An aerosol optical depth product for NOAA's SURFRAD network

The MultiFilter Rotating Shadowband Radiometer (MFRSR) is used for visible spectral measurements at NOAA's Surface Radiation Budget (SURFRAD) network. Calibration of the MFRSR channels for aerosol optical depth (AOD) is obtained from 0-airmass linear extrapolations of Langley plots (the natural log of the output voltage versus air mass) that are derived from cloud-free MFRSR measurements. To automatically identify pristine measurements for calibration Langley plots, the SURFRAD algorithm cross references times identified as cloud free by the Long-Ackerman clear-sky identification method, which uses SURFRAD broadband solar measurements, with the times of MFRSR measurements. Such plots represent first-cut Langley calibrations that are generally free of cloud contamination. The few bad calibration Langley points that survive the first cut are rejected by deviation-from-linearity tests that follow. Time series of Langley calibrations and associated errors are constructed for the duration of a particular MFRSR's deployment. Several channel-specific Langley calibrations computed within a two-month period are normalized to unit solar distance and then averaged to derive mean channel calibrations and errors that represent that period. A standard "propagation of error" method is applied to compute error associated with the mean Langley calibration. The rejection of outliers using standard statistical methods further minimizes that error.

To ensure smoothly varying channel calibrations for a particular instrument, the two-month "representative" Langley calibrations are analyzed in time series over two-year periods. Because channel sensitivities drift slowly, linear fits to the two-year time series of two-month Langley calibrations were attempted. However, a slight periodicity about the best-fit line resembling the annual temperature cycle consistently appeared for all MFRSRs and all channels. We concluded that the perceived variance around the linear fit was not random error; rather, it was a consequence of the instrument's temperature dependence. Therefore, a periodic function was fit to the time series. The time series of error associated with the two-month Langley calibrations did not show the same annual periodicity and therefore was fit to a linear expression. All two-year fits were overlapped by four months on either end to ensure smooth transitions of the Langley calibrations and associated error.

Daily AOD files are produced for each station. The first step in the AOD calculation is the retrieval of channel-specific calibrations for the day being processed from the best-fit equations that represent their drift and intra-annual variation. Channel calibrations for that day are then corrected to the earth-sun distance (that was removed in the calibration process) appropriate to the day being processed. The calibrations and each MFRSR measurement are then applied to the Beer-Lambert law to compute a daily time series of total optical depth for each channel. Aerosol optical depth for each channel is computed by subtracting the contributions of molecular scattering and ozone absorption specific to each channel from the total optical depth. Ozone absorption coefficients are chosen based on the central wavelengths specific to the MFRSR that made the measurements. Total ozone for the location and date being processed is automatically obtained from a NASA/TOMS web site. Molecular scattering is computed based on the MFRSR central wavelengths using the station pressure measured for the time of the AOD calculation. Contributions by nitrogen dioxide absorption in the spectral channels of the visible MFRSR are negligible and thus ignored. Last, the daily time series of AOD is subjected to cloud screening because AOD can only be computed for times that the sun's beam is cleanly sensed by the MFRSR. The daily AOD product for a particular SURFRAD station is organized in local standard time. Each row represents a two-minute MFRSR measurement time and contains AODs and errors for each spectral channel. Nominal errors for the 500 nm channel AODs are ±0.01 (one standard deviation), but those errors vary systematically with solar zenith angle. Typically, the smaller the solar zenith angle, the larger the error. A quality control parameter is also listed on each line; a 0 indicates the AODs on the line are acceptable, whereas a 1 means that the AOD's on that line are likely contaminated by clouds and thus not usable.

The SURFRAD AOD analysis method includes several unique features that minimize error and facilitate the calibration and AOD calculation procedures. The once painstaking task of identifying calibration Langley plots is uniquely automated in the SURFRAD algorithm by cross-referencing the MFRSR measurements with the SURFRAD clear-sky product. Use of the clear-sky product ensures that only the most pristine sky conditions are used for Langley calibrations. Identifying mean channel-specific Langley calibrations that represent two-month periods and grouping them over two-year periods allows for the resolution of, and removal of, the temperature dependence of the MFRSR, thus further reducing error in the ultimate product. Finally, the use of SURFRAD station pressure for accurate molecular scattering calculations at the resolution of the measurements, and the automatic acquisition of daily total ozone over each station for spectral ozone absorption calculations, further improve the accuracy of the SURFRAD AOD algorithm. Provisional AOD data for SURFRAD stations can now be computed. However, before a final product can be released, the central wavelengths of the MFRSRs that have been used at SURFRAD stations must be checked and verified.