CIMEL CE318

O

ur sun photometer instrument, CIMEL Sun Photometer CE318, is #237 in the NASA AERONET (AErosol RObotic NETwork.) This microprocessor controlled, stepper motor positioned robot has a two component optical head containing the sun collimator without lens and the sky collimator with lenses and filter wheel. Sun tracking is controlled with a 4-quadrant detector. Data is temporarily stored in memory and once every hour uploaded to NASA via a GOES-E (geosyncronous) satellite uplink.

The instrument directly measures the incoming solar energy (sky radiance) at selected wavelengths of 340, 380, 440, 670, 870, 1020 nm (aerosols), and 936 nm (water vapor). Light is absorbed and scattered by atmospheric gasses, water vapor and aerosols. The concentration of different atmospheric components can be determined by the attenuation at wavelengths which are strongly absorbed or scattered. Total Optical Depth (TOD) is the sum of the Rayleigh Optical Depth (from atmospheric gasses, e.g. Nitrogen, Oxygen, Argon), the Ozone Optical Depth, the Mixed Gas Optical Depth (e.g. Carbon dioxide, oxides of Nitrogen), the Water Vapor Optical Depth and the Aerosol Optical Depth (AOD). AOD is of the most interest because it cannot be directly measured by other methods, and is determined by subtraction of the other quantities. Aerosol phase functions and size distribution can also be inferred. The sun photometer measurements are also important as standards for satellite-derived values of AOD. 

Accurate knowledge of the spatial and temporal extent of aerosol concentrations and properties has been a limitation for assessing their influence on satellite remotely sensed data (Holben et al., 1992) and climate forcing (Hansen and Lacis, 1990). With the exception of the AVHRR weekly ocean aerosol retrieval product (Rao, et al., 1989), the voluminous 20 year record of satellite data has produced only regional snapshots of aerosol loading and none have yielded a data base of the optical properties of those aerosols which are fundamental to our understanding of their influence on climate change. With the advent of the EOS era, new algorithms for global scale aerosol retrievals and their application for correction of remotely sensed data will be implemented (Kaufman and Tanré, 1996). However the prospect of fully understanding aerosols influence on climate forcing is small without validation and augmentation by ancillary ground-based observations as can be provided by radiometers historically known as sun photometers. Following is a description of a new sun-sky scanning radiometer system that standardizes ground-based aerosol measurements and processing, can provide much of the ground-based validation data required for future remote sensing programs and may provide basic information necessary for improved assessment of aerosols impact on climate forcing.

Background

Ground-based atmospheric aerosol measurements using sun photometry has changed little since Volz (1959) introduced the first handheld instrument more than three decades ago. Modern units can collect data more accurately and quickly but the method remains largely the same, that is a filtered detector measures the spectral extinction of direct beam radiation according to Beer's Law:

V[lambda] = Vo[lambda] d2exp-([tau][lambda]m)*ty (1)

where

V=Digital voltage

Vo=extratrestrial voltage

m=optical airmass

[tau]=total optical depth

[lambda]=wavelength

d=ratio of the average to the actual earth-sun distance

ty=transmission of absorbing gases

The digital voltage (V) measured at wavelength ([lambda]) is a function of the extraterrestrial voltage (Vo) as modified by the relative earth sun distance (d) and the exponent of the total spectral optical depth ([tau][lambda]) and optical airmass (m). The total spectral optical depth is the sum of the Rayleigh and aerosol optical depth after gaseous correction.

The multi-filter rotation shadowband radiometer (which is commercially available) employs a different strategy. It measures spectral total and diffuse radiation to obtain the direct component from which aerosol optical thickness is computed using Beer's law. The instrument nominally measures at one minute intervals and has been shown to be reliable over long periods of time. The measurements are networked to a common server by a modem interface and the data processed by a common analysis system (Harrison et al., 1994). It is widely used in the United States principally for the DOE ARM sites. As the number of measurements from the shadowband network increases, the impact of aerosol loading on the radiation balance should be more clearly understood especially when taken in concert with other ground, airborne and satellite measurements.

Sky scanning spectral radiometers, that is radiometers that measure the spectral sky radiance at known angular distances from the sun, have expanded the aerosol knowledge base most importantly through inversion of the sky radiances to derive aerosol properties such as size distribution and phase function (Nakajima et al. (1983, 1996), Tanré et al. (1988), Shiobara et al. (1991), and Kaufman et al., (1994)). This technique requires precise aureole measurements near the solar disk and good straylight rejection. Historically these systems are rather cumbersome, not weather hardy and expensive. The CIMEL and PREDE (French and Japanese manufacturers respectively) sun and sky scanning spectral radiometers overcome most such limitations, and provides retrievals from direct sun measurements of aerosol and water vapor abundance in addition to aerosol properties from inversion of spectral sky radiances. Since the measurements are directional and represent conditions of the total column atmosphere, there are direct applications to satellite and airborne observations as well as atmospheric processes.

As has been demonstrated by the shadowband network and satellite remote sensing in general, prompt delivery of the data for analysis is fundamental for obtaining a comprehensive, continuous data base and allows assessment of the collecting instruments health and calibration. To achieve this goal, minimize costs and expand the coverage globally, we use the simple and inexpensive Data Collection System (DCS) operating on the geosynchronous GOES, METEOSAT and GMS providing nearly global coverage in near real-time at very little expense (NOAA/NESDIS, 1990).

Finally there are the very contentious issues of processing the data archive. Although Beer's law is very straightforward, its implementation has as many variations as there are investigators who use it. The central problem being agreement on the accuracy which the aerosol optical thickness is derived. The uncertainties in computation of the airmass (m), the calculations for the Rayleigh and ozone optical depths ([tau]r, [tau]o) and water vapor expressed as total column abundance or precipitable water (Pw) as well as strategies for calibration of the instruments and monitoring the long term change in calibration all combine to preclude any globally accepted processing scheme. Perhaps even more debatable are the aerosol properties derived from inversions of the sky radiances with the radiation transfer equation. Our solutions make the raw data and calibration data available to the user and provide a basic processing package (of published, widely accepted algorithms) with sufficient friendliness and flexibility that all data may be accessed globally through common forms of electronic communication on the internet.

Following is the Aerosol Robotic Network (AERONET) version of a ground-based aerosol monitoring system that offers a standardization for a ground-based regional to global scale aerosol monitoring and characterization network. We have assembled a reliable system and offer it as a point of focus for further development of each component. As an example of the system's performance under a variety of conditions, we present data collected in the Brazilian Amazon during the dry season and Mauna Loa Observatory, Hawaii. Owing to the fundamental importance of these and similar data for basic aerosol research, aerosol forcing research and validation of retrievals from space based platforms, we are emphasizing this system for a regional to global scale network of these observations. Our philosophy is for an open, honor system whereby all contributed data may be accessed by anyone but publication of results requires permission of the contributing investigators. We have designed and implemented a system that promotes these goals.

Automatic Sun and Sky Scanning Spectral Radiometer

Most if not all sun photometer networks have had limited success when people are required to make routine observations. Therefore an automatic instrument is a fundamental component for routine network observations. The measurement protocol must be reasonably robust such that unwanted data may be successfully screened from useful data, data quality and instrument functionality may be evaluated and the instrument should be self-calibrating or at the least collects data for its calibration. Following is our assessment of the CIMEL CE-318 instrument that meets these criteria of a field hardy, transmitting, sun and sky scanning spectral radiometer which is used in the AERONET program.

General Description

The CIMEL Electronique 318A spectral radiometer manufactured in Paris, France is a solar powered weather hardy robotically pointed sun and sky spectral radiometer. This instrument has approximately a 1.2 degree full angle field of view and two detectors for measurement of direct sun, aureole and sky radiance. The 33 cm collimators were designed for 10-5 straylight rejection for measurements of the aureole 3 degrees from the sun. The robot mounted sensor head is parked pointed nadir when idle to prevent contamination of the optical windows from rain and foreign particles. An additional internal collimator is protected by a quartz window allowing observation with a UV enhanced silicon detector with sufficient signal-to-noise for spectral observations between 300 and 1050 nm. The sky collimator has the same field of view but an order of magnitude larger aperture-lens system allows better dynamic range for the sky radiances. The components of the sensor head are sealed from moisture and desiccated to prevent damage to the electrical components and interference filters. Eight interference filters are located in a filter wheel which is rotated by a direct drive stepping motor. A thermister measures the temperature of the detector allowing compensation for any temperature dependence in the silicon detector. A polarization model of the CE-318 is also used in AERONET. This version executes the same measurement protocol as the standard model but takes additional polarized measurements in the solar principal plane hourly at 870 nm (Table 1 and 2).

The sensor head is pointed by stepping azimuth and zenith motors with a precision of 0.05 degrees. A microprocessor computes the position of the sun based on time, latitude and longitude which directs the sensor head to within approximately one degree of the sun, after which a four quadrant detector tracks the sun precisely prior to a programmed measurement sequence. After the routine measurement is completed the instrument returns to the "park" position awaiting the next measurement sequence. A "wet sensor" exposed to precipitation will cancel any measurement sequence in progress. Data are downloaded under program control to a Data Collection Platform (DCP) typically used in the geostationary satellite telemetry system, (see Data Transmission section).

Measurement Concept

Since the instrument was first available in 1992, the measurements protocols have evolved to a point in which we feel maximum information content is achieved within the constraints of the hardware and software available for the network system and the goals of the aerosol climatology data base. The radiometer makes only two basic measurements, either direct sun or sky, both within several programmed sequences. The direct sun measurements are made in eight spectral bands (anywhere between 340 and 1020 nm; 440, 670, 870, 940 and 1020 nm are standard) requiring approximately 10 seconds. A sequence of three such measurements are taken 30 seconds apart creating a triplet observation per wavelength. Triplet observations are made during morning and afternoon Langley calibration sequences and at standard 15-minute intervals in between (Table 1). The time variation of clouds are typically greater than that of aerosols causing an observable variation in the triplets that can be used to screen clouds in many cases. Additionally the 15-minute interval allows a longer temporal frequency check for cloud contamination.

Table 1. Measurement sequences of the CIMEL Sun/Sky scanning spectral radiometer. Note, that m = Rayleigh optical airmass, AOT = aerosol optical thickness, Pw = precip water, a = Angstrom wavelength exponent, and P([theta]), = phase function.


Spectral Range nm

Target
No. Obs.
Obs.

Interval

Application
BASIC DIRECT SUN
340 to 1020
Sun
1 each [lambda]
~ 8 sec. for. 8 [lambda]
AOT, Pw, a
Triplet Observation
340 to 1020
Sun
Three direct sun
3 @ 30 sec. apart, 1 min total
AOT, Pw, a & cl[omicron]ud screening
Standard Measurement Sequence
340 to 1020
Sun
Variable: depends on day length
Ea. 15 min m=2 AM to m=2 PM
AOT, Pw, a
Langley
340 to 1020
Sun
16, AM & PM between airmass (m) 7 & 2
m=7 - 5, incr of.5 m

m=5 - 2, incr. of.25

Langley, Cal., AOT, Pw, a
BASIC SKY
440 to 1020
Sky
1 each [lambda]
none
Sky Radiance
Langley sky
440 to 1020
Sky
16 between

m 7 & 2

m=7 - 5,.5;

m=5- 2,.25

Stability of Langley Plot
Almucantar
440 to 1020
Sky
72

(Table 2)

>8/day: m= 4, 3, 2, 1.7 hrly 9AM to 3PM
Size Dist. and P([theta]), AOT, a
Polarization
870
Sky
42

(Table 2)

hourly

m=3 AM to m=3 PM

Size Dist. and P([theta])
Principal Plane
440 to 1020
Sky
42

(Table 2)

hourly

m=3 AM to m=3 PM

Size Dist. and P([theta]) AOT, a

Sky measurements are performed at 440, 670, 870 and 1020 nm (Table 1). A single spectral measurement sequence of sky radiances (Langley sky) is made immediately after the Langley airmass direct sun measurement, 20 degrees from the sun. This is used to assess the stability of the Langley plot analysis according to O'Neill et al., 1984. Two basic sky observation sequences are made, the "almucantar" and "principal plane". The philosophy is to acquire aureole and sky radiances observations through a large range of scattering angles from the sun through a constant aerosol profile to retrieve size distribution, phase function and aerosol optical thickness (AOT). An almucantar is a series of measurements taken at the elevation angle of the sun for specified azimuth angles relative to the position of the sun. The range of scattering angles decrease as the solar zenith angle decreases thus almucantar sequences made at an optical airmass of 2 or more achieve scattering angles of 120 degrees or larger. Scattering angles of 120 degrees are typical of many sunsynchronous viewing satellites thus a measure of the satellite path radiance is approximated from the ground station. During an almucantar measurement, observations from a single channel are made in a sweep at a constant elevation angle across the solar disk and continues through 360 degrees of azimuth in about 40 seconds (Table 2). This is repeated for each channel to complete an almucantar sequence. More than eight almucantar sequences are made daily at an optical airmass of 4, 3, 2 and 1.7 both morning and afternoon. If an optical airmass of 2 or less is not achieved, an almucantar is made hourly between 9 AM and 3 PM local solar time for the standard instrument and skipping only the noon almucantar for the polarization instrument. A direct sun observation is made during each spectral almucantar sequence.

Table 2. Almucantar and Principal Plane sequences for the standard and polarization instruments.


Sun

Sky (deg.)
ALMUCANTAR

Azimuth angle relative to sun

0deg.
6.0, 5.0, 4.5, 4.0, 3.5, 3.0, 2.5, 2.0, -2.0,-2.5, -3.0, -3.5, -4.0, -4.5, -5.0, -6.0, -8.0, -10.0, -12.0, -14.0, -16.0, -18.0, -20.0, -25.0, -30.0, -35.0, -40.0, -45.0, -50.0, -60.0, -70.0, -80.0, -90.0, -100.0, -110.0, -120.0, -130.0, -140.0, -160.0, -180.0

Duplicate above sequence for a complete counter clockwise rotation to -6

PRINCIPAL PLANE:

Standard

Scattering Angle from sun (negative is below the sun)

0deg.
-6.0, -5.0, -4.5, -4.0, -3.5, -3.0, -2.5, -2.0, 2.0, 2.5, 3.0, 3.5, 4.0, 4.5, 5.0, 6.0, 8.0, 10.0, 12.0, 14.0, 16.0, 18.0, 20.0, 25.0, 30.0, 35.0, 40.0, 45.0, 50.0, 60.0, 70.0, 80.0, 90.0, 100.0, 110.0, 120.0, 130.0, 140.0
PRINCIPAL PLANE:

Polarization

Scattering Angle from sun (negative is in the anti solar direction)

-
-85.0, -80.0, -75, -70, -65.0, -60.0, -55.0, -50.0, -45.0, -40.0, -35.0, -30.0, -25.0, -20.0, -15.0, -10.0, -5.0, 5.0, 10.0, 15.0, 20.0, 25.0, 30.0, 35.0, 40.0, 45.0, 50.0, 55.0, 60.0, 65.0, 70.0, 75.0, 80.0, 85.0

The standard principle plane sequence measures in much the same manner as the almucantar but in the principal plane of the sun where all angular distances from the sun are scattering angles regardless of solar zenith angle. This measurement sequence begins with a sun observation, moves 6 degrees below the solar disk then sweeps through the sun taking about 30 seconds for each of the four spectral bands.(Table 2). Principal plane observations are made hourly when the optical airmass is less than 2 to minimize the variations in radiance due to the change in viewing optical airmass.

Polarization measurements of the sky at 870 nm are an option with this instrument. The sequence is made in the principal plane at 5 degree increments between zenith angles of -85 and +85 degrees. The configuration of the filter wheel requires that a near-IR polarization sheet is attached to the filter wheel. Three spectrally matched 870 nm filters are positioned in the filter wheel exactly 120 degrees apart. Each angular observation is a measurement of the three polarization filter positions. An observation takes approximately 5 seconds and the entire sequence about 3 minutes. This sequence occurs immediately after the standard principle plane measurement sequence.

Instrument Precision

We define the precision of the instrument as its ability to accurately reproduce results from multiple measurements under constant conditions using standardized techniques. Three methods will be used to assess the radiometric precision: (1) The variability of the digital numbers (DN) from the spectral response acquired from the 1.8 meter diameter integrating sphere at Goddard Space Flight Center which is used to determine the gain and offset calibrations of the sky radiance channels, (2) examination of dark current values taken during each sky radiance measurement and (3) the triplet variability of the DN's taken from Mauna Loa Observatory Langley observations was used to evaluate the sun channels.

All instruments are routinely calibrated with Goddard's 1.8 meter integrating sphere at least twice per year and the reference instruments approximately monthly. Each calibration session consists of three sequential measurements at four lamp levels (radiance levels). The sphere's precision is not well known however the absolute accuracy is ~5% (Walker et al., 1991). Assuming the sphere has perfect precision we may use these data to estimate the precision of the sky channels. The percent deviation from the mean of each sequence was averaged from all the sequences since 1993 for each of the three reference instruments. In all but 1 case, the variability was much less than 1% of the mean value (Table 3A). Given these results, some of the variability in Table 3A could be attributed to the uncertainty in the precision of the integrating sphere and the potential for variability in the data collection procedure.

Over 3000 dark current values were examined for each instrument and the average values computed by wavelength for the sun and both sky (aureole: 2 to 6 degrees = sky1 and dark sky: 6 to 180 degrees=sky2) observations. The dark current values for the sun observations averaged less than 1 count compared to typical measurement values of 2000 to 15000 counts depending on wavelength, optical depth and airmass (Table 3B), thus for typical conditions the dark current is insignificant. The sky observations have a higher dark current value ranging from 2 to 14 counts with standard deviations of approximately the same magnitude. Typically this is about 1% of the signal and is subtracted prior radiance computation.

Table 3. The DNs were used to compute (A) the % variation from the mean radiance measuured at the integrating sphere for the sky channels, (B) the mean Dark current values for all measurement conditions and (C) the % variation of the mean triplet values during selected Mauna Loa Langley Plots for three field and reference instruments.

(A) Integrating Sphere


Inst. #2
Inst. # 13
Inst. # 32

Mean % Var.

Mean % Var.

Mean % Var.

[lambda] (um)
1.02
0.87
0.67
0.44
n
1.02
0.87
0.67
0.44
n
1.02
0.87
0.67
0.44
n
12 Lamps
-
-
0.1
0.3
3
-
-
0.1
0.1
9
-
-
0.5
0.3
4,8
6 Lamps
2.7
0.8
0.7
0.4
7

0.5
0.1
0.1
9
0.1
0.1
0.1
0.2
8
2 Lamps
0.2
0.3
0.2
0.3
8
0.1
0.1
0.1
0.2
9
0.3
0.2
0.2
0.3
8
1 Lamp
0.1
0.1
0.1
0.4
7
0.1
0.1
0.1
0.4
9
0.1
0.1
0.1
0.4
8
(B) Dark Current

Inst. #2
Inst. # 13
Inst. # 32

Mean DN

Mean DN

Mean DN


sun
sky1
sky2
n
sun
sky1
sky2
n
sun
sky1
sky2
n
1020 nm
1.17
11.98
7.16
3201
1.29
6.01
4.00
3889
0.43
14.04
8.00
2703
940 nm
0.64
-
-
3201
0.22
-
-
3889
0.05
-
-
2703
870 nm
0.73
8.07
4.36
3201
0.59
3.62
2.87
3889
0.21
9.17
6.17
2703
670 nm
0.56
4.52
2.02
3201
0.15
1.93
1.14
3889
0.11
6.40
4.15
2703
440 nm
0.60
4.94
2.10
3201
0.15
2.02
1.16
3889
0.10
5.57
3.31
2703
380 nm
0.56
-
-
3201
0.01
-
-
3889
0.06
-
-
2703
340 nm
0.77
-
-
3201
0.23
-
-
3889
0.05
-
-
2703
Sky1=small aperature collimator for measurements from 2 to 6 degrees from sun

Sky2=large aperature collimator for measurements from 6 to 180 degrees from sun


(C) Mauna Loa Langley Plots


Inst. #2
Inst. # 13
Inst. # 32
Sun
Mean Var. (%)
n
Mean Var. (%)
n
Mean Var. (%)
n
1020 nm
0.2
288
0.3
168
0.1
264
940 nm
0.2
288
0.3
168
0.2
264
870 nm
0.3
288
0.4
168
0.2
264
670 nm
0.3
288
0.3
168
0.2
264
440 nm
0. 3
288
0.3
168
0.2
264
380 nm
0.7
288
0.5
168
0.6
264
340 nm
0. 9
288
0. 7
168
1.0
264

Langley plots from NOAA's Mauna Loa Observatory have been made to determine the spectral extraterrestrial voltage (Vo[lambda]) for these instruments since 1994. The observatory's high altitude and isolation from most local and regional sources of aerosols provides a very stable irradiance regime in the mornings (Shaw, 1983). The Langley Plot is a logarithm of the DN taken during these times plotted against the optical airmass between a range of 5 and 2 (between 3.5 and 2 for 340 nm). The intercept is the calibration coefficient (zero airmass DN) and the slope the optical thickness. If the aerosol loading is constant with time, these points plot as a straight line. The deviation of these points from the linear regression line is a measure of the precision of the instrument although it does include atmospheric variation which we are assuming is negligible at Mauna Loa during the selected Langleys. Table 3C shows the average variability of a triplet is less than 1% and is most typically 0.3% for all three instruments. This is in agreement with the precision estimated from the integrating sphere analysis.

Instrument Calibration

Calibration refers to the determination of the calibration coefficients needed to convert the instrument output (DN) to a desired output, in this case aerosol optical thickness (AOT), preciptable water, and radiance (W/m2/sr/um). The calibration accuracy is the level of accuracy with which a desired output is achieved using defined comparison procedures. Calibration is frequently traced back to the variability with which the calibration coefficients are determined to achieve that unit output. Thus instrument calibration is a combination of the instrument precision, the calibration procedure and the algorithms used. In this section, we will discuss the variability of the calibration coefficients we determine for the sky channels from the 1.8 meter integrating sphere, the spectral Vo from the Mauna Loa Langleys and the change in the calibration coefficients as a function of time. We will also discuss the intercomparison procedure for transferring the Vo calibration coefficients from a reference instrument to a field instrument and the computation of the resultant variability.

The sphere calibration procedure given in the previous section allows us to compute a gain and offset for each sky measurement wavelength. The mean dark current DN is typically between 0 and 14 counts (the median DN is zero for the sun channels) (Table 3b) which is subtracted from the DN thus giving an offset of zero. The Instrument DNs are plotted against the exitant radiance from the integrating sphere and a gain is computed from the linear regression fit through the origin. The mean gain is computed from three regression gains made for each session. The accuracy of the sphere is reported as +/-5% (Walker et al., 1991) thus the calibration coefficient accuracy can be no better than 5% plus the variability of the three regressions (precision) or conservatively +/-~ 5.5%.

The Vo calibration coefficients are typically computed from an average of three to seven Langley plots obtained at the Mauna Loa Observatory, 3500 m altitude. The variability of the retrieved mean Vo as measured by the coefficient of variation (CV, standard deviation/mean) indicates the combined stability of the atmosphere, instrument and the repeatability of the calibration procedure. The averaged Mauna Loa Langleys Vo obtained during all calibration sessions have a standard deviation of ~0.25 to 0.50% of the mean for the visible and near-IR wavelengths, ~0.5 to 2% for the UV to ~1 to 3% for the water vapor channel (Table 4). These values of Vo repeatability by Langley analysis are similar to those reported by Schmid and Wetrl: (1995).

The Mauna Loa (MLO) calibration is conducted with two simultaneously operating master instruments. Comparisons are made between ratios of spectral raw voltages as a check for instrument repeatability. A diurnal variation of less than 1% of the ratioed voltages is considered acceptable. Approximately monthly the MLO master instruments are swapped with two reference instruments located at GSFC. The GSFC reference instruments are used for intercomparison with field instruments. Monitoring voltage ratios is continued for all master instruments.

Table 4, The mean CV in percent by wavelength (nm) of the Mauna Loa derived Langley Vo for all of the wavelengths used in the reference CIMEL sun photometers.

Inst.No.

1020
940
870
670
500
440
380
340

CV %
CV %
CV %
CV %
CV %
CV %
CV %
CV %
2
0.19
2.39
0.14
0.18
0.22
0.22
0.35
2.10
13
0.27
0.89
0.29
0.44
0.90
0.40
0.77
0.63
32
0.26
3.19
0.19
0.24
0.23
0.29
1.10
0.48

With respect to the long term stability of the calibration coefficients, the optical interference filters are the limiting factors. Periodic sphere gains and mountain top Langley calibration coefficients have been determined since 1993. The results are typical for interference filters. On average, for non-UV channels there has been a decrease from 1 to 10% per year and, after 2 years, there has been a rapid decay in some filters (Table 5). Since the percent decrease in the time dependent calibration coefficients is usually greater than the uncertainty of a semiannual Vo determination we use a linear interpolation of the Vo between calibration dates. This requires that the instrument calibration coefficient be followed closely. Thus, until more information is available, we calibrated instruments on a 6 month rotation and change filters after two years of field use. Therefore the percentage changes which occur between Vo calibrations are actually a factor of 2 to 3 smaller than shown in Table 5 since these values are on a percent change per year.

Most instruments cannot be calibrated at Mauna Loa and a well calibrated integrating sphere with sufficient radiometric output is not common, therefore most instruments are calibrated at Goddard Space Flight Center with the 1.8 meter integrating sphere and with intercomparisons against the Goddard reference instrument with a Mauna Loa derived Vo. Intercomparisons are made by solving Eq. 1 for the field instrument Vo based on the reference instrument [tau]a during simultaneous observations (time difference of less than 6 seconds), under clear stable atmospheric conditions ([tau]a typically less than 0.1). One day of simultaneous comparisons are usual. The CV of the Vo computed from these comparisons is typically larger than the reference instrument uncertainty. The total error is the uncertainty attributed to the field instrument calibration coefficient due to transfer of calibration from the reference instrument plus the error from the reference instrument calibration defined from the Mauna Loa Langleys. As with the reference instruments, calibration coefficients are then linearly interpolated between the calibration tie points unless independent information suggests a different method as in the case of a change in filters at which time new calibration comparisons must be made.

The spectral voltage ratios of the field instrument is compared to the reference instruments during several days. Diurnal variations of less than 1% are considered acceptable.

Table 5, The decay rate of zero airmass voltages, Vo, (%/yr) is shown for filters less than two years old for each reference instrument.

Wavelength (nm)

Instr.

Number

1020

940

870

670

500

440

380

340

# 2
-2
-1
2
2
3
-4
11
3
# 13
5
-31
2
0
ND
2
23
11
# 13
10
5
10
11
ND
15
20
15
# 32
4
6
7
2
2
4
26
5

Measurements of the spectral temperature sensitivity of the instrument in a temperature controlled chamber showed agreement with the manufacturers published temperature sensitivity of the detectors. To date only the 1020 nm channels showed significant temperature variation (0.25%/oC +/-0.25%/oC ) warranting a correction to a reference temperature in the processing. Each instrument in the network is evaluated for temperature sensitivity.

Data Accuracy

Various instrumental, calibrational, atmospheric and methodological factors that influence the precision and accuracy of optical depth determination have been pointed out clearly in a series of publications (Shaw, 1976; Reagan et al., 1987 and Russel et al., 1993) and attempts to account for or minimize these are described in previous sections. Instrument uncertainty due to electro-optical precision is for all practical purposes insignificant (Table 3) for a properly operating instrument. The variability of the atmosphere is characterized by the variability of the triplet optical thicknesses which may at times be cloud contaminated. This uncertainty is computed, can be used as a screening tool, and may be retrieved from the AERONET data base. Additionally the uncertainty due to calibration is tracked with all time dependent data and may also be retrieved from the data base. Typically the total uncertainty in AOT from a newly calibrated instrument under cloud free conditions is +/-0.01 for [lambda] >440 nm and 0.02 for shorter wavelengths. Determination of the uncertainty in the water vapor retrieval is limited by the uncertainty of the radiosonde intercomparisons, typically less than 10%.

The uncertainty of the sky radiance data is more difficult to ascertain since these only constitute single observations and no absolute self-calibration procedure is yet implemented between the sphere calibrations. Based on the sphere calibration, the uncertainty in the sky radiance at the time of calibration is approximately +/-5% for all four channels at the time of calibration. We performed a sensitivity study of the retrieved scattering optical depth at 670 nm based on increases of 1, 5, 10 and 20% in the calibration coefficients and compared them to Langley extinction AOT retrievals assuming similar errors in Vo (Table 6). Since the scattering aerosol optical depth is directly related to the aureole brightness, the accuracy is largely a function of the sky calibration. It is clearly seen that sky retrievals have tolerable errors to approximately 10% in this example where as the direct approach would generally be acceptable with a calibration error of less than 2 to 3 percent. The sky absolute errors decrease with longer wavelengths. This property is leading to the development of an in situ sky calibration procedure that may be used to calibrate the direct sun observations, (Nakajima et al., 1996).

Table 6 The extinction optical thickness ([tau] ext ) and scattering optical thickness ([tau] sct ) for an airmass of 2 and wavelength of 670 nm and their absolute differences are computed referencing an exact aerosol optical depth of 0.0528 assuming various calibration coefficient errors. The extinction optical thickness increased by 0.024 and sky optical depth 0.005 with the 5% error according to a sensitivity study. A calibration error of 5% is frequently encountered with interference filters.

Calibration error

0%
1%
5%
10%
20%
[tau] sct
0.0541
0.0549
0.0582
0.0622
0.0703
[Delta] [tau] sct
0.0013
0.0021
0.0054
0.0094
0.0175
[tau] ext
0.0528
0.0578
0.0772
0.1005
0.1440
[Delta] [tau] ext
0.0000
0.0050
0.0244
0.0477
0.0912

Data Transmission

Data are transmitted from the memory of the sun photometer microprocessor via the Data Collection Systems (DCS) to either of three geosynchronus satellites GOES, METEOSAT or GMS and then retransmitted to the appropriate ground receiving station. The data can be retrieved for processing either by modem or internet linkage resulting in near real-time acquisition from almost any site on the globe excluding poleward of 80deg. latitude. The DCS is a governmental system operated for the purpose of transmitting low volume environmental data from remote sites for various institutions and government agencies.

Each station on the GOES, GMS and METEOSAT networks has been assigned a user ID and transmission time window passing up to 30 kbytes per day in either 24 or 48 individual transmissions at hourly orhalf-hourly intervals. During each transmission, a packet of data and status information are time stamped by the radiometer, the transmitter and the central receiving station (Wallops Island, VA, USA for GOES; Darmstadt, Germany for METEOSAT; and Tokyo, Japan for GMS). Typically the data are maintained in the receiving station computers for 3 to 5 days before they are overwritten. The data are retrieved daily from the central receiving station which we term near real-time.

Processing System

A fundamental component of the AERONET system is a package of user friendly UNIX software that provides near real-time information on the status and calibration of the instruments, provides data processing with referenced and generally accepted processing algorithms, provides an orderly archive of the data and provides convenient electronic access for all users to the raw and processed data base. We shall discuss these aspects of the current operational state of the software and future enhancements.

Instrument and Network Status

The radiometer data stream includes date, time, temperature, battery voltage, wet sensor status and time of transmission as well as several levels of identification numbers. The DCP adds a time stamp at the time of transmission as does the DCS receiving station plus checks for parity errors and signal strength of the transmission. After data are downloaded from the central receiving station, a status report and a trouble shooting report are automatically generated and e-mailed to appropriate system and instrument managers. The status report provides a comprehensive assessment of the operation of the radiometer and DCP for the data transmitted with the current download. Network managers then have sufficient information to assess the operation of individual stations. To more quickly identify trouble spots, a troubleshooting report is generated that lists by instrument only information that fails to meet normal operating thresholds i.e. low battery voltage, transmission time error, missed transmission etc. This approach can identify remote station problems quickly often leading to same day resolution. Documentation of the status report is available under the AERONET home page http://spamer.gsfc.nasa.gov.

Data Processing

There is lack of agreement on corrections, calibration procedures, data analysis procedures etc. often caused by divergent error tolerances or specific requirements of various investigators. We have implemented a series of processing algorithms on a UNIX server that have been published in the open literature and/or are generally accepted by the scientific community (Table 7). These algorithms impose a processing standardization on all of the data taken in the network facilitating comparison of spatial and temporal data between instruments. The archival system allows the user community to access either the raw or processed data via internet for examination, analysis and/or reprocessing as needed. The archival browse algorithms are collectively known under the program name "demonstrat" which graphically provides access to all aspects of the data base. The program operates on an Hp 735 UNIX server called "spamer.gsfc.nasa.gov". The algorithms within "demonstrat" comprise three principal categories; time dependent retrievals such as AOT and Pw, calibration assessment, and sky radiance retrievals. There are a growing number of sub processing algorithms within each of these. A full description of `demonstrat' is beyond the scope of this paper however a help session can be found under the AERONET Home Page. As importantly, `demonstrat' allows all data to be retrieved through "FTP" and e-mail access for personal computer analysis and/or reprocessing as the user requires. As new and improved approaches and models are accepted within the community the processing may be applied uniformly to the network wide data base. Additionally access to the data base through demonstrat provides an opportunity for testing new algorithms and models for an increasingly diverse set of measurements for a variety of locations and conditions. The following figures were obtained directly from the `demonstrat' output to illustrate the access to the data base.

Table 7, the algorithms, inputs, corrections and models used in computing the aerosol optical thickness, preciptable water (Pw), spectral irradiance, and sky radiance inversions are referenced.

Variable, algorithm or correction

Comments
References
Basic Computations


Rayleigh Optical Depth, [tau]r

refractive index of air

depolarization factor


Penndorf, 1957

Edlen, 1966

Young, 1980

Bucholtz, 1995

Solar Zenith Angle, [theta]o

Michalsky, 1988
Earth sun distance, d

Iqbal, 1983
Ozone amount, O3
Table lookup by 5deg. lat. long.
London et al., 1976
Aerosol optical air mass, ma

Kasten and Young, 1989
Rayleigh optical air mass, mr

Kasten and Young, 1989
O3optical air mass, mo

Komhyr et al., 1989

Corrections


Temperature, T
~0.25%/deg.C for 1020 nm

specific for each inst.

Hamamatsu Inc. and Lab measurements
Water Vapor for 1020 AOT
from Pw retrieval, Lowtran
Kneizys et al, 1988
Rayleigh, all wavelengths
for elevation (standard profile pressure dependance)

O3 abs. coef. [lambda] > 350 nm

Vigroux, 1953
O3 abs. coef. [lambda] < 350 nm

Bass and Paur, 1984
Time, t
Cimel, UTC, DAPS time stamps, +/-1 second
Refer to Homepage
Retrievals


Spectral direct Sun AOT, Langley Plots
Beer's Law
Shaw, 1983
Pw: (a, k, Vo)
Modified Langley
Bruegge et al., 1992 Reagan et al., 1992
Scattering AOT
From spectral sky radiance
Nakajima et al., 1983
Size Dist., Phase function
From spectral sky radiance
Nakajima et al., 1983
Size Dist.
From spectral direct sun AOT
Twitty, 1975; Halthore and Fraser, 1987, King 1978
Models


Spectral2 (irradiance)
parameterized spectral Radiation Trasnfer (RT)
Bird and Riordan, 1986
6-S (Linkage)
analytical RT
Vermote et al., 1995
Procedures


Cloud Screening
Thresholds, [lambda] AOT & t
Refer to Homepage
Climatology, Direct Sun
AOT, Pw, Wavelength Exp.
Refer to Homepage
Climatology, Sky
Size Dist., Phase function, g
Refer to Homepage

Archival Browser (`demonstrat')

The custom browser `demonstrat' allows a comprehensive method of viewing and screening the data in either raw or processed form. Following are a few of the options available in demonstrat that we feel are important for use in a network data base.

Time Dependent AOT Retrievals. The time dependence window serves as the access point for all other windows. The aerosol optical thickness, precipitable water, Angstrom wavelength exponent and calibration coefficient trends as well as the status indicators (battery voltage, temperature and wet sensor) may be plotted as a function of time in this window. For a particular instrument and location, all or part of the data may be displayed by interactive cursor subsetting. For example, the dry season data (June to October) from Cuiaba, Brazil clearly shows the increase in aerosol optical thickness as the burning season begins in August (Figure 1). Subsetting to eight days of data or less, the UTC time scale and a local time bar are drawn, the mean 15 minute. direct sun AOT observations are plotted and almucantar (triangles) principle plane (squares) and successful inversions (o and x, respectively) are shown under the time scale (Figure 2). A hatched line above the time scale indicates Langley data and vertical bars inside the plot (not shown in this example) indicate that the wet sensor has been activated and no sun data are available. Individual points may be rejected in these windows.

Figure 1, The aerosol optical thickness dry season record for Cuiaba, Brazil showing the increase in aerosols in August-September 1993 due to regional burning.

Figure 2, The aerosol optical thickness in Cuiaba on August 14, 1993 (top) shows significant aerosol loading in contrast to June 23 (bottom). Note the addition of time dependent information on the abscissa including almucantar ([Delta]),principal plane([), inversion (o or x) and Langley data (|).

Calibration Assessment. Historically, uncertainty due to calibration of sun photometers has limited their wide scale deployment and long term use. No new methods are offered however "demonstrat" imposes a standard computation of aerosol optical thickness and Pw calibration coefficients and in so doing renders a simple method via a graphics window for the user to assess the quality of these calibration coefficients interactively from the on-line data base. Two windows were implemented for standardizing the direct sun calibration coefficient procedure and assessing their quality control. The first is the traditional Langley Plot with the modified Langley method used for water vapor retrieval. A second method is a simple intercomparison.

The radiometer acquires a Langley data sequence each morning and afternoon between a Rayleigh optical airmass of 2 and 7. The interactive calibration routine allows manual rejection of data points and automatically computes a table of Vo's for each wavelength. Tabled Vo's are recomputed and displayed after each rejection of data outliers. The Vo's may be applied to the original Langley data and aerosol optical thickness plotted as a function of time or airmass in two additional windows for further inspection of the quality of the Langley Plot. The water vapor calibration coefficient determined by the modified Langley method (Bruegge et al., 1992 and Reagan et al., 1992) is performed in much the same way. The water vapor transmittance is modeled from each 940 nm filter function using MODTRAN and has been shown to be largely independent of temperature and water vapor profiles (Halthore et al., 1997). Both Langley methods are typically used only for absolute calibration analysis with more restrictive airmass ranges from high mountain top acquisitions for our reference instruments . This is particularly important for the UV and 940 nm (water vapor) wavelengths.

Intercomparison of two automatic instruments is accomplished with an algorithm that searches a specified portion of the data base for a space and time coincidence (Fig. 3). Sun and backscatter sky data are automatically intercompared by spectral aerosol optical thickness and sky radiance. A table of old and new calibration coefficients are generated. The analyst then decides the necessity of further calibration assessment.

The history of the calibration coefficient determinations for each instrument is easily tracked on demand by a calibration tree showing the date, location and reference instrument from which each intercomparison was made, back to a mountain top Langley or sphere calibration. Additionally a time dependent plot of the calibration coefficients shows the trends over time for the instrument in question.

Fig. 3 shows the intercomparison window which allows recomputation of Vo from a time series of AOTs from a reference instrument (top plot to a field instrument (bottom plot). The resulting AOT is displayed and a table of old and new Vos can be displayed. Options to examine time dependent voltages may also be accessed from this window.

Sky Radiance Inversions. The almucantar window displays the four channel sky radiances as a function of scattering angle, volume size distribution from 0.1 to ~8.0 um, scattering phase function and a table of the aerosol optical thickness and wavelength exponent computed from both direct sun and the aureole measurements (Fig 4). Additionally the spectral asymmetry factor is computed from the phase function. From the radiance data, a window may be opened with zoom capabilities which separates the four spectral sky radiance bands into single color coded bands allowing close inspection of the data. The program automatically checks the quality of the almucantar data by examining the symmetry of the aureole radiances about the sun. The angular asymmetry is defined as |(l-r)/(l+r)*0.5| (where l=left side and r=right side radiance pairs) and if it exceeds 10%, those pairs are removed from the inversion process. If the standard deviation of the difference between aureole pairs divided by the averaged value of the angular pairs exceeds 10% or there are not a sufficient number of data points remaining with symmetry (10), the data are not inverted. The inversion routine used is that of Nakajima et al., (1983) and has a number of options that will be implemented over time. This will include size distribution inversions by combining the spectral optical thickness from direct sun measurements and aureole data and in cases where the almucantar or principle plane data are not available an interactive inversion from the spectral AOT data can be made but the retrieved size range will be smaller due to reduced sensitivity to large particles.

The principle plane data are processed using the same inversion; however only data on the zenith side from the solar disc are used in the inversion due to asymmetry induced by the ground reflectance and an increasingly large optical airmass . The principle plane window has identical capabilities as the almucantar window. The test for the quality of the data is simply the smoothness of the curve.

(a)

(b)

Figure 4 A successful inversion of almucantar radiances during low aerosol loading (4a) and high aerosol loading (4b) is possible when the radiance data are symmetric about the sun (upper left plot within window). Inversions produce a volume size distribution with good accuracy from 0.1 um to about 8 um aerosol radii (lower left window). The aerosol optical thickness and wavelength exponent are computed (upper right window) and compared to that measured by direct sun observation. The spectral phase function and asymmetry factor (lower right side of window) from the aureole inversion are also computed using Nakajima et al.'s (1983) pakrad code.

Radiative Transfer Model Interface. We have incorporated a parameterized spectral cloud free irradiance model SPECTRAL2 (Bird and Riordan, 1986) to compute the total, direct and diffuse down welling flux in the solar spectrum at 122 wavelengths from the measured aerosol and water vapor parameters. Single scattering albedo is the only required parameter which the instrument does not measure and thus must be supplied by the user. The interactive computations are made for any instantaneous or time dependent measurements. The window displays the spectral flux curves for the total, direct and diffuse irradiance and a summary box gives integrated values for each component of the broad band (0.3-4.0um) and Photosyntheticallly Active Radiation (PAR) (0.4-0.7 um). The model is applied to the time dependence creating a data set of integrated fluxes. Options exist to compute coincident fluxes for user specified background conditions. Ratios of ambient vs. background conditions are computed and displayed in a summary box.

An interface to the more rigorous 6S model has been developed. The size distribution parameters ( ) deduced from the almucantar inversion as well as the index of refraction (imaginary and real) can easily be input to the 6S model (Vermote et al., 1996, 1997) and used to compute the phase function, extinction and scattering coefficient at any wavelength between 0.25 and 4.0um. These quantities are then used to generate a large set of atmospheric parameters in addition to the simulation of the signal observed from aircraft or space by a variety of sensors. The computation of the phase function and extinction and scattering coefficients is done by the MIE subroutine (described in details in Vermote et al., 1996). Computations are restricted to the case of the scattering of electromagnetic waves by a mixture of homogeneous isotropic spheres, the physical properties of particles whose sizes are comparable to or larger than the wavelength. These assumptions are in accordance with those used in the sun photometer size distribution retrieval algorithm.

Cloud Screening. Data are taken by the automatic instruments under all non precipitation conditions causing significant cloud contamination in some of the raw data. The philosophy for using the data base within "demonstrat" is to provide for the user simple cloud screening tools based on the variability of the triplets and for continental non dust aerosols the spectral dependence of the AOT. Despite these screenings, some cloud contaminated data will be displayed. Further refinements to the screening process may be made by adjusting a filter threshold for the windowed data based on the triplet variability and/or the 15 minute variability on the time dependent window allowing the data set to be screened and downloaded as desired by the user. This process does not alter the original data base.

Automatic cloud screening of the almucantar and principal plane data are by symmetry and smoothness checks respectively of the data about the solar disc as explained under `sky radiance inversions'.

Downloading Data. Labeled spreadsheet export files may be created during a `demonstrat' browser session of all raw or processed data in the data base and all data processed during a session e.g. modeled fluxes. Data for export may be selected by location, time, and the type of raw or processed data desired. The data may be downloaded to any computer with internet access using a guest account or may be e-mailed directly during a "demonstrat session". Homepage data access is under development.

Global Perspective

Through 1996 approximately 60 instruments were deployed world wide on various Islands, North America, South America, Europe, Africa, and the Middle East, fostered by collaboration between international, national, and local agencies private foundations and individuals (Fig. 5). As the data base continues to expand, the processing system becomes more sophisticated and more users have access to the data base, the need to provide better access to and quality assurance of the data base becomes more critical. To aid in that effort, the reference data base is located on 'spamer.gsfc.nasa.gov' at Goddard Space flight Center in Greenbelt, MD, USA or `loaser.univ-lille1.fr' (IP number is 134.206.50.10) at Lab. d'Optique Atmosphérique, U.S.T. de Lille, 59655-Villeneuve d'Ascq,France for European access. A third supported data base will be established in Tokyo, Japan to support access to the data from eastern Asia. Identical clones of these systems have been established at various locations to facilitate access to the data for local activities. All processing changes are made to the entire spamer reference data base to maintain uniform processing.

An automatic, computerized quality assured data base is under development to provide a screened data set to the scientific community. It will be accessed by a simplified version of the `demonstrat' browser, `demonstrat II', available through the AERONET homepage. The data must exceed specified optical, radiometric and calibrational specifications and well as incorporating screening algorithms for cloud contamination that are functionally related temporal and spectral behavior of the aerosol optical depth. Further details will be included in the homepage. Preliminary screening tests have eliminated approximately 50% of the original data.

The network is expected to provide characterization of aerosol optical properties, a data base for atmospheric correction, validation of satellite based aerosol retrievals, and atmospheric corections to satellite observations of ocean color and numerous opportunities for aerosol research. The simple technology and international collaboration that has produced AERONET can be adapted to accommodate complimentary data sets of BRDF, automatic lidar systems, and radiation networks.

Figure 5, The approximate location of instruments is represented by the colored circles. Measurements are made at permanent sites year round. Data are taken seasonally at high latitudes and/or when cloud cover permits. In 1996 nearly 60 locations contributed to the data base.