The multifilter rotating shadowband radiometer (MFRSR) takes spectral measurements of direct normal, diffuse horizontal and total horizontal solar irradiances. These measurements are at nominal wavelengths of 415, 500, 615, 673, 870, and 940 nm. The measurements are made at a user-specified time interval, usually about one minute or less. The sampling rate for the Atmospheric Radiation Measurement (ARM) Climate Research Facility MFRSRs is 20 seconds. From such measurements, one may infer the atmosphere's optical depth at the wavelengths mentioned above. In turn, these optical depths may be used to derive information about the column abundances of ozone and water vapor (Michalsky et al. 1995), as well as aerosol (Michalsky et al. 1994) and other atmospheric constituents.
A silicon detector is also part of the MFRSR. This detector provides a measure of the broadband direct normal, diffuse horizontal and total horizontal solar irradiances. A MFRSR head that is mounted to look vertically downward can measure upwelling spectral irradiances. In the ARM system, this instrument is called a "multifilter radiometer (MFR)". At the Southern Great Plains (SGP) there are two MFRs; one mounted at the 10-m height and the other at 25 m. At the North Slope of Alaska (NSA) sites, the MFRs are mounted at 10 m. MFRSR heads are also used to measure "normal incidence" radiation by mounting on a solar tracking device. These are referred to as normal incidence multi-filter radiometers (NIMFRs) and are located at the SGP and NSA sites. Another specialized use for the MFRSR is the narrow field of view (NFOV) instrument located at SGP. The NFOV is a ground-based radiometer (MFRSR head) that looks straight up.
The direct normal component is calculated from the difference of the global and diffuse measurements. At the start of a measurement series, a global measurement is first taken with the shadow band in the home position; fully rotated counter-clockwise without hitting support arm. The shadowband is then rotated from the home position and stops in three positions before returning home. The first and third stops are just before and just after shading the diffuser. At the second stop the diffuser is completely shaded (see picture below). Measurements at the first and third stops are used to correct the error introduced by the shadowband shading a portion of the sky in addition to the sensor.
This is an image of the NSA C2 MFRSR correctly set up to shade the sensor on the 2nd stop. Notice how the white disk in the middle of the sensor head is completely shaded.
Filter Definitions
Filter1 & Filter5 are for aerosol measurements.
Filter2, Filter3 & Filter4 are for ozone measurements.
Filter6 is for water vapor measurements.
Here is a nice plot of the spectral response of each filter for the MFRSR. Notice how each has a curve centered around the wavelengths we describe. It also shows the magnitude difference expected between the different "colors".
The metrics for the MFRSR will have a lot of missing data overnight. Since there is no incoming radiation, the instrument does not collect the data. There are also a lot of nuisance flags that trip with this instrument. The valid min for a lot of the radiation variables is 0 W/m2/nm, but given the noise that can creep in from the instrument, logger, data cables, etc, the values can drop below 0 at times.
The top plot is the Hemispheric Irradiance for all bands. A sunny day is the best time to determine if the instrument is operating correctly. The hemispheric data should follow a nice smooth curve, peaking at solar noon (dashed-yellow line).
Diffuse Irradiance
The 2nd plot is the diffuse irradiance. This should follow a similar trend as the hemispheric data, but may be a little noisier.
Direct Normal
The 3rd plot is the Direct Normal Irradiance. It will have a little broader curve, but should trend similar to the other 2 plots on a clear day!
Head Temperature and Logger Voltage
This is a diagnostic plot of the head temp and the logger voltage. The voltage will appear noisy due to the small y-scale it is plotted on.
MFRSR Cloudy Day
A cloudy day may look like bad data, but if you know the variables it should not be hard to pick out! The Hemispheric and Diffuse should follow the same trend, albeit a little noisier or with spikes. The big clue is that the Direct Normal Irradiance is mostly near 0 W/m2/nm. We are not getting any direct shortwave radiation from the sun, because it is getting diffused by the clouds.
Currently, the only comparison plots are at SGP and NSA C1 as these are the only 2 sites that also have a co-located NIMFR. These pull in data from the nearby MFRSR E13 instrument as well as the NIMFR. Each filter is plotted out along with the broadband comparison. The general trends should be very similar, but the actual values may differ a little depending on the filter.
Known issues for this instrument that MAY NOT need to be mentioned in your DQA's:
High Latitude 24 hour data comparison
At high latitudes the MFRSR is not able to record correct data for a full 24 hours/day. During days without sunrise/sunset (NSA Summer) the shadow band is not able to make a full rotation because the band would strike the support arm. To ensure the instrument is not damaged, the MFRSR is programmed to stop recording data and enter sleep mode +/- 2 hours of local solar midnight. The NIMFR does not have this problem and will continue to record data. This means the comparison plot of the two will not match +/- 2 hours of local midnight.
Head Temperature - NSA
The head temperature is consistently failing QC at NSA. This is a known issue and does not need to be reported.
Notice that the head temp stays between 41° C and 43° C throughout the day. The head temp is set to keep it at 40° C and this is about as good as one can expect. A prime example of excellent head temperature stability. Though 40° C is the desired temp, head temps in tolerance can range from between 35° C and 45° C. In some particularly hot climates, temps can go above 45° C. Consistency between 35° C and 45° C is desired, but often the daily range fluctuates significantly more than in the above plot. Head temperature is an indicator of instrument health. For the most part, if it stays between 35° C - 45° C things are acceptable.
Logger voltage should ideally be flat between 12 and 14 Volts. It can range a bit higher or a little lower and is sometimes a little noisy. The above example is as good as it gets. Most instruments will have a little noise with an occasional dip or spike.
Head_temp example with problems
Compared to the above good example, this temperature plot goes way out of the 35°C - 45°C range. It has significant noise early in the day. A head_temp like this should be reported.A not too uncommon occurrence. These happen from time-to-time and can safely be ignored. If the occur frequently they should be reported.
Clock reset with brief missing data at C1
The SGP MFRSR at C1 has a daily spout of missing data at approximately 0608UTC. This is due to the clock resetting, and is perfectly normal. See DQPR 4928 or DQR D141207.3 for more information.
Spikes to negative values
During days with changing sky conditions, negative values may appear in the diffuse irradiance data. These are an artifact of how the data are collected. From the IM from the ASI MFRSR: "One measurement set takes a bit less than 10 seconds to complete. There are four measurements: Nadir, 1st-side-band, sun-blocked, and second-side-band. When there are partly cloudy conditions, with fast moving distinct clouds, a side band measurement can be shaded by clouds, and the other measurements can be in full sun, or at least "stronger" sun (or vice-versa). This can lead to the negative (and positive) spikes seen." As such, these spikes do not need to be mentioned in your DQAs unless unusually large. Check out DQPR 5390 for more information.
Large Irradiance Difference Spikes
Sometimes, you may notice a spike larger than 60 W/m^2 between the down_short_hemisp and derived_down_short_hemisp irradiances that occurs at multiple facilities at around the same time each day. This is likely due to a dusty rainy gust front, that brought with it dirt and debris as the front moved through. This leads to dirty PSP domes, and thus, the spikes. This can be fixed by cleaning the domes, which already occurs every two weeks as part of preventative maintenance. As such, these spikes do not need to be mentioned in your DQAs unless unusually large.
Past problems that do need to be mentioned in DQAs:
Oscillation in a channel
This example shows an oscillation in the 500 nm (green) channel. It is most obvious in the direct normal radiation data. Mentor indicated that it is possibly an issue with the cosine correction data and the instrument should be re-characterized and calibrated. See DQPR 3982
Spikes to Negative Numbers
These spikes to negative numbers could be a true issue, or could be something we will just have to live with. ARM decided that to better calibrate the MFRSR's it needed to take data continuously throughout the day including the night time values. These "dark" values would be used to help compute the offset of the instrument. To get the dark values the data logger manufacture was changed from Yankee to Campbell and the software was updated. Previously the Yankee data logger did not allow any negative numbers to be written to the raw data file, but the new format does allow negative numbers. To better analyze the data, the DQO has requested that all the data remain in the file even when it's obviously incorrect to better facilitate the detection and determination of problems. Therefore it's possible for 'good' data to contain these negative values, even thought they are physically impossible. Although, if the instrument is showing negative values during completely clear sky days or the frequency of negative values gets unreasonable (you will need to determine what that level is) then there is most likely a problem. See DQPR 1710
Shading Problems
The most common problem with the MFRSR is a misaligned shadowband. There are numerous reason why the band may be off. Below are two pictures of the shading band misaligned.
Here is an example of the shadowband misaligned. The first picture is of the 2nd stop position where the sensor should be fully shadded. The second picture is of the 3rd stop position where the band is not intended to shadow the sensor. A problem with the shadowband at one position will typically result in a problem at all three positions.These are plots of the misaligned shadowband. The band was intentionally set off to the side to analyze what response the data would show with the band off. The two pictures above show the band alignment for the worse time period between 17.5 - 18.1 GMT (Hours). The band was then adjusted to be shading the sensor better but still slightly off for the time period between 18.1 - 18.5 GMT. At 18.5 the band was aligned correctly to its original/correct position. At 21:15 the band was adjusted again to better center the shadow.
Obvious shading problem
If only all shading problems were as easy to spot as the following direct and diffuse examples from the same instrument on the same day. The data are from a clear day so the plots should look like the normal plots above.
Obvious bad shading of direct_normal_narrowband_filter# for all filtersObvious bad shading of diffuse_hemisp_narrowband_filter# for all filters
Less obvious shading examples
More common than the obvious shading problem above, are subtle shading problems as shown below. In reality, however, to a trained eye the following examples are actually fairly obvious. Once a person is familiar with identifying shading issues, even those that truly are subtle can be readily identified. The examples below also show why it is important to view both the direct and diffuse plots. In these examples from the same day, the shading issue is much more prominent in the diffuse data than the direct.
Note that between the dashed lines the data across all channels are noisy in comparison to data outside the dashed lines. A single occurrence like this on one day without similar features on prior or subsequent does not mean there is a shading problem. Look for similar features on prior and subsequent days, and across the same hours, to confirm a shading problem.
Figure shows a partial shading problem, found at SGP E4 on 01/13/2005. This plot is obviously different than the “good data quality” plot shown above. The diffuse hemispheric is very noisy in the afternoon hours. The direct normal irradiance is (to a lesser degree) also noisy during this time. Note the distinct between shading “noise” and cloudy “noise”. Shading noise is generally a broadening of the irradiance readings whereas the cloudy noise is much spikier (more erratic).
An example of normal data that appears to be a shading problem
At times, the diffuse irradiance values may appear a bit noisy and assume a shape similar to what one may expect for a shading problem. For example, check out the plot from the E12 SGP MFRSR from October 2015. While the diffuse irradiance values are indeed noisy during the day, we don't see the same feature reciprocated in the direct normal irradiance data. Thus, the shadow band is likely still in the correct position and accordingly no DQPR was issued. When interrogating data in suspicion of a shading problem, it's always a good idea to consult the other plots, as well as other nearby sites on the same day.
Note how even though the diffuse irradiance values are a wee bit noisy, the direct normal irradiance values appear to be fine. A shading problem likely does not exist.
Two Channels Overranging
Below are a few examples of other problems that arise. Generally though, if all the channels do not have the same characteristics as the rest, e.g., spikes or dips caused by clouds, there is likely a problem and it should be reported. Note that 940 (filter 6) can be an exception. This channel is sensitive to water vapor which can be highly variable over the course of a day, while the other channels are not affected by water vapor. Even so, it will never be radically different than the other channels.
Note how the purple and green traces are flat around solar noon in comparison with the other channels. This is a plot of the hemispheric data.
Filter Failure
Occasionally, a filter will go squirrely-squirrel. This means that the filter behaves quite anomalously compared to the same filter at surrounding sites AND other filters at the same site. Some problems include excessive noise, extended drop-offs to very low values, and frequent reports of -9999 values. Two examples are shown.
Above shows erratic behavior of the 673 nm channel at SGP E1. This problem is normally associated with the data logger, a loose connection/cable issue, or a degrading sensor. Also see DQPR 633 for a similar case at SGP E8.Figure above indicates a filter (in this case, the 415 nm channel at SGP E8) reporting values close to 0 W/m2 for the whole day. Sources of this problem are similar to those of the previous example.
Ingest/Calibration File Problem
Figure shows such an example. In this case, the instrument was just brought online. The raw data of the MFRSR was OK. However, the info file, which the ingest uses to process the data, had errors. The result is unusual behavior of a filter (in this case, a strange rise and fall of the 500 nm channel that appears like an upside down bowl). DQPR #223 has more information on this example.
Data Logger Problems Affecting All Wavelengths
Figure shows an excellent example of a data logger problem, completely destroying the data quality of an instrument. When a plot like this is seen, submit a DQPR immediately and suggest logger replacement. See DQPR 153 or DQPR 1620Figure shows a step change in the filter values that coincide with a tech visit. Typically the filter values will have a spike associated with the step change if it is a result of a tech visit to upload new configuration files, change an instrument, ... This may or may not be an indication of an issue. All changes should be reported in DQA reports. If the step change appears drastic a DQPR should be filed. It would be good to send the mentor a note about the change to double check if new calibration coefficients were uploaded to the instrument. DQPR 1640
Broken or bad signal cable
The 415 nm channel had a bad signal cable that was affecting the signal transfered from the MFRSR head to the datalogger. This is seen by the 415 nm channel not behaving any where near the same as the other 5 channels. See DQPR 11
Obvious Problem After Re-Install of Instrument
After data returned from a CM visit to the site on 5/29/2003 the data is obviously not correct. 610, 862 & 940nm channels are not responding like the other channels, and the other channels have step changes that are not correlating. Typically a cloudy sky that causes the data to have sudden changes will cause all channels to jump around at the same time. In this case the MFRSR head overheated and caused some damage to the electronics. See DQPR 74
Extra noisy single channel
The 415nm channel is extra noisy in this plot. The cloudy skies are causing all channels to show some varriation in the strength of the diffuse and direct sunlight, but the 415nm channel is showing an additional variability. A few days after this all the channels flatline to zero and it was determined that the MFRSR head had a major failure. See DQPR 75
Intermittent Spikes
A failing data logger was causing all channels to spike intermittently. This problem is difficult to discover on cloudy days, but the very large diffuse values are larger than possible changes due to clouds. See DQPR 92
“One of these things is not like the other"
The 415nm channel was not behaving like the other channels. This is a clear sky day where the difference is quite obvious, but even on cloudy days the 415nm channel was showing step changes and noise not similarly see in the other channels. See DQPR 199
Shadowband not working at all
The two plots above demonstrate a case where the shadowband at E9 was not covering any part of the sensor. This is determined by comparing the Hemispheric and Diffuse measurements, which at E9 are exactly the same. Also, the shape of the Hemispheric data suggests a clear day, but the Direct Normal shows no data. For comparison, the plot on the right shows the same day but for E11. Notice how the Hemispheric and Diffuse curve shapes and order of the lines are different. Also the Direct Normal is showing plenty of direct beam. See DQPR 220 , DQPR 243 , DQPR 604
Periodic spikes in all channels
The periodic spikes in the data were caused by a failing stepper motor. With the spikes occurring for all channels and only in the Diffuse and Direct values the correct original guess was the band motor. See DQPR 264 or DQPR 300 or DQPR 1211respectively
MFRSR Head Failing
The instrument began to fail in February of 2004. In this case the failure resembles a typical shadowband misalignment with the noisy diffuse data, but this case also has noisy hemispheric data. The shadowband is not used to take hemispheric data so this indicates a more serious problem with the instrument. See DQPR 357
Lightning Strike Damages Logger
A lightning strike near the instrument on 4/11/2005 damaged the logger. The strike most likely occurred near 07:00 GMT, where the instrument was able to restart after some down time between 07:00 - 08:00 GMT but is obviously not correctly working. See DQPR 513Lightning strike at SGP E15. See DQPR 580
Incorrect Lat/Lon set for band motor position
The latitude position of the instrument was incorrectly set, resulting in the band improperly shading the sensor. This may account for the shadowband correctly functioning for part of the day but incorrectly shading in the morning and afternoon. Around local solar noon the band may be able to properly shade the sensor because the sun is closest to directly over head. See DQPR 534
Flat Line around solar noon
The 610nm and to a lesser extent the 415nm channels flat lined around local solar noon due to the instrument saturating. The saturation in hemispheric values then causes a downward dip in the 610nm values. This required a replacement of the MFRSR head. See DQPR 541
Shading issue with 415nm channel
There was a very slight shading problem with the instrument. With the sun angle being shallow and intensities being low in the Arctic the MFRSR has a more difficult time than the NIMFR calculating the direct value. In this case the 415nm channel shows the slight shading issue more than the other channels. This might be due to the larger gain applied to that channel than others. See DQPR 624
Noisy 673 nm channel
This plot shows the 673nm channel noisier than the other channels. Sometimes the cable or connectors could be the problem, but 673nm has been a problematic channel in the past. In this case the sensor head was sent in for refurbishing. See DQPR 626
Bad calibration File after re-installation
The 673nm channel was incorrectly reporting due to a bad calibration file after the instrument was re-installed. See DQPR 1108
Broken Signal Cable
A broken signal cable resulted in all the channels to flat line near zero. The plots auto-scale to match the size of the data, so the plot above looks like the channels have some sort of signal but in reality they are all zero. This broken cable happened sometime during transition at sunrise/sunset or during the night when the instrument is sleeping. See DQPR 1154
Clipping data under cloudy conditions
The first plot shows a case where the 500nm channel was being clipped around solar noon during cloudy conditions. The second plot shows the 415nm channel being clipped in clear sky conditions. Both of these problems are a result of the signal saturating limits of the detector and may be due to the filter allowing too much light past or the instrument having an improperly set gain. See DQPR 1168
Just all messed up data
The instrument was suffering from multiple problems.1) Spikes in the data. The normal spikes in MFRSR data will spike down during cloudy conditions. These are large upward spikes.2) Shading issue seen between sunrise and 15:00 in the morning.3) Detector Temperature not constant throughout the day. The temperature is not too large or tripping flags, but it is following the daily temperature rise. This is caused by solar heating and there's currently no solution.See DQPR 1180
Strange Bumps??
The bump near 19:00 GMT on the first Hemispheric and Diffuse plot is strange and not expected. Also the dip around 22:00 GMT in the Hemispheric plot is unexpected. The primary reason the dip is suspicious is that it appears again three days later at the same exact time (second plot). Also, the Direct Normal component is essentially zero. This is a good indication that the instrument is not correctly shading and the band could have slipped on the stepper motor. See DQPR 1656
Detector Temperature too low & failing min
The head_temp (Detector Temperature) began to fail the MIN threshold around 16:00 GMT.The head_temp continued to fall until it reached "equilibrium" with the ambient temperature, and started to trend with the local Meteo temperature. This indicated that the heater fuse had blown, and the temperature was restored after a site visit repaired the blown fuse. See DQPR 1695
Stepped Jump in Data
The image above demonstrates what is though to be a processing error commonly seen in the data for the MFRSR. In this case the analyst noted a stepped jump in the data around solar noon, most present on clear days and in the diffuse measurements. The mentor has noted this is not an error actually present in the raw data, but is indicative of some sort of processing error. See DQPR 2816
Negative Direct Normal Data
The image above shows negative direct normal irradiance data near the end of the day, which is incorrect. There also appears to be shading occurring. In this particular instance, the shadowband was loose on the motor shaft and not turning with the motor, and the shadowband set screw was stripped out. The shadowband and set screw were removed and replaced, and the shadowband was aligned. See DQPR 6983.
Overnight Spikes in Data
The image above shows overnight irradiance data, which is incorrect. In this particular instance, replacing a blown 1/2 amp fuse and the heater board solved the problem. See DQPR 7436.
Spikes at Sunrise and Sunset
The image above shows large spikes in the direct normal narrowbands at sunrise and sunset. These spikes were both positive and negative and occurred almost daily. This was likely caused by the cosine correction for the 1625 nm channel. See DQPR 10075 and DQPR 10085.
