Amateur Weather Observations Opportunities and Challenges Simon Bell1, Dan Cornford1, Lucy Bastin1, Mike Molyneux2 1Aston University, Birmingham Office, Exeter 2Met Contents 1. The current state of amateur observations 2. Possible applications 3. Challenges facing amateur observers 4. Intercomparison field study 5. Estimating observation bias and uncertainty The current state of amateur observations State of Amateur Observations What is an Amateur Observer? By ‘Amateur’ I do not mean Amateurish!! Rather they regularly collect weather observations due to a passion for meteorology rather than because it is their job. Who you calling ‘Amateur’!?! Automatic and/or Manual Observations State of Amateur Observations Automatic Weather Station Setup Step 1. Buy your weather station. Step 2. Install your weather station Step 3. Dedicate a computer to your weather station Step 4. Connect the display console (datalogger) to the computer. Step 5. Install the appropriate software. Notable software packages include: • Weather Display • Cumulus (Windows only) • Weathersnoop (Mac only) Step 6. Decide where to send your data… State of Amateur Observations Where does the amateur data end up? Met Office Weather Observations Website (WOW) - Launched early 2011. - Near real-time observations. Weather Underground - Longer established than WOW. - Not weather ‘under’ the ground. Climatological Observers Link - Greater focus on daily measurements. - Monthly Bulletins. - Over 40 years old! State of Amateur Observations Spatial Resolution Met Office MMS • ~250 stations • Professional network Met Office WOW • 592 stations (as of last week) Weather Underground • 1554 stations (as of last week) • • • • Amateur network ~60% also upload to WU, Amateur Network ~20% also upload to WOW (based on 100m proximity) • • Climatological Observers Link, COL, ~350 active members Citizen Weather Observing Program, CWOP, 491 ‘Active Members’ State of Amateur Observations Temporal Resolution • Majority are sub-hourly • Most commonly at 5 minute intervals • Near real-time observations State of Amateur Observations Weather Stations models Weather stations used to automatically upload data to Weather Underground in February 2012. A total of 1353 stations were investigated, of which 16.6% were of unknown type and have been excluded from the diagram. Possible Applications Possible Applications Assimilation into NWPs or Post-Processing Could resolve more grid cells within high resolution NWP model. Could help correct site-specific forecasts, by learning from a sequence of past observations. Station locations (various networks) for Reading area. Grid = 1.5 by 1.5km Met Office Android App Possible Applications Most useful variables Useful Not so useful Temperature – Urban Climate studies Pressure Humidity – Deep Convection Wind speed & direction Precipitation – Flood forecasting Snow depth/cover Boscastle, Cornwall. 16th Aug 2004. Birmingham heat island derived from MODIS satellite images. Tomlinson et al, 2012. Int. J. Clim. uksnowmap.com, snow as recorded by Twitter users on 28th Jan 2013 Challenges facing Amateur Observers Challenges facing Amateur Observers Exposure and Siting Fair Isle MMS station (Dave Wheeler) Well exposed rural professional Met Office site Sheltered urban Amateur site The urban nature of many amateur stations presents the following issues: • Nearby trees and buildings can alter the catch of the rain gauges. • Warm or reflective surfaces nearby can bias temperature readings. • Making representative wind measurements within the turbulent urban boundary layer is very difficult. Challenges facing Amateur Observers Representativity Scales Scales Representative Microscale : < hundreds of metres Yes Local scale: one to several kilometres Maybe Mesoscale : tens of kilometres No Image via NASA Challenges facing Amateur Observers Ventilation and Radiation Shielding Affects temperature and humidity readings. Consequence of: 1. Sheltered sites 2. Poor radiation shield design. Challenges facing Amateur Observers Calibration • Calibration may be expensive, e.g. calibrating humidity sensors. • Or time consuming, e.g. rainfall. • Thermistor may not be waterproof, ice/water bath impossible. • COL do have a calibration initiative with a tinytag sensor. Can collocate with amateur kit. Challenges facing Amateur Observers Other design flaws Tiny buckets will be sensitive to accumulating debris. Snow easily accumulates on them, wind vanes can freeze. However could be spotted earlier at an amateur site. Challenges facing Amateur Observers Other design flaws Some components are painted black, therefore prone to overheating. Wireless transmitters and batteries can often fail leading to missing data. Intercomparison field study Intercomparison field study The field site Winterbourne No.2 MMS site, Birmingham Data from August 2012 - Present Intercomparison field study The Stations Davis Vantage Pro 2 With wireless transmission and FARS. (x2) Name Davis Vantage Vue (x2) Price (approx.) Oregon Scientific WMR200 (x2-ish) Accompanying Software Software used for data retrieval La Crosse WS2350 Maplin USB Wireless Weather Forecaster (N96FY). (Fine Offset WH1080, Ambient Weather, Watson.) Observation Frequency (minutes) Memory at this obs. freq. Davis VP2 £1000 Weatherlink Weatherlink 10 ~ 2.5 weeks Davis Vue £450 Weatherlink Weatherlink 10 ~ 2.5 weeks Oregon Scientific £350 Weather OS Virtual Weather Station 10 291 days La Crosse £100 HeavyWeather HeavyWeather 60 7.29 days EasyWeather EasyWeather 10 ~ 30 days Maplin £60-125 Intercomparison field study Temperature Bias Average Bias : +0.19 Residual Variance: 0.04 Average Bias : +0.18 Residual Variance: 0.06 Davis Vantage Pro 2 • Slight warm bias at night, fairly neutral during the day. • Hints that the aspirating fan should really be on continuously. Intercomparison field study Temperature Bias Average Bias : +0.15 Residual Variance: 0.07 Davis Vantage Vue • Potential longwave radiation bias Average Bias : -0.11 Residual Variance: 0.08 Intercomparison field study Temperature Bias Average Bias: +0.44 Residual Variance: 1.27 Oregon Scientific WMR200 • Shortwave radiation bias • Poorly ventilated shielding Average Bias: +0.75 Residual Variance: 1.33 NB: change of colour scale! Intercomparison field study Temperature Bias Average Bias: +0.75 Residual Variance: 4.34 La Crosse WS2350 • Shortwave radiation bias Average Bias: +0.45 Residual Variance: 0.66 Maplin Weather Station • Shortwave radiation bias Intercomparison field study Temperature Bias as a function of Radiation NB: Be aware of small sample sizes. Images taken by a Flir i5 thermal imaging camera, the colour-scale is consistent. Intercomparison field study Humidity A typical time series plot In wet situations: Amateur stations tend to read drier than Met Office sensors. In dry situations: VP2’s read close to or wetter than mms. Vue’s and Maplin normally wetter. OS drier, often dramatically so! La Crosse often drier, but has a mind of it’s own. Intercomparison field study Humidity Intercomparison field study Precipitation Intercomparison field study Precipitation Corrected Intercomparison field study Precipitation • Measuring long term totals = poor • Picking up time of rainfall events and indicating intensity = good Estimating observation bias and uncertainty Estimating observation bias and uncertainty What do we need to find out? So we have an observation from an amateur weather station, for which we need to know the following: • Is the data biased? If so we need to correct it. • How certain are we the observation is correct? - Need an estimate of the residual variance. In order to provide a value for the bias and residual variance we need an initial estimate of what the weather is at the amateur location. Estimating observation bias and uncertainty First guessing the weather at amateur location Two options 1. Interpolate observations from nearby professional sites. 2. Use a short range forecast from a high resolution NWP. Estimating observation bias and uncertainty Interpolation Model – the predictors Elevation Coastality Land Cover (urban?) Roughness Temperature Foehn Wind Exposure/Shelter Aspect Soil Composition Data Acknowledgements: UGS GMTED, CEH LCM2007, CEH Soil Parent Map. Estimating observation bias and uncertainty Interpolation Model – the predictors Elevation Coastality Easting Land Cover (urban?) Roughness Northing Temperature Foehn Wind Exposure/Shelter Aspect Soil Composition Data Acknowledgements: UGS GMTED, CEH LCM2007, CEH Soil Parent Map. Estimating observation bias and uncertainty UKV Model vs Interpolation Model For Air Temperature 12th – 25th July 2012 Verification with MMS stations only Estimating observation bias and uncertainty For Dew Point 12th – 25th July 2012 Verification with MMS stations only Estimating observation bias and uncertainty Interpolation Model – planned update • Regression Model • Primarily with linear relationships, but could handle quadratic, etc… • Will smooth in time, not just space, should hopefully improve our prediction power significantly. • Dynamic clustering, the tricky bit! Once you have an estimate of the truth at an amateur locations then what…. Estimating observation bias and uncertainty Distinguishing natural differences from artificial errors Differences due to geospatial changes should already be captured Estimating observation bias and uncertainty Distinguishing natural differences from artificial errors Consistent bias • E.g. ‘Out of the box’ sensor bias. • Can try to learn this during periods when synoptic dependent biases are negligible. Estimating observation bias and uncertainty Distinguishing natural differences from artificial errors Bias dependent on synoptic situation • Siting issue (e.g. near wall or road) • Design flaw (ventilation, radiation shielding) Crucial to learn these dependencies (e.g. radiation bias) so we can parameterise the bias in the model. Estimating observation bias and uncertainty Distinguishing natural differences from artificial errors Natural differences (we want to capture!!) • Station is in frost hollow? But, only want to capture this natural variation if our weather model is competent to resolve it! i.e. is it representative? Estimating observation bias and uncertainty How metadata can help Good metadata would give us some prior information regarding the artificial bias we’d expect to see. WOW star rating helps indicate how well exposed a site is, the quality and siting of the temperature, rainfall and wind devices. At present the only quality control on WOW comes from the amateurs themselves. However, metadata is often missing or unstructured, will have to rely heavily on the data itself to inform us about the characteristics of a site. Estimating observation bias and uncertainty How does it all fit together? Synoptic Situation Geospatial information (e.g. elevation) Met Office Observations (e.g. radiation levels) Interpolation Model Short-range model forecast First guess of weather at amateur location Clusters Estimating observation bias and uncertainty How does it all fit together? Knowledge from intercomparison field study Metadata (e.g. station type) Synoptic Situation Geospatial information (e.g. elevation) Met Office Observations (e.g. radiation levels) Prior information about the synoptic and instrument dependent bias Interpolation Model Short-range model forecast First guess of weather at amateur location Clusters Estimating observation bias and uncertainty How does it all fit together? Knowledge from intercomparison field study Metadata Synoptic Situation (e.g. station type) (e.g. elevation) Met Office Observations (e.g. radiation levels) Prior information about the synoptic and instrument dependent bias Amateur Weather Observation Geospatial information Interpolation Model Short-range model forecast First guess of weather at amateur location Clusters Model Bias Correction Amateur weather observation (corrected for artificial bias) Residual Variance Estimate of uncertainty Estimating observation bias and uncertainty How does it all fit together? Knowledge from intercomparison field study Metadata Synoptic Situation (e.g. station type) (e.g. elevation) Met Office Observations (e.g. radiation levels) Prior information about the synoptic and instrument dependent bias Amateur Weather Observation Geospatial information Interpolation Model Short-range model forecast First guess of weather at amateur location Clusters Model Bias Correction Amateur weather observation (corrected for artificial bias) Residual Variance Estimate of uncertainty Data Assimilation Scheme Initial conditions in NWP model In Summary: Opportunities • Freely available, near-real time weather observations available from >1700 amateur station locations in GB alone. • Sparser professional networks (+high-res forecast models) can help quantify the bias and uncertainty associated with these amateur observations. • Amateur observations, once bias corrected and given a reliable uncertainty estimate, could have a significant impact on data assimilation schemes, post-processing, flood forecasting, urban climate studies, etc… Challenges • Amateur equipment prone to significant biases due to poor siting, calibration and shielding design, and/or because poor exposure --> unrepresentative. • Identifying whether an amateur site is well exposed, calibrated and maintained can be difficult to ascertain from just the sporadic metadata. • Disentangling natural spatial variations in weather from the artificial biases present within amateur observations could prove very tricky. • Working out over what scale an amateur, or even a professional, weather station is representative is a difficult undertaking. • Any Questions?