Most Americans may not realize it, but we really don’t know with a lot of accuracy how much snow there is sitting in the mountains during winter. We also don’t always have a precise picture of where the snow level is when a storm moves in, or how much will run off when the snow melts.
One reason for this is that, in most areas, the weather sensor network in the mountains simply isn’t very dense. Gauges that measure rain and snow are often placed for convenient access. The highest elevations and forested areas often have no sensors, leaving huge data gaps in many watersheds.
This creates a host of problems, from estimating flood risk accurately to figuring out how much water is available for summer farm irrigation.
A team from the University of California and the United States Department of Agriculture (USDA) recently completed a project to fill the data gaps. With funding from the National Science Foundation, they developed a new wireless network consisting of 140 sensor pods installed across 830 square miles of the American River Basin. Previously, the area had only 27 sensors.
This critical California watershed drains through Sacramento, one of the most flood-prone metropolitan areas in the nation. And it now has the most data-rich precipitation monitoring network anywhere. This could vastly improve forecasting, especially as climate change disrupts previous expectations about storm behavior.
To learn more about the project, Water Deeply talked to Roger Bales, a distinguished professor of engineering at the University of California, Merced, director of the Sierra Nevada Research Institute, and project leader of the new sensor network.
Water Deeply: Why did you undertake this project? What was the need that drove it?
Roger Bales: We don’t do that good of a job of measuring the basic water balance in the mountains. Think of the water balance as being like this formula: precipitation = evapotranspiration + runoff. Water is coming in as precipitation, and it’s leaving either because the vegetation is putting it back into the atmosphere, or because it’s draining away as runoff. Some of that’s subsurface, but some of it comes out in runoff.
In the mountains, we don’t really measure precipitation very well. On any given day, in any given storm, you actually only know how much rainfall is occurring. There may be only a few rain gauges, and they are not at the highest elevations. You have snow pillows at the higher elevations, which are telemetered, and they give you the amount of snow falling in a few generally flat, open areas (think of a meadow). But that’s not representative of the landscape. Rain gauges also tend to be put in convenient locations to get to.
There are a lot of decisions that depend on knowing how much precipitation is occurring, and how much water is in storage. The two main storage reservoirs in the mountains, besides the dams at the base of the mountains, are the snowpack and the soil water storage. The seasonal snowpack changes every time there’s a storm or every time there’s snowmelt. It changes daily.
So the amount of precipitation and snow that falls is statistically related to runoff, but not in a mass balance sense. That is, you can’t take the sum of all the rain gauges or snow pillow data and say that’s how much rain fell or how much snow fell, because the rain gauges are too sparse and not representative. The operational network tends to be located at mid- and low-elevation clearings, not so much at the high elevations and not so much in forests.
We designed a network that is more representative of the landscape and has a denser set of sensors. We scaled this up from a small headwater catchment that we did in the Kings River Basin, going from about 400 acres to the whole snow-covered part of the American River Basin. We basically replicated what we did in the southern Sierra 14 times, and in that way we’ve covered the whole basin – at least the topography and land cover of the whole basin.
Water Deeply: What technical problems were involved in developing these sensors?
Bales: I would say our main innovation is the wireless network.
When we started doing this, the technology wasn’t mature enough to use. So my colleague at U.C. Berkeley, Steve Glaser, developed the electronics needed to do this. We purchased the sensors for measuring temperature, relative humidity, snow depth and, at some locations, we also measure solar radiation and soil moisture. Then he developed the data logger and multiplexer radio combination. So on one little board or chip, we have the electronics that will take the electronic signal from the sensors and, at 15-minute intervals, store them and then radio-telemeter those out to the mother station, which you then send out via cellphone to get those out to a computer where you can actually use them.
So, in the American River Basin, we have 14 clusters of sensors. Think of a cluster as being typically 10–12 sensor nodes and each node has several sensors and is solar powered. There’s a solar panel, a radio antenna sticking above the mast, then on a long arm is the snow sensor.
Then think of the radio signal from each of those, and call one of them the mother pod or the base station. So each of those 10 signals comes to a base station and that’s where you have your cellphone transmitter, and you’re sending that out to the cloud. To get the radio signal between them, you have some hoppers (signal repeaters) because it’s hilly and forested. So to get 10 measurements we may have 25 poles in the ground with radios on them to get the signal to the mother station. Then replicate that 14 times around the American River Basin.
Water Deeply: How much does it cost to install a network like this?
Bales: We have a new project going up in the Feather River right now, and we’re going to a different snow sensor there. We’re actually doing four clusters in the North Fork Feather River Basin. I believe the cost of this research network is about $40,000 per cluster. If you’re going to harden it for a robust operational network, you might want to spend more than that.
The paybacks are, I think, a no-brainer here. Just consider forest health. Knowing how much snow is out there can give you some predictability. During a drought year, if there’s not much snow, we should have been able to predict the tree mortality had we known the precipitation accurately. We know the amount of water trees use every year is about the same. If the precipitation falls below that number, you’re either going to have to suck it out of the ground from deeper storage or the trees are going to die.
Water Deeply: What results have you seen so far?
Bales: We’re researchers. We don’t do water operations or make forecasts. But we’ve demonstrated you can use it in real time. You can tell where the rain-snow transition is. That is, the elevation where it’s raining versus snowing. That was difficult to tell in the past because there aren’t enough sensors out there. This would help inform flood forecasting so you’d know: “Hey, it’s raining pretty high up.” You would know much more precisely how much rain is falling, which is obviously a concern for flood forecasting, whereas if it’s snow, you know it’s not going to melt right away.
The second thing we can get a better estimate of is how much snow is out there in storage, because that’s cumulative from storm to storm. We’re actually measuring how much snow is on the ground, and we’re doing it continuously. It’s been done at the existing snow pillow sites, but remember they’re only in clearings at mid-elevation. We’re extending it to more representative measurements.
Water Deeply: What are some real-world applications where your data network could have helped?
Bales: In the 2017 case of Oroville Dam (the spillway disaster), there was some uncertainty as to how much of the precipitation was rain versus snow in the basin above the reservoir. With a network like this, you will more accurately know how much rain versus snow is falling, and where. So you get a better estimate of how much water is going to be coming in today versus after the snow melts. That’s related to flood control.
Secondly is water allocations. Downstream users of water want to know how much they’re going to get as early as possible in the year. With a better quantitative estimate, we can provide decision makers with good data earlier in the year. So if you get that information two weeks earlier or a month earlier, I think that’s probably real money to somebody.