, and Southern South America are most susceptible to deposited PM which causes PV efficiency losses that are comparable to the maximum PV efficiency close to 0.3 achieved elsewhere. Coastal regions are also significantly affected by deposited PM, e.g. countries around the Caribbean and the Mediterranean, and over New Zealand. The main regions where PV efficiency losses due to atmospheric PM are East and South Asia, particularly over highly polluted regions such as North China and the Indo-Gangetic Plain.

In Chapter 3, I focus on developing strategies to mitigate PM-induced PV efficiency losses. From the perspective of reducing emissions, I calculate the aforementioned three PV efficiencies by halving anthropogenic source sector emissions. By subtracting each of these quantities from baseline values and comparing these differences across source sectors, I find that that reducing residential emissions is the most effective approach to reduce PM-induced PV efficiency losses, and that the biggest PV efficiency gains are over East and South Asia. Using 2019 PV capacities as a baseline, I find that a 50% reduction in residential emissions would lead to an additional 7,687 and 1,823 GWh yr−1 produced in China and India, respectively. The corresponding economic benefits would amount to US$653 million and US$144 million per year, respectively. From the perspective of removing deposited PM from solar panels either manually or by robots, I calculate the aforementioned first PV efficiency by taking into account yearly, quarterly, monthly, weekly, and daily sweeping of panels. By subtracting the baseline value from this quantity and comparing these differences across frequencies, I find that routine sweeping of panels is effective at reducing PV efficiency losses due to deposited PM, and that even an annual sweeping routine could lead to an approximately 40% PV efficiency recovery in Central Asia and Southern South America. The work described in previous and current chapters are collectively and succinctly summarized into a manuscript that is under review for Environmental Science & Technology.

In Chapter 4, I go a step further by fully quantifying the benefits to air quality and PV power output from reducing residential fuel emissions. I run the integrated model at a finer resolution of 0.5◦ latitude × 0.625◦ longitude with a focus on an Asian wintertime (January 1 to February 29, 2008) which is the main spatiotemporal domain where reducing residential emissions substantially benefit PV power output. I run the integrated model with original emissions to output baseline ground-level PM2.5 concentrations, column-integrated aerosol optical depth (AOD) at wavelength of 550 nm (AOD550nm), and the aforementioned three PV efficiencies. I then run the integrated model again but with reduced residential fuel emissions to determine the benefits to air quality and PV power output from reducing these emissions by taking the difference of baseline and sensitivity simulations. I find that Eastern and Northeastern China, and the Indo-Gangetic Plain are the three key regions where reducing residential particularly solid biofuel emissions leads approximately linearly to reductions in PM pollution and improvements in PV efficiency. Completely removing residential emissions would result in 20–30% and 30–40% reductions in column-integrated AOD550nm and ground-level PM2.5 concentrations, respectively, and an approximately 30% improvement in PV efficiency. I attribute these approximately linear benefits to the large volume of carbon emissions, primarily owing to the low combustion and thermal efficiencies of residential devices and the general absence of any end-of-pipe controls, that typically form into carbon aerosols in an approximately linear way in the atmosphere after aggregating to temporal and spatial mean values.

This work is being prepared for submission.

In Chapter 5, I use the GEOS-Chem 3-D model of atmospheric composition to improve our ability to use satellite-retrieved column-integrated AOD and statistical and machine learning models to infer ground-level PM2.5 concentrations. In particular, I use the GEOS-Chem global 3-D model of atmospheric composition to explore how changes in the vertical distribution of aerosol extinction coefficients affects the relationship between ground-level PM2.5 and column-integrated AOD and how I can use that information to improve the robustness of ground-level PM2.5 estimates inferred from satellite-retrieved AOD over eastern China. I define a metric, ΓAOD P BL , that describes the fraction of AOD that resides in the planetary boundary layer compared to the total columnar AOD. I determine physically-meaningful PM2.5:AOD relationships using data for which ΓAOD P BL ≥ 50%, a criterion based on sensitivity analyses on data clusters that I identify using a hierarchical clustering method. I use statistical and machine learning models to describe these PM2.5:AOD relationships, and use a Monte Carlo approach to quantify the improvement after the selection of more physically relevant data records.

Benefiting from the improved representativeness of AOD for ground-level PM2.5, the new method effectively reduces bias in inferred estimates of ground-level PM2.5 by 10–15% and 9–12% for space-borne sensors passing over in the morning and afternoon, respectively. It also captures more variations in ground-level PM2.5 by up to 8% and 5% for space-borne sensors passing over in the morning and afternoon, respectively. This work is published in Atmospheric Environment. In closing, the above findings advance understanding of PM pollution and PV power output particularly their links, and provide motivation for future studies of evaluating impacts of real emission controls and climate change on past, current, and future power output and intermittency of more types of PV panels and vice versa using improved models.