Interim Report 1996: USDA Ultraviolet Radiation Monitoring Program

Transcription

1 Interim Report 1996: USDA Ultraviolet Radiation Monitoring Program Introduction Network Expansion Instrumentation Calibration and QA Data Management Research Meetings Future Work References Appendix A Appendix B Appendix C Appendix D Appendix E D.S. Bigelow, J.R. Slusser, J.H. Gibson Natural Resource Ecology Laboratory Colorado State University Fort Collins CO January 1997 USDA Agreement # Executive Summary The discovery of the Antarctic ozone hole in 1985, accompanied by large increases in surface ultraviolet-b (UVB) radiation, has raised serious questions about the continued protection of the earth's living systems from the harmful effects of UVB radiation. To assess the potential for damage that increased UVB radiation might have on agricultural crops and forests, the USDA established a UVB monitoring network and began measuring UV radiation with broadband UVB-1 pyranometers in Because of the limited information available from broadband instruments a new multi-spectral instrument has been developed to provide spectral UVB data at network sites in support of biological and atmospheric science research. This 7 wavelength UV Multi-filter Rotating Shadowband Radiometer (UV-MFRSR), manufactured by Yankee Environmental Systems, has now been installed at 10 of the network's 20 sites. Plans call for an expansion of the network to at least 26 sites, each equipped with a UV-MFRSR, UVB-1, and VIS-MFRSR. Twelve of the UVB broadband instruments were recalibrated at least once since 1994 allowing comparison of their spectral responses over time. The comparisons indicate a non-linear drift in the spectral response of the detectors. This suggests possible non-reproducibility in the laboratory characterizations and underscores the need to have calibrations carried out at the newly established NOAA interagency supported calibration facility. The stability of the multi-spectral instrument's filters transmission and center wavelength is a factor critical to obtaining accurate multi-spectral data. In a test of the stability of the multi-spectral instrument's filters, Langley plots for VIS-MFRSRs were obtained for clear days. It was determined that many units exhibited a decline in the Langley intercept which is indicative of either changes in filter spectral response or degradation of the diffuser. Work in progress will permit routine adjustments for common wavelength mis-registration and drift. The application of cosine corrections and laboratory calibrations to the raw VIS-MFRSR and UV-MFRSR data has been implemented in a prototype stage and is slated for general use in early Finally, the success rate of querying sites by phone has been increased by developing new software. Because the current laboratory calibrations of the UV-MFRSR are in some doubt an alternative means of checking calibration is being investigated. The method requires knowing the extraterrestrial irradiance, currently measured to within 5% agreement by different satellites, the relative response of the detector, and the Langley voltage intercept. Initial comparisons at Table Mountain CO of UV-MFRSR irradiances calibrated using this method agree to an average of 7% with a radiative transfer model and to within an

2 average of 14% of a Brewer spectroradiometer. It is recommended that further investigation of the Langley method of calibration be moved to a research site at Mauna Loa Observatory Hawaii. The network continues to intercompare different instruments at its Central Plains Experimental Range CO site. Two new UV-MFRSRs and a Smithsonian 18 channel SR-18 will be added to the site in During June 1996 the network co-sponsored a third Annual Spectroradiometer Intercomparison at Table Mountain CO. The intercomparison brings together instrumentation from various federal programs in an attempt to identify bias in calibrations, angular response and spectral response of the various instruments. An increase in the number of data requests has necessitated the streamlining of codes and the acquisition of new, more efficient database management software. The transfer of network data to interested clients has been facilitated by the establishment of an ftp directory which is updated every two weeks. Daily plots of multi-spectral UV-MFRSR and broadband UVB-1 data are also routinely made available on the World- Wide-Web ( ) the day after the data is measured. To further increase the efficiency of data dissemination and storage, the network has acquired a CD ROM writer. Ozone has been successfully retrieved using data from the newly deployed UV-MFRSR under clear skies, and work is underway to retrieve ozone columns during cloudy weather. We are also collaborating with NASA Goddard to provide ground truth from our site at Table Mountain CO for their TOMS/ADEOS satellite data. This collaboration uses a model to predict surface UV irradiances for all of the US and may result in effectively extending the range of our network beyond 26 sites. Conferences attended by the network personnel included the NASA/TOMS Intercomparison in Greenbelt MD, the International Radiation Symposium in Fairbanks AK and the Synthetic Spectra Workshop, Fort Collins CO. Presentations were made by network staff at all these conferences. The network recommends expanding to at least 26 sites to provide the minimum grid-based network originally proposed. It is recommended that wherever possible the network continue to collocate with established UV instrumentation to allow comparisons of measurements. The NOAA Calibration Facility in Boulder CO should continue to be given full support by the network as it will eventually perform all of the network's UV calibrations including cosine corrections and recharacterizations after field use. Monitoring filter stability through continued analysis of Langley intercepts and frequent recalibration should also be maintained. Given the network's recent advances in UV instrumentation and the successful demonstration of the use of models to generate synthetic spectra we believe it is timely to convene a group of biologists and botanists studying the effects of UV radiation on biological systems for the purpose of reaching a consensus on exactly what sort of data products will be needed from the network to further their research. 2

3 Interim Report: USDA Ultraviolet Radiation Monitoring Program, 1996 D.S. Bigelow, J.R. Slusser, J.H. Gibson Introduction The USDA Ultraviolet Radiation Monitoring Program was established in 1993 to provide the US Department of Agriculture with the information necessary to determine if changing levels of ultraviolet light have an effect on food and fibre production in the United States. Prior to the establishment of the program only limited information was available to make such an assessment and the geographic distribution and quality of this information was insufficient to meet the requirements of the agency (Gibson, 1991; UVB Monitoring Workshop, 1992). Two different but complimentary actions were taken by the agency to begin to obtain the information necessary to make its assessment. The first solicited proposals for the development of an improved spectroradiometer (see Appendix A) and the second action which is reported in this document, established an ultraviolet radiation monitoring program. The primary objective of the USDA Ultraviolet Radiation Monitoring Program is to provide information to the agricultural community about the geographic and temporal climatology of UVB irradiance. Its data is intended to assist scientists in relating changes in stratospheric ozone to changes in ultraviolet light and to improve the understanding of the factors which control ultraviolet light. Both are critical in assessing the impacts of changing UV light on agricultural systems. Since the establishment of the network the data has found use with model developers, human health effects researchers, ecosystem scientists and those seeking a ground truth measurement for satellite systems. The initial network of twelve stations was established with broadband meters and ancillary measurements of temperature, humidity and seven wavelengths of visible light produced by the Multi-Filter Rotating Shadowband Radiometer (MFRSR) (Harrison et al., 1994), the latter instrument serving as a intelligent data logging and communications system for all of the measurements taken at a single site. Since that beginning the network has expanded to 20 stations and the broadband meters have been supplemented with a seven wavelength ultraviolet rotating shadowband radiometer referred to in this document as a UV- MFRSR. Accomplishments in 1996 Climate Network Expansion It has been estimated that monitoring locations are necessary to adequately define a useful national UV climatology (Gibson, 1992). During 1996 the network expanded from 12 to 20 sites which included the establishment of a research site near Boulder, Colorado at Table Mountain, the site of the annual North American Spectroradiometer Intercomparisons. Locating at the Intercomparison site provides the network with continuous access to a spectroradiometer which can be used to validate proposed routine products of the USDA monitoring program such as synthetic spectra and optical depth retrievals (see Research Section below). 1

4 Other sites established in 1996 include Grand Canyon, Arizona; Baton Rouge, Louisiana; Wye, Maryland; Grand Rapids, Minnesota; Mead, Nebraska; Big Bend, Texas; Underhill, Vermont and Lake Dubay, Wisconsin. Current sites in the network are shown in Figure 1. Additional sites chosen but not yet installed include Poker Flat, Alaska; Mauna Loa, Hawaii; Fort Peck, Montana and Everglades National Park in Florida. It is anticipated that the network will be complete and fully instrumented by the end of Instrumentation (Hardware) Figure 1 Locations of Sites in the USDA UVB Monitoring Program in 1996 Visible Multi-Filter Rotating Shadowband Radiometers (VIS-MFRSR) - The VIS-MFRSR is a 7 channel, 10 nm full-width, half-maximum (FWHM) passbanded shadowbanded radiometer that measures visible light at 415, 500, 610, 665, 860 and 960 nominal wavelengths. Yankee Environmental Systems (YES) of Turners Falls, Massachusetts became the sole vendor of the VIS-MFRSR instrument during 1993 through an exclusive licensing agreement with the instrument's developers at the State University of New York at Albany (SUNY) and the Pacific Northwest Laboratory (PNL). This change in vendor resulted in a number of improvements to the shadowband instruments which were not entirely compatible with the network's existing polling activities and the instruments originally purchased from PNL. The acquisition and deployment of the new YES instruments beginning in 1996 required that many of these differences be addressed. The YES shadowbands support higher modem speeds which lead to reduced phone bills and faster data transfers, thereby reducing the risk of polling failures due to line interruptions. Coincidently the network was required to move away from the Hayes 2400 baud Pocket Modems 1 originally supplied as standard equipment with the PNL shadowbands. Not only were they no longer manufactured but new Federal Communications Commission guidelines specified that modems were to no longer be powered by telephone line power as is the case with the pocket modem. Finding a replacement modem however, proved difficult because of the wide variety of phone line quality throughout the network and the quirkiness of the shadowband s communication system. Four brands of modems were evaluated in various combinations of host and remote machine configurations. A US Robotics External 28,800 Sportster 1 model proved to be the most robust for the network's applications and is now being used throughout the network. New versions of firmware controlling the operation of the shadowband also appeared with the YES shadowbands. In addition to fixing a few obscure operational bugs, the new firmware gave the instruments more functionality (band angle control, expanded memory management, automatic head and board identification coding) and a new access password. Unfortunately, the new firmware was not backwardly 1 The naming of products in this report does not constitute an endorsement by the USDA. Products are identified solely for the convenience of the reader. 2

5 compatible with older PNL equipment and required that clock crystal speeds of the older PNL units be boosted from 5Mhz to 12Mhz. The network is in the process of making this upgrade but has been hampered by new problems identified in the upgraded equipment. These include a band wobble, solved with a firmware modification, and a time keeping problem requiring frequent clock adjustments (twice per week) to the upgraded units. Modifications are significant enough that the network must maintain a troubleshooting inventory of both PNL and YES shadowband parts. PNL and YES shadowband hardware components are not considered to be interchangeable. Ultraviolet Multi-Filter Rotating Shadowband Radiometers (UV-MFRSR) - In addition to the new locations, the network began expanding the instrumentation at each site to include a new seven channel ultraviolet version of the visible Multi-Filter Rotating Shadowband Radiometer. This new shadowband instrument manufactured by YES contains 2 nm FWHM filters at 300, 305, 317, 325, 332 and 368 nm nominal wavelengths. Ten sites were equipped with the UV-MFRSR in Prototype instruments were first made available to the network by YES during the 1995 North American Spectroradiometer Intercomparison and then throughout the following year. The first production instrument was deployed to the Davis, California site in August of With the addition of a second shadowband at each polling location (the UV-MFRSR) the network faced the prospect of adding a second phone line to each site. However, additional phones lines are not always available in remote locations. Phone switches were investigated for these locations and found to be successful in most circumstances but it was discovered that the line voltage requirements of phone switches vary widely. Numerous switches were tested before an acceptable one was found to work in most of the networks locations. It is anticipated that the network will be able to meet its polling requirements with a combination of phone switches and a minimal number of extra phone lines. Broadband UVB Pyranometers - Original network monitoring locations were equipped with only a VIS- MFRSR and a broadband UVB pyranometer. The pyranometers, YES model UVB-1 measuring broadband UV between 280 and 320 nm, were only envisioned as a temporary UVB monitoring instrument until a higher quality instrument became available. Because of this they were only deployed at the network's original 12 sites. In the spring of 1996 the network decided to retain the broadband pyranometers at its original sites and add the broadband to its standard array of instrumentation. This decision was based upon the observation that broadband meters continue to be used by many researchers including those involved in human health effects of ultraviolet light. The continuation of broadband meters in the network also helps to maintain an historical reference to past studies which have used the meters. Ancillary Instruments - Each site continues to be equipped with a temperature/ humidity sensor and a downward looking LI-COR 1 photometer for indicating the presence or absence of snow cover. A barometer has additionally been installed at the Table Mountain research site to allow pressure to be factored into the ozone retrieval and synthetic spectra models being evaluated at this site. Smithsonian SR-18 UVB Radiometer - The network took delivery of its first SR-18 in late May of 1996 and added it to the instrumentation at its Colorado site at the Central Plains Experimental Range in June. This instrument is a 2 nm FWHM, multi-filter radiometer with 18 channels that span the range of nm. The instrument has run continuously at the site with only minor interruptions due to a failed power supply. Five additional units are to be delivered in Work with this instrument has been curtailed because of its inability to communicate effectively with the network's host computer. Currently an operator must be sent to the site to copy data onto floppy disks. These disks are then carried or mailed to the 3

6 network for processing. Additional software also needs to be developed to handle undocumented inconsistencies with returned fixed length records before any of the data can be routinely evaluated by the network. High Resolution Spectroradiometers - The Competitive Research Grants Office of USDA-CSREES and this project have supported the development of a high resolution spectroradiometer to measure UVB radiation. This development is under the direction of Dr. Lee Harrison at SUNY. Current plans call for the establishment of approximately six "research" or reference sites using these high resolution scanning instruments. Installation and operation of these instruments including site support and data acquisition will be performed by Harrison. These instruments are designed to have a wavelength range of nanometers and a band pass of 0.1 nanometers (Gibson 1991). The development of the instrument has been delayed by problems of acquiring a commercial monochromator meeting desired specifications. Such a monochromator is now available and will be available for testing at the 1997 North American Spectroradiometer Intercomparison. Information on the progress on the development of this instrument is detailed in Appendix A. Table 1. Instrumentation Available at Each USDA UVB Monitoring Location: End of State/Site Name VIS-MFRSR UV-MFRSR Broadband Arizona, Grand Canyon pnl X California, Davis pnl X X Colorado, Central. Plains Exp. Range X X X Colorado, Table Mountain X Georgia, Griffin pnl X Illinois, Bondville pnl X X Louisiana, Baton Rouge pnl Maine, Presque Isle X X X Maryland, Wye pnl Michigan, Douglas Lake X X Minnesota, Grand Rapids X X X Nebraska, Mead pnl New Mexico, Jornada pnl X New York, Geneva pnl X Ohio, Oxford pnl X Texas, Big Bend pnl X X Vermont, Underhill X X X Utah, Logan pnl X X Washington, Pullman pnl X X Wisconsin, Lake Dubay X 4

7 Table 1 above lists the instrumentation installed at each network site at the end of Unless otherwise noted equipment has been manufactured by Yankee Environmental Systems (YES). The network anticipates that a complete set of multi-filter, broadband and ancillary instruments will be available at each site by the end of Calibration and Quality Assurance Visible Multi-Filter Rotating Shadowband Radiometers - The most significant achievement with VIS- MFRSR calibration and quality assurance during 1996 was the locating of the original board calibrations for the 20 units purchased from PNL in Unfortunately, corresponding head calibrations are still missing. The board calibrations however, permit an approximate calibration of the first two years of network data by matching recent YES head calibrations with the older board calibrations. Software developed by the instrument design group at SUNY which applied calibration to the VIS-MFRSR data streams was useless prior to this time. Seven of the 20 original PNL heads were re-calibrated by YES in 1996 permitting calibration to be applied to 7 of the 12 sites that were operating prior to However, because the network rotates VIS-MFRSR heads throughout the network, it will not be able to provide calibrated data from any of its sites that continue to use PNL heads until the remaining 13 heads cycle through the YES calibration facility. This is expected to be completed in early The lack of available laboratory calibration cited above prompted the network to explore the use of an alternative method for calibrating its VIS-MFRSR data. In the absence of laboratory-provided calibration data, VIS-MFRSR calibration may be derived through a Langley analysis of each instruments specific channels (see the Research Section below for a more thorough discussion of the Langley technique). However, this method depends upon the stability of the passbands of the VIS-MFRSR over time. Using Langley analysis software supplied by the SUNY-Albany VIS-MFRSR group (Harrison and Michalsky, 1994), the network examined all of its data for suitable Langley events and arranged the returned intercepts by head for further analysis. In addition to the review of filter stability the Langley calculations provided a quality check of the integrity of the application software which matches cosine characterization results with individual data records. Mismatched record keeping or improper software logic prevents the execution of the Langley analysis program. Figure 2 illustrates the results of this stability check for head # This head was first located at the network's Colorado site and then at the network's Utah site hence the break in the data record. The downward drift of V o at each wavelength suggests that the filters are exhibiting some instability, especially early in their life. Though most heads do not as yet have as extensive an operating history as the one displayed, most appear to exhibit similar behavior. In some cases, where all filters appear to drift together, the downward drift may be due to the deterioration or soiling of the diffuser. In those cases where there does not appear to such uniformity in the decline, the 610 and 665 nm wavelengths typically exhibit the strongest downward trends. Preliminary indications suggest that a more extensive Langley history will need to be developed before it is clear how to compensate for the apparent drifts in filter response. Further analysis of filter stability is planned for Early in 1996, the instrument development group at SUNY reported that some units were displaying a time-keeping problem when they were collecting data at a high rate and were being polled frequently. 5

8 Vo Vo for Unit 8725 through December 1997 (using 500 nm as a KEY Channel) 415nm 500 nm 610 nm 665 nm 860 nm Date USDA UVB Monitoring Network: January 1997 Figure 2 Time Series of Langley Regression Intercepts for USDA Head # 8725 When it appears, the problem shifts the time of a 3 minute average forward or backward by one or more records. Days then appear to have more or less 3-minute records than would be expected based upon the minutes of daylight computed for each day at each site. This introduces an error in all downstream data processing that relates the angular corrections of data to the time of day at individual sites. A computer program was developed by the group to check for this error. When the program was applied to the historical USDA data files there appeared to be time discrepancies in most of the USDA raw data files. Through some extensive detective work it was found that the time problem was actually caused by a particular sequence of network polling commands that instrument firmware was supposed to prevent. The command sequence has since been amended and updated firmware is now available which performs as designed. After a preliminary evaluation of the affected data the network concluded that the error introduced is not serious enough to warrant a correction to the data at this time. During the above upgrading of firmware in PNL manufactured VIS-MFRSRs the network discovered that the upgrade caused an intermittent band motion problem. The problem was further identified in all newly manufactured YES shadowbands using this new "K" series prom. As the band returns to it index (home) position it intermittently shakes, sometimes causing the instrument to lose track of its rotation index. This results in an instrument shutdown. The problem was resolved with the release of a second prom ('M') 6

9 which all but two network sites have received. Since the problem was discovered early the affect on the network has been minimal. Ultraviolet Multi-Filter Rotating Shadowband Radiometers - Calibration and quality assurance of the UV- MFRSR follows the procedures outlined for the VIS-MFRSR instruments. Because of its later deployment in the network it has not suffered the problems noted in the VIS-MFRSR discussions above. Laboratory calibration has been available for the instruments and there do not appear to be any time-keeping errors in the data files. Filter stability, though having even less Langley history than the VIS-MFRSR appears to be comparable to that noted in the VIS-MFRSR data records. Finally band motion problems were resolved with the installation of the series "M" prom. Broadband UVB Pyranometers - Extensive reviews of the broadband data were conducted in 1996 in preparation for the release of the data to the network's ftp archive. One such review was an examination of the nighttime values or dark counts returned by the instrument when the signal should be zero. It was discovered that data from four of the nine initial sites and all of the sites the network shared with the Quantitative Links monitoring program (Maine, Illinois and Ohio), had non-zero dark counts as a result of a ground loop in the initial wiring scheme used at these sites. Table 2 lists the brief periods when dark count corrections were deemed necessary and presents the median nighttime dark count value for each time period listed. Multiple time periods at the Maine and Washington sites result from attempts to remediate the problem which only achieved partial solution. Table 2. Periods of Dark Counts Corrections at USDA UVB Monitoring Sites Sites Approximate Dates Median dark counts Griffin, GA August September Bondville, IL August November Howland, ME October July July period of record 16 Presque Isle, ME October November Geneva, NY August September Oxford, OH August December Pullman, WA July August August September September October

10 counts Year and Month Figure 3 Monthly Distribution of Dark Counts Resulting from a Ground Loop at the Howland, Maine Station Figure 3 illustrates the problem at the Howland, Maine Quantitative Links site using monthly distributions of daily nighttime data. As shown, dark counts display a strong seasonal (actually temperature) dependency. Though similar at each site, regression analysis suggested that a separate regression be used to correct data at individual sites. Separate linear regressions were developed of the form S c (C N P50 ) [(T T P50 ) slope of the regression of T P50 daily versus N P 50 daily ] 1 where the corrected counts (S c ) are equal to the count of a 3-minute (or 1-minute) record (C) minus the median dark count of the time interval being corrected (N P50 ), less a temperature dependency correction. The temperature dependency correction is computed as the product of the temperature of the 3-minute (or 1-minute) record (T) minus the median nighttime temperature of the time period (T P50 ) times the slope of a regression of the daily nighttime median temperatures versus the daily median dark counts. A second review, conducted in preparation of placing broadband data into the ftp archive, involved identifying the explicit calibration lineage of the broadband meters. This was deemed necessary to document how the network calibrated, collected and processed its broadband data. This resulted in the creation of the broadband measurement document given in Appendix B and in the determination that the calibration constant used by the network to convert raw signal voltages to calibrated units of watts/meter 2 was in error. Prior to the review (April 1996) the network had been dividing instead of multiplying its raw counts by a predetermined conversion factor. This resulted in a necessary correction of to all 8

11 previously released data. All clients were notified of the mistake and necessary correction, and the problem and solution were also posted to the network's World-Wide-Web site. A final aspect of the review was a comparison of the spectral responses of those broadband meters that had been recharacterized since their original deployment to sites. By 1996 twelve of the broadband meters had been recalibrated at least once and four had been through the calibration facility 4 times. Although only preliminary conclusions can be drawn from such a small data set, it would appear that characterizations performed in 1994 and 1995 are likely biased towards the shorter wavelengths. In all of the twelve paired recharacterizations the second spectral response displayed a marked shift towards the shorter wavelengths. This shift appears to reverse however in those meters that had cycled through the facility two more times. Most likely, these changing responses represent imprecision in the recharacterization process rather than actual changes in the meters themselves. There does not appear to be any other explanation which would support non-linear trends in these meters. The spectral responses of individual meters will continue to be tracked until it can be determined what, if any, corrections may be necessary to fully qualify the network's broadband data. Ancillary Instrumentation - Temperature and humidity measurements continue to lack calibration. A hand held, portable temperature /humidity probe was acquired as a field reference instrument during 1996 and field technicians began using it to adjust local instruments. However, there has been no attempt to maintain a traceable calibration of this instrument. This will be accomplished at some future date. Calibration constants printed on the calibration certificates of the downward looking LI-COR photometers were averaged to create a standardized unit conversion constant which is now being applied to all LI-COR raw voltages in the network. Because the use of this instrumentation is only to determine the presence and absence of snow the network has concluded that any further calibration of this instrumentation in unwarranted. Smithsonian SR-18 - Nothing has been done to establish calibration or quality assurance procedures for the SR-18. The development of calibration and quality assurance procedures for this instrument is one of many tasks facing the network in Intercomparisons - The network's Central Plains Experimental Range site (Colorado) continues to be the network's primary site for prototyping, testing and characterizing network instrument configurations. In addition the site serves as the primary location for quality assurance studies of network precision and bias. During 1996 there were 6 broadband meters, two new seven channel UV-MFRSRs, one VIS-MFRSR and one new Smithsonian SR-18 Radiometer located at the site. An additional filter instrument, a GUV-511, manufactured by Biospherical Instruments 1 is scheduled to be installed in early Duplicated and collocated instrumentation data produced in 1996 will be used in 1997 to establish network precision and bias estimates. Calibration Facility - A critical issue in the establishment of the monitoring program, and one that has plagued previous efforts to monitor UVB radiation, is an understanding of the spectral characteristics of the instruments employed using a National Institute of Standards and Technology (NIST) traceable calibration. The Radiometric Physics Division of NIST produced the initial characterizations of the network's broadband meters and Harrison's group at SUNY produced the characterizations of the initial PNL manufactured VIS-MFRSRs. However, it became apparent very early in the life of the network that these facilities did not have the capacity to handle the volume of instrumentation that would be deployed in 9

12 the USDA network. As a result all new and previous instrumentation is now routinely characterized and calibrated by YES. YES calibration facilities however are also limited and it is anticipated that they too will not be able to meet the needs of the network over the long-term. The USDA is not alone in its desire to establish a more permanent, NIST traceable calibration facility. Because of this need and the availability of expertise in the radiation measurement programs conducted by the NOAA-ARL labs in Boulder, CO, NOAA was selected to construct a calibration facility under the direction of John DeLuisi and Patrick Disterhoft. This calibration facility is supported by the USDA (subcontract to NOAA), USEPA, NOAA and NSF. To assure the highest possible quality, NIST is providing radiometric standards, technical guidance and oversight. As a part of this facility NOAA has established a field site on Table Mountain north of Boulder, CO where various instruments can be compared against the solar irradiance. This site also serves as a permanent location for instrument intercomparisons including the Annual North American Interagency Spectroradiometer Intercomparisons. In June of 1996 the network helped co-sponsor a third annual North American Spectroradiometer Intercomparison. Results of the intercomparison will be reported at a later date. Results of the 1 st North American Spectroradiometer Intercomparison (1994) will be published in late spring of this year by NIST (Thompson, et al. 1997). Intercomparison data sets are available through the NIST ftp server (lasulite.nist.gov). An additional field site is planned at a high altitude site west of Boulder, CO operated by the University of Colorado s Institute for Arctic and Alpine Research INSTAAR. This will provide clearer sky conditions for instrument evaluation and Langley calibrations. Presently the calibration facility is modifying a McPherson double monochromator to perform responsivity scans in the UV region which will determine the wavelength response characteristics of broadband and narrowband filter instruments. For the YES UVB-1, four orders of magnitude of signal range have been obtained with a 0.8 nm bandpass. This will permit the determination of the top of the long wavelength plateau. The determination of filter functions for the VIS-MFRSR is still limited. Using the same bandpass and acquiring the signal straight from the preamplifier, only two orders of magnitude are currently available. When an improved amplifier is installed, another 1.5 to 2 orders of signal range should be possible. Since calibration and cosine measurement facilities have been completed, full characterization (filter functions, cosine response, and calibration) of the broadband and VIS-MFRSRs should be available in late April of With the acquisition of a new high resolution, high throughput double monochromator in early 1997, work to develop the capability to measure the filter functions of the UV-MFRSR will begin. The UV instruments with 2 nm passband interference filters will require the monochromator to operate at a slit width of less than the 0.8 nm available with the McPherson. Ideally the slit width of the monochromator should be one fortieth that of the interference filter passband or in the case of the UV-MFRSRs, one fortieth of 2 nm. When the incorporation of the new monochromator is complete, full characterization of the UV instruments can begin. 10

13 Data Management Over the past year data management and programming tasks have been organized into four major categories. These include software support for polling individual sites, the application of calibration information to network data, improved accessibility to network data, and data archiving. A fifth category, that of systems administration of the project s more than a dozen mixed platform computers, printers and tape systems, continues to occupy two-thirds of a person's time. This is largely due to the project s location off-campus which necessitates the self-maintenance of both the computers on the local network and the offcampus network itself. Support for Polling Sites - Data management tasks in support of site polling activities are given the highest priority in the network. Failures in polling or corruption of data files either during the transfer of information or within the shadowband s data logging system immediately trigger an investigation and remedial action involving two or three of the project s staff. If you do not first collect the data, there will be nothing to show for the time spent measuring the solar radiation. The updating of data tables that hold and track each instrument s location and configuration was completed in time for the deployment of the network s first UV-MFRSRs. Software previously prototyped in the Perl scripting language was re-written into C and the entire polling process was made more efficient, faster and automated thereby taking full advantage of the network's data base management system (Ingres 1 ). Now as soon as data is entered in the system by the field technicians it immediately becomes available to the entire data system. This has the further benefit of providing the information necessary to keep the Web pages current each day. During the transition from Hayes modems to US Robotics modems and while gaining experience with the new modems the network experienced brief periods of less reliable polling than was previously enjoyed. To compensate the network implemented a dual modem/machine, smart polling system that tracks the results of the polling through each modem and attempts to re-poll sites when failures have occurred. This strategy restored the reliability of network polling and it has now been implemented as a permanent feature of the network s polling strategy. The success rates of retrieving VIS-MFRSR data from each of the network's sites is given in Appendix C. Application of Calibration Information to Data Records - Cosine characterizations performed first by PNL and SUNY and later by YES were moved into a data base management system during 1996 thereby allowing the network to automate the process of selecting and applying the most recent (or appropriate) cosine correction constants to any selected USDA network data. The automation of this selection process enabled the network to include cosine corrected data in its World-Wide Web data presentation and to support cosine correction software previously developed for the Unix, IBM and MacIntosh computing platforms, all of which are in use by our data users. More efficient C programs replaced prototype Perl language applications during the streamlining of the cosine correction data processing applications. A similar processing strategy has been used to automate the application of calibration data to both the VIS- MFRSR and UV-MFRSR data streams and it is anticipated that this process will move from the prototype to production stage in early Presently all available calibration data for both VIS-MFRSR and UV- MFRSR instrumentation have been moved into the data system and prototype programs (in Perl) are being tested which apply calibration to selectable network data sets. 11

14 Improved Accessibility to Network Data - The primary means of distributing network data continues to be via the World-Wide-Web site maintained by the network ( The number of information inquiries via other means ( , telephone, letter, etc) however, also continues to grow (see Appendix D). Information "clients" have increasingly been interested in the network's historical data, both broadband and VIS-MFRSR, but the historical data was only available via a formal data request to the network. To remedy this the network created a permanent ftp archive for broadband data which is updated twice per month. An easy to use link from the Web page to the ftp directory gives data users access to all of the network s broadband data. Users may also access the data directly through standard ftp connections. The network also expanded its daily display of VIS-MFRSR and broadband data during 1996 to give the scientific community the ability to plot, on demand, all of the network s data by the day, week, or month. Ancillary data (temperature, humidity and surface reflectance) along with the new UV-MFRSR data are also now available alongside the shadowband data and all shadowband plots now routinely display cosine corrected data. Documentation of the broadband measurement and calibration procedures have been added to the Web site as well as a listing of all project personnel. It is anticipated that further documentation of VIS-MFRSR and UV-MFRSR measurement and calibration procedures will be added to the site in With increased usage of the Web site by a wider variety of clients (see Appendix D) execution speed and efficiency have become more important to the network. To meet this challenge the network has replaced much of its Perl computer code with faster, more efficient, C-language code and has modularized much of its computer code to eliminate redundancies and cross program dependencies. The network has also begun a transition from an Ingres to an Oracle 1 data base management system (DBMS). After careful consideration of the network's ongoing and future needs, it was concluded that the Ingres product was not keeping pace with advances in software inter-connectivity. The network appeared to be increasingly creating work-arounds for data transfer among software packages and graphical user interfaces. While there will be a learning period to become familiar with the new DBMS, the network expects to experience long-term gains in data programming and processing efficiency. Data Archiving - With more than 12,000 raw data files and more than 200 calibration characterization files now collected, it has become important for the network to effect a long term strategy for archiving its data. Primary raw data files are now used only infrequently as the network utilizes more highly processed monthly data sets for its primary data files. Appendix E contains a portion of the network's overall data system design indicating how both raw and processed data are arranged in the network's computer system. The network has acquired a CD ROM writer and will begin moving data files currently contained in the raw/ data directories to this media. Other materials to be moved to this archive include polling log files (archive/ directory) and original scanned images from each of the network's sites. As the data base management system becomes the source of primary network data, monthly data files now used as the primary source of network data will also be moved to CD ROM archives. 12

15 Research The Langley Method of Calibration - It has been assumed that in the absence of explicit calibration measurements or as a complementary check of lamp calibrations performed by YES and soon to be conducted by a newly established NOAA Calibration Facility, a Langley calibration technique may be used as an alternate method of calibration. This has not however been well tested in the UV. The attenuation of the direct radiation as it passes through the earth s atmosphere is described by the Beer-Lambert Law (Craig, 1965). I 8 ' I o,8 exp(& j J 8,i m i ) 2 where I 8 is the direct irradiance at the ground, I o,8 is the extraterrestrial irradiance, J is the optical depth for the ith absorber and m is the slant path (or airmass) through the atmosphere. The equation assumes unchanging atmospheric optical conditions throughout the day and uniform horizontal mixing of the absorbers, conditions typically met only at clean, high elevation sites. Taking the natural log of both sides ln I 8 ' ln I o,8 & mj 3 and, if instead of irradiance we measure uncalibrated voltages, the equation is ln V 8 ' ln V o,8 & mj 4 where V 8 is the measured voltage of a particular channel and V o, 8 is the extrapolated voltage intercept at zero airmass. A plot of the natural log of voltage due to the direct beam at one filter wavelength versus the optical path length or airmass (typically the secant of the solar zenith angle) results in a straight line whose slope is the optical depth of the atmosphere at that wavelength and whose intercept at zero airmass is the voltage the detector would register if it were pointed towards the sun at the top of the atmosphere. To transform this measured voltage at the ground (V 8 ) to a calibrated irradiance we must know the extraterrestrial solar irradiance I o, 8 and the filter function or relative response of the filter/photodetector combination, F 8 (Tug and Baumann, 1994). Recent measurements of I o, 8 reviewed by Cebula et al., (1996) reveal differences of approximately 4% in the region between 300 and 350 nm. Compared with other errors of the Langley method of calibration, these differences in the measured extraterrestrial flux are small. The irradiance at the ground then is: I 8 ' V 8 m I 8 F 8 d8 V o,8 m F 8 d8 5 13

16 Below in Figure 4 is a plot of ln voltage versus airmass for October 10, 1996 at Table Mountain for the channel. Langley plots were analyzed using the Harrison and Michalsky (1994) algorithm on clear days at the Central Plains Experimental Station, CO (July and August 1996) and at Table Mountain, CO (September and October 1996). Figure 1 Langley plot for the channel October 10, 1996 at Table Mountain, CO. The range of the airmasses is 2 to 6. The slope is the atmospheric optical depth, J, at this wavelength. Table 3 gives the average Langley voltage intercept for the 6 channels used along with the standard deviation and percent standard deviation of the 16 Langley events. No Langley plots are available for the 300 nm channel because of insufficient signal to noise. Table 3. Average of Sixteen Langley Voltage Intercepts July through October, 1996 Filter Vo Std. Dev % S.D. 7% 7% 8% 8% 10% 8% Using these voltage intercepts, the SUSIM extraterrestrial solar flux (VanHoosier et al., 1988) and the measured filter functions in equation 5, calibration factors were developed. These factors were then used to determine global irradiances for noon values measured by the UV-MFRSR at Table Mountain, CO. To check the validity of these irradiances, the results were compared to both a model (Stamnes et al, 1988) run for the appropriate conditions (solar zenith angle [SZA]), albedo, aerosols, elevation) and the irradiances 14

17 measured by a collocated Brewer spectrophotometer (Bais et al., 1996). The comparison is summarized in Table 4. The Brewer only measures out to 365 nm so no comparison is made at nm. Table 4. Comparison of UV-MFRSR Using Langley Calibration with Model and Brewer Spectrophotometer October 10, 1996 Table Mountain, CO SZA=47.8E Filter CW (nm) UVRSR Model Brewer (Watts/m 2 /nm) The irradiances of the UV-MFRSR using the Langley calibration factors and the radiative transfer model agree to within 11% at all wavelengths and with an average difference of 7.7%. Shaw (1982) demonstrated that at the very clean high altitude Mauna Loa Observatory extrapolations of V o, 8 were constant to within a standard deviation of 1.2 % at 380 nm and 0.4% at 415 nm. This is due to extremely stable optical conditions of the observatory. Extrapolations to zero airmass in the ultraviolet below 380 nm are expected to be less constant for two reasons: changes during a day in the ozone column amount (which absorbs strongly below about 330 nm) will alter the optical depth, and more scattered light around the sun s disk at shorter wavelengths (McKenzie and Johnston, 1995). Whether the technique will produce Langley derived irradiances as good as Shaw (1982) will only be known with additional work. These results however, do suggest that the application of the Langley technique to wavelengths in the UV is promising. Further work should also address the reasons why irradiances measured by the Brewer are consistently 8 to 17% lower than the UV-MFRSR or the model. There may be a bias in the Brewer's irradiance calibration. Retrieving Ozone from the UV-MFRSRs - Measurement of ozone is critical in constructing synthetic spectra from the 7 filter measurements of the UV-MFRSR as well as to quantify the relative contribution of ozone, clouds and aerosols to UVB attenuation. We have investigated a method for extracting ozone column amounts for clear skies from the 311 nm and 317 nm channels of the UV-MFRSR using the Langley method described above. Presently it is only valid during clear days when the ozone and aerosols are constant temporally and spatially. The UV-MFRSR returns the direct normal irradiance, I, at 7 ultraviolet wavelengths. A plot of ln(v) versus airmass yields a straight line (Langley plot) with a slope of the optical depth, J, which is comprised of J ' J ray % J ozo % J aer 6 15

18 where J ray is the Rayleigh or scattering optical depth, J ozo is the ozone optical depth, and J aer is the aerosol optical depth. Two channels with center wavelengths close together (311 and 317 nm) were chosen and total optical depths, J 1 and J 2, at these wavelengths were computed using the Langley method outlined above. The aerosol optical depth for both channels were assumed to be equal and the Rayleigh optical depth computed as per methods outlined by Stephens (1993) allowing a solution for (J ozo1 - J ozo2 ). J 1 ' J ray1 %J ozo1 %J aer1 7 & (J 2 ' J ray2 % J ozo2 % J aer1 ) 8 J 1 &J 2 ' (J ray1 & J ray2 ) &' (J ozo1 & J ozo2 ) 9 Next, the effective ozone cross section, "& was computed, weighted by the extraterrestrial solar flux, S(8), and the filter function of the photometer, F(8). " ' m"(8)s(8)f(8)d8 m S(8)F(8)d8 10 The ozone optical depth, J ozo, is the product of the ozone column, i, and the effective ozone cross section "&. J ozo ' P" ozo 11 Thus combining equations 9, 10, and 11 the ozone column, i, can be determined. P ' ((J 1 &J 2 )&(J ray1 &J ray2 )) " ozo1 & " ozo

19 Using this technique ozone column abundances were determined from direct normal solar irradiances measured by the UV-MFRSR located at the Central Plains Experimental Range, CO on the mornings of July 11, 22, and 23. These values are compared with those measured at 10:00 a.m on the same dates by the NOAA Dobson spectrophotometer located at Boulder, CO about 100 km SW of the site (Table 5). The agreement is to within an average of 7% suggesting that ozone retrievals are possible from the UV-MFRSR channels and therefore the network has promise for using synthetic spectra as a means of establishing standard wavelength reporting of its multi-filter measurements. Table 5. Comparison of Column Ozone Derived from the UVRSR and a Dobson Date UVRSR (DU) Dobson (DU) July July July NASA /USDA-UVB Cooperation in Model Validation - NASA has expressed a desire to utilize the USDA UV-MFRSR measurements as a set of ground-based measurements upon which they can evaluate their model-based calculations of UVB in their TOMS satellite program. As an early indication as to how the comparisons might be made we made a number of tests using the Stamnes (1988) model in parallel with the NASA model as a general means of isolating the causes of differences in predicted irradiances. The radiative transfer models were run with identical input parameters (scattering optical depth, albedo and SZA) with the goal of obtaining irradiances that agree to within 1% for various input conditions. Preliminary results using data from the Table Mountain, CO site suggest that TOMS overflights may be used as a daily reference to USDA instrument stability. Once perfected the NASA satellite measurements and algorithm will extend the range of coverage of UV measurements and predictions to cover the entire continental United States. Conferences/ Workshops/Meetings Project staff participated in a number of conferences, workshops and symposia in Participation in these events as well as more informal meetings with colleagues and collaborators provides the network with peer review and discussion of its methodology and results, and ensures that network products are perceived as high quality and useful to the community whose needs it is striving to fulfill. Central Plains Experimental Range Third Annual Symposium; Colorado State University, Fort Collins, CO. January 11, This annual symposium brings together all of the scientists who conduct research at the Central Plains Experimental Range (CPER) in eastern Colorado. The CPER is the primary site used by the UVB Monitoring Network to test new and additional instrumentation and instrument configurations. The exchange of information between project staff and others working on the CPER provides the network with an invaluable source of site characteristics as well as furnishes the network with a wealth of new ideas for the use and presentation of its data. 17

20 1996 North American Spectral Radiometer Intercomparison; Boulder, CO. June 18-25, This annual intercomparison is sponsored in part by the network and serves to establish the comparability and bias of network data to those projects using alternate types of instrumentation. In addition, the intercomparison provides the supporting documentation necessary to determine the causes of the differences between the various instrumentation that participates the intercomparison. Two of the network's new UV- MFRSRs were included in the intercomparison to determine their comparability to spectroradiometers. Current Problems in Atmospheric Radiation: International Radiation Symposium (IRS) 1996; University of Alaska, Fairbanks, AK. August 19-24, The IRS Symposium is a once every four year, international conference of radiation scientists. Dr. James Slusser, Dr. James Gibson and David Bigelow presented a poster at the conference titled USDA UVB Monitoring Program, which introduced the radiation community to the monitoring program and presented some examples of what might be accomplished through the use of the data. An abstract of the presentation will be published along with others included in Session 8 titled"uv Radiation and Modeling". WWW Access to CA-OpenIngres Databases. CAWorld; New Orleans, LA. August 25-29, This workshop was developed and conducted by Bill Davis and another Colorado Sate University staff member to acquaint data base management practitioners with methods and Ingres tools available for use with Common Gateway Interface (CGI) scripts. These scripts are used to create on-the-fly HyperText MarkUp Language (HTML) forms and documents. The presentation was cited in the trade journal DBMS Magazine. Measuring Ultraviolet Light With the Shadowband Radiometer: the Effects of Ozone, Clouds and Aerosols; Colorado State University, Natural Resource Ecology Laboratory, Fort Collins, CO. September 27, This seminar given by Dr. James Slusser, introduced researchers at Colorado State University and our department to some of the work being done within the USDA Ultraviolet Radiation Monitoring Program. TOMS/ADEOS UV Data Product Validation Product Workshop; NASA/Goddard Space Flight Center, Greenbelt, MD. October 8-9, NASA invited scientists with a history of involvement with groundbased UV instruments to this workshop to discuss how ground-based instruments might be used to validate UV estimates derived from the US/Japanese TOMS/ADEOS satellite program. The workshop served to introduce the UV scientists to the satellite program and introduce NASA scientists to the nuances of ground-based UV measurement, calibration and data interpretation. Dr. James Gibson and Dave Bigelow made presentations to the attendees outlining the status and availability of network data. UVB Monitoring Program Project Retreat; Estes Park, CO. October 15-17, With the recent increase of project staff and the imminent expansion of the network, both in the number of sites and amount of instrumentation at each site, it was concluded that a short retreat would be beneficial in reestablishing program priorities and bringing all of the project staff to a common level of understanding of the projects scope, goals and vision. Synthetic Spectra Workshop; Colorado State University, Fort Collins, CO. December 5-6, Scientists from NCAR, NOAA, NASA, Canada's AES, and SUNY Albany met in Ft. Collins to discuss the ways to construct the entire solar spectrum reaching the earth's surface from a set of 7 UV-MFRSR filter measurements. Two major approaches were presented. 18

21 1. Derive ozone from the ratio of two channels, e.g. 311 and 368 nm (similarly described above in the Research Section). Derive effective cloud optical depths by taking the ratio of the observed to theoretical clear-sky irradiances at 368 nm. Then iterate cloud thickness at 368 in the model until the observed ratio is achieved. Retrieved ozone and cloud amounts can then be used as input to radiative transfer models to generate synthetic spectrum. This approach is described by Dahlback (1996) and similar to Stamnes et al. (1991). 2. Using all 7 channels, measure the global transmission, which for each channel is the ratio of the measured global voltage to the Langley zero airmass intercept. Then, using a newly established model by Min and Harrison (private communication) construct synthetic spectra attenuating the extraterrestrial solar flux by multiplying it by a function using 6 wavelength dependent parameters. The filter functions are multiplied by the first guess synthetic spectrum to generate 7 model transmissions which are compared with the measured transmission. The 6 parameters are adjusted until the differences between the measurements and the model are at minimum. Initial results presented using both techniques are very promising. Starting with filter "measurements" constructed from a synthetic spectrum, however, the second algorithm returned a virtually identical synthetic spectrum. The algorithm is fast and reliable, and can be run "as needed" meaning there is no need to store the synthetic spectra. A critical use for the algorithm is to transform measured filter irradiances to common wavelength values. Currently each channel registers irradiances measured at slightly different center wavelengths, making comparisons of different data sets impossible. By using this algorithm and standard set of filter functions, all measured irradiances may be transformed to a standard set of wavelengths. Recommendations and Future Work Much of the effort of building a UV monitoring program over the past years has centered around familiarizing project personnel with the operation of VIS-MFRSRs and identifying promising locations for network instrumentation. Along the way a number of "improvements" have been introduced into basic VIS- MFRSR technology and the network has by and large been requested to absorb the cost of implementing these changes into the routine procedures of network operations. While these improvements are appreciated and encouraged they have highlighted the differences between technology proven in the visible spectra and new challenges of working in the UV.! The network must continue to investigate those procedures that have been developed for the visible spectra to ensure UV results reported by the network are valid. As outlined in the preceding pages a number of outstanding issues have been and will need to be brought to resolution to achieve this.! The network needs to continue to expand into at least the 26 site, grid-based, network that was originally proposed if it hopes to enjoy the spatial coverage it set out to define. Some locations that offer special opportunities such as the Table Mountain and Central Plains Experimental Range in Colorado and the Mauna Loa, Hawaii sites need to be enhanced to provide the network with specific research and quality assurance information.! The establishment of a research/network site at Mauna Loa Observatory appears to be the only

22 way the network will be able to credibly establish that Langley methods can be applied to the UV and to demonstrate that synthetic spectra will indeed be reflective of actual measurements. We recommend setting up 2 UV-MFRSR instruments atop the 11,300 foot elevation Mauna Loa Observatory. One will be the normal network shadowband instrument. The second instrument will be mounted on a sun-tracker with a collimating tube which limits the field of view to about 2.0E (the sun subtends 0.5E). This tracker follows the sun's motion through the sky with great precision. Setting 2 UV instruments atop the Mauna Loa Observatory will allow study of the UV photometer performance under exceptionally clean and aerosol free skies (Shaw, 1979) enabling the network to: 1) reduce standard deviation in Langley voltage intercepts which will improve Langley calibrations and allow comparisons with laboratory results; 2) compare results with tracker and shadowbands including testing of the cosine response of the shadowband; 3) compare calibrations derived from Langley methods with the New Zealand spectrometer (Bodhaine et al., 1996); 4) study aerosol optical depth in ultraviolet! Standardization and equity of peer calibration laboratories need to be initiated and supported to establish network accuracy and to ensure equity through time of the UV measurements made by the network. Laboratory capacity for instrument calibration and characterization is minimal and the network has already been forced to use facilities at SUNY-Albany, Yankee Environmental Systems and soon the NOAA Calibration Facility in Boulder, CO to complete its quality assurance initiatives. Both calibration techniques and facilities need to be evaluated and standardized if the network is to continue to provide a quality product to its scientific customers. Continued support for the NOAA Calibration facility is critical to these needs.! The tracking of filter stability needs to continue in order to establish a firm basis for implementing routine algorithms for correcting and normalizing instrument response.! Work in progress with ozone retrievals and synthetic spectra generation needs to continue so that the network can establish a routine procedure for achieving commonality in the wavelengths it reports for its filter instruments. To bring these procedures to fruition the network needs to advance its work with ozone retrievals to include cloudy skies and expand its capability of producing synthetic spectra. It is more typical for there to be clouds than clear sky so ozone retrieval under cloudy conditions must be considered. To a first approximation clouds are spectrally neutral. Using a discrete ordinate radiative transfer code, Stamnes et al (1991) demonstrated that ozone column could be effectively retrieved from the ratio of measured spectral (0.5 nm resolution) global irradiances at 340 nm and 305 nm. We are confident that a method of retrieving ozone columns based on the ratio of measured filter global irradiances at 368 nm and 311 nm can be achieved and tested by comparing ozone columns derived from the UV-MFRSR at Table Mountain with those measured by the Dobson at Boulder 10 miles to the south.! Collocation with both peer and identical instrumentation needs to continue so that network precision, bias and its comparability to complementary research programs is established. The installation of two GUV-511 Biospherical Instruments filter instruments at the Central Plains Experimental Range, CO, will provide the network with an ongoing independent check of UV-MFRSR calibrations and help establish a direct relationship between the second most used filter instrument currently used in the world. There are over 30 of these instruments in operation throughout North and

23 South America as well as Europe and Antarctica. Unlike the UV-MFRSR, the GUV-511 has no shadowband, so it measures only the total horizontal irradiance. Daily comparisons of both instruments 305 nm channels and the UV-MFRSR 368 nm to the GUV nm channels are a high priority. The Smithsonian SR-18 instrument has been running smoothly at the CPER since July However there have been data transfer problems which prevent polling the instrument and downloading the data. These need to be resolved but since most likely there will be no more than 6 of these instruments in operation, comparison of the SR-18 with the UV-MFRSR must be considered a lower priority than comparison with the GUV-511. We will continue to collect data from the SR-18 in anticipation of having the resources to make a detailed analysis of this important instrument.! We recommend that a workshop for biological scientists who use UV data be conducted in We recommend that scientists investigating the effects of UV radiation on plants and animals as well as humans and other living organisms, be invited to discuss their current research and discuss what form of data from the USDA UVB Monitoring Network might be most useful to them. This could for example involve writing algorithms to transform the 7-channel, 3-minute averages into time and wavelength energy integrals and weighted biological doses. Alternately there could be the need for daily maxima or other measures that can be extracted from the data stream with suitable algorithms. The UV effects scientific community is still fairly new in addressing problems of calibration, experimental design, and comparison of results. It is expected that the workshop will further all of these important issues.

24 References Bais, A.F., C.S. Zefros, and C.T. McElroy, Solar UV measurements with the double- and singlemonochromator Brewer Ozone Spectrophotometers, Geophys. Res. Lett. 23, , Bodhaine, B.A., R.L. McKenzie, P.V. Johnston, D.J. Hoffmann, E.G. Dutton, R.C. Schnell, J.E. Barbes, S.C. Ryan and M. Kotkamp, New ultraviolet spectroradiometer measurements at Mauna Loa Observatory, Geophys. Res. Lett., 23, , Cebula, C.B., G.O. Thuiller, M.E. VanHoosier, E. Hilsenrath, M. Herse, G.E.. Brueckner, and P.C. Simon, Observations of the solar irradiance in the nm interval during the ATLAS-1 mission: A comparison among three sets of measurements - SSBUV, SOLSPEC, and SUSIM, Geophys. Res. Lett., 23, , Craig, R. A., The Upper Atmosphere: Meteorology and Physics, 509 pp., Academic Press, New York, Dahlback, A., Measurements of biologically effective UV doses, total ozone abundances, and cloud effects with multichannel, moderate bandwidth filter instruments, Appl. Opt., 35, , Gibson, J.H. (ed), Criteria for Status-and-Trends Monitoring of Ultraviolet (UV) Radiation: Recommendations of the UV-B Monitoring Workshop (March 1992), National Atmospheric Deposition Program, Natural Resource Ecology Laboratory, Colorado State University, Fort Collins, CO, Gibson, J.H. (ed), Justification and Criteria for the Monitoring of Ultraviolet (UV) Radiation: Report of UV-B Measurements Workshop (April 1991). National Atmospheric Deposition Program, Natural Resource Ecology Laboratory, Colorado State University, Fort Collins, CO, Harrison, L., J. Michalsky and J. Berndt, Automated Multi-Filter rotating shadow-band radiometer: an instrument for optical depth and radiation measurements. Appl. Opt. 33, Harrison, L., J. Michalsky, Objective algorithms for the retrieval of optical depths from ground-based measurements. Appl. Opt. 33, , McKenzie, R.L. and P.V. Johnston, Comment on Problems of UV-B radiation measurements in biological research. Critical remarks on current techniques and suggestions for improvements by H. Tüg and M.E.M. Baumann, Geophys. Res. Lett., 23, Shaw, G.E., Aerosols at Mauna Loa: Optical Properties, J. Atm., Sci. 36, , Shaw, G. E., Solar spectral irradiance and atmospheric transmission at Mauna Loa Observatory, Appl. Opt. 21, , Stamnes, K., C. Tsay, W. Wiscombe and K. Jayaweera, Numerically stable algorithm for discreteordinate-method radiative transfer in multiple scattering and emitting layered media. Appl. Opt., 27, ,

25 Stamnes, K, J. Slusser and M. Bowen, Derivation of total ozone abundance and cloud effects from spectral irradiance measurements, Appl. Opt., 30, , Thompson, A., T. Early, J. DeLuisi, P. Disterhoft, D. Wardle, J. Kerr, J. Rives, Y. Sun, T. Lucas, T. Mestechkina and P. Neale, 1994 North American Interagency Intercomparison of Ultraviolet Monitoring Spectroradiometers. J. Res. NIST. (To be published May/June 1997). Tug, H. and M. Baumann, Problems of UV-B radiation measurements in biological research. Critical remarks on current techniques and suggestions for improvements Geophys. Res. Lett., 21, , UV-B Monitoring Workshop: A Review of the Science and Status of Measuring and Monitoring Programs, Science and Policy Associates, Inc., Washington, DC, March 10-12, VanHoosier, M. E., et al., Absolute solar spectral irradiance 120 nm nm (results from the Solar Ultraviolet Spectral Irradiance Monitor - SUSIM - Experiment onboard Spacelab 2), Astrophys. Lett.. Commun., 27, ,

26 Appendix A. Testing of an Instrument's SA U-1000 Double Monochromator Testing of an Instruments SA U-1000 Double Monochromator Lee Harrison & Jerry Berndt, ASRC The US Military Academy, West Point NY, owns an early Instruments-SA U-1000 double monochromator. Capt. Augustus Fountain at West Point has been very gracious in letting us impose on him to do the tests reported here. This instrument is equipped with 1,800 g/mm gratings, and a cosecant drive (so that it's natural mechanical scale is wavenumber rather than wavelength). Thus it is not an exact duplicate of the model variant we would use for UV spectroradiometry (that would use 3,600 g/mm gratings and a sine-drive), but we were unable to locate an instrument with the sine-drive that we could gain access to, and so elected to test the West Point instrument as certainly being more instructive than testing none at all. Previously in our laboratory here we had developed a variant of the electronic drive-control system used for the RSI Monochromator-based field instrument that is more flexible in the range of stepping motor systems it can handle. (The one we fabricated for the RSI-Instrument was optimized [and limited] to that specific application.) We tested this drive system against a 1 meter Acton monochromator we have at ASRC that is a fairly good surrogate of the ISA in so far as drive electronics a-re concerned. This drive system also included the signal acquisition and digitization from a silicon photodiode detector; we wished to provide our own drive system to guarantee that motion/sampling sequences were completely controlled and understood, and substitute our own detector to avoid both potential interface difficulties and any chance of destruction of our host's PMT detector. A Si photodiode would not be the detector of choice for atmospheric UV spectroradiometry, but is ideal for laboratory use when testing against bright line sources, as it possesses linearity over wide dynamic range, effectively instantaneous recovery from high signal levels, and can't be damaged by inadvertent large exposures. For the testing the instrument was illuminated with a low-pressure Cd & Hg discharge lamp to provide a spectrum with Figure 2 Raw Cd & Hg line spectrum used for testing. accurately known line positions. The raw line spectrum as observed by the U-1000 is shown in figure 1. In this figure wavenumber is decreasing to the right; the line features used for subsequent wavenumber/wavelength testing are listed in table 1 that follows. This spectrum shown does not include several features that were discarded because they were multiple lines too close to resolve. All of the wavelengths cited in the table are taken from standard tables for the lines measured in air at I bar and 15' C, and the wavenumbers are then directly computed. 24

27 Because these lines have been so commonly uses as secondary wavelength standards we are confident that the cited values are accurate to one least significant digit shown in the wavelength column. Table 1: Cd & Hg emission lines used for our tests Wavenumber 1 Wavelength, nm Stepcount The stepcount in table 1 is arbitrary, but was established by the number of steps required to move from a position close to the short-wavelength limit of the instrument. The entrance and exit slits were adjusted to 100 µm, and for simplicity the two intermediate slits between the stages were left wide open. The 100 µm slits yielded optical slitwidths ranging from approximately nm for the shortest wavelengths, to for the 643 nm line. These narrow optical slitwidths make accurate line centroid retrieval a relatively easy task given reasonable signal-to-noise. Figure 2: 61 repeated scans of the nm emission line 1-9 in black, in red, in green, in blue, in purple. The optical slitwidth is =0.030 nm FWFM 1 Note that these are wavenumbers in air computed from the referenced wavelength. This is appropriate for the analysis done here, as the monochromator is air-filled, and the fundamental physics of the grating diffraction are controlled by the wavelength at the grating surface. This is in contrast to the common (but not uniformly applied) convention that optical properties expressed in wavenumber are for the reciprocal lengths in vacuum. 25

28 We checked the primary focus adjustment, and were pleased to see that it had remained so accurately focused during two years without use that no adjustment on our part improved the slit-flinction. Visual inspection suggested minor slit-misalignment, that we did not try to remove. For the tests the entire spectrum range shown in table 1 was scanned in the order shown, with a 1000 step domain approximately centered on each peak being digitized and stored with a sample every 2 steps. Following the measurement of the longest wavelength peak the drive would retrace to a position 1000 steps below the first stored position, and repeat the cycle. All stepping was done with a period of 3 msec between steps. This is much slower than the instrument can be operated 2 but was chosen to ensure no loss of motion. The repeated 61 scans of the fourth of the emission lines at nm is shown in figure 2. Figure 4 Detail of figure 2. Note the recurring feature at 301,870 steps. Examination of figure 2 makes it immediately apparent that the wavelength reproducibility clusters, with sequences of scans grouping together. The first 10 scans are most erratic, and have been discarded for further analysis on the presumption that an instrument that has remained idle for so long should be permitted a warm up. (These 10 discarded scans represent slightly more than 12 hours of continuous running.) Within the best clusters reproducibility was excellent (as good as nm 1 sigma), but then there would be a "jump" to a new domain. Obviously on seeing a behavior like this the immediate hypothesis that comes to mind is that somehow we are "losing" steps... either due to failure of the driving/counting system to accumulate them properly, or that for whatever reason the motor could not actually make them (driven too fast, etc.). 2 Our drive operated the instrument's SLO-SYN M061FDO8 stepping motor with a conventional L/4R drive at the manufacturer's rated current of 3.8 ampere. At this current the motor was only mildly warm to touch, while either stepping or stationary. ISA's spectra-drive system uses much higher currents, and the motor becomes too hot to grip. We are uncomfortable with the thermal stress on this motor for a field instrument, and would consider modifications if the ISA instrument is selected. Our conservative power to the motor coupled with a wish to operate at speed where we were sure steps would not be lost produced a stepping rate where a full scan and retrace took approximately 80 minutes. 26

29 We present figure 3 to demonstrate using the evidence within the data that neither of the above simple explanations make sense: the signal resolution and noise floor are so good that you can see small repeating "glitches" observable on the shoulders of the peaks where the derivative of the signal is large. Note that these features stay in exactly the same place on all scans relative to step count, while the peaks wander back and forth. They can't be due to noise. There are several possible mechanical explanations; we think the most probable to be very small local machining imperfections of the cosecant bar. Alternative explanations include local cyclic error in the mechanical reduction from the motor to the leadscrew, or a local binding spot or "drunken" turn on the leadscrew. For any of these causes the data remain a- compelling demonstration that the stepping count and motion input to the drive mechanism were accurate. If our assignment of the physical cause to imperfections of the cosecant bar is correct then the fidelity of the drive/leadscrew mechanism itself is excellent. The wander in the wavelength reproducibility would then be due to angular shifts either in the connection of the cosecant bar to the grating shaft, or to grating shifts relative to the grating shaft. (Note that with respect to the latter we were operating the instrument with the intermediate slits wide open, so that a grating could make a small shift without affecting the throughput.) All following analyses are based on the retrieval of the effective center of the line shapes present in the data. There are a variety of common algorithms to do so, and the optimal use and true accuracy of such methods remain vigorously argued issues among optical metrologists. Given the high signal-to-noise in these signals, narrow optical slitwidth, relatively flat baselines, and the absence of overlapping peaks the accurate recovery of the effective line centers is much easier for these data than it can be in the general case. All data presented here were analyzed by two methods, and results compared as a test of potential method-dependent error. The first method is commonly known as the "centroid method" or "method of moments." It consists of computing C = S( x) xdx 0 0 S( x) dx where S(x) is the measured signal at position x. This method is simple to code and fast, and makes no assumptions about the peak shape or symmetry thereoe However it does maximally weight the impact,of the wings of the peak, and hence also is maximally sensitive to baseline removal (if done), or baseline magnitude 3 Figure 5 Linear Center-fit technique line 3 It also demands a choice about the domain of the integration, that obviously cannot be 0 to - in practice. T'his interacts with the choice of baseline removal. 27

30 The second method is the double linear-fitting technique shown in figure 4: the regions from 20% to 80% of the maximum amplitude to either side of the peak were individually fitted with a least-squares line, and then position of the intercept of the two lines was taken as the center fit. This algorithm works well when (as is the case here) the peaks are isolated, signal-to-noise is high, and the instrument slit-function is a reasonable approximation to the ideal symmetric triangle. For both methods linear baseline removal was done on the data before the centroid or center-fit was computed. The results were then compared, and all peaks of all scans except three peaks in scan 7, had centroids and center-fits that agreed 4 to nm or better. Two of the three discrepant peaks showed anomalous noise in the tails that would preferentially bias the centroid algorithm. The third had a asymmetric peak that would cause the accuracy of either method to be in doubt. The question of why scan 7 was peculiarly affected remains moot, but we think that major linepower transients in the building at its time are the likely explanation. In any event this scan is one of the first ten that we will not include in the linearity and reproducibility analysis. Figure 5: Temperature and centroid ª stepcount vs. scan number Figure 5 show the difference of the centroid positions for peaks four through eleven (404 through 643 nm) for each scan from scan 3 (arbitrarily chosen) and the temperature of the instrument casting. During the three days of testing the instrument temperature ranged from 20.5 to 21.7 'C with a time-series that shows only a weak diurnal component. The data in figure 5 make it apparent how anomalous the first 10 scans are relative to all the rest. Neglecting the first ten an obvious anticorrelation of temperature vs. centroid is apparent for all the wavelengths. Using multivariate regression it explains = 40% of the variance in the 404 nm position, and less than 25% of the variance at 643 nm. The same data are replotted in figure 6 as a classic scattergram of A stepcount vs. temperature. The time-line trajectory through these is also shown for the sixth peak. Again this plot emphasizes how anomalous the first 10 scans are relative to the others. 4 The conversion from step-scale to wavelength is discussed subsequently. 28

31 Figure 6: Centroid ª stepcount vs. temperature We then analyzed the centroid data as follows: 1. A linear fit of centroid stepcount vs. the wavenumber (ftom table 1) was performed to obtain the coefficients m, b for centroid_stepcount = m * wavenumber + b this establishes the wavenumber calibration scale for the instrument. Note that wavenumber is the independent variable in this least-squares problem because the variance to be minimized is that of the centroid step-count from the fitted line. 2. Given m, b then the following can be trivially computed: M wavenumber = (centroid_stepcount -b)/m - wavenumber; M wavelength = 10,000,000 /((centroid_stepcount -b)/m) - wavelength; The M quantities are the deviations from the linear fit in question. This process was done twice, once allowing each scan to have an individual linear fit, and then using a single linear fit from scans for all the data. The results of the deviations with respect to wavelength are shown in figures 7 and 8 respectively. 29

32 Figure 7: Deviation from linearity of individually fitted scans. Figure 8: Deviation from linearity given a single fit for all the scans. 30

33 Figures 7 and 8 contain the basic information about the wavelength reproducibility and linearity of the instrument. T'he deviations from linearity are systematic and reproducible; they are less than 0.01 nm for all observed lines shorter than that at 643 nm, and 0.02 nm in that case. The cosecant mechanism has a control slope that must go to infinity as wavenumber goes to zero, and so such a mechanism will be expected to get in trouble at longer wavelengths (this is one of several reasons why a sine mechanism is preferred). The total wavelength-assignment error budget of the instrument can be seen in figure 8. Your eye notes the extremal range of the variances in this figure; were one-standard-deviation error bars plotted rather than all the points then the deviations around the means would appear half as large. Again, momentarily neglecting the 643 nm line, the total error budget is made up of roughly equal contributions from the deviations from linearity and reproducibility. This worst-case error budget is less than 0.01 nm, except for the linearity deviation of the 643 nm line. Readers should note that this apparent error budget could be reduced by applying temperature dependent corrections for the deviations, that as noted above can explain somewhat less than half the variance in the reproducibility over the range of temperatures tested. Further, we can attempt to remove the nonlinearity residuals; a fourth-order fit is shown in figure 8. An optimist applying both of these can reduce apparent standard deviations of the residuals below nm at all wavelengths, with only rare outliers exceeding nm. While I find this tempting, I think it is a bit too optimistic for my taste given only the data presented. In the case of the temperature corrections one would like to see a somewhat larger temperature range, and series that better decouple the temperature from other possible time-dependent effects. In the case of the linearity one would wish to see a greater density of line features in the spectrum, so that one could be sure that the fit was truly removing a function that was a fairly large scale feature of the system, rather than simply aliasing a fit through errors that might either be locally random or periodic with high spatial frequencies. Figure 9: Optical Slit-Widths vs. Wavelength The bold magenta line is the theoretical FWHM in wavelength for 100 µm slits, on a double additive I m monochromator with both gratings turning "tip toward output" Figure 9 shows the apparent FWHM optical slit-widths retrieved from the data by the double linear-fit algorithm. At the shorter wavelengths the agreement with the theoretical prediction is excellent. The discrepancy at the nm line is expected because this emission line is an extremely broad one (listed in the line catalogs as "nebulous"). The large discrepancy at 546 nm is due to the fact that this line saturated the A/D converter, so that the apparent line shape 31

34 is trapezoidal and the amplitude at which to determine the FWRM is underestimated. (Under these conditions the double linear-fit algorithm can still retrieve the line center with some loss of accuracy.) However we don't understand the discrepancies at 576 and 643 nm. Both are narrow emission features; the 643 nm Cadmium line is one of the narrowest discharge emissions known, and was the standard for interferometry before the advent of lasers. We emphasize that these optical slit-widths were extremely reproducible. Figure 10 shows the fractional variation of slit-width determined for each emission line, vs. the scan sequence. (These are color-coded as black, brown, red... blue, violet, and then black with dots, and brown with dots, for the emission lines in short to long wavelength order.) The slitwidths for all of the lines except the two longest appear to be stable to ±2%. The time variation of the stepcount position for the 435 nm line is superimposed; there are no significant correlations with it. Figure 10: Variation of Optical Slit-Widths vs. Scan An alternative explanation for the variation in the apparent line widths is to assume that the slitwidths were perfectly stable, and that the noise seen in figure 10 is a conservative test of the reproducibility of the linear-fit retrievals, since the width is computed by differencing two fits for each emission line. Given this assumption, then the uncertainty of the individual line center retrievals used for figures 7 and 8 would be approximately 2 1/2 x0.02x0.03 nm (at I sigma) = nm. Thus the reproducibility of the line centers is an instrument effect rather that an artifact of the retrieval, but this basic premise is clear simply by inspection of figure 2. Probably there is some variation in slitwidth, but it is less than 2% for all wavelengths except the longest (where the motion of the cosecant bar is stressed, and greater irregularity is expected and seen here), and below our ability to resolve. In any event these data suggest that variation in the slit position is not a likely explanation of the clustering line positions. The first-order theory of monochromator aberrations is wavelength independent, and hence would not explain the apparent broadening of the optical slit-widths at longer wavelengths. This remains a minor mystery to be sorted out, but we don't preclude analytical error on our part. 32

35 Appendix B. Quality Assurance Procedures for BroadBand Measurements (Excerpt from the network's World-Wide-Web Application Pages: QA: Characterizations of USDA Broadband UV-B Pyranometers) Quality Assurance The USDA UV-B monitoring program recognizes the importance of instrument stability in establishing a long-term climatological record. The network documents the stability of its broadband instruments through annual calibrations and annual recharacterizations of each instrument's spectral response. Initial instrument characterizations were first established by submitting each of the network's initial broadband meters to the National Institute of Standards and Technology (NIST) for an evaluation of their cosine and spectral response. Plots of the NIST characterizations are available at the end of this document. The cosine response of a broadband meter is a measure of the departure of the angular response of the instrument from that of Lambert's cosine law. This law states that the response of an ideal detector to constant and uniform light should decrease in proportion to the cosine of the angle of incidence of the light. In practice the response is a function of the design geometry and manufacturing of the meter. It is anticipated that once determined, the cosine response characteristic of an individual instrument will not change unless the instrument becomes damaged. In field applications the stability of cosine response, once characterized, should only be dependent upon maintaining it in a level plane. In the USDA monitoring program this is checked and adjusted annually. The spectral response of a broadband meter is a measure of the instrument's response to light at specific wavelengths - typically generated with a high resolution scanning monochromater and xenon arc source. Two important quality attributes of the instrument can be determined and tracked through this characterization; the instrument's central wavelength stability and stability of its characteristic shape. Because of the importance of documenting uniformity in these parameters over the life of the USDA's monitoring program, spectral characterizations are remeasured approximately annually when each instrument is returned to the manufacturer for recalibration. Calibration of the USDA broadband meters follows the theory of Grainger et al., That is, a calibration constant for a selected broadband meter was derived from a regression of the integrated spectral response of a spectroradiometer against the signal produced by the selected meter. This meter serves as the primary reference for the network. Results of the regression yielded a relationship of ± 0.11 (Watts/meter 2 )/Volt. The relationship is referenced to the 296 nm monochromatic radiation peak of the broadband's spectral response function and integrates energy over the range of nm. It should be noted that referencing to other peak wavelengths and ranges will result in a different relationship (constant). Annual calibrations follow ASTM E_824 methodology ending with the adjustment of each test meter's signal to the signal of a reference instrument maintained according to ASTM method E_816 by the calibration facility. The merits of this approach are discussed by DeLuisi et al. (1992). Presently, the calibration facility is located at Yankee Environmental Systems, Inc. in Turners Falls, MA. 33

36 Broadband Cosine Response Figure 1 illustrates a typical cosine response of a Yankee Environmental Systems UVB-1 Pyranometer used in the USDA UV-B monitoring program. The black continuous line of the plot represents an ideal cosine response. The colored line is the actual cosine response function of meter serial number which was determined on the date listed in the legend in the center of the plot. Figure 1: Typical Cosine Response of the YES UVB-1 Broadband Pyranometer It should be noted that the network does not apply cosine correction to its broadband measurements. Grainger et al. (1993) have noted that the R-B type meters are "nearly azimuthally independent" citing standard errors of relative cosine responses of less than 0.1% for solar zeniths angles < 65 degrees and less than 2.7% everywhere else. Broadband Spectral Response Figure 2 illustrates three successive spectral response characterizations of a single Yankee Environmental Systems UVB-1 Pyranometer. A legend in the upper right corner of the plot indicates the serial number of the meter being displayed, its characterization date and the agency that performed the characterization. The black line of the plot represents the original spectral characterization of the instrument by NIST. Differences in the dynamic range of subsequent overlays reflect differences in laboratory capabilities. Most characterization facilities are not able to achieve the 10-5 measurement thresholds that have been achieved at NIST. This plot also illustrates the differences in spectral response that will be noted when differing materials are used for the meter's transparent dome. The red line (the plus symboled plot and highest line in the shorter 34

37 wavelengths) represents a dome constructed with UV-grade fused silica whereas the blue line (triangle symbol and line with the shortest spectra), one with Schott 280 glass. Initial characterizations of the USDA meters at NIST (black line and open circle symbol) indicate that the meters were constructed with Schott 280 glass. It should be noted that all instruments currently used by the network have UV-grade fused silica domes. Figure 2: Sample Spectral Response of a YES UVB-1 Broadband Pyranometer NIST characterization of the initial broadband instruments deployed by the USDA network indicated that 8 of 13 instruments exhibited peak spectral responses at 298 nm while 3 instruments had a peak response at 296 nm. Two instruments peaked at 300 nm. Calibration of Broadband Meters The USDA monitoring program applies the (Watts per m 2 )/Volt calibration constant uniformly to all its raw voltages to convert them to units of energy. It has been suggested that calibration accuracy may be improved through the use of zenith specific correction factors (Yankee Environmental Systems, 1995). The following table presents these corrections for a Yankee broadband meter calibrated using a referenced integrated wavelength range of nm (1.968 (W/m 2 )/V)) with a peak spectral response at 296nm. It should be noted that referencing to other peak wavelengths and ranges will result in different correction factors. 35

38 Table 1. Correction Ratios For Various Solar Zeniths Angles Solar Zenith Angles Ratio [From Yankee Environmental Systems, Inc] Because this correction is based on clear sky measurements and specific assumptions about the relationship of direct to diffuse radiation (Green et al., 1980) the USDA has decided not to apply zenith specific corrections to its broadband measurements. At this time, the network recommends that individual data users determine the validity of the above assumptions in the context of their use of the data and then if appropriate and necessary, apply the specific correction. NIST Characterizations of USDA UVB Broadband Pyranometers Ten of the Yankee UVB-1 Broadband Pyranometers that are deployed in the USDA UV-B Monitoring Program were characterized by the National Institute of Standards and Technology (NIST) for their spectral response, cosine response and linearity prior to their field installation. Since the original characterizations, the instruments have been reevaluated by Yankee Environmental Systems to determine their stability. Documentation of this stability (or instability) can be demonstrated by viewing the preceeding plots. Figures 3 and 4 display the original cosine and spectral responses of the networks original broadband meters. Variability in the plots reflects the between instrument variance of the measurements. 36

39 Figure 3: NIST Cosine Characterization of the USDA's YES UVB-1 Broadband Pyranometer Figure 4: NIST Spectral Characterization of the USDA's YES UVB-1 Broadband Pyranometer 37