Overview of Met Office Vaisala FD-12P Present Weather Sensor
Consistency Trial

P D Shearn

Met Office Beaufort Park Easthampstead Wokingham Berkshire RG40 3DN United Kingdom

Tel +44 (0)1344 855830 Fax +44 (0)1344 855897 e-mail:

Abstract

Intercomparisons of Vaisala FD-12P Present Weather sensors will be described. These instruments were selected for network use by the Met Office and the intercomparisons were carried out at Beaufort Park near Bracknell in the UK. Instruments were tested close to each other in batches of eight. It was found that a period of at least twenty days with various meteorological conditions was necessary to give a useful result. This work was carried out to show the levels of agreement between identical units and to ensure that all were "soak" tested before starting operational use. The experimental procedure, logging system and data analysis methods will be described and results from one of the batches processed will be discussed to provide the overall conclusion to the work.

Introduction

The purpose of this trial was to ascertain the range of differences between FD-12P instruments in a comparison with their mean present weather code in a large variety of different weather types. This allows an assessment of the accuracy of each instrument, even though no observer was present on site. This method is validated through comparison with one of the proven instruments that had already been on trial and that already had a history of observations associated with it. The results would add confidence in the instruments by examining the consistency of the reports between the sensors. Due to the operational demand for instruments, the same instruments could not be made available continuously throughout the trial.

Experimental Site and Hardware

The final form of this system is described as follows:

Eight SensorsEight Vaisala FD-12P present weather sensors were set up on the Met Office trial site at Beaufort Park. Their locations and associated identifiers can be seen on Fig. 1 on the following page, and the minimum spacing between adjacent sensors was 18m. Three instruments of long standing stability (ISU10, ISU14 and ISU17) were used to provide a ‘virtual’ standard instrument by taking the mode of their readings.

Logging systemEach sensor was connected to an Intelligent Sensor Unit (ISU). The ISUs were all connected together on a network. Data elements were recorded once per minute and stored on a PC in daily files.

Data ProcessingOnce per week, the daily data files were transferred to the data processing PC. An Excel macro was then run on each daily file. The macro replaced any missing data with nulls or deleted any duplicate data and sorted the dataset into sensor and weather element type. It then generated a one minute mode of the present weather code from the three standard instruments’ outputs and compared the test instruments’ output to this ‘virtual’ standard instrument. A similar procedure was also carried out for the other measured elements (i.e. visibility, rain rate). The macro also produced graphical and tabular results for the data and for minute on minute corrections to the mode (or mean) for each instrument and element. A mean daily value for the corrections for each instrument and element was also calculated. Another Excel spreadsheet was then used to take these data and show how these mean daily corrections (when normalised compared to the number of occurrences in the day) varied over the period of the experimental run. Only significant weather types (i.e. WMO code 40 and above) were used for the purposes of this trial as in operational use the Semi-Automatic Meteorological Observing System (SAMOS) present weather arbiter would use other sources of data to produce an output for mist or fog.


fig. 1 : Site Plan /
fig. 2 : FD-12P Sensor Head

FD-12P Operating Principle

The FD-12P (see fig. 2 above) consists of an infra-red transmitter and a receiver for forward scatter measurement, a temperature sensor, a capacitive precipitation sensor and a controller. The multiple functions of the sensor (producing WMO automatic present weather code, visibility, rain and snow rates) are based on combining the three independent measurements with a Vaisala patented method to produce an accurate output. To ensure correct operation the instrument must be calibrated (both optically and for its temperature sensor) and have the correct reference values set in its on board software. The instrument self corrects for low levels of optical surface contamination over time.

Results and Discussion

A case study on a batch of instruments tested between the 1st March 2002 and the 28th April 2002 will be examined. This period contained only 25 days with significant weather occurences, and three of the instruments (ISU12, ISU13 and ISU16) that were installed part way through the run period only experienced 10 days of significant weather. The data from this run is summarised by the spreadsheet in two forms, a graph (fig. 3) showing the spread of daily average corrections to the WMO Present Weather Code for each instrument (these averages are weighted against the proportion of the day taken up by significant weather), and Table 1 showing the mean of the average daily corrections, and their standard deviation. After a year of investigation, the characteristics of instruments outside the normal population have become apparent, and can be easily identified from the above mentioned graph and table. From analysis, limits have been set for the maximum allowable difference from the mode for the output from each instrument. This allows the ready identification of instruments that have been set up incorrectly, need optically calibrating, or have component failures that may or may not have been logged by the instrument’s on board diagnostic software.

To demonstrate the validity of this analysis, the results from one day (28th April 2002), are shown in fig. 4 with a frequency analysis on the corrections to the mode WMO Present Weather code for each instrument.

fig. 3

Table of RMS, Mean and Standard Deviation values for the Average Daily Corrections to the WMO code Mode
ISU10 / ISU11 / ISU12 / ISU13 / ISU14 / ISU17 / ISU15 / ISU16
RMS / 0.05 / 0.22 / 0.06 / 0.06 / 0.08 / 0.11 / 0.13 / 0.27
Average / 0.02 / -0.10 / 0.02 / 0.00 / -0.06 / 0.08 / -0.05 / 0.20
SD / 0.06 / 0.20 / 0.06 / 0.06 / 0.06 / 0.08 / 0.12 / 0.18

Table 1

From a years worth of data it has been found that the population of normally behaving instruments lies within an RMS correction limit of 0.1. It can be seen that from Table 1 that instruments ISU11, ISU15, ISU16 and even Standard Instrument ISU17 do not in this case meet the criteria. Further investigation revealed that for ISU17 and ISU15 the problem was resolved by cleaning the instruments optics and recalibrating them (excessive dirt had built up on the optics, possibly due to recent grass cutting). ISU11 and ISU16 did not however reveal any physical problems, and further investigation was carried out on individual daily data sets (N.B. Further summary data for other instrument runs as above is available from the author).

fig. 4

As can be seen from fig. 4, Most instruments show little variation from each other. However ISU11 produces notably different results for 28th April 2002. The open squares represented by ISU11 can be seen to be higher during the period of measurement on the day. This indicates a higher WMO code reading of snow and rain, rather than rain or drizzle. From this it can be postulated that there is a possible error in the instruments temperature sensor causing false snow reporting. However this graph provides very little evidence of error in ISU16 apart from a very small number of outliers. To find this evidence we must look at a frequency analysis of the corrections for that day, as shown in fig. 5 below.

fig. 5

From the above graph outlier peaks are evident for both ISU11 and ISU16s cases, a small peak to the left for ISU11 and a smaller peak to the right for ISU16, with also a depressed central peak for ISU11. The ISU11 peak corresponds to the higher reading values shown on fig. 4, for ISU16 the correlation with fig. 4 is less obvious, possibly the lower readings around a 1000 minutes and at 1250 minutes into the day. These two instruments were therefore not deployed and put aside for further investigation later.

Conclusion

Over the period of a year, 38 instruments have been tested with 8 instruments initially falling outside our acceptance criteria. Six of those instruments met our requirements in later runs after simple maintenance or the correction of setup values in the instruments’ on board software. Only the two instruments in the case study above have shown problems resistant to a simple solution. Further investigation into these two instruments will be carried out once the resources are available. Recalibration is rarely needed with checks being performed every six months, however excessively polluting environments can cause severe contamination of the optical lenses and so prompt cleaning and recalibration is required in these circumstances. If the Vaisala Present Weather sensor is setup correctly, passes the acceptance criteria of the consistency trial and receives proper maintenance as required, then it can be considered acceptable for Met Office purposes.

Acknowledgements

Thanks go to Mike Molyneux, Stuart Goldstraw, Jon Buck, Stuart McRobbie, Bernie Ryley and Vanessa Willot of the Met Office and to Mike Brettle at Vaisala.

References

Vaisala 1998: Weather Sensor FD12P User’s Guide