Citizen Scientists Serve Vital Role in Gathering Water Information

Some water professionals, however, are wary of data collected by volunteers.

Gary Woodard, associate director of SAHRA (Sustainability of semi-Arid Hydrology and Riparian Areas), University of Arizona, contributed this Guest View.

For water resource managers seeking data, these are the best of times and the worst of times. Never before has so much data been available; never before has the need for more data been as acute.
This apparent contradiction stems from two factors. First, Arizona’s growing population and economic development are increasing municipal water demands. A resurgent copper industry, growing electricity demand, and a desire to preserve remaining surface water flows and their riparian ecosystems are creating new water demands as well.

These new and growing demands have triggered the need for more intensive water resources management. Researchers have responded with improved scientific understandings, better physical models that integrate relationships between atmosphere, land surface, surface water, and groundwater at the basin scale, and sophisticated decision support systems. But these new management tools are data hungry. And many key components of basin-scale water budgets remain largely unknown, such as aggregate precipitation, soil moisture, ET, and pumpage from the state’s 100,000 exempt domestic wells.
Simply put, improved water management requires more water data.

Second, global change researchers have declared the end of stationarity, the fundamental assumption underlying water resource planning. Climate change, and land cover changes triggered by development and climate change, will alter patterns of precipitation, runoff, surface flows, and recharge. Calculations based on historical data about where the flood plain is, and what constitutes an assured water supply, may be seriously in error. What is certain is that in a non-stationary world, more observations and long-term, continuous observations are critical.

Simply put, our historic data are, at best, less predictive of the future; at worst, they are dangerously misleading.

Increasingly, researchers conducting field-intensive studies with limited resources are recruiting citizen scientists to gather critical data at low cost. Networks of volunteers also can make qualitative observations of changing environmental conditions that require human observers, and they often become effective advocates for research.

The number and diversity of citizen science networks has increased sharply in recent years. For example, RainLog.org was created by the University of Arizona’s SAHRA Center and Cooperative Extension in 2005 to gather monsoon data in the Upper San Pedro for researchers developing a watershed model. Since then, it has expanded enormously in numbers and geographic coverage. Today, over 1,400 active RainLoggers across Arizona report precipitation from backyard gauges.

The uses of RainLog data have multiplied as well, and now include urban runoff researchers, watershed modelers, drought monitors, weather reporters, master watershed stewards and K-12 educators. In addition, over 2,400 homeowners subscribe to RainMapper.org, a service that provides neighborhood-specific precipitation estimates.

We designed RainLog, to sit at the intersection of three trends: growing numbers of potential volunteers, inexpensive instruments capable of gathering research-quality data, and increasingly ubiquitous high-speed internet access. Citizen scientists are recruited, educated as to the scientific issues being addressed, provided with basic information on how to make observations, and shown how to report those observations.
Successfully recruiting and retaining hundreds of volunteers requires a seamless system of gauges, web apps, databases and user support that is flexible, scalable, reliable, low-cost, and easy to use. Retention also requires that volunteers receive positive feedback, including communication with researchers, seeing their data entered into research databases and graphically visualized, and receiving regular updates on the progress of the research and its implications for society.

Despite the apparent success of RainLog and other citizen science networks, there is a fly in the ointment. Some water professionals are reluctant or unwilling to use data collected by volunteers. And there is no denying that data collected citizen scientists inherently have potential problems.
In any of citizen science network, the level of training received by volunteers and the type of instruments used to collect data can vary widely. Some experienced volunteers may need very little training to correctly collect and report data, while others may need more extensive, hands-on training. Our RainLoggers range from retired Weather Service staff and irrigation district employees to middle schools students and a troop of Brownies.

Volunteers often are allowed, even encouraged to select their own instrument based on personal preference and affordability. Some 57 percent of RainLoggers have large, wedge-shaped gauges, and 16 percent have tipping buckets, but a few use tuna fish cans or jelly jars. This heterogeneity in volunteers and instruments within a network may result in significant reporting biases that decrease confidence and utilization of the data for some purposes.

RainLog.org is addressing these concerns by systematically investigating five significant QA/QC issues that apply to many other volunteer data collection networks: 1) instrument siting may be sub-optimal; 2) gauges vary in accuracy and precision; 3) experience and skill of observers varies; 4) missing data are not randomly distributed; and 5) volunteer networks may exhibit clustering

While the research is ongoing, we have reached some preliminary conclusions. First, gathering metadata on volunteers and their instruments can help spot problems and allow for adjustments and corrections. For example, a tipping bucket gauge that tips every 1 mm (0.04?) will under-record rain by an average of 0.02? per event. Central Tucson averages 75 events per year, meaning that uncorrected data will under-report annual precipitation by about 1.5?, or 12 percent.

Second, if the network is dense and the time series is long enough, interpolating rainfall amounts for points within a volunteer network can be a highly useful way to identify suspect data. Large, persistent deviations between reported rainfall amounts and interpolated values using nearby gauges can be evidence of suboptimal instrument siting, inaccurate instruments, or data reporting errors. Often, patterns in the deviations will suggest the likely source of error.

Recent research by Garcia, Peters-Lidard & Goodrich (Spatial interpolation of precipitation in a dense gauge network for monsoon storm events in the southwestern U.S., 2008, Water Resources Research special issue, “50 Years of Walnut Gulch”) has shown that the standard one-over-distance-squared (1/d2) interpolation approach is inappropriate for convective storms such as monsoon events. Weighing by one-over-distance-cubed (1/d3) produces a better fit, but both approaches fail when the instrument network has clusters of gauges. Volunteer networks often exhibit such clustering due to volunteers recruiting their neighbors.

A multiquadric biharmonic (MQB) interpolation approach fits well and is impervious to clustering. Unfortunately, MQB imposes major computational requirements for large networks that vary over time. Volunteer networks vary constantly, and the Tucson basin alone has over 500 active volunteers. Thus, we are working to develop a modified MQB approach that uses only the nearest few dozen gauges to interpolate each point.

A final conclusion is that even official networks of expensive gauges are not immune to QA/QC issues. Some networks of rain gauges are largely co-located with stream gauges, putting them in topographical low spots. Others are located near trails in mountainous areas, resulting in most of them being on or near ridgelines. Gauges for flood warning purposes selected to work reliably in heavy downpours and strong wind gusts may under-report precipitation in moderate events. And while clusters of gauges are rarely seen in official networks, strings or filaments are, and these raise interpolation issues as well.

The bottom line is, we need all the data we can get, from official gauges, remote sensing, and networks of volunteer citizen scientists. The challenge is to identify potential sources of bias and error in each type of data, employ automated methods for data screening, and develop better approaches for incorporating different types of data into our models and forecasts. QA/QC methodologies are fairly well developed for data from official gauges and remote sensing instruments. Developing similar approaches for data from citizen scientists is doable and should be a priority

Water Center Home -- AWR Home -- Search