Can “virtual” weather stations replace real weather stations?
Volume 10 Issue 3
Joe Zagrodnik (AgWeatherNet Postdoctoral Reseacher)
Dave Brown (AgWeatherNet Director)
Growers rely on weather data to manage their crops and make decisions about irrigation, pest and disease prevention, frost mitigation, labor allocation, and more. Traditionally, meteorological observations have been obtained through automated weather stations such as those maintained by AgWeatherNet, but it is cost prohibitive to place a full professional weather station in every field and orchard so growers must account for differences in weather conditions between their location and the nearest weather station.
One proposed solution that has recently received attention is the concept of “virtual” weather stations. A virtual weather station uses gridded weather data from commercial providers interpolated to any latitude and longitude. Virtual weather data is inexpensive and ubiquitous with many different commercial providers delivering weather data to online tools and mobile apps. Data providers start with a “grid” output from global weather models, then apply various proprietary statistical corrections and interpolations to deliver estimated weather data down to the nearest 3 miles or less.
AgWeatherNet meteorologists have completed an intensive analysis of virtual weather data from a leading commercial provider (DarkSky) and found the accuracy for air temperature to be inadequate for many agricultural decision support needs. For 8.5 years of AgWeatherNet station data with replicate air temperature sensors at 156 locations (445,102 daily comparisons):
• 24% of DarkSky daily high temperatures were off by more than 2°F;
• 21% of DarkSky daily low temperatures were off by more than 4°F, with a systematic 1°F warm bias;
33% of 966 different station-season combinations experienced a DarkSky estimated 375 growing degree day error of calendar 5 days or more (Jan. 1 start, 50°F base temperature);
Errors were higher in areas with complex topography; and
Site-specific DarkSky bias often changed from year to year, so growers attempting to correct for virtual station bias will often be chasing a moving target.
It would be great if inexpensive virtual weather data could replace weather stations. And virtual weather data might well be better than using weather station data from 20 miles away or at a very different elevation. But growers should be aware that virtual weather data is not nearly as accurate as a weather station. The mean error for AWN weather station air temperature is < 0.3 °F, and generally < 1 °F for quality commercial sensors from companies like Davis Instruments, METER or Pessl. Washington State topography, nocturnal inversions, and cold air flows combine to generate complex microclimates that make weather data interpolation challenging (particularly at night).
AgWeatherNet meteorologists are actively researching weather data interpolation methods. But our understanding and experience suggest that for any location in Washington where a virtual weather product is to be used, at least some ground truth data should be collected for site-specific calibration and validation.
For many locations, interpolation and virtual weather stations will not work, and site-specific instrumentation provides the only accurate weather data solution. With the support of the Whatcom and Skagit County Conservation Districts, AgWeatherNet has added additional stations for these two counties. And AgWeatherNet now offers a free “Tier 3” service, ingesting data from approved private all-in-one weather stations and affordable temperature/relative humidity sensors (https://weather.wsu.edu/?p=115550) and making data from these private installations available on AgWeatherNet and AWNfarm platforms.