Over the last decade I have noticed a shift from using our historical data to plan, to collecting in-season data to best manage our growing crop. Through the process of this shift, we as growers have the ability to collect insurmountable amounts of data. This much data can have the ability to stop us in our tracks, not knowing which way to go next. In Episode 4 of Precision Points, Rachel Leonard walks us through a new technology that Geosys is working on to help us prioritize the pile of data we gain each day.
Rachel is the Development Program Lead at Geosys, an Urthecast Company. She has spent years heading up development for tools designed to get satellite imagery into growers’ hands so they can make decisions, both in and out of season. Over the last several years however, there has been a push for more and more in-season imagery from satellites. As the collection and delivery of this imagery has increased, there has become a need for a way to sort through the images and prioritize which data pieces are the most important. Enter, change detection.
Change Detection is an algorithm that highlights which fields have seen the most significant (or least significant) changes over a 5-15 day period. So as images come in at an ever more rapid pace, this technology can flag those that could be of the more importance. This could mean a change across the entire field, but also within the field boundary.
“That’s really the game of Change Detection, to call out those images in which change has occurred” Rachel said. “Change detection analysis is based on the distribution of the NDVI across the field.”
There are three different changes that this technology highlights: correlation, spatial variation and distribution. Correlation looks at the change within the field boundaries and identifies to what degree the different areas correlate in regards to their overall change, asking, “did the field experience the same change?” Spatial variation looks at how the field spatially varies from one image to the next to highlight if the changes were uniform across the field or not. Distribution looks at how the NDVI change is distributed across the field, noting if this distribution is the same from one image to the next. These three factors determine the scores any given field is given for change over that time period.
Prioritizing imagery then allows us to prioritize which fields need our attention. Growers and ag retailers still need to be hands-on in the field, but knowing where to look first makes this data much more user-friendly. The change detection index is looking just at a single field’s change, so it's powerful to combine with the Field Monitoring Tool that compares different fields to each other.
The development and release of this technology is coming at a great time as Geosys is ramping up storage capacity and processing power in anticipation of UrtheDaily. UrtheDaily is a satellite constellation that will provide daily calibrated high-resolution imagery, with a target launch in 2022.
Have you used NDVI Satellite Imagery? Leave a review here.
Transcription:
Host: Morgan Seger
Guest: Rachel Leonard, Development Program Lead with Geosys
Morgan Seger (00:22):
Welcome back to Precision Points. I'm your host Morgan Seger. And today on the podcast, I have Rachel Leonard from Geosys and UrtheCast Company. Now I've had the opportunity to work with Geosys over the last decade or so. And what they do is bring satellite imagery to the marketplace so growers can access it. So they're collecting and processing satellite imagery. And one thing that we've noticed over time is there has been a shift from using this imagery, satellite imagery, to create a historical archive of a field and use that to build a plan. And that shift has happened now to where we're collecting that imagery more real time in season, and trying to use that imagery to pinpoint areas that could use some attention throughout the growing season, and really call an audible on our plans and use that live feedback from the satellites to determine how to move forward throughout the growing season.
Morgan Seger (01:21):
When I first started working with satellite imagery, we were hard pressed to get a couple of images a season. And when I say that, I mean, the satellites are going over all the time, but where I live, and in the area I was working, we have lots of cloud cover we were competing with and other variables that were making the images that we were able to collect, not that applicable for how we were trying to use them to influence management decisions. Now, as this technology has continued to evolve and change, Geosys is now collecting quite a few images. In a recent article I read, they shared that the average field got 24 cloud free maps across the United States last year. So if you're thinking, as a grower who has multiple fields, or even as an ag retailer, who's maybe managing hundreds of fields, that is a lot of data to sort through.
Morgan Seger (02:13):
So what Rachel and I talked about today is a new technology called change detection. And what that is, is basically a way for them to prioritize what fields are making the most impactful changes, so we can prioritize where we're looking first. I think you're going to really enjoy this interview.
Morgan Seger (02:32):
Welcome back to Precision Points. Today on the show I'm joined by Rachel Leonard from Geosys and UrtheCast Company. She is the development program lead. Rachel, welcome to the show.
Rachel Leonard (02:43):
Yes, thank you for having me.
Morgan Seger (02:45):
I'm so excited to get to catch up. I know we've we worked together over the last couple of years, in and out, but would you mind just taking a couple of minutes and introducing yourself to our audience and giving them some of your background?
Rachel Leonard (02:58):
Sure. Well, as you mentioned, I work as the dev program lead for the R7 tool I'm on the Geosys team, but I interface very closely with the Winfield United business team working on R7, as well as our offshore development teams. So we have a team in France and a team in India that we work with. My role is responsible primarily for requirements management, mockups, functional specs, things like that, and working with the teams on technical feasibility. So the Winfield business team will come to us and say, "Hey, we'd really like to incorporate a new feature into R7. How do we go about doing that?" And it's my job to present those business requirements to our technical teams and get an understanding of what we can and can't do, of course, relay that back to the business, and we refine from there.
Rachel Leonard (03:43):
As far as Geosys, we were purchased in late 2018, early 2019 from Land O'Lakes. We were originally an independent company, but we were part of the Land O'Lakes family for about four years. And for us, the ownership change really made sense. Although our roots are in agriculture and Winfield United is our largest customer, there are many verticals for which Geosys analytic capabilities could really be employed in conjunction with UrtheCast's existing business and customers on the satellite imagery side.
Morgan Seger (04:11):
So do you guys then have your own satellites or are you mostly collecting data from satellites and processing that information?
Rachel Leonard (04:20):
Yeah, we use what we call a satellite constellation, which is supplied by a number of providers, some of which are public, some of which are private, but that allows us to maximize the imagery that we can promote our remote sensing algorithms. So the medium resolution imagery providers that we use are USGS, the European Space Agency, as well as UrtheCast, who owns another small company called Deimos Imaging, who operate their own satellites as well.
Morgan Seger (04:52):
Gotcha. Now I know one of the things that always made Geosys kind of stand out was how they process that imagery and made it usable to farmers. How are you seeing growers access the imagery and then use it to actually make decisions on their farms?
Rachel Leonard (05:09):
Yeah. From our providers, of course we have a small variation in resolution depending on the imagery provider, but we supply all maps and process them to calibrate to the same resolution. So the user really doesn't know which source the image came from. We constantly measure the calibration between sources to ensure the highest quality of maps output. So day to day, your maps should provide a seamless progression of growth on your field. And from there, growers can really take a look over the course of the season and see that progression, recognize risks as they happen, and make management practice decisions as a result.
Morgan Seger (05:47):
Gotcha. I always really liked working with satellite imagery because it would give us a consistent data set. So having that processing that calibrates everything to the same level was always really important because if you're picking up a new field and you don't have historical data, or if your historical data isn't calibrated, or if you're a smaller farmer like me, who doesn't actually have a way to get that field level or within the field variability information, I always really appreciated that.
Morgan Seger (06:16):
I saw an article you shared on LinkedIn the other day about change detection. And so one of the things I wanted to talk about today is what change detection is. And just a little background for our listeners. These satellites are going over all the time, collecting imagery all the time. And at first, when we were first rolling these tools out, it was like, Oh, we can't get cloud free images, but now we're getting tons of images. And as a grower, it was really hard to sort out, is this one important? Do I need to be looking at it? Especially if you're getting back to back days or a couple within a week or two's timeframe. So would you mind kind of walking through what change detection is and how that can help us determine which images are going to be the most powerful?
Rachel Leonard (07:03):
Yeah. And you're spot on. We went from a few years ago, getting one or two images per field per month to getting three, four or five a week. And so for a given grower, agronomist who manages upwards of 50 to a hundred fields, it's really hard to focus your attention when every field is getting an image every couple of days. You don't know which ones to reference, to look at, which fields might have some risks growing on them. So that's really the aim of change detection is to call out those images in which change has occurred.
Rachel Leonard (07:34):
Now, that doesn't mean that all change is bad. Of course, we expect change to happen on a field over the course of the season, otherwise we'd have bigger problems. But certainly change detection analysis is based on the distribution of the NDBI across the field. And we analyze images between a five to 15 day window, as we know that any less than that, only nominal changes can be observed, any more than that, we've determined so much change can occur that the output isn't valuable for the user. So we really want to recognize the time sensitivity of trying to identify those risk areas.
Rachel Leonard (08:08):
Now with our rollout of change detection this year, there really isn't any context behind the change detection and change index value that we provide on the images. It's truly to identify based on the gamut of new images that you've received on your fields on any given day, which fields might require some attention, even if that's just to look at the weather data and say, "Yep, last week, this field got a lot of rain. Of course, it's going to grow a lot more this week."
Rachel Leonard (08:35):
So it's a very complex algorithm behind the scenes. But from there, as I mentioned, we're able to calculate a change index value, which is an integer between one and 10, 10 being the most observable change, one being very little change. And in the R7 tool, we've integrated the first change detection tools this year. So that includes the index on the in season images that we integrate, as well as a field ranking dashboard to call out for that grower or that agronomist that manages many, many fields, which ones he should focus on that day.
Morgan Seger (09:08):
Gotcha. So what would a typical change within a field look like for it to register as a high impact change?
Rachel Leonard (09:18):
Yeah, it's hard to contextualize it at this point. Ultimately it's up to the agronomist to make the determination. It could have been a application of a product on a field that caused some heavy growth. Also at the end of the season, senescence or burn down would cause high change index values, more rapidly if it's very dry conditions. But again, it's up to that agronomist or precision ag lead to form those opinions. And that's what we're really trying to garner this year is a lot of feedback on contextualizing that change based on the crop site, the region, the growth stage. And from there, we hope to build in additional change index features to help contextualize that change.
Morgan Seger (10:00):
Gotcha. Yeah. I mean, you're covering a huge area and tons of different crops, so I could see it really being specific to a grower. When you're looking at these numbers, is it able to look within the field, changes within the field, or is it just looking at that overall index for the entire field?
Rachel Leonard (10:18):
Yeah, there are a few components of the change index and one is the spatial variability. So we can say if there's an area that suddenly is now not performing as well as the rest of the field, we'll call that out. There are some features in the tool now, the ability to compare two in season maps. So in real time, the user can view them both and determine those areas that may have changed more than others. And again, that might prompt a scouting opportunity, a tissue sample, what have you, but it's dependent on, of course, the conditions. As well, we've incorporated a change hotspots feature for our pilot users, where again, doing an analysis of the difference, they can see which pixels encounter the most change.
Morgan Seger (11:02):
Okay. So when you were first talking about your overview of the imagery that you're using, you said that there could be some small differences in resolution, even though you guys calibrate all of that back. Do you see that impacting the results you're getting from change detection?
Rachel Leonard (11:19):
Oh, we did last year, when we were running our first pilot of this project. And so as a result, we don't include a few of our sensors in this calculation. We noted that they were notorious for giving us false positives or high change index values where they weren't warranted. So it does depend on the provider, on the source image, but yes, you're right.
Morgan Seger (11:43):
Okay. Interesting. Now, do you anticipate this replacing some of the other things you guys have built in to help us sort out fields now, like the field comparisons or the trends? Or do you see this adding on?
Rachel Leonard (11:55):
Yep. That's a big question we get is how is this any different from field monitoring tool that's benchmarking my fields as well. And the difference, big difference is that field monitoring tool is not only based on low resolution imagery, which is different, but also the benchmark is calculated based on other fields, other similar fields in your cohort. So fields that are planted around the same time with the same crop, whereas change detection is really aiming to benchmark your field against itself. Again, in that five to 15 day window where we can say, "Okay, how much change is occurring across this field? And where is it occurring?"
Morgan Seger (12:32):
Gotcha. Well, I really appreciate the work you guys are doing to help sort this out because I think satellite imagery is a really powerful tool. But I remember not that long ago, and I was like, it would be really cool if we could get a push alert every time we got a new image and now you would go crazy. So I think this is definitely important work that you guys are doing. One thing I had a question for you on is Earth daily. Do you have any updates for us? I know a while ago I heard that maybe it was going to be ready by 2022.
Rachel Leonard (13:06):
Yep. And I think that's still a date that's in our mind, but certainly securing the funding and building the satellites is a long process. So that's where the UrtheCast team is now. On the Geosys side, we are of course, very excited at the prospect of daily, calibrated, high resolution imagery. The work that we're doing on our side is to standardize the APIs used to power our image processing engines. As well, we're beefing up for lack of a better term, our storage capacity and processing power in anticipation of that daily imagery.
Morgan Seger (13:40):
Yeah. I mean, I could see that being a real game changer, especially for the areas that even though we do have this influx of imagery, there are still geographical areas that struggle depending on their climate and things like that. So I think that that'll be a really powerful tool for them. So where could growers who are interested in learning more go to follow along with both Geosys and UrtheCast?
Rachel Leonard (14:06):
I would say follow us on LinkedIn. We often put out a blog post on our new features and tools. And as we garner some more feedback from users, hope to have some testimonials on how it's being used in the field. And then I would say stay tuned for any updates on our website as well.
Morgan Seger (14:23):
Awesome. Awesome. I appreciate that. So one question that I like to ask our guest is, is there one technology outside of your own that you are really excited about?
Rachel Leonard (14:34):
I feel like most of my time is spent focusing on the R7 tool, but I think the work that Winfield United is doing on sprayer technology is really interesting. And with their research facility and the wind tunnel is super cool, but I really also love the integration of robot technology on the farm. I think this is pretty exciting because from there, there are so many possibilities to incorporate AI, machine learning, so on and so forth.
Morgan Seger (14:56):
Yep. We get that a lot. I think there's a lot of anticipation around AI and automation and things like that. So that's definitely one that I think a lot of people are keeping their eyes on and could really change our entire way of farming, for large growers and small growers alike. So, interesting. Awesome. Well, I appreciate you taking the time today. I look forward to catching up soon.
Rachel Leonard (15:24):
Thanks again for having me. It was great to speak with you.
Morgan Seger (15:26):
Well, I hope you enjoyed that interview with Rachel. Like I said, I really appreciate satellite imagery for its ability to give us a calibrated data set regardless of where we are at historically with that field. The other thing is it's interesting to think about is the systems approach that you have here. So we have all of these images and change detection is now going to help us see which fields are making the most impactful change today, so we can make sure we have eyes on those fields.
Morgan Seger (15:57):
But it goes one step further because you're still getting the satellite image. So once you have that identified, the fields that are making the biggest changes, you can still use the imagery then to look at that in boundary variability, to target where you're going within that field, to make sure you're seeing those spots that are maybe slow in development, or that have changed a lot over the last five to 15 days. I think it's a really powerful system for growers of all sizes. And it definitely sounds like companies are going to continue to ramp up their ability to collect process and serve growers with this type of data.
Morgan Seger (16:36):
And with that, I am so grateful that you decided to tune in and listen to another episode of Precision Points. Don't forget to rate, review and subscribe to our podcast, so you don't miss an episode.
Host: Morgan Seger
Morgan Seger spent ten years working in ag retail, specifically in ag tech. She lives and farms in western Ohio, where she has four children with her husband Ben. Morgan has her own blog called Heart and Soil where she talks about her experience farming, gardening, and raising her family.
Guest: Rachel Leonard, Development Program Lead with Geosys
Rachel Leonard is the Development Program Lead on the Geosys team for the R7 Tool. Her role is to act as a conduit between the Winfield United business team and the offshore development teams in France and India. She has been working on the R7 Tool project for 7 years, beginning in support and growing her role into operations and business analysis. She is responsible for the development of business requirements, functional specifications, and mockups for R7 and the traceability of these requirements throughout the project’s iterative life cycle: from development, user acceptance testing and finally, release and operational support of the software to users. She also participates in service delivery of the R7 Tool’s In-Season Imagery program. Her background is in GIS and previous to her work with Geosys she worked as a GIS Analyst for a small business supporting local and county public works operations.
תגובות