Login / Signup

Emerging investigator series: quantifying the impact of cloud cover on solar irradiance and environmental photodegradation.

Michelle G NevinsJennifer N Apell
Published in: Environmental science. Processes & impacts (2021)
Environmental photodegradation is dependent on the solar irradiance that reaches the Earth's surface, and photodegradation half-lives of contaminants are typically estimated assuming clear sky (i.e., cloudless) conditions. In this work, the effect of cloud cover on solar irradiance was investigated. Data from the National Renewable Energy Laboratory (NREL), which spanned 3 years of observations (10/2017 to 12/2020), were used to train two machine learning models to predict irradiance based on three inputs - day of year, time of day, and percentage of the sky that was cloudy. Results showed a non-linear relationship between cloud cover and irradiance. Solar irradiance was minimally impacted up to ≈50% cloud cover but decreased by ≈67% at 100% cloud cover. Both random forest and artificial neural network models performed well with relative root mean squared errors of 26-31%, which varied depending on the source of cloud cover data and the spectral region being modeled. Daily irradiance values for a whole year were predicted for varying cloud conditions using the machine learning models; this result was approximated using a quadratic fit of y = 1 - 0.00243x - (4.24 × 10-5)x2 where y is the fraction of clear sky irradiance expected and x is the percentage of cloud cover in the sky. In addition, the model results supported that there was no wavelength dependence for the effect of cloud cover. Therefore, decreases in both direct and indirect photodegradation rates should be proportional to the decrease in irradiance, which has a non-linear dependence on cloud cover.
Keyphrases
  • machine learning
  • neural network
  • big data
  • artificial intelligence
  • emergency department
  • climate change
  • computed tomography
  • magnetic resonance
  • high speed