Comparing observed readings with other stations
You may be thinking of checking and even trying to calibrate your own AWS against other weather data which may be available in the same locality as your base, for example airfield observations for which updates are usually posted publicly at frequent intervals. A little caution is needed in making these comparisons! By all means cross-check your data against any other relevant observations - that's part of the fascination of running a weather station - but don't jump to conclusions too quickly.
One obvious problem is defining what the 'same locality' means in practice. We're all aware that weather conditions can change over relatively short distances, especially in undulating and hilly country, and some considerable caution is needed in choosing a valid reference site. Even in relatively flat country this can be a problem under certain weather conditions, for example when dense fog may linger over water courses while the land a few hundred yards away is enjoying bright sunshine.
And even if you're comfortable that your AWS and the reference site are comparable, unexpected differences in observed values between the two sites can still be misleading. You may be caused unnecessary concern about your AWS and waste time in fruitless investigation, if the limitations of such comparative exercises are not recognised. Above all, it is vital to compare like with like. Here are some examples of the three main considerations when comparing data from two sites:
- The values of some weather parameters can vary dramatically over relatively short distances. For example, the standard height at which to measure air temperature is 1250mm and, under certain weather conditions, there can be a difference of 5C or more between the temperature at this height and at ground level. And wind speed near to ground level is strongly influenced by every kind of physical obstruction, notably the lie of the land and nearby tall buildings and trees. For many observers there is therefore often no practical sense in which there is a definitive value of wind speed around a particular locality at a given time.
- For data comparisons to be valid, the weather sensors
must have very similar exposure and be carefully set up. For
example, rain gauges must be in a surprisingly large open
area if they are not to under record by virtue of being in
the rain shadow of a nearby building, wall, fence, tree etc.
Also, most automated rain gauges are now of the swinging
bucket type and it is essential that the gauge is mounted on
a shelf which is accurately in a horizontal plane for the
correct rainfall values to be measured.
It's evident that unless you are fortunate enough to be able to locate your AWS sensors in a large open area such as a farm or an airfield and at an appropriate height then you simply can't make definitive comparisons of certain weather parameters with other 'official' data which may have been measured in your locality. This is especially true for windspeed measurements - most amateur observers just don't have access to a 10m high tower set in open, unobstructed land and have to accept a compromised location. Instrumentation may also vary in how it defines and measures wind gusts. This doesn't mean that the wind speed data from such a location is without value, merely that its main usefulness is for relative comparisons at that specific base location.
- All AWS systems can measure weather parameters only to a limited accuracy and this really opens up the whole issue of the numerical validity of scientific data measurements and the consequent validity of comparisons. We can only touch on this issue here, but it's important to distinguish between accuracy, precision (reproducibility of a measure - values can be highly repeatable but inaccurate) and resolution (a display might show temperatures to 0.1°C resolution, but temperature measurements are unlikely to be repeatable to 0.1°C, let alone accurate to 0.1°C). As a generalisation, the data from a typical non-professional weather station will have an accuracy of ±5-10%, with temperature values accurate to ±1°C or a little better. Crucially, when comparing values from two stations you have to allow for error in the observations from both stations. For example, estimates of a particular value (eg rainfall on a given date) from two nearby stations, each specified as measuring to ±5% accuracy, must differ by 10% (to a first approximation - a rigorous statistical analysis is much more complex, but the sum of the two errors is a useful practical guide) before it becomes likely that the difference is genuine and therefore deserving of further investigation. Note also that error increases towards the threshold (the smallest value measurable by the AWS) of certain weather parameters, such as wind speed and rainfall. An automatic rain gauge will typically measure in steps of 0.2mm or 0.25mm, but, for all sorts of practical reasons, measurements amounting to 1-2 steps (ie up to 0.5mm) will be relatively inaccurate and up to 10 steps (2-3mm) may be needed before the specified accuracy is attained. So comparisons of rainfall can only usefully be made on days with significant (>2-3mm) rain.
The relative accuracy which might be expected for the main weather parameters when comparing observations between two nearby stations is shown qualitatively in the following table:
|Temperature||Reasonable||Exposure (eg height) must be comparable
Protection from direct sunlight essential
Due account of microclimate
|Humidity||Reasonable||Free air circulation around sensor essential
Humidity only measurable to limited accuracy
|Pressure||Good||Comparator site must be at similar altitude|
|Wind speed||Poor||Usually difficult for amateur sites to achieve sufficient exposure for valid comparison with 'official' values|
|Wind direction||Reasonable||Sensor must be relatively well exposed and clear of all obstructions which might deflect wind flow.|
|Rainfall||Reasonable, but beware localised showers||Sensor must be well exposed and accurately installed.|
Assessing accuracy and calibrating your AWS
As you can see, using data from other stations in your general locality to check the correct operation of your AWS can only provide a rough overall guide to its accuracy. So, if you do suspect that your instrumentation is reading incorrectly, and have already carefully checked the sensor siting and installation, how you do go about assessing its accuracy more specifically? The obvious answer is to use a trusted sensor/instrument placed at the same location as your main AWS sensor and to compare readings.Care needs to be taken over four points:
- Sensors from the two instruments must be placed at genuinely identical locations. For example, if the primary temperature sensor is protected against direct sunlight inside a Stephenson screen then so must the test sensor. Rain gauges must be positioned at the same height and with identical exposure. If the sensor locations are not genuinely identical, the whole comparative exercise is likely to be a waste of time.
- The principle that all measurements have finite accuracy still applies. Two sensors, each with a nominal accuracy of ±5% and identically located may differ by up to 10% and still each be within specification. Remember also that traditional manual instruments have intrinsic errors, which must be taken into account.
- The reference instrumentation must be of equivalent or superior accuracy to the AWS system. There is little point using, for example, a single reference thermometer accurate to only ±1°C to test an AWS temperature sensor with a specified accuracy of ±0.5°C, except as a check against gross error. (As ever, don't be fooled by the display resolution of electronic instruments. Many such thermometers are accurate to only ±1°C, but display temperatures to a resolution of 0.1°C. The decimal part of such displays has little real meaning). In some situations, it may be possible to use multiple reference instruments to increase the nominal accuracy. For example, using two reference thermometers of nominal ±1°C accuracy, but whose readings agree to perhaps 0.5°C, does provide a rather better indication of the true temperature, provided the two thermometers are not of the same make and type.
- In particular, reference instruments used to calibrate an AWS must be of known high accuracy. ('Calibrate' meaning here to calculate adjustment or scaling factors which can be applied to the AWS readings in order to improve on nominal accuracy. For example, a reference thermometer of known ±0.25°C could be used to calibrate an AWS with a specification of ±1°C.) But, in practice, unless you have professional connections or a deep pocket, it's often difficult to get access to instrumentation which is significantly more accurate than a good amateur AWS such as the Davis range. Even a simple mercury thermometer with a calibration certificate to confirm accuracy of ±0.25°C can cost £300.
The take-home message is perhaps to recognise and accept that even professional equipment is not capable of measuring perfectly accurate weather data. There is a certain level of error on all observations. So don't search for accuracy which is unrealistically or unachievably high - you will end up disappointed.
Exactly what degree of accuracy you do aim for will depend on your own interests and circumstances. There are also other considerations beyond the intrinsic accuracy of the instrumentation. For example, it might be argued that reporting temperature to better than 0.5°C accuracy and resolution starts to lose meaning, because the exact value measured will depend more on the precise environment around the probe, for instance was the grass on the ground beneath the probe long or short at the time of measurement? By all means check the accuracy and confirm that you are comfortably within the specified limits, but improving on the nominal accuracy may prove expensive and time-consuming.
Here are some comments on checking specific parameters:
Mercury thermometers designed to be accurate to ± <0.5°C, but not individually checked or calibrated, can be bought for around £50. Don't forget to place the bulb at exactly the same point in space as the AWS probe.
Humidity sensors in a typical AWS systems do not have high accuracy (eg ±5%) and usually don't read at all accurately at humidity levels >90% (in fact many displays are limited to a maximum reading of 90-95%). There is therefore only limited value in trying to check or calibrate this parameter unless you suspect it to be grossly out. If there is an airfield nearby, the temperature and dewpoint temperature values may be helpful in checking humidity readings. (see pressure below)
Typical instrument error ±7°, though any possible mounting error in the anemometer fixing need to be added to this.This parameter can obviously be checked visually, though some definitive means of establishing North accurately is essential.
Probably the most difficult parameter to check. Only the availability of a trusted anemometer which can be mounted exactly alongside the existing AWS anemometer is likely to be successful. Representative nominal specification is ±5%.
This is a parameter which (corrected for any difference in altitude) doesn't vary significantly across the local area, especially under conditions of established high pressure. It's also relatively easy to find an accurate reference value. Many websites (such as XCWeather) show current pressure patterns across the country. Schools and laboratories often have a reasonably accurate barometer which can be used for reference, provided the AWS is checked promptly. Alternatively airfields provide this information if you can access it. With a suitable radio, instructions to landing aircraft will often be heard to pass the current local pressure (be careful to note the difference between 'QFE' (pressure at the airfield ground altitude) and 'QNH' (pressure corrected to sea level) though). When high pressure is really settled and stable, even TV and radio forecasts will mention the current pressure value.
This is readily checked by making comparisons with a manual rain gauge. Measurements should be made over several 'rain days', ie days when there is >2-3mm of rainfall. An inexpensive rain gauge from a garden centre should suffice for an approximate check, but to investigate discrepancies of <10-15%, an official BS rain gauge is recommended, costing from £50-150, depending on accuracy