Smart Irrigation: A Supercomputer Waters the Lawn
Satellite control of company and municipal irrigation systems, based on custom weather data, saves billions of gallons of water nationwide
|
Image: Flickr / KumaravelIn Brief
Businesses and municipalities waste billions of gallons of water each year in poorly timed lawn irrigation.
Smart controllers on water lines, instructed by satellite, can deliver just the right amount of water at the most effective time.
HydroPoint's custom-programmed supercomputer models the daily water needs of every square kilometer across the U.S. and Canada, remotely driving clients' irrigation systems.
In Silicon Valley the Campbell Union School District's sprinklers used to dutifully water the soccer fields and gardens at 12 campuses even during spring showers. Temporarily shutting off each of the 45 irrigation control boxes, by hand, wasn't worth the custodians' time. But in 2009 the district installed new "smart" controllers that automatically adjust daily watering to the weather. Each box, fitted with a microprocessor and antenna, receives local real-time weather information by satellite from the WeatherTRAK climate center supercomputer run by Petaluma, Calif.–based HydroPoint Data Systems, Inc.
On one April morning after a three-day rainstorm, Campbell Union's facilities supervisor, David Radke, checked the status of the controllers by logging in toWeatherTRAK.net The service had put the district’s irrigation network on pause when the storm began, and according to WeatherTRAK's math, the network would not need to resume watering for eight to 11 days.
If needed, Radke can reprogram the controllers from his Web account, but because his crew has carefully customized each box's settings, he finds that "there's really not a whole lot to do. That's the beauty of it." The soccer fields look better than ever, yet in 2009 WeatherTRAK slashed Campbell Union's water use by 39 percent and cut its utility bills by $108,000, recouping the installation expenses nearly twice over.
With most sprinkler systems, property owners set the traditional controller—basically a timer—to irrigate at specific intervals. Often, too much water is lost to evaporation during hot weather or to runoff during cool weather, which can also carry chemicals into the local watershed or ocean. Because outdoor irrigation can suck up 50 percent or more of urban water consumption, smart irrigation services have caught on in drought-prone western states like California, where water prices are relentlessly rising. (Occasional big floods don't help the long-term problem.) HydroPoint now has more than 8,000 clients using 24,000 of its smart controllers, including Walmart, Coca-Cola, Hilton, Jack in the Box and the University of Arizona as well as the cities of Charleston, S.C., Houston and Santa Barbara. In 2011 customers are projected to save 64.4 billion liters of water and over $111 million in water expenses, as well as 68 million kilowatt-hours of water-pumping electricity.
The technology is still going through growing pains, but HydroPoint has solved several daunting weather-forecast and modeling challenges that help it deliver just the right amount of water at the most effective time.
Water by satellite
Founded in 2002 in Petaluma 65 kilometers north of San Francisco, HydroPoint has engineered what appears to be the most sophisticated weather-based irrigation system among a growing list of competitors. The company built its primary climate-modeling center outside of Salt Lake City, programming a supercomputer to simulate local weather for every square kilometer across North America—all just to water the grass. The center communicates via a two-way satellite link with control boxes that operate distinct zones of a client's irrigation system. Wires running underground from the outdoor boxes open and close valves in the water lines. Every night the climate center broadcasts local weather-related data to a microprocessor inside each controller, which runs software that uses the information to compute precisely how much and when to water its zones, customized to one of 18 plant types as well as other factors like soil type and ground slope.
The boxes can talk back, too. For example, Campbell Union installed additional wireless rain sensors (fixed to an exterior building wall) that, on detecting moisture, signals controllers, which put irrigation on hold and notify the climate center.
A WeatherTRAK prototype went live in 2003, starting with one-way satellite service in California. Two-way Internet-based communications became available in 2006, allowing a client to adjust thousands of controllers across multiple properties, all from one Web account. Monthly subscription fees today range from $4 to $18.75. The service is a prime example of the nascent field of water information systems that analyze real-time data to tackle water management issues—data that until recently has been sorely lacking in the industry. "You can't manage what you don't see," says HydroPoint co-founder Chris Spain.
Really local forecasts, in real time
Spain and his colleagues jumped into the business by buying a local irrigation-controller company and bringing on board its owner, Mike Marian, a self-taught engineer. Marian (who subsequently left HydroPoint in 2006) had patented the idea of using paging signals to transmit information to controllers about evapotranspiration(ET)—the amount of water soil loses to evaporation as well as transpiration by plantsgrowing there. Marian's system used free ET data from the nearest state weather stations, maintained for farmers by the California Irrigation Management Information System.
The entire state, however, had only around 120 stations. Simply averaging ET values from the two nearest stations (as farmers often did) could be misleading, because microclimates even a few kilometers apart can produce substantially different values. And although property managers could buy their own mini weather stations, those were less reliable.
The fledgling HydroPoint team instead boldly explored a cutting-edge alternative: using measurements from public weather stations across the U.S. to model an accurate ET for any location, down to a square kilometer—without having an actual station or sensors on-site. And doing it every night.
Dan Dansereau, a computer scientist who worked on high-resolution weather modeling at Utah State University in Logan thought it could be done. But he says he told Marian and Spain "You're going to need a big-ass computer," before he joined them as HydroPoint's chief scientist. Dansereau recalls one prominent colleague warning, "It's never going to work."
Indeed, modeling proved a complicated task. The team wanted to use the most accurate method for estimating ET, the Penman–Monteith equation, which factors in temperature, wind, solar radiation and relative humidity. But the trick to deriving real-time ET everywhere would be creating those four input parameters for each of the 19.2 million one-square-kilometer grids across the continental U.S. and Canada.
Dansereau's plan was to model the parameters using the same highly sophisticated (and public) forecasting software that the U.S. Air Force relied on, called theMesoscale model, or MM5, created by Pennsylvania State University and the National Center for Atmospheric Research (NCAR) in Boulder, Colo. Running MM5 at a daily one-square-kilometer resolution nationwide would take such tremendous computing power that no one had attempted it. Every day the climate center had to download millions of real-time data points via satellite from the National Oceanic and Atmospheric Administration (NOAA) and other public sources. That included surface observations collected hourly from ground weather stations, ocean buoys and weather balloons around the world, global satellite-imaging data every 15 minutes, and hourly Doppler radar data for the U.S. In addition, the MM5 program required 256 categories of local ground information such as topography, soil, vegetation and land use.
To manage the data avalanche, the HydroPoint engineers wrote proprietary code to organize and shovel it into databases. But they were still "really stymied" because MM5 required a lot of manual data entry, recalls co-founder and vice president Peter Carlson, who along with Dansereau literally typed in parameters every day. To streamline the process, they devised automated programs so the model could run itself from a single script.
Making forecasts accurate
A second major obstacle was verifying accuracy. How would the engineers know if WeatherTRAK's ET calculations were right? "This whole idea of 'ground truth' is a pretty complicated one," Spain says.
First, MM5 itself "was susceptible to garbage in–garbage out," Spain notes, because ground weather readings can themselves be wrong. The answer was writing a heuristic algorithm that crawls 24/7 through the incoming data and weeds out anomalous readings. For example, if the program detects an impossibly big swing in temperature or humidity readings from one hour to the next from the same weather station, that data is disqualified. Dansereau and Carlson later programmed similar verification steps into a more advanced successor to MM5, the Weather Research and Forecasting model (developed by NCAR, NOAA and partners), which the NationalWeather Service relies on. The result of HydroPoint's version of that model, which it runs today, is a "virtual weather station" for each square kilometer in North America. The outputs for temperature, wind, solar radiation and relative humidity are then funneled into the calculations of ET values, which are transmitted to customers' controllers by 8 P.M. Pacific time.
The second part of the accuracy problem was validating whether WeatherTRAK's numbers were on target. The engineers, focusing initially on California, realized they could check their daily results against actual ET measurements reported by the state's 120 weather-station array, which were published the following morning. "We process our data and publish the answers," Dansereau explains. "Then we get to check tomorrow to see how accurate our answers were." If WeatherTRAK finds an error or gets better information after the fact, it re-runs the model for past days, and can send retroactive corrections to controllers to continually ensure accuracy.
A glass half full—or more
A third key hurdle for HydroPoint was converting ET values into an irrigation schedule that would keep grass and plants in optimal health. The company developed scheduling software for the smart controllers that factors in ET plus 19 ground parameters entered by the client: from plant and soil types to sprinkler-head model. If a lawn's maximum water supply were the equivalent of a full glass of water, the goal is to let the level dwindle to half-full before topping the glass off again, Carlson says.
The WeatherTRAK system has continued to evolve. One ongoing challenge was rain. The company has always advised installing a rain switch that turns off the controllers when precipitation is detected. Since 2007 WeatherTRAK has also analyzed Doppler radar data to approximate the amount of rain that has accumulated on each square kilometer. The climate center roughly predicts the moisture depletion rate by subtracting the estimated daily ET loss for each grid, thus determining how long the irrigation pause should last—information that is sent to the controllers.
Supercomputing the savings
HydroPoint today runs WeatherTRAK on a blade supercomputing cluster with two teraflops of number-crunching power. Jimy Dudhia, an atmospheric physicist at NCAR, who helped develop the Weather Research and Forecasting model, says it is surprising that HydroPoint can calculate real-time ET at such a fine-grain resolution across the U.S. The calculations require more computing power than the National Weather Service's 12-kilometer-resolution forecasts. "In terms of science, it's on the front end," Dudhia says.
Others also see ET modeling as the way to go. Managers of California Irrigation Management Information System, for example, have developed their own method for simulating ET values statewide. They post the free spatial data online, but it only covers California, and only at a two-square-kilometer resolution.
As with any weather modeling results, HydroPoint's won't be perfect. But does the computing footwork pay off? Michael Dukes, a University of Florida agricultural engineer who has studied various smart irrigation systems, says HydroPoint can be accurate under optimal conditions, but notes that other leading smart irrigation companies have also devised innovative solutions. For example, Rain Bird Corp.'s ET Manager controller—which, like HydroPoint, earned top scores on a standardized irrigation industry test—estimates ET from hourly data received wirelessly from private local weather stations. Dukes believes ET Manager might do a better job than WeatherTRAK of factoring in rainfall, using a tipping-bucket gauge on a client's site that quantifies precipitation more reliably than most rain sensors.
Still, HydroPoint can quote two dozen independent studies showing that WeatherTRAK curtailed water use by anywhere from 14 to 82 percent. But how much you save depends on how much you were wasting. A major study that tracked 2,294 California sites using 14 different brands of smart controllers found only a disappointing 6 percent average decline in water consumption in the first year. Nearly 57 percent of the sites saved water—but another 42 percent actually used more.
One explanation is that some landscapes were previously under-irrigated, so their usage rose, says civil engineer Peter Mayer of Aquacraft, a Boulder-based firm that conducted the California study. And if an irrigation system has leaky valves, high-end controllers may not help much. The study noted that HydroPoint's controllers, when averaged, reaped no savings the first year. Spain says that's because most of the devices were installed incorrectly. After Los Angeles's water agency fixed installation problems, the WeatherTRAK controllers reduced water use by roughly 16 percent.
The bottom line is that landscape contractors and property owners have to be educated about setting up and adjusting smart irrigation systems. Installation troubles are a real stumbling block, Mayer says. "We're kind of in the VCR-era of smart controllers, where nobody knows how to program it properly." Nevertheless, Mayer sees the evolving technology as a key conservation tool. The U.S. Environmental Protection Agency's Water Sense program plans to begin issuing labels for smart controller brands that pass efficiency certification by the end of this year.
From HydroPoint's perspective, the biggest barrier is the status quo of old habits that waste cheaply priced water. Now that rates are going up, attitudes are changing. The water crisis, Spain says, "is truly going to be the biggest issue of the century."
On one April morning after a three-day rainstorm, Campbell Union's facilities supervisor, David Radke, checked the status of the controllers by logging in toWeatherTRAK.net The service had put the district’s irrigation network on pause when the storm began, and according to WeatherTRAK's math, the network would not need to resume watering for eight to 11 days.
If needed, Radke can reprogram the controllers from his Web account, but because his crew has carefully customized each box's settings, he finds that "there's really not a whole lot to do. That's the beauty of it." The soccer fields look better than ever, yet in 2009 WeatherTRAK slashed Campbell Union's water use by 39 percent and cut its utility bills by $108,000, recouping the installation expenses nearly twice over.
With most sprinkler systems, property owners set the traditional controller—basically a timer—to irrigate at specific intervals. Often, too much water is lost to evaporation during hot weather or to runoff during cool weather, which can also carry chemicals into the local watershed or ocean. Because outdoor irrigation can suck up 50 percent or more of urban water consumption, smart irrigation services have caught on in drought-prone western states like California, where water prices are relentlessly rising. (Occasional big floods don't help the long-term problem.) HydroPoint now has more than 8,000 clients using 24,000 of its smart controllers, including Walmart, Coca-Cola, Hilton, Jack in the Box and the University of Arizona as well as the cities of Charleston, S.C., Houston and Santa Barbara. In 2011 customers are projected to save 64.4 billion liters of water and over $111 million in water expenses, as well as 68 million kilowatt-hours of water-pumping electricity.
The technology is still going through growing pains, but HydroPoint has solved several daunting weather-forecast and modeling challenges that help it deliver just the right amount of water at the most effective time.
Water by satellite
Founded in 2002 in Petaluma 65 kilometers north of San Francisco, HydroPoint has engineered what appears to be the most sophisticated weather-based irrigation system among a growing list of competitors. The company built its primary climate-modeling center outside of Salt Lake City, programming a supercomputer to simulate local weather for every square kilometer across North America—all just to water the grass. The center communicates via a two-way satellite link with control boxes that operate distinct zones of a client's irrigation system. Wires running underground from the outdoor boxes open and close valves in the water lines. Every night the climate center broadcasts local weather-related data to a microprocessor inside each controller, which runs software that uses the information to compute precisely how much and when to water its zones, customized to one of 18 plant types as well as other factors like soil type and ground slope.
The boxes can talk back, too. For example, Campbell Union installed additional wireless rain sensors (fixed to an exterior building wall) that, on detecting moisture, signals controllers, which put irrigation on hold and notify the climate center.
A WeatherTRAK prototype went live in 2003, starting with one-way satellite service in California. Two-way Internet-based communications became available in 2006, allowing a client to adjust thousands of controllers across multiple properties, all from one Web account. Monthly subscription fees today range from $4 to $18.75. The service is a prime example of the nascent field of water information systems that analyze real-time data to tackle water management issues—data that until recently has been sorely lacking in the industry. "You can't manage what you don't see," says HydroPoint co-founder Chris Spain.
Really local forecasts, in real time
Spain and his colleagues jumped into the business by buying a local irrigation-controller company and bringing on board its owner, Mike Marian, a self-taught engineer. Marian (who subsequently left HydroPoint in 2006) had patented the idea of using paging signals to transmit information to controllers about evapotranspiration(ET)—the amount of water soil loses to evaporation as well as transpiration by plantsgrowing there. Marian's system used free ET data from the nearest state weather stations, maintained for farmers by the California Irrigation Management Information System.
The entire state, however, had only around 120 stations. Simply averaging ET values from the two nearest stations (as farmers often did) could be misleading, because microclimates even a few kilometers apart can produce substantially different values. And although property managers could buy their own mini weather stations, those were less reliable.
The fledgling HydroPoint team instead boldly explored a cutting-edge alternative: using measurements from public weather stations across the U.S. to model an accurate ET for any location, down to a square kilometer—without having an actual station or sensors on-site. And doing it every night.
Dan Dansereau, a computer scientist who worked on high-resolution weather modeling at Utah State University in Logan thought it could be done. But he says he told Marian and Spain "You're going to need a big-ass computer," before he joined them as HydroPoint's chief scientist. Dansereau recalls one prominent colleague warning, "It's never going to work."
Indeed, modeling proved a complicated task. The team wanted to use the most accurate method for estimating ET, the Penman–Monteith equation, which factors in temperature, wind, solar radiation and relative humidity. But the trick to deriving real-time ET everywhere would be creating those four input parameters for each of the 19.2 million one-square-kilometer grids across the continental U.S. and Canada.
Dansereau's plan was to model the parameters using the same highly sophisticated (and public) forecasting software that the U.S. Air Force relied on, called theMesoscale model, or MM5, created by Pennsylvania State University and the National Center for Atmospheric Research (NCAR) in Boulder, Colo. Running MM5 at a daily one-square-kilometer resolution nationwide would take such tremendous computing power that no one had attempted it. Every day the climate center had to download millions of real-time data points via satellite from the National Oceanic and Atmospheric Administration (NOAA) and other public sources. That included surface observations collected hourly from ground weather stations, ocean buoys and weather balloons around the world, global satellite-imaging data every 15 minutes, and hourly Doppler radar data for the U.S. In addition, the MM5 program required 256 categories of local ground information such as topography, soil, vegetation and land use.
To manage the data avalanche, the HydroPoint engineers wrote proprietary code to organize and shovel it into databases. But they were still "really stymied" because MM5 required a lot of manual data entry, recalls co-founder and vice president Peter Carlson, who along with Dansereau literally typed in parameters every day. To streamline the process, they devised automated programs so the model could run itself from a single script.
Making forecasts accurate
A second major obstacle was verifying accuracy. How would the engineers know if WeatherTRAK's ET calculations were right? "This whole idea of 'ground truth' is a pretty complicated one," Spain says.
First, MM5 itself "was susceptible to garbage in–garbage out," Spain notes, because ground weather readings can themselves be wrong. The answer was writing a heuristic algorithm that crawls 24/7 through the incoming data and weeds out anomalous readings. For example, if the program detects an impossibly big swing in temperature or humidity readings from one hour to the next from the same weather station, that data is disqualified. Dansereau and Carlson later programmed similar verification steps into a more advanced successor to MM5, the Weather Research and Forecasting model (developed by NCAR, NOAA and partners), which the NationalWeather Service relies on. The result of HydroPoint's version of that model, which it runs today, is a "virtual weather station" for each square kilometer in North America. The outputs for temperature, wind, solar radiation and relative humidity are then funneled into the calculations of ET values, which are transmitted to customers' controllers by 8 P.M. Pacific time.
The second part of the accuracy problem was validating whether WeatherTRAK's numbers were on target. The engineers, focusing initially on California, realized they could check their daily results against actual ET measurements reported by the state's 120 weather-station array, which were published the following morning. "We process our data and publish the answers," Dansereau explains. "Then we get to check tomorrow to see how accurate our answers were." If WeatherTRAK finds an error or gets better information after the fact, it re-runs the model for past days, and can send retroactive corrections to controllers to continually ensure accuracy.
A glass half full—or more
A third key hurdle for HydroPoint was converting ET values into an irrigation schedule that would keep grass and plants in optimal health. The company developed scheduling software for the smart controllers that factors in ET plus 19 ground parameters entered by the client: from plant and soil types to sprinkler-head model. If a lawn's maximum water supply were the equivalent of a full glass of water, the goal is to let the level dwindle to half-full before topping the glass off again, Carlson says.
The WeatherTRAK system has continued to evolve. One ongoing challenge was rain. The company has always advised installing a rain switch that turns off the controllers when precipitation is detected. Since 2007 WeatherTRAK has also analyzed Doppler radar data to approximate the amount of rain that has accumulated on each square kilometer. The climate center roughly predicts the moisture depletion rate by subtracting the estimated daily ET loss for each grid, thus determining how long the irrigation pause should last—information that is sent to the controllers.
Supercomputing the savings
HydroPoint today runs WeatherTRAK on a blade supercomputing cluster with two teraflops of number-crunching power. Jimy Dudhia, an atmospheric physicist at NCAR, who helped develop the Weather Research and Forecasting model, says it is surprising that HydroPoint can calculate real-time ET at such a fine-grain resolution across the U.S. The calculations require more computing power than the National Weather Service's 12-kilometer-resolution forecasts. "In terms of science, it's on the front end," Dudhia says.
Others also see ET modeling as the way to go. Managers of California Irrigation Management Information System, for example, have developed their own method for simulating ET values statewide. They post the free spatial data online, but it only covers California, and only at a two-square-kilometer resolution.
As with any weather modeling results, HydroPoint's won't be perfect. But does the computing footwork pay off? Michael Dukes, a University of Florida agricultural engineer who has studied various smart irrigation systems, says HydroPoint can be accurate under optimal conditions, but notes that other leading smart irrigation companies have also devised innovative solutions. For example, Rain Bird Corp.'s ET Manager controller—which, like HydroPoint, earned top scores on a standardized irrigation industry test—estimates ET from hourly data received wirelessly from private local weather stations. Dukes believes ET Manager might do a better job than WeatherTRAK of factoring in rainfall, using a tipping-bucket gauge on a client's site that quantifies precipitation more reliably than most rain sensors.
Still, HydroPoint can quote two dozen independent studies showing that WeatherTRAK curtailed water use by anywhere from 14 to 82 percent. But how much you save depends on how much you were wasting. A major study that tracked 2,294 California sites using 14 different brands of smart controllers found only a disappointing 6 percent average decline in water consumption in the first year. Nearly 57 percent of the sites saved water—but another 42 percent actually used more.
One explanation is that some landscapes were previously under-irrigated, so their usage rose, says civil engineer Peter Mayer of Aquacraft, a Boulder-based firm that conducted the California study. And if an irrigation system has leaky valves, high-end controllers may not help much. The study noted that HydroPoint's controllers, when averaged, reaped no savings the first year. Spain says that's because most of the devices were installed incorrectly. After Los Angeles's water agency fixed installation problems, the WeatherTRAK controllers reduced water use by roughly 16 percent.
The bottom line is that landscape contractors and property owners have to be educated about setting up and adjusting smart irrigation systems. Installation troubles are a real stumbling block, Mayer says. "We're kind of in the VCR-era of smart controllers, where nobody knows how to program it properly." Nevertheless, Mayer sees the evolving technology as a key conservation tool. The U.S. Environmental Protection Agency's Water Sense program plans to begin issuing labels for smart controller brands that pass efficiency certification by the end of this year.
From HydroPoint's perspective, the biggest barrier is the status quo of old habits that waste cheaply priced water. Now that rates are going up, attitudes are changing. The water crisis, Spain says, "is truly going to be the biggest issue of the century."
ABOUT THE AUTHOR(S)
Ingfei Chen is a freelance writer in the San Francisco Bay area. Her articles on neuroscience, medicine and ecology have appeared in The New York Times, Smithsonian and Science.
No comments:
Post a Comment