Estimating Solar Radiation at the Ground from Space (Clouds, Turbidity); Measuring it Directly at the Ground (Invited Presentation)

Broadband and spectral solar irradiance measurements and surface estimates from satellites are necessary to understand the spatial and temporal variability of solar energy at the surface. The only absolute instrument for solar irradiance is an Absolute Cavity Radiometer (ACR). The World Radiometric Reference (WRR) in Davos, Switzerland is represented by six ACRs. Expense and impracticalities prevent ACR use for routine monitoring, however, calibrations of all monitoring solar radiometers should be traceable to the WRR. The few U.S. high-quality surface radiation networks with such traceability and quality instrumentation (e.g., SURFRAD) are deficient in regional climate representation. All efforts to improve this situation have been rejected. Nevertheless, high quality surface solar measurements have been successfully applied to remedy NWP problems, solar prediction model deficiencies, satellite validation, climate studies, etc.

Solar radiometers have either thermopile or silicon cell detectors. Thermopile detectors are best because their spectral response is limited only by the transmissivity of the radiometer’s protective dome or window, which is flat and >90% over the solar spectrum at the surface (~260 to 2500 nm). Silicon cell detectors have a faster response but are only sensitive from 300 to 1100 nm and their spectral response is not flat; it is mound-shaped, peaking at ~940 nm. In practice, radiometer calibrations are set at 45° solar elevation, however, when the solar beam is present radiometer responsivity varies with solar zenith angle (cosine response). The practice of applying the 45° calibration to all measurements is error free for overcast conditions, but induces “cosine” errors under clear and partly cloudy skies. Infrared cooling of the detector under clear skies is another error source. Cosine and thermal offset errors nearly vanish when summing direct and diffuse thermopile-based measurements for global horizontal irradiance (GHI). Silicon cell detectors also have offset and cosine errors as well as a temperature dependency that is usually not accounted for. Further, since the diffuse light of a clear sky is short-shifted (blue), the poor sensitivity of the silicon cell at short wavelengths invokes error when applying a broadband calibration to a shaded (diffuse) measurement. Correction schemes for this problem carry associated uncertainty. Consequently, silicon cell detectors have greater uncertainty (±6.6% for GHI and ±10.5% for direct + diffuse), whereas thermopile-based direct + diffuse measurements are less uncertain (±2.3%) and are preferred.

The seven U.S. SURFRAD sites have one-minute daytime sky cover measurements from Total Sky Imagers (TSI). Their images’ 160° FOV produces a high bias in cloud fraction of approximately 10% due to cloud overlap near the horizon. Reprocessing TSI images for a 100° FOV centered on zenith would be more representative of the actual cloud fraction and a better match to cloud fractions from the satellite perspective.

NOAA polar orbiter, GOES, and NASA CERES imagery have been used to estimate surface solar irradiance. Comparisons to surface measurements have shown that they do well for clear conditions but are less accurate for partly cloudy, overcast, and snow cover scenes. GOES surface solar estimates (GSIP) have a bias of 7 Wm^-2 with a standard deviation of 87 Wm^-2, and show systematic overestimates for cloudy scenes and underestimates for clear scenes. The current GOES imager has only one channel in the solar spectrum, precluding the retrieval of atmospheric and surface properties. Its imagers are degrading and adversely affecting GSIP quality. The Advanced Baseline Imager (ABI) on the new GOES-R satellite has six solar channels that will permit derivation of atmospheric and surface properties. Tests performed using 10+ years of MODIS data as an ABI proxy show similar uncertainties as GSIP, but reduced high and low biases for cloudy and clear scenes, respectively. In anticipation of GOES-R, spectral albedo and more accurate solar beam measurements have been added to SURFRAD sites in 2015 and 2016.

Satellite-based aerosol optical depth (AOD) is available from GOES, MODIS, MISR, AVHRR on POES, and VIIRS on NPP/JPSS. Historically, AOD from AVHRR has been limited to the oceans, but lately it has been improved to include spectrally dark land areas. Though, based on comparisons with AERONET, satellite-based AOD over land has twice the uncertainty of that over the oceans. The current GOES produces 550 nm AOD for the CONUS that is accurate to 30% for AOD > 0.1. Traditionally, bright surfaces such as deserts and snow-covered regions have been masked out because their surface properties do not conform to AOD retrieval assumptions. However, NASA has added a separate “Deep Blue” algorithm to MODIS AOD processing for bright land surfaces that has uncertainties comparable to other MODIS AOD. Similarly, the NOAA VIIRS algorithm also has the capability to retrieve AOD over bright desert regions. None of these new AOD algorithms for land areas work over snow. The ABI on GOES R will produce AOD at higher temporal and spatial resolution than the current GOES. Spectral information will enable calculation of the Angstrom Exponent for particle size and aerosol typing, but only over water. There is no known solution for that shortcoming.

NCEP does not currently assimilate surface radiation measurements into its NWP models. However, the Rapid Refresh Model (RAP, 13 km resolution) and its high-resolution version HRRR (3 km resolution) have benefitted from surface radiation measurements. Specifically, during the NOAA/DOE Solar Forecasting Improvement Project, comparisons to SURFRAD measurements played a major role in resolving a positive surface air temperature bias in those models. In a feedback loop, erroneously high model solar irradiance led to a deeper and drier PBL (which affected the low level wind), a reduction of model clouds, spurious high-based convective initiation, and greater surface temperatures. Mitigating steps included a change from the Goddard SW radiative transfer algorithm to the more efficient RRTMG SW scheme, improved “aerosol-aware” cloud microphysics with coupling to radiation, better accounting for radiative effects of subgrid scale clouds, and better land surface physics. These reduced the mean solar radiation bias of the models by about half. Future work with the Wind Forecasting Improvement Project will be to evaluate the role of radiation in PBL formation and turbine-height winds for better wind forecasts for renewable energy applications.