Developing and exploiting automated methods to estimate spectrally resolved direct normal irradiance for multi-junction concentrator photovoltaic applications
File(s)
Author(s)
Choi, Tsz Hei
Type
Thesis
Abstract
This thesis presents the development of a set of practical tools useful for anyone who may want to develop, deploy or estimate the performance of highly efficient multi-junction (MJ) solar cells in concentrator photovoltaic systems (CPV). These tools use open-source datasets and models to estimate the spectral direct normal irradiance (DNIλ) for any location and for any period. The estimation schemes developed here differ from existing solutions in that a reliable method for properly treat- ing aerosols is established, which is shown to be essential under clear-sky conditions. In conjunction with a solar cell simulator, the schemes enable the accurate performance prediction of MJ CPV systems, critical to their proliferation and the lowering of cost for solar energy.
Amongst the proposed schemes, the Spectral Aerosol Classification Scheme with Circumsolar Factor (SACS + CSF) shows the closest agreement with observations. Under this scheme, the spectral variation of aerosol extinction is represented by an optimally selected aerosol type, whilst the circumsolar irradiance is accounted for using a physics-based look-up table. The simulated broadband DNI carries biases ranging from −0.27 % to 0.52 % (−2 W m−2 to 3 W m−2 ), averaged at +0.068% (< 1 W m−2 ), when validated across five sites. The average spread of error over the five sites is about 2.5 %. When validating the schemes against spectral observations, the average bias ranges from −0.6% in the 868 nm channel to +0.3 % in the 500 nm channel. When coupled with an MJ CPV design in a simulator, it is demonstrated that using SACS + CSF statistically significantly reduces the systematic bias in the power output estimation by order of 10 W m−2 (6 %) when compared to the climatological approach. This is important in the context of assessing long-term MJ CPV resources as power outputs are integrated over multidecadal timescales.
Although the schemes have been developed with AERONET observations, they are designed to use satellite and reanalysis model inputs. This is necessary to enable the estimation of DNIλ globally based on long-term records. However, retrievals from satellites carry higher uncertainties. This motivated an integrated analysis of the non-linear error propagation chain from aerosol products to DNIλ, and to quantities relevant to the performance of MJ CPV cells. It is discovered that factors like surface reflectance and observation zenith, typically thought of as crucial in understanding error in satellite retrievals, become less critical further down the error chain. For MJ CPV resources, it is shown that it is the aerosol typing that primarily dictates the error distribution. This has important implications for the selection of aerosol products for solar resource assessment. For example, products like MODIS and ECMWF MACC may yield better statistics in aerosol optical depth (τ ), but produce less accurate MJ CPV efficiency (η) estimation than, for example, AATSR / ORAC.
The error propagation analysis also revealed that high aerosol loading increases the sensitivity of MJ CPV η to error for coarse aerosols. Continental polluted and urban aerosols with high aerosol loading are also conditions where η is more susceptible to errors in τ and Angstro ̈m exponent (α). η has minimum sensitivity to error under maritime aerosol conditions. Ultimately, it is concluded that for most climatologies, typical uncertainties of α and τ in aerosol products propagated into η produce an error that is less than 10 %, the typical magnitude of the spectral effects for MJ CPV. As such, the use of satellite observations for estimating MJ CPV resources is suitable in most cases.
To further illustrate the importance of spectral DNI, the impact of the COVID-19 lockdown and air pollution changes on MJ CPV resources is studied over China using satellite retrievals of NO2 and τ. It is estimated that the combined changes in atmospheric species lead to a 19.8 ± 8.5 % (62 W m−2 ) increase in broadband surface DNI, along with a shift in the spectral distribution towards bluer wavelengths. Feeding these changes into an MJ CPV simulator results in a 29.7±8.5% (29 W m−2) increase in the power output. Around one-third of the increase in power output arises from the improved spectral matching. These results indicate that cleaner atmospheres can have disproportionate co-benefits for MJ CPV technologies, which has important implications for air pollution and climate change policies.
Amongst the proposed schemes, the Spectral Aerosol Classification Scheme with Circumsolar Factor (SACS + CSF) shows the closest agreement with observations. Under this scheme, the spectral variation of aerosol extinction is represented by an optimally selected aerosol type, whilst the circumsolar irradiance is accounted for using a physics-based look-up table. The simulated broadband DNI carries biases ranging from −0.27 % to 0.52 % (−2 W m−2 to 3 W m−2 ), averaged at +0.068% (< 1 W m−2 ), when validated across five sites. The average spread of error over the five sites is about 2.5 %. When validating the schemes against spectral observations, the average bias ranges from −0.6% in the 868 nm channel to +0.3 % in the 500 nm channel. When coupled with an MJ CPV design in a simulator, it is demonstrated that using SACS + CSF statistically significantly reduces the systematic bias in the power output estimation by order of 10 W m−2 (6 %) when compared to the climatological approach. This is important in the context of assessing long-term MJ CPV resources as power outputs are integrated over multidecadal timescales.
Although the schemes have been developed with AERONET observations, they are designed to use satellite and reanalysis model inputs. This is necessary to enable the estimation of DNIλ globally based on long-term records. However, retrievals from satellites carry higher uncertainties. This motivated an integrated analysis of the non-linear error propagation chain from aerosol products to DNIλ, and to quantities relevant to the performance of MJ CPV cells. It is discovered that factors like surface reflectance and observation zenith, typically thought of as crucial in understanding error in satellite retrievals, become less critical further down the error chain. For MJ CPV resources, it is shown that it is the aerosol typing that primarily dictates the error distribution. This has important implications for the selection of aerosol products for solar resource assessment. For example, products like MODIS and ECMWF MACC may yield better statistics in aerosol optical depth (τ ), but produce less accurate MJ CPV efficiency (η) estimation than, for example, AATSR / ORAC.
The error propagation analysis also revealed that high aerosol loading increases the sensitivity of MJ CPV η to error for coarse aerosols. Continental polluted and urban aerosols with high aerosol loading are also conditions where η is more susceptible to errors in τ and Angstro ̈m exponent (α). η has minimum sensitivity to error under maritime aerosol conditions. Ultimately, it is concluded that for most climatologies, typical uncertainties of α and τ in aerosol products propagated into η produce an error that is less than 10 %, the typical magnitude of the spectral effects for MJ CPV. As such, the use of satellite observations for estimating MJ CPV resources is suitable in most cases.
To further illustrate the importance of spectral DNI, the impact of the COVID-19 lockdown and air pollution changes on MJ CPV resources is studied over China using satellite retrievals of NO2 and τ. It is estimated that the combined changes in atmospheric species lead to a 19.8 ± 8.5 % (62 W m−2 ) increase in broadband surface DNI, along with a shift in the spectral distribution towards bluer wavelengths. Feeding these changes into an MJ CPV simulator results in a 29.7±8.5% (29 W m−2) increase in the power output. Around one-third of the increase in power output arises from the improved spectral matching. These results indicate that cleaner atmospheres can have disproportionate co-benefits for MJ CPV technologies, which has important implications for air pollution and climate change policies.
Version
Open Access
Date Issued
2021-07
Date Awarded
2021-10
Copyright Statement
Creative Commons Attribution NonCommercial Licence
License URL
Advisor
Brindley, Helen
Sponsor
Imperial College London
Natural Environment Research Council (Great Britain)
Publisher Department
Physics
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)