Combining in situ Soil Moisture with Satellite Data for Improved Irrigation Recommendations

How researchers are combining in situ, drone, and satellite measurements to extract key information and how these data can be connected across scales.

Improving irrigation requires smart data gathering to help growers make better choices in the field. Measuring in situ creates high-resolution, temporal data enabling us to see clearly what’s happening over time—but only at a single point. Satellites show data across a large spatial scale but are hampered by revisit frequencies, clouds, and resolution limits.

Often we see information in a silo, looking at one type of data or another. The challenge to researchers is how to connect across these scales and combine the information to make better irrigation decisions. In this webinar, Dr. Colin Campbell explores the future of irrigation and research he’s been doing with collaborators at Brigham Young University. Learn:

  • How researchers are combining in situ, drone, and satellite measurements to extract key information
  • How these data can be connected across scales

Next steps


Our scientists have decades of experience helping researchers and growers measure the soil-plant-atmosphere continuum.


Dr. Colin Campbell is a senior research scientist with METER Group and serves as the Vice President of Environment. He is also an adjunct professor with the Dept. of Crop and Soil Sciences at Washington State University where he teaches a class in environmental biophysics. Following his PhD. in Soil Physics at Texas A&M University, where he studied field scale carbon flux, Dr. Campbell has spent the last 18 years developing sensors and instruments to make measurements in the soil-plant-atmosphere continuum. One of the highlights of his career was working together with METER colleagues to design and build the Thermal and Electrical Conductivity Probe that measured multiple parameters including soil moisture and thermal properties on the surface of Mars as a part of NASA’s 2007 Phoenix Mission. His latest work has been focused on developing in situ moisture release curves and perfecting a new all-in-one mini-weather station


See all webinars

Lab versus in situ soil water characteristic curves—a comparison

Lab-produced soil water retention curves can be paired with information from in situ moisture release curves for deeper insight into real-world variability.


Rapid, in situ Thermal Conductivity Measurements, Even in Moist Insulating Materials

Dr. Gaylon Campbell shows how to measure the thermal conductivity of wet insulation with a 5-minute read time. These previously impossible measurements are now possible through advanced electronics and analytical methods.


Water Management: 3 Important Tools You Might Be Missing

Dr. Colin Campbell explores the latest water management research and real world examples to answer the questions: Does water management work? What are challenges and best practices? And what should we do next?


Case studies, webinars, and articles you’ll love

Receive the latest content on a regular basis.


Hello, my name is Dr. Colin Campbell. I’m a senior research scientist here at METER Group, and also an adjunct faculty member in the Department of Crop and Soil Sciences at Washington State University. And today, I want to talk more about the future of irrigation and some work that I’ve been doing with the collaborators that I’ve noted here at the bottom of the screen. When we think about the need to improve irrigation, quickly coming to my mind is gathering data that will help us make better choices in the field. For many years, I’ve worked on projects where we’ve cheaply measured in situ water content, that has the value of being able to create high resolution, temporal data so that we can see what’s happening very, very clearly over time. And yet, this happens at a single point. Other products out there, like satellites, help us to see things across a large spatial scale. And yet, because of clouds, and because of satellite passovers and limits to resolution, we can’t do the same things that we do with in situ sensors. So we got to thinking, how can we use some of these different pieces of data to try to help us understand how to improve our irrigation? This study that we’ll talk about today is just a small part of some larger scale work we’ve been doing for the past few years, and I’m going to talk about that as we go.

So as I mentioned, one of our key goals is really connecting across scales. The challenge that I’ve talked about is that often we see things in kind of a siloed approach. I’ve made in situ water content measurements for many years. And there are others, when I go to conferences, who often talk about the great things about satellite products. And when we usually talk about one thing or another thing, we may talk glowingly about the fact that we can see changes in water content very clearly over time, for example, or matric potential. But we don’t necessarily talk deeply about the fact that at points just a few meters away, it may differ in water content. And in the same way, people who do remote sensing may talk about how the satellite gives them some information about the water content, but may not dwell on the fact that they can only get about the top, sensitivity from the top five centimeters. So it occurs to me, and to many, that our challenge really relates to how do we measure across these scales? Because of course, in one situation, we’re really happy with the resolution in time, but not in space, and the other we’re happy with just the opposite. So the obvious answer that we came up with was, why don’t we combine them together to extract the key information from each of these things? From in situ measurements — I haven’t mentioned drones, but drones should be added into the conversation — and from satellite products. Because being able to combine them together, this will drive our overall knowledge. But things that we don’t entirely understand right now are how are these scales connected and will combining them actually provide meaningful results in the end?

So I’m going to talk a little bit about these ideas in the context of a project we’ve been working on now since 2016. A grower collaborator of ours purchased a variable rate irrigation system. I’m showing that here on the right. And this is kind of an interesting picture. This actually shows the prescription that he laid out for his variable rate irrigation system on this one particular field. I once asked him Hey, what is this red down here mean? Well, that was actually some alfalfa hay that he was growing that he turned the irrigation off on because he was harvesting. This area down in the south half the field that’s irrigated might look kind of familiar and in fact, if you look closely, it probably starts looking like a golf course, which it is. It’s the tiny community of Grace, Idaho, it’s their golf course. And the collaborator is very interested in golf, so he made on the south half of this field a nice little a nine hole golf course. And it’s interesting that this nine hole golf course is something that our collaborator knows how to irrigate. All those different colors represent different you irrigation values. So you can see we irrigate this fairway less, for example, then maybe this is the green, I don’t know, for sure. And or maybe a tee box here. And here’s the green down here, I’m not sure. But you can see he has a pretty comfortable idea of how he wants to irrigate that field. If you look at the colors in the north half of the field up here, what you’ll notice is there’s not much variation at all. This is because he doesn’t really know how to irrigate this field that in the year of this, the first year of this study, in 2018, it was planted in potatoes. Now sorry, the first year that we’re going to talk about here in a minute that I did mention it started in 2016.

So really, the goal of the study was to try to understand how to provide the grower with information where he could use this precision irrigation technology that’s been rapidly advancing for years to use it with a very clear recommendation of how much water and where it should be applied. Over the years from 2016, we started this project in a little bit of different way than we’re actually running it now. So the first effort we used was to intensively soil sample this entire field — we have over 100 sites — and then use those soil samples with measurements of water content at several depths. And these were just samples in time. And then evapotranspiration, to predict crop water loss, then using that information, this very pretty dense spatial sampling, we could then be able to see how much yield was created in the field with a yield monitor and then compare it with this water loss through ET and soil sampling, and then find a productivity index for these locations across the field. Well, we weren’t able in that first effort to connect the dots in terms of yield very well. So in 2018, we decided to incorporate soil sampling with continuous monitoring at strategic sites across the field. And then we use a satellite to upscale these data. And I’m going to talk briefly about that. That’s actually something I taught more in depth in another virtual seminar, which we’re going to link here, so that you can actually go in there and watch that in more detail if you’re interested.

Once we finished with that, the grower was so excited with what we’d learned in 2018, and I’ll show you why in a moment, that he wanted to extrapolate that learning to multiple fields with now a single monitoring site and try to use satellite to upscale that. So instead of the multiple monitoring sites in this one field, he chose to try to use a single site and try to upscale the best we could. So I just want to kind of provide a quick overview of our 2018 results. Again this was presented in much more detail in another seminar that we’ve linked. But in 2018, this was out in a single field, we were measuring matric potential, using the METER Group TEROS 21 at 45 centimeters. We buried them at 15, 45, And then well we have water content sensors also at 15 and 45 centimeters and then one water content sensor at 75 centimeters. It turned out that the 45 centimeter sensors were the most interesting because they were where we found the majority of things happening and the best connection between what was happening in the soil and what was happening over satellite. So the matric potential here is graphed on the y axis over here. And time is now on the x axis here running from the beginning of the experiment early June all the way till they killed the vines in late August. So we deployed in this field six total sites, we named them 906 all the way through 912. And sometimes in a slide that I’m going to talk about in a moment, we sometimes chopped the nine off because it was ubiquitous here. So site 6, 7, 9, 10, 11, 12.

And this graph shows you very clearly that three of the sites, not on purpose, but just as it happened, went dry over the season, that they dropped out of this plant optimal range that’s basically down to about negative 100 kilopascals matric potential, and enter this range of stress, which lasts from about negative 100 to negative 1500. Now these are not exact numbers. These are for this negative 1500. That’s just an estimate and the negative 100, again, these are numbers that are in the literature and we could talk about, but let’s assume that they’re approximately close to when the plant is in optimal water and in stress and then eventually, down very low would be permanent wilting point. So we see some things in stress, we see some things that maintain a well water condition. And this site seven dried just a little bit. Now, when we were actually running this experiment, the grower went out to the field and said, No, no, this is showing that some of these locations are dry, but my field is doing perfectly well. It was interesting that he actually was walking out to this site 906, which was closer to his access. And so it’s not surprising that it seemed like it was doing quite well. But the question was, is there a tie between between the stress condition, yield, and also, can we scale it using satellite data?

So we’re going to answer those, we answer that in the other virtual seminar, but I’m going to talk about it briefly here and try to tell you why we decided to upscale this experiment. So we broke the data up so that each site, we simply measured how long below this 100 kilopascals level was each of those sites, what was the time that they were below negative 100 kilopascals. And what these data show is that grabbing the yield from our yield monitor system, those sites that were not in stress very long, they yield up around 40 megagrams per hectare, but those that were in stress showing right here, and these were in stress, about 40 to 50 days, these clearly, according to our yield monitoring system, showed much less performance, much less yield from those locations. And when I talked with the grower about this, he was pretty excited, because he said Man, within that one field that I thought was so uniform, we clearly got, just on accident, some large range of yield differences. So upscaling with the satellite, we used our satellite data to try to predict yield just from this connection between matric potential change over the season, and the yield that was generated. So we made a model with this delta matric potential with our satellite data, which we included a water content surrogate with SAR data, we also included a temperature component as well, a canopy temperature component with near infrared. And when we tried to scale it across the field, we were actually pretty excited because there was a weak but significant correlation between the two. And we continue to work on evaluating these data when we have time, and believe that adding in things like water content, drone flyvers, and other things will help us really refine this relationship.

But that’s not what I’m going to talk about today. Today, I’m going to talk more about how we decided to upscale this relationship in other fields, because the grower got really excited about this and said, Hey, why don’t we try to do this on other fields. But this time, I don’t have time and I don’t have the money to put in a bunch of different sites in each field, I’d like to do this on one field. So the first thing we did was notice a correlation between a different index that we didn’t use in the data I just showed and making that model, we used a index called the Normalized Difference Water Index, which we also got from satellite information. And this shows, here, over time, so time on the x axis now that the water index here on the y axis, that our site 12 or 912, that was our driest site, deviated strongly from the other sites over time in this August period. And it gave us a sense that that may be quite a sensitive index to help us judge when there was stress. So the question for 2019 was, can we expand our effort? Because the grower was anxious to use all that we learned in this initial project to really help him with all the other fields full of potatoes because this particular field that had the VRI was actually going into its wheat rotation, so two years of wheat, and it’ll be another from then three years went before it goes back into potato rotation. Interestingly enough with all these other fields, he has fun names for them. So I wrote them here: Cemetery, Max’s Pivot, Barbara’s West Mini, and you can see the rest there. And so I’ll be using those field names as we talk about some of our results. So it was also kind of exciting that there was enough interest in the system that the grower we work with actually purchased all the systems, installed them, and monitored them through the whole season multiple times a day. So he put in water content sensors at 15 and 30 centimeters, and also matric potential sensors at 30 centimeters, he also threw out a rain gauge as well as the METER Group, ZL6, which is a cloud connected data logger. So he used an app on his phone to pretty much. multiple times a day, monitor his sensors in the field. And of course, as I mentioned, he wanted to do this himself. So his limited funds meant that he was only going for a single site. And that and just the time he had to actually do the installation.

So the first question that he asked, and it’s one that we get over and over is, Hey, okay, where do I put these systems? And it came into this discussion right away, because these in situ data systems are really temporally rich, but spatially very poor. So the in situ measurement sites were actually selected by us just using estimates of seasonal wetness across each of the fields that he wanted to put them in. And there are six shown here, there’s actually a seventh, I’m not sure exactly where that is. But one of the collaborators on the project Ryan Smith went through and analyzed the seasonal wetness across each of these fields over the last several seasons. And when he did that, he was able to find a very dry point in the field, and an average point in the field that represented the entire field. So in red, on this field, you’ll see these red dots, that’s the driest location that the satellite sensed over the last several years. And the green represents the average location, the average moisture across the season. And then we actually talked to the grower because it was his system and asked him, Where do you want to actually put this in? And his choice was definitely the average. That was where he wanted to go. So here’s the volumetric water content in the seven fields that we installed sensors in, starting here on the x axis in mid June, now it was a cooler, wetter spring here in southern Idaho, and the fields were not planted till relatively late in the season. And here’s the volumetric water content on the y axis going from around 10%. So it’s not all the way down at zero, all the way up to about about 40% Each field started a water content that was quite different. This field Reed’s Middle, for example, seem to be quite a bit lower in water content than some of the others. But because of the nature of water content, we can’t really say whether or not that was an adequate amount of water for the potatoes in that field compared to any other field. Because these fields definitely differed by soil type.

We can see some interesting things in these curves though. We can see irrigation events for example here in Reed’s Middle, and here and here. Here, we see a very clear irrigation event to start here and we see them throughout this trace for example. And here as well, in homeplace southwest, we see them. Now depending on the soil type, sometimes we have some very pronounced irrigation events and sometimes we don’t. I think that the take home point here is that although we have some great water content data, and we can even see generally the crop taking up water and also being irrigated, that it’s not entirely clear how much water or whether or not the plants are stressed in this graph. Now this is a graph that comes straight out of the ZENTRA cloud software that METER produces, and that Ryan was actually using, the grower, during the year.

Here’s also the 30 centimeter sensors in each of the fields, and now, this is the matric potential, instead of the water content. We already described the matric potential in another side, but this graph goes on the y axis from zero, and matric potential is negative. So it goes down to negative 500 kilopascals. And once again, we have the same timeframe here on the x axis from mid June all the way into September. Now, this graph looks a little bit messier. But I would put to you that actually, it’s a little bit nicer than the other graph. So here’s the planting, each one, sensor came in dry, it quickly equilibrated. And then across all these fields, there’s only a range of about plus or minus 10 kilopascals difference between any one of the fields. And so what this tells you is that all those fields were starting in a water potential range that is optimal. So the question that came up in the last slide that was one of the site’s not watered enough, is clearly not an issue here, because it’s at the right water potential, it’s in the optimum range. What I did with the grower, and we worked together and just created a simple range inside that green bar that he needed to keep his water. So he simply looked at these data every day, and made sure to try to keep as best he could and within the context of how much water he had available to put on these fields, just keep those bars or each trace within that green area. And you can see overall, that he actually did a pretty good job of that. Although maybe, as we look at it, we might say, Oh, I see some periods of time where he’s well outside that. But what he did then was actually turn on the center pivot. It took some time to go out to that location and to bring it up to the optimum level. And in the end, he ended up keeping the water in a very good range on all these fields.

Now, why do I say that was such a definitive statement? Well, what we wanted to know is to figure out whether there was crop stress in these fields and how it affected the yield. So I just put up for comparison the 2018 potato data here. And so here are those sites, again, and here are the days in stress for the sites where we saw a concerning drop in yield compared to the yields in these other fields. Now, here in this big table, these are all the fields from 2019. And we essentially did the same thing, where we looked at how many days was each field below negative 100 kilopascals. And when you look at that, we have 18 14, 17, two times where it never got below that. And then this Reed’s Middle we saw that had some of the lower water content, water potential didn’t look too concerning as we looked at it, but that was 21.

So if we looked over here, where we saw no yield change, we have this connection to the 16 days below negative 100. So none of the fields in 2019, were ever approaching the 40 to 50 days in stress. In fact they didn’t do that much at all. So what we saw was that because of that, there was really no loss in yield. As we looked across there, all we see was for quite good yielding sites. These are different varieties, by the way, so something like 32 is not necessarily concerning. And interesting it has the lowest yield, but it had the fewest days below negative 100 kPa. So it could be just a different yielding variety, which I think it was, or it could be another issue like disease or pest pressure that actually caused the lower yield. It was interesting though, to observe something, so here’s the yield at the sensor location. And look at how close that yield was to the average yield in the field so the difference between the yield of the sensor and the average field yield, and then compare that to the field standard deviation of yield. So this is from our yield data, our yield monitoring system going through the field in every case, but one, this case and it was dead on here, the difference was less than one standard deviation of the yield we picked up, which I think is really exciting. It tells us that the satellite data we use to decide where to put our sensors, actually, was really good, that we were able to choose a good spot to represent the average of this field.

So we took all these data, and we tried to connect them into the ecostress satellite data. So a couple of our team members, some data scientists, took our data and tried to fit our water potential using ecostress satellite data, and ecostress only, it only goes back to the middle of 2018, so we still need to do some work to even use that for last year’s experiment. We could use it for this year. But ostensibly, I’m not going to go into data here that we were able to get a decent R squared when we did the training set on the data. But when we actually tried to test our data, we got a really low R squared and a high root mean squared error, which means that we essentially, and this is pretty clear from this graph, we couldn’t use the satellite data to model the data in the field. At least not yet. So we only tested this, we just barely started this work, we only tested on matric potential, we’re hoping that we can gather a whole bunch of these products, we have drone data still, we have water content data, we also have canopy— well, we have have drone data, sorry, we don’t have canopy temperature. We have drone data, we have water content data, and a few other pieces that we can gather together to hopefully model this a little bit better. But right now, that didn’t work very well. And when we tried to do the Normalized Difference Water Index, with the correlation with potato yield and ET—not good. So we weren’t that excited about these data. We need to do a little bit more work and try to see what’s behind the lack of correlation.

So overall, we struggle getting a yield connection to our satellite data. So from ecostress, we didn’t see that ability to upscale from our point measurements up to the field using the satellite. And the connection we saw with yield last year was not kind of pulled through and connected to this year. We still hope that there might be some things that we can glean out of this, but that’s so far what we’ve learned. But what we also learned was that from the grower, his sense is that he used much less water. In a typical year, he used between 50 and 60 centimeters of water. This year he was well below the bottom end of that. Now it was a cooler year. And so we’re working on ET analysis, actual infield ET not satellite ET, to see what the difference was this year to other years. But the thing he was most excited about was that the average potato size that we picked out by satellite, were pretty much right on, both what we saw there in the yield data and also when he actually went out and physically dug potatoes and compared weights at three sites in each field, that average was always sitting at the location that we buried the sensors. So he felt really confident that we were actually putting our system right on the average of where it should be. And that idea, just the opportunity to actually answer that question of, Where do I install my sensors if I’m just going to install at one location in that field? The idea that we could answer that question is pretty exciting.

So in summary, picking a single measurement site with satellite data seemed to work quite well. And that I’m pretty excited about. I’ve been working in this area for many years, and got that question over and over. And I feel like we’re just a little bit closer to an answer with that. And irrigating to in situ water potential at that average site that was generated by satellite got great potato yields and seem to use less water and and made it so that the number of days in stress was clearly limited such that there was there was no apparent yield loss due to stress of the potatoes. But right now scaling up using a single season satellite data was just not successful. And I think that was one of the big problems as I mentioned. The ecostress only goes back a little over a year, and trying to match those was difficult. We’re going to test out some of our other things. You notice we had the Normalized Difference Water Index, we’re going to work on that a little bit. But there are still other sources of data that we can work with and see if we can find correlations that will let us scale up these point measurements more effectively. And that’s no surprise because right now, this connection between in situ data, satellite data, and drone data, I mean, we see a lot of people working on it, but there’s still a lot of work to be done to try to understand this. But from Ryan’s perspective, the grower we’re working with, he was extremely excited that we’ve made all this progress and he felt like he was able to manage his irrigation water much better this year with this kind of system. So I think that’s a great take home message as well. With that, we’re going to end. Thank you for spending time with me talking about this exciting new effort we’re engaged in and look forward to any questions you might have. You’re welcome to email me at [email protected]. Have a great day.

icon-angle icon-bars icon-times