Generation of Langley Voltage Intercepts

Last Updated: September 12, 2008

During the first week of each month new Langley voltage intercept data are generated by submitting QC-screened data, collected over the previous month, to a program named la (Langley Analyzer). The program originated at ASRC, Atmospheric Solar Radiation Group at SUNY-Albany NY, and uses the methodology described in Harrison and Michalsky (1994). In short, the method performs regressions of bias- and cosine-corrected direct normal voltages, expressed as ln(mV), on airmass and subjects them to a sequence of statistical tests. A Langley analysis is attempted for each morning and afternoon period corresponding to the time of a selected airmass range. From each successful final regression, comprised of points that pass all of the tests, a Langley voltage intercept value (V0) is taken as the intercept of the line at airmass 0; the slope of the line is the average optical depth.

The program has parameters that can be set by the user. These are defined as:

  • LSfitSD: Limit of the standard deviation of the final regression
  • LowAM: Lower limit of the airmass range
  • HighAM: Upper limit of the airmass range
  • OutLimit: Number of standard deviations allowed in each least squares test
  • CloudSlop: Flexibility in the cloud passage test
  • FracPts: Fraction of points in the airmass range that must remain after other tests have completed
  • Key: If a key channel is specified, all tests are performed only on that channel. If enough points remain for a Langley plot to be produced, those same points are used to derive plots for the other channels.

Some of the settings we use vary by wavelength range: 
	Wavelengths 	Airmass Range 	LSfitSD
	300 - 317 	1.2 - 2.2 	0.009
	325 - 368 	1.5 - 3.0 	0.009
	415 - 940 	2.0 - 6.0 	0.006
Four other parameters are set as follows for all wavelengths:
  • FracPts = 0.33333
  • CloudSlop = 0.0
  • OutLimit = 1.5
  • Key Channel = none

The results of analyses with 12 or more points remaining in the final regression are entered into a database table. The following values are stored:

  • Raw V0
  • Normalized V0 (normalized to remove variation due to Earth-Sun distance)
  • Optical depth
  • Standard deviation of the final regression
  • Number of points in the period (morning or afternoon)
  • Number of points in the selected airmass range
  • Number of points used in the final regression
  • Start and end times (site local standard time) of the airmass range

Estimation of Daily Langley Voltage Intercept for Langley Calibration and Other Computations

Langley Voltage Intercepts(V0s) are needed to convert measured voltages from a given MFRSR instrument to irradiances using the Langley calibration method and to compute aerosol optical depths and column ozone. Ideally, Langley analysis of the MFRSR data would produce two V0 values (one for morning, another for afternoon) for each channel for each day of instrument operation. Also ideally, since it represents the voltage that would be measured by the instrument if it were placed at the top of the atmosphere, V0 would be constant. In reality neither of these ideals is met. The density of V0 points varies greatly from site to site, being quite sparse at the more turbid sites; and changes in both atmospheric conditions and instrument function can result in significant variation and drift.

Since V0 values are needed for every day, and since variability must be considered, estimated daily V0 values are predicted from a timeseries of Langley-generated V0s, accumulated over the entire deployment period of a given instrument at a given site. To be included in the timeseries, we require that V0s be:

  • normalized (to remove variation due to Earth-Sun distance)
  • from a morning period
  • from a successful Langley analysis comprised of at least 12 points that passed all of the Langley tests
To accomplish this for the entire history of the network, the period of operation of each site is subdivided according to the deployment periods of the various MFRSR units used. When necessary a given deployment period is further subdivided upon significant changes in board gains. Then to each time period a suitable prediction method is assigned. Currently, the default method is linear regression.
To predict V0s using linear regression, a sequence of two Least Squares tests is performed. Points that fall outside two standard deviations from the first regression are omitted; then a second Least Squares fit is applied to the remaining points. The resulting linear equation is used to predict a V0 value for each day. No predictions are made in cases where there are fewer than four points in a timeseries.
Each month after the Langley Analyzer program (la) has produced a new set of V0s from the previous month's data, the predicted daily V0s for all instruments currently operating are recomputed with the additional Langley points included.
Note: Before they are used for calibration or computational purposes, the daily predicted V0 values are converted back to the raw form (as opposed to normalized for Earth-Sun distance).