The Weather Service Radar-1988 Doppler (WSR-88D) network within the United States has recently been upgraded to include dual-polarization capability. Among the expectations that have resulted from the upgrade is the ability to discriminate between different precipitation types in winter precipitation events. To know how well any such algorithm performs and whether new algorithms are an improvement, observations of winter precipitation type are needed. Unfortunately, the automated observing systems cannot discriminate between some of the more important types. Thus, human observers are needed. Yet, to deploy dedicated human observers is impractical because the knowledge needed to identify the various precipitation types is common among the public. To most efficiently gather such observations would require the public to be engaged as citizen scientists using a very simple, convenient, nonintrusive method. To achieve this, a simple “app” called mobile Precipitation Identification Near the Ground (mPING) was developed to run on “smart” phones or, more generically, web-enabled devices with GPS location capabilities. Using mPING, anyone with a smartphone can pass observations to researchers at no additional cost to their phone service or to the research project. Deployed in mid-December 2012, mPING has proven to be not only very popular, but also capable of providing consistent, accurate observational data.
An app for smartphones allows citizen scientists to provide observations about winter precipitation type at the surface at least equivalent in quality to human-augmented Automated Surface Observing System (ASOS) observations.
WHY THIS IS SUCH A GREAT IDEA.
Late in 2011 a planned upgrade of the Weather Service Radar-1988 Doppler (WSR-88D) radar network began in earnest (www.roc.noaa.gov/WSR88D/PublicDocs/DualPol/DPstatus.pdf). This upgrade adds vertical polarization information to the existing horizontal polarization information (Ryzhkov et al. 2005a). The overarching focus for the dual-polarization upgrade is improvement in quantitative precipitation estimation (QPE) and, in this, there has been some success (Cocks et al. 2012; Berkowitz et al. 2013), including data quality improvement (Ryzhkov et al. 2005a), discrimination of the rain/snow line using the ρHV field (Ryzhkov and Zrnic 1998), hail detection (Ryzhkov et al. 2005a), and tornado detection via the debris signature (Ryzhkov et al. 2005b). But, even beyond these successes, dual-polarization radar offers far more capabilities, especially when merged with environmental data.
Perhaps chief among the added benefits of polarimetric radar is the ability to help discriminate between different precipitation species or types in winter weather. Precipitation type information is useful for various reasons. For example, forecasters need knowledge of winter precipitation type because it helps inform them whether or not the thermodynamic profiles are developing as expected. Winter weather precipitation type affects surface transportation support and road maintenance since precipitation type affects decisions about whether to treat roads and, if treatment is needed, what process to use. Aviation ground deicing operations are heavily affected by precipitation type, but certain types of precipitation (e.g., ice pellets) also indicate freezing rain aloft and thus flight conditions that should be avoided. Electric utility infrastructure suffers during freezing precipitation events, which means that knowledge of where freezing precipitation is occurring helps utilities plan how best to maintain the power grid.
Within the suite of dual-polarization algorithms fielded with the upgraded radars is the hydrometeor classification algorithm (HCA; Park et al. 2009), which is used mostly for QPE enhancement (Giangrande and Ryzhkov 2008; Berkowitz et al. 2013). Because the HCA was developed with warm season convection in mind and because it assumes (among other things) a monotonic temperature profile with height and is fundamentally intended to provide hydrometeor type within the radar pulse volume (not at the surface), its performance is compromised in winter weather, especially in the presence of warm, elevated layers (Elmore 2011). Any failure in the monotonicity assumption for temperature with height is a significant issue with HCA in winter weather because the HCA depends upon the existence of only one freezing level through which precipitation falls. Within winter weather, this assumption is often invalid.
The inadequacy of the current HCA when misapplied to diagnose winter surface precipitation type has been noted by operational meteorologists within the NWS and the broadcast media, with the strong desire for improved surface HCA output expressed by both groups. To address the specific need for surface hydrometeor type information in winter weather, the winter surface hydrometeor classification algorithm (WSHCA) is being developed (Schuur et al. 2012). To both develop and also validate such algorithms and other dual-polarization algorithms, described in Ryzhkov et al. (2013) and Lakshmanan et al. (2014), high-quality surface observations of precipitation type are needed. The current automated observing systems do not provide information about some types, such as ice pellets. Yet, these types have important operational ramifications. Thus, a better source of precipitation type data is needed.
Observing precipitation requires no advanced education in meteorology and the general public can distinguish between rain and snow; different forms of frozen precipitation (e.g., snow versus ice pellets); and, within limits (discussed below), the difference between nonfreezing and freezing precipitation. Because such knowledge is common, it seems only natural to use it. The new generation of web-enabled portable devices (“smart” devices) offers an ideal platform for laypeople located almost anywhere to contribute their knowledge toward improving dual-polarization algorithms. To help laypeople identify different precipitation types, the mobile Precipitation Identification Near the Ground project (mPING) maintains a web page with descriptions of the various precipitation types (www.nssl.noaa.gov/projects/ping/types.php). Precipitation types are also internally documented within the app itself.
To employ these devices requires an application, or “app,” that reports back only the required data. Meteorological citizen scientist projects are not new: the Community Collaborative Rain, Hail and Snow (CoCoRaHS) was introduced in 1995 (Cifelli et al. 2005). Other examples exist outside of meteorolog— for example, Project Budburst (http://budburst.org/). However, mPING is unusual, if not unique, in that participants are intentionally kept anonymous and so need not register and, in fact, cannot register because there is no registration process.
Among the requirements are that the observations must be compact—free-form comments and photographs of precipitation fail in this regard because of their sheer volume, but also because photographs, in particular, cause an enormous increase in required bandwidth. Another requirement is to use the device's intrinsic GPS location and time for tagging observations. Perhaps the final requirement is that the app should keep the reporter anonymous to ensure privacy.
The mobile Precipitation Identification Near the Ground project (now changed to meteorological Phenomena Identification Near the Ground in a recent upgrade to the app) originated in 2006 as a way to gather validation information to assess the performance of the HCA as a surface precipitation type classifier (Elmore 2011). In the project's initial form, observations were entered through a web page interface (Fig. 1). Observations were requested within a 150-km radius of the KOUN (Norman, Oklahoma) test bed radar because, at the time, it was the sole WSR-88D-based dual-pol prototype. Users provided their latitude and longitude, based on either their own knowledge or through any of a number of web-based geolocation services, the time of the observation, and, through radio buttons, the precipitation type. The resulting data were added to a large database system maintained at the National Severe Storms Laboratory (NSSL). While data collection through the web form continues, it has become clear that with the nationwide dual-pol upgrade to the WSR-88D, a more effective data gathering means is both needed and attainable.
This led to a program based on the Severe Hazards and Verification Experiment (SHAVE; Ortega et al. 2009) wherein students actively probe areas of winter weather via telephone calls, seeking observations of precipitation type. While the winter SHAVE was successful, it became clear that targeting areas of transitional precipitation types, such as mixes, freezing precipitation, and ice pellets, is not straightforward; standard surface observations are inadequate; radar clues are ambiguous; and such regions are relatively small and transient in nature.
One of us (Flamig) has substantial experience developing weather-based apps for iOS devices and offered to help develop one that would support widespread, easy submission of precipitation type observations. The iOS development of mPING and the Android version are functionally identical but follow different operating system guidelines and so look very different. So far, apps exist only for the iOS and Android platforms, as these make up about 80% of the devices currently in use. Versions for other platforms may be developed in the future.
Among the key features of mPING are immediate feedback to users that their submission has been accepted and the ability to display and even download all submissions using a web-based display (viewable from within the apps). Up to 24 h of reports from across the continental United States and for any day back to November 2006 can be displayed. While users remain anonymous, the report density and frequency is such that when the display is centered on the user's location and magnified (zoomed in), individual reports are easily seen when they appear. The display can be seen using a desktop browser at www.nssl.noaa.gov/projects/ping/display/ (Fig. 2). A simplified display (with zoom capability) is used for mobile devices (www.nssl.noaa.gov/projects/ping/display/phone.php).
We paid particular attention to simplicity. The user interface had to be very simple (Fig. 3), and data entry had to also be simple and intuitive, not because users lack sophistication, but because the app must remain unobstrusive. Users are extremely concerned about battery life, so the app has to be smart about the way it uses the GPS engine, which is a significant power drain. To both avoid confusion and to standardize the various types that can be reported, users choose from a limited number of precipitation types with a pull-down menu (Fig. 4). These types are test, none, hail, rain, drizzle, freezing rain, freezing drizzle, snow, wet snow, mixed rain and snow, mixed rain and ice pellets, mixed ice pellets and snow, ice pellets/sleet, and graupel/snow grains. Descriptions of the various precipitation types are internally documented within the app itself and also described on the mPING website at www.nssl.noaa.gov/projects/ping/types.php. For hail only, an additional parameter (size to the nearest 0.6 cm or 0.25 in.) is also required. Location and observation time (in UTC) are gathered from the device's internal GPS engine. Thus, only the precipitation type is provided by the user; all else is automatic. The WSHCA research at NSSL is focused exclusively on precipitation type so no intensity estimates are requested.
To avoid rapid, inadvertent data submission while the device is being carried in a pocket or purse, a 5-min lockout timer is enforced so that observations can be entered at no higher frequency. The 5-min lockout timer also suppresses malicious attempts to rapidly enter misleading data. The most recent release of the app has relaxed the lockout timer to 30 s so that rapidly changing convective phenomena can be better captured.
Both the mobile apps and the web page submit information via HTTP to a common database that validates user input (to prevent malicious attacks, but not to quality control the observations) and provides persistent storage of the public reports. All quality control is done in postprocessing. We have so far found that these crowd-sourced data are very high quality when measured by internal temporal and spatial consistency. It is clear to us that the vast majority of entries are made with the best intentions. Even so, mistakes occur and the occasional misleading report appears. Fortunately, misleading reports in particular are very obvious (e.g., 20-cm hail reports in the absence of convection, rain in midst of large-scale snow, reports of precipitation in areas known to be clear, etc.) and are easy to remove by hand through simple inspection.
WHAT WE HAVE LEARNED AND WHERE WE MAY GO NEXT.
We have become convinced that immediate feedback to the user is very important and figures largely in the success of the mPING app. Not only are users rewarded by seeing that their data are actually being ingested, but they report an overall increased interest in weather and the project by simply watching the reports as they come in and change with time. In addition, the data are open and publicly available for text download in 24-h increments via the main display web page.
When these apps were initially released, the announcement was limited to only social media (i.e., Facebook, Twitter, etc.). The formal press release occurred much later, on 6 February 2013. Yet, we found that once mention was made within social networks, word spread rapidly about both the apps and the mPING project among those who are interested in weather but are not necessarily professional meteorologists; evidence is apparent in the download history of both the iOS and Android versions of the app (Fig. 5) and in the ~209,000 reports received between 19 December 2012 and 23 April 2013 (Fig. 6; Table 1). During this time, we have occasionally seen areas around cities become very active within the span of about an hour following mention of the app and project, often in cooperation with a local National Weather Service Forecast Office. Even before the formal media announcement, several media articles were published about mPING as well as at least one favorable editorial (e.g., http://idealab.talkingpointsmemo.com/2013/02/mping-noaa-storm-app.php, www.npr.org/blogs/alltechconsidered/2013/02/25/171715999/this-app-uses-the-power-of-you-to-report-the-weather, and www.bostonglobe.com/editorials/2013/02/08/the-folks-behind-national-weather-service-are-now-crowdsourcing-nemo/5MD65k88EfDUA30iV8DY3K/story.html).
Among the informal comments made on various social networks and in e-mails to the authors, users find two favorable characteristics that stand out. In no particular order, the first is the simplicity of the interface. Users appreciate how easy mPING is to use and how quickly they can enter observations and then be about their business. The second is immediate, uncluttered feedback, which both satisfies users' basic curiosity and helps retain their interest, even when winter weather is not occurring in their immediate vicinity. Both of these characteristics, taken together, may constitute a fundamental dual requirement for future efforts like mPING. The simple observation entry interface avoids tedium and immediate feedback keeps users' interest.
We suspect that allowing users to submit a “test” report and then see the report appear on the real-time display satisfies a reasonable desire to use and test the app immediately upon installation. Test report submission also strengthens users' confidence that the app does what is claimed. While we have no proof, we also suspect that the ability to submit test reports helps users resist the temptation to falsely report precipitation to test the app and see a report when no precipitation is occurring.
Even though the vast majority of observers are not trained in meteorological observations, we find that the observations appear to be of remarkably consistent and of high quality. In several instances, one of us (Reeves) polled professors of meteorology in regions experiencing complex winter precipitation, such as ice pellets, freezing precipitation, or mixed precipitation. In every case, these trained meteorologists validate the reports that are nearest to them in both time and space.
Transitional precipitation types that can be reported by Automated Surface Observing System (ASOS) stations are freezing rain and, when augmented by a human observer, ice pellets. These are among the most variable winter precipitation types in both space and time owing to the complex thermodynamic profiles required to generate them (Baldwin and Contorno 1993; Bourgouin 2000; Czys et al. 1996; Ramer 1993). To help quantitatively assess the reliability of mPING observations within these transitional precipitation types, observations of ice pellets and freezing rain are compared to manually augmented surface observations made by trained observers at the sites shown in Fig. 7. Only explicit mPING observations of ice pellets and freezing rain between 1 December 2012 and 31 March 2013 are used in this comparison; all other categories, including graupel and freezing drizzle, are excluded. This yields a total of 2382 observations by trained observers that can be matched to mPING reports. A comparison of trained observations to mPING observations is provided in Fig. 8. In this figure, both the length of time and the distance between the trained observation and surrounding mPING observations are varied. For freezing rain, as the amount of time between the mPING observation and the trained observation is decreased, the percent of mPING observations that agree with a given trained observation increases (Fig. 8a). Those mPING observations that are within about 5 km and 12 min of a trained observation have rates of agreement in excess of 80%. The picture is somewhat different for ice pellet (PL) observations (Fig. 8b). Here, the rate of agreement is maximized (at 70%) for distances between 15 and 30 km. For this precipitation type, the sampling of PL observations is comparatively limited (761) and there are relatively few mPING observations that are within 15 km of trained observations. Nevertheless, these rates of agreement are rather good, given that these forms of precipitation often occur in narrow zones or in mixes with other forms (Crawford and Stewart 1995; Robbins and Cortinas 2002; Cortinas et al. 2004) and suggest that most of the time the untrained observers participating in the mPING program are providing high-quality observations.
Even in the face of this evidence, mPING observation quality has limits. Based on real-time intercomparisons between mPING observations and comparisons to human-augmented ASOS observations, we have limited confidence that most people can distinguish between ice pellets and graupel, or that all observers use the same definition for “wet snow”; thus, the latest versions of the app no longer support these precipitation types. However, we have relatively high confidence that people properly identify mixes. While the version of the app used in this work contains 13 different precipitation types and all of the categories are always retained, for purposes of developing the WSHCA classifier and for comparing forecast precipitation type to mPING observations (Baldwin and Contorno 1993; Bourgouin 2000; Czys et al. 1996; Ramer 1993), these 13 categories are collapsed to only four: snow, rain, freezing rain, and ice pellets full in the understanding that these four “collapsed” types do not imply homogeneous precipitation type, but rather are in the spirit of major components. These four types are also the primary precipitation types developed by the various precipitation type algorithms used in numerical model post processing. Early work intercomparing mPING report self-consistency (not shown) indicates that the most consistent results are created when mixes with a rain component are collapsed to the nonrain component; that is, rain/snow mixes are collapsed to snow. Similarly, snow/ice pellet mixes appear most consistent when collapsed to ice pellets. These methodologies remain a topic of continuing research.
These data are potentially invaluable for the development of precipitation type algorithms that work with the upgraded dual-polarization WSR-88D radars and also for hail-size algorithms planned for the WSR-88D dual-pol radars. These data may also prove useful for additional studies and works, including (but not limited to) precipitation type algorithms for numerical models, ground icing for road maintenance and aviation operations, and even aviation in-flight icing work.
The app itself is not static: enhancements have already been made and additional mobile platforms may be considered in the future. New categories will be added, some will be dropped, and some categories that describe meteorological affects (such as flooding) and nonprecipitating weather (such as storm damage and obstructions to visibility) will be added. Plans are in motion to add resulting data stream as an Advanced Weather Interactive Processing System (AWIPS) data feed.
This work was supported by the NEXRAD Product Improvement Program by NOAA/Office of Oceanic and Atmospheric Research. Funding was provided by NOAA/Office of Oceanic and Atmospheric Research under NOAA–University of Oklahoma Cooperative Agreement NA11OAR4320072, U.S. Department of Commerce. The statements, findings, conclusions, and recommendations are those of the authors and do not necessarily reflect the views of NOAA, the U.S. DOC, or the University of Oklahoma.