Adapting WRF for Solar Forecasts Clear sky conditions - more often, less error, most easily fixed Cloudy conditions – less frequent, more error All errors.

Adapting WRF for Solar Forecasts• Clear sky conditions - more often,

less error, most easily fixed• Cloudy conditions – less frequent,

more error• All errors worse for DNI than GHI

Clear sky conditions• Error mostly a function of water vapor and aerosols• Models do not do a good job handling aerosols• Background aerosols, locally generated dust, and then both are magnified by

increased humidity.• Aerosol source for Tucson mostly locally generated by human activity or wind• Wind direction also matters• Current version of algorithm to correct for this reduces clear sky error from a

monthly average of about to about (10-3.3%)• Nearly eliminates bias in both clear and cloudy sky conditions• Should improve as study gets more data with time

May clear sky average error by time of day Forecast-Observed

-100

-50

0

50

100

150

200

250

Sunrise Noon Sunset

Erro

r ()

June average error by time of dayclear sky, no smoke

-100

-50

0

50

100

150

200

Sunrise Noon Sunset

Erro

r ()

Cloudy Conditions

• Significantly more variability• Can break up clouds into different cases with

different physics• Dust algorithm nearly eliminates bias in July,

suggesting total cloud mass is proper• Error mostly related to time and location of

convective clouds• Slight underprediction of winter clouds? We

will get more data on this later this year.

Current all sky(cloudy+clear) dust/dewpoint correction stats

Forecast minus actual, units

Month Bias(total error) Average error:

uncorrected May 83.1398 138.59

corrected May -4.48947 81.2545

uncorrected June 91.3892 141.603

corrected June -0.0650794 78.5955

uncorrected July 96.5335 272.241

corrected July -18.3734 221.625