2021 Nebraska Cover Crop and Soil Health Conference - Kristen Veum
Soil Sensing and Soil Health - Kristen Veum, Research Soil Scientist at USDA-ARS Cropping Systems and Water Quality Research Unit, University of Missouri-Columbia
Toggle between list and paragraph view.
- [00:00:21.110]All right. So I'm Kristen Veum.
I'm with the ARS in Columbia,
- [00:00:26.540]And today I'm going to talk about soil
sensing applications in soil health,
- [00:00:32.090]and I think I'll be able to help
make out some of this time too.
- [00:00:35.420]So I'm a research social
scientist, and most of the time,
- [00:00:39.140]my research involves
lab work and evaluating
- [00:00:43.820]soil properties. So methods,
lab methods, and, you know,
- [00:00:48.080]things along those lines.
- [00:00:49.640]And so these are some pictures from my
lab and the awesome staff that I have
- [00:00:53.870]that helped me do all of that, that work.
- [00:00:58.230]We all know, okay. You know,
the soil health concept. Well,
- [00:01:02.990]we need some science to,
have as the foundation of that.
- [00:01:07.580]So as a soil scientist, the way I
contribute to that is through these,
- [00:01:13.190]biological and physical soil
measurements and evaluating
- [00:01:18.110]how they work. But as you probably know,
- [00:01:23.270]there are a lot of different soil
properties that we can measure.
- [00:01:27.740]And in fact,
- [00:01:28.280]it's probably unlikely that you would
ever be able to measure everything.
- [00:01:32.570]And we have to think about, well,
what are we going to measure?
- [00:01:35.570]How are you going to choose
- [00:01:38.300]And we like to lump these into three
categories, right? Like chemical,
- [00:01:42.680]biological, and physical measurements.
And so if you look at this list,
- [00:01:46.940]you may be familiar with
some of these things.
- [00:01:49.040]Maybe you've measured some of these
things on your own farm or in your own
- [00:01:53.030]experiments. and maybe you're
not familiar with some of the others,
- [00:02:00.350]but with, within that research area,
- [00:02:03.380]one of my primary interests is
working on assessments for producers.
- [00:02:08.450]And that's, that's a goal of mine.
- [00:02:09.860]But if we're going to measure
soil health for producers,
- [00:02:13.580]we need measurements that are accessible
so that they can actually get it
- [00:02:18.200]somewhere, right.
Somebody has to offer it.
- [00:02:20.660]It has to be affordable and
something that we can interpret.
- [00:02:26.150]There are a lot of
challenges related to that.
- [00:02:29.000]And a lot of people have
been working on that,
- [00:02:31.550]but today the soil sensing work
really pertains more to that
- [00:02:36.290]affordability question.
- [00:02:39.680]So this is just the list of the
cost of a lot of these different,
- [00:02:44.390]so health analyses.
- [00:02:46.220]And of course the cost varies depending
on which lab you're talking about.
- [00:02:50.720]Ones that I circled here are more
of the standard fertility analysis.
- [00:02:55.880]But if you look at these costs,
you can tell that right away,
- [00:02:58.760]when you start adding different
sole measurements together,
- [00:03:02.740]the costs on that get
pretty high, pretty fast,
- [00:03:08.970]but this is what our fields look
like, right? I mean, I, I don't know.
- [00:03:13.290]Maybe somebody out there has a really
homogenous field where everything's all
- [00:03:17.190]the same, but at least here in Missouri,
- [00:03:19.830]we have a lot of variability in our
fields. And some of that comes from
- [00:03:25.890]erosion. You movement of soil
over time and deposition.
- [00:03:30.330]But if we're going to
understand soil health,
- [00:03:34.530]now we're going to manage our
fields at the field spiel.
- [00:03:37.830]We really want to get a grip on
how the soil properties vary,
- [00:03:43.170]you know, within that single field. Right.
- [00:03:45.240]But when you combine that with the fact
that these lab costs are really high,
- [00:03:50.040]we have a problem.
- [00:03:53.880]So how are we going to reduce
costs? Well, in the lab,
- [00:03:56.700]there are a lot of ways that we
try to reduce costs for, you know,
- [00:04:00.210]for producer testing, right?
- [00:04:03.120]So we have combination
testing where you can do,
- [00:04:05.220]there are some things that we can
try to measure simultaneously.
- [00:04:09.750]We try to use cheaper supplies.
We try to use fewer supplies.
- [00:04:14.700]We try to come up with methods
that are shorter and faster,
- [00:04:18.720]and therefore a lab goods you'll
make a better profit on that,
- [00:04:23.340]move it through more quickly, we
try using smaller sample sizes.
- [00:04:27.480]So you don't need as much soil
or reducing overall volume,
- [00:04:32.070]or we can use less sophisticated or
less expensive instruments in the lab.
- [00:04:37.410]But even if we do all of this,
- [00:04:40.170]you still have to collect a
sample and send it to the lab.
- [00:04:43.590]And your results only tell you about
where that one sample came from.
- [00:04:50.550]So this.
- [00:04:50.880]Is where proximal soil sensing comes
in. So Andrea was just talking about,
- [00:04:56.040]using remote sensing, right?
So whether it's satellite data,
- [00:05:00.990]drone data feeds where
you have sensors that are,
- [00:05:06.510]you know, up in the sky, somewhere up
in the atmosphere and getting imagery.
- [00:05:10.530]We also have sensors that are
directly in contact with the
- [00:05:15.510]soil, and that's what we
call approximal soil sensors.
- [00:05:18.090]So they're either on or in the soil.
And in comparison with some of the lab,
- [00:05:23.040]methods that we have,
- [00:05:25.230]the sensors are generally non-invasive
and non-destructive, which is great.
- [00:05:30.810]Once you get past that capital
equipment costs of actually buying the
- [00:05:35.460]sensor, which can be very expensive,
- [00:05:37.740]actually using it and getting the data
in is inexpensive. And then with that,
- [00:05:42.480]you can get this high
resolution data, right?
- [00:05:44.520]So you can cover and get a lot of
information over space and over time,
- [00:05:48.570]we'll keep coming back with it.
- [00:05:50.670]And then you're not collecting more soil
samples and sending them off to the lab
- [00:05:55.200]and they're pretty easy to
operate. So they're low tech,
- [00:05:59.810]in the sense that if you have to
train somebody to do it, it's a lot,
- [00:06:03.620]it's a lot easier for someone to operate
this equipment than it is to be fully
- [00:06:07.820]trained on laboratory procedures.
- [00:06:13.220]So the techniques,
- [00:06:15.050]approximal techniques that we use
the most in my unit and in our
- [00:06:19.430]research are these three.
- [00:06:21.410]So there's the visible
near infrared spectroscopy.
- [00:06:25.040]And I'll just call that NIR and
I are spectra on talking and it's
- [00:06:30.020]been used and pretty
successful for estimating.
- [00:06:33.380]So organic carbon texture,
- [00:06:37.580]And in some cases, although
it's in sort of rare,
- [00:06:41.240]aggregate stability
and some nutrients,
- [00:06:45.920]electrical conductivity is
another one we use a lot of,
- [00:06:49.490]and that's pretty good, pretty
successful for soil texture, mineralogy,
- [00:06:54.050]CEC, and so moisture. And
then the penetrometer,
- [00:06:58.760]you've probably all heard of that.
- [00:07:00.560]there are even handheld
versions of a penetrometer,
- [00:07:03.950]but that relates more to those
soil, physical attributes, like,
- [00:07:07.040]so I'll text her bulk
density and compaction,
- [00:07:10.520]and I've got some pictures here of
some of those pieces of equipment.
- [00:07:15.440]So Andrea already did this for me.
- [00:07:17.390]She talked about the
- [00:07:20.450]So I don't need to go into much
detail here, but each sensor,
- [00:07:24.950]if it's a spectroscopy sensor,
- [00:07:27.500]it's going to give you information from
different parts of the spectr right?
- [00:07:31.790]So you can have sensors.
- [00:07:33.860]so we have these passive gamma detectors
that tell us about gamma rays that are
- [00:07:37.850]coming out of the soil. And we have,
- [00:07:40.490]the ability to take soil and subjected
to NMR and get specter for that.
- [00:07:45.770]But most of the things that we use out
in the field are in that middle range,
- [00:07:49.760]they're the visible and
near infrared range.
- [00:07:54.890]This is what some of these spectrum
will look like. in this case,
- [00:07:59.300]these were all essentially the same soil,
- [00:08:01.790]but they had been under
different long-term management.
- [00:08:05.210]So there was a perennial Timothy
system continuous till corn,
- [00:08:10.220]degraded oil that was,
reconstructed into a Prairie system.
- [00:08:15.500]And then we have this native
Prairie view at the bottom.
- [00:08:19.250]And on the left is one type of
spectral data. And on the right,
- [00:08:23.690]it's a different type of spectral data,
- [00:08:26.270]but all I want to show you here
is that if you look carefully,
- [00:08:30.240]and maybe you put on
your glasses for this,
- [00:08:32.090]but there are slight
differences in these spectra.
- [00:08:37.010]And so that's what we're trying
to, to analyze. And you know,
- [00:08:41.120]why are the spectrum different
from one sample to another? Well,
- [00:08:45.860]in spectroscopy usually
relates to organic matter.
- [00:08:48.800]So you can have differences in the
quantity and the quality of the organic
- [00:08:53.720]matter in the soil and then the
inorganic fraction. So those minerals,
- [00:08:58.410]like the sand Sultan play contribute
to those spectral features and
- [00:09:03.390]make the spectrum look different. And
then Erin water also has an impact.
- [00:09:09.300]The main challenge with this type of
sensor data is that it's subject to some
- [00:09:13.650]environmental effects,
differences in temperature,
- [00:09:16.800]and particularly moisture can
cause a lot of interference
- [00:09:21.630]in the spectrum. And then certain
things just have a really weak signal.
- [00:09:25.860]And so we can't really
pick it up very well
- [00:09:30.390]when it comes to soil
- [00:09:33.870]And this is the most widely useful sensing
technology and precision agriculture.
- [00:09:39.450]I mean,
- [00:09:39.660]it's been used pretty extensively and
there are several different commercial
- [00:09:43.140]sensors out there that you can,
- [00:09:45.300]you can use everything
from a small handheld unit,
- [00:09:49.560]which is this orange thing in the middle.
- [00:09:51.000]You can just carry it in your hands
instead of down and take a reading.
- [00:09:55.080]There are also some that you can
pull behind an ATV with a wagon or a
- [00:09:59.850]sled, or you can even,
- [00:10:03.540]attach these to your equipment and pull
them through the field with a tractor.
- [00:10:08.940]And potentially even when you're
engaging in other field operations,
- [00:10:15.130]here's an example of where you see
data was used to map soil texture.
- [00:10:19.300]It was a pivot field and they
- [00:10:24.390]took that ENC sensor and they rub
it back and forth across this field.
- [00:10:29.280]Then they went in there and I'm going
to set it all at the end is here maybe
- [00:10:34.410]20 locations.
- [00:10:35.430]They took samples and sent
that off to the lab and got a
- [00:10:40.140]texture analysis.
- [00:10:41.520]And then they compared that UC readings
to the sand content and there was a
- [00:10:45.900]pretty strong relationship there.
- [00:10:47.880]So then they translated this
NC data into a texture map or
- [00:10:52.860]a sand present sand map. And you can,
- [00:10:56.970]you can imagine how this might be useful,
- [00:10:58.890]especially if you're doing something like,
variable irrigation, for example.
- [00:11:06.420]And do you mind to ask this question,
okay, this is supposed to be out,
- [00:11:09.390]so I'll help. Why is she talking
about soil texture? And even though,
- [00:11:13.310]so I'll text her,
- [00:11:14.040]isn't considered necessarily a
traditional soil health measurement or a
- [00:11:18.840]dynamic soil property
it's is fundamental to,
- [00:11:23.820]the soil health characteristics,
the potential for soil health.
- [00:11:29.130]so for example,
- [00:11:31.620]texture really controls how much water
and how much carbon organic matter as
- [00:11:36.480]well can hold, right? So
it really is important,
- [00:11:40.110]in the grand scheme of soil health that
we understand so little texture and that
- [00:11:43.800]we'll take, take that into account.
- [00:11:46.680]So I'm gonna play a quick video for you.
This is us in action, out in the field.
- [00:11:51.570]So here's a,
- [00:11:52.890]an NIR sensor it's being
pulled across the field.
- [00:11:56.740]So you're getting a lateral map of, and I,
- [00:11:59.440]our spectra and following behind
that is Scott with our sled with,
- [00:12:04.330]ECG sensor.
- [00:12:07.900]And so we'll get a map that
has both of those sensor types.
- [00:12:12.760]And then that truck right there
has our probe mounted on it.
- [00:12:19.690]And we're going to see that in action
here in a moment. So in this case,
- [00:12:23.920]we're getting profile data. So the
sensors are here at the end of this probe,
- [00:12:28.600]and it's going down into the soil.
- [00:12:30.160]We can get it's reading
constantly as it moves,
- [00:12:32.830]and we're getting data down
to a depth of about one meter.
- [00:12:36.340]And so we're getting that and
I are spectroscopy electrical
- [00:12:42.880]and those penetrometer
readings all at the same time.
- [00:12:46.900]So that window is a
Sapphire window for the NIR.
- [00:12:49.750]And you can see the context
there at the bottom for EDC.
- [00:13:03.480]So I'm going to give you a couple case
studies where we've applied the sensor
- [00:13:07.860]data technique to try to
estimate soil health indicators,
- [00:13:12.120]the soil properties.
- [00:13:13.710]And most of this work was done at our
longterm agro ecosystem research site,
- [00:13:18.630]which is in central, Missouri,
and out there, we have
- [00:13:24.270]12 different row crop
systems and some CRP systems.
- [00:13:27.750]So we really have a pretty wide
range of soil health values.
- [00:13:34.080]So the first thing we did was look at
using NIR spectroscopy to estimate soil
- [00:13:38.820]carbon. And it worked really well,
- [00:13:41.430]but I'll say that was
not really novel insight.
- [00:13:44.880]A lot of people have done
work like that in the past.
- [00:13:47.790]And so we weren't really surprised,
and it wasn't really new information,
- [00:13:53.580]but then we wanted to know, well,
- [00:13:55.200]can we use NIR to estimate all
these other soil properties,
- [00:13:59.820]all this other soil health stuff
that we were measuring in my lab
- [00:14:04.260]and lo and behold,
- [00:14:06.240]everything that was in the
biological category did pretty well.
- [00:14:10.350]So the microbial biomeds,
carbon beta glucose today's,
- [00:14:14.910]but all that stuff at the top,
- [00:14:16.470]which were the chemical and nutrient
measurements performed pretty poorly.
- [00:14:21.210]And the physical measurements
down here in bottled water,
- [00:14:24.000]stable aggregates and water-filled
pore space did not do well.
- [00:14:27.600]That also is not very surprising.
- [00:14:30.060]And then you have to ask
yourself the question,
- [00:14:33.270]we know that organic carbon bonds are
producing these spectral signatures,
- [00:14:38.760]but in this case,
- [00:14:41.580]am I really able to capture the
beta glucose a days or the microbial
- [00:14:46.440]biomass content, or is this just a result?
- [00:14:51.110]The fact that these measurements are
highly correlated with each other.
- [00:14:55.520]So I'm really just
estimating them by proxy.
- [00:14:59.030]And I think it's just the proxy
measurement there, which is fine,
- [00:15:03.020]as long as you understand that.
- [00:15:05.450]So then we wanted to know what happens
then if we stick the sensors together and
- [00:15:10.070]sort of do a mashup, right, we've got
NIR, you know, it works for some things,
- [00:15:13.610]but not others. So how
can we improve on that?
- [00:15:16.580]So then we tried to sensor data
fusion. And in this case, again,
- [00:15:21.530]we're using NIR spectra,
- [00:15:23.840]but we collected that spectra in
the lab because of the issue with,
- [00:15:28.360]with water content that
I mentioned before.
- [00:15:30.950]And then we used infield NC
data and penetrometer data.
- [00:15:35.570]And with those which you
can see in the bottom,
- [00:15:37.340]we just carved off the data
from that upper surface soil
- [00:15:43.220]and matched it up with our lab data.
- [00:15:46.310]And in this case,
- [00:15:48.560]we also calculated these
soil health scores.
- [00:15:52.580]So I'm sure you've probably heard of the
soil management assessment framework.
- [00:15:56.480]So in this case, we took all
that lab data, we sport it.
- [00:15:59.960]And then we tried to estimate the lab
values of the actual indicator value plus
- [00:16:05.990]these scores.
- [00:16:07.460]And what we found again was that,
- [00:16:12.230]what did the best with
that biological category?
- [00:16:15.740]And we expected that because we
already knew we could do that,
- [00:16:18.650]do that fairly well.
- [00:16:21.530]But we did see a bump
in our performance on
- [00:16:26.510]the physical side.
- [00:16:27.620]So that added ISI information and
the penetrometer was giving us a
- [00:16:32.540]little more information related to the
soil, physical properties. Although,
- [00:16:36.650]you know, in this case, I would still
say, that's not very good. I mean,
- [00:16:39.500]that's an R squared of just barely
over 0.5. So it's an improvement,
- [00:16:43.970]but it's not really stellar. Now,
- [00:16:48.020]everything I've shown you so far was
looking at surface oil properties.
- [00:16:51.650]So in the top, maybe 15
centimeters or six inches,
- [00:16:56.630]but what about the soil
profile? Right? I mean,
- [00:16:58.400]we know that it's all profile
is really important. Most,
- [00:17:02.180]most soil health assessments
really focused on the surface soil,
- [00:17:05.750]but we know that the profile
can control a lot of things,
- [00:17:09.950]including the movement of water.
And, that relates a lot to,
- [00:17:15.200]crop productivity and everything else.
- [00:17:17.240]So how are we going to understand
what's happening below the surface
- [00:17:22.490]without having to go out there
and dig soil pits all the time.
- [00:17:26.630]And so that probe sensor that I
showed you a few minutes ago, we took,
- [00:17:30.430]took that out and we collected
easy data on penetrometer
- [00:17:35.270]data and added that back to that NIR
- [00:17:40.130]data, but we've chopped
the soil up into horizons.
- [00:17:44.590]So we had this full profile
information and we got pretty good
- [00:17:49.440]results, I would say for soil
texture. So sand, silt, and clay,
- [00:17:53.790]and then also, well, density did
better than I had expected there,
- [00:18:01.710]but the real test, this last, last
thing that I'm going to show you here,
- [00:18:05.610]which was all right. Ultimately,
- [00:18:08.160]we want to get all of
this data in the fields.
- [00:18:11.580]We want to collect
everything in the field.
- [00:18:13.740]Maybe we have a couple of calibration
samples that we send back to the lab,
- [00:18:17.400]but we want all these sensors
to work in field conditions.
- [00:18:21.420]So they would have to work with more
soil and we want that profile data.
- [00:18:27.420]So we went to 153 locations,
22 different fields.
- [00:18:31.110]In this theory in Indiana, we collected
data down to one meter using that probe.
- [00:18:35.400]I just showed you.
- [00:18:37.050]And then we collected a sample and
took it back to the lab and scanned it
- [00:18:41.850]there so that we could compare
our results. And we use this lab,
- [00:18:46.800]lab measurement or lettuce mation
as our, sort of our gold standard here.
- [00:18:53.220]And if you're curious, this
is, this is what those are.
- [00:18:57.180]The data looks like when we plot it,
after it comes off the P four thousands,
- [00:19:02.280]it's on the left, that's at sand and
clay data from that's from the lab,
- [00:19:07.320]but then we, we can take out
those spectr basically
turn it into a heat map,
- [00:19:11.940]which this was not my ideas.
And Scott Drummond's idea,
- [00:19:14.590]I thought was really cool.
- [00:19:16.020]So you can see the change in the
spectrum keeps based on the color here.
- [00:19:21.450]And then we have easy data and then
kind of traumas or data on yen.
- [00:19:25.110]And so the top of this thought
is the surface of the soil.
- [00:19:28.380]And then it shows you
the data as it goes down.
- [00:19:31.200]So this would be data from
one location in the field.
- [00:19:36.270]What we do this all over the
place and then put it together.
- [00:19:40.170]So this is a really ugly plot.
- [00:19:42.810]don't bother pulling out your
reading glasses, just to look at it.
- [00:19:45.810]I'll tell you the important things here.
- [00:19:48.570]So on the left for carbon, there's
a lot of scatter in that data.
- [00:19:52.320]It was a really messy plot.
This is what happens when,
- [00:19:55.890]when moisture is in the soil
and we're trying to use NIR
- [00:20:01.740]but people have come up with
some pretty fancy techniques now,
- [00:20:05.730]different transformations and
statistical techniques essentially
- [00:20:10.410]remove the effect of
moisture. And so on the right,
- [00:20:14.850]you can see the improvement
that we have there. in fact,
- [00:20:18.630]that that improvement was a lot
better than I thought it would be,
- [00:20:21.630]but I would still say there's room for
improvement before we get to the point
- [00:20:26.220]where we can really say this is working
well. And then for clay content, I mean,
- [00:20:30.900]that graph was really ugly before we
applied the EPO transformation and
- [00:20:35.850]did some other shenanigans
with the data. But,
- [00:20:39.120]so that's where that stands. Now.
- [00:20:42.360]The long term goal really is to put all
these sensors together so that we can
- [00:20:46.650]conduct what I call the
soil health assessment.
- [00:20:50.440]And it's just capturing these
different aspects of soil properties,
- [00:20:54.790]and it puts high spatial
and temporal resolution.
- [00:21:01.300]And if you look at the types of
sensors that I showed you today,
- [00:21:04.570]plus other information from
the literature, you know,
- [00:21:07.600]for the physical properties,
- [00:21:09.790]either conductivity or mechanical
resistance sensors work fairly well,
- [00:21:14.530]but there's still room for
improvement. The biological side,
- [00:21:17.680]the optical sensors are pretty good.
- [00:21:21.670]And then the chemical and nutrient
categories where we need the most work,
- [00:21:26.770]this, fabulous table,
it was put together by Dr.
- [00:21:29.950]Adam talk really summarizes where
we are now with using soil sensors
- [00:21:34.870]to estimate so properties. Some,
- [00:21:38.170]we can do a fairly good job of right
now. Most of them still need work.
- [00:21:42.670]And again, it's that nutrient.
- [00:21:45.730]And so all chemical
property category where,
- [00:21:50.140]we can do a lot better and I
know people are working on this.
- [00:21:53.440]So I'm hoping in the near future, we'll
have a lot of progress in this area.
- [00:21:59.530]So that's it.
- [00:22:02.400]Thank you.
Log in to post comments