Rapid Soil Sensing for Precision Agriculture and Carbon Market

Yufeng Ge, Professor, Advanced Sensing Systems Engineer, Deparment of Biological Systems Engineering, University of Nebraska Lincoln Author

05/24/2024 Added

13 Plays

Description

Dense, quantitative soil data are increasingly demanded for applications like precision agriculture, climate modeling, and carbon crediting. Our lab has been working on enabling tools and systems for rapid, low-cost measurement of soil properties. In this seminar, I will cover a few projects we have worked on, key results, and lessons learned.

Searchable Transcript

Search:

[00:00:00.750]The following presentation
[00:00:02.220]is part of the Agronomy
[00:00:03.570]and Horticulture seminar series at the University
[00:00:06.690]of Nebraska Lincoln.
[00:00:09.120]Okay, so let's get started with this part.
[00:00:11.370]If anyone wants to come in a little bit later,
[00:00:12.840]they can always miss part of my introduction
[00:00:14.130]and still have plenty of time for you folks talk.
[00:00:17.220]So it is my real pleasure this morning
[00:00:19.260]to be introducing Professor Yufeng Ge,
[00:00:22.080]the Eberhard Endowed Professor in the Department
[00:00:26.790]of Biological Systems Engineering here at UNL.
[00:00:29.550]In addition to being the director
[00:00:30.780]of the Plant Fenomics Facility,
[00:00:34.680]Dr. Ge is a professor in biological systems engineering.
[00:00:37.470]He's also affiliated with the Center
[00:00:39.360]for Plant Science Innovation.
[00:00:41.250]I have known Yufeng for about 10 years.
[00:00:44.970]And I just wanna note that, you know,
[00:00:46.140]so he got his PhD from Texas INM.
[00:00:47.820]He came up here as an assistant professor,
[00:00:49.650]and in less than 10 years, rose through the ranks
[00:00:52.380]to the rank of full professor.
[00:00:54.150]If you are a graduate student,
[00:00:55.110]you may not realize just how exceptional that is
[00:00:58.170]in that period, he
[00:00:59.460]or no, in his career to date,
[00:01:01.200]he has published more than a 100 peer reviewed papers.
[00:01:03.540]He holds multiple patents.
[00:01:05.220]He's had funding from a wide range of federal agencies,
[00:01:07.620]including the Department of Energy,
[00:01:09.120]US Department of Agriculture from commodity boards.
[00:01:12.483]And one of the striking things about Yufeng,
[00:01:15.120]so we are in an Agronomy Department,
[00:01:16.350]we will appreciate plants
[00:01:17.790]and soil are very different things.
[00:01:20.250]Yufeng is a very successful researcher,
[00:01:22.920]both in studying plants,
[00:01:24.630]but also studying soils,
[00:01:26.250]which is what he's gonna be telling us about today.
[00:01:28.710]So with that, I'm looking very much forward to this
[00:01:30.780]'cause I know very little about soils.
[00:01:33.240]Yufeng, take it away.
[00:01:34.170]Okay, all right.
[00:01:35.130]So thank you so much, James for the introduction
[00:01:38.700]and can everybody hear me okay?
[00:01:42.270]All right, and so, online folks is alright.
[00:01:45.780]Okay, so like James said,
[00:01:50.623]I started 2014 as assistant professor,
[00:01:54.030]and I think I was joined this seminar
[00:01:57.030]in 2017, maybe each time.
[00:02:00.450]And then that back in the time they were talking about
[00:02:03.237]the research for this seminar series, right?
[00:02:05.460]Then after that,
[00:02:06.630]I really want to come back,
[00:02:07.830]and every beginning of the semester was me, me, me right?
[00:02:10.677]But they never invited me again until this time.
[00:02:13.980]So, yes.
[00:02:15.540]So for next time, you are lacking a name,
[00:02:17.670]you know who to reach out.
[00:02:19.209](Ge laughs)
[00:02:20.042]But, so I am going to talk about some of the,
[00:02:25.024]one of the research branch that I always try
[00:02:27.060]to keep in my research group, right?
[00:02:28.860]So, you know, if you look at my job description,
[00:02:32.370]so soil isn't really it isn't emphasized there,
[00:02:36.060]but I carry on from my PhD study and I love it.
[00:02:40.050]So, therefore that's what I'm,
[00:02:42.270]and it's very closely relates to
[00:02:45.360]precision agricultural, site specific management,
[00:02:48.600]and digital agricultural to this day.
[00:02:53.280]Some of it is also relates to carbon sequestration
[00:02:56.880]and MRV of the soil, carbon in the soil.
[00:03:00.770]So that's what I'm going to talk about.
[00:03:03.124]So managing soil, I'm gonna,
[00:03:08.740]let's see, I'm gonna stand here, okay?
[00:03:10.920]So that I can look at the screen.
[00:03:14.520]So managing soil variability, the essential part
[00:03:17.490]of the precision agriculture, right?
[00:03:19.260]So, when we talk about the precision agriculture,
[00:03:21.900]you want to know the infield of variability.
[00:03:24.690]Now, there are different ways that you can know this.
[00:03:27.870]You can always measure plants directly
[00:03:30.210]through remote sensing, right?
[00:03:32.010]But, we always know that soil is
[00:03:36.660]a significant part of that
[00:03:38.130]because, you know, plants get a lot from the soil,
[00:03:41.457]and soil in agricultural fields is highly variable, right?
[00:03:44.940]And then, so at the field scale,
[00:03:47.670]so certain soil properties can be yield limiting,
[00:03:51.150]and that can be your pH,
[00:03:53.460]can be your organic matter, right?
[00:03:55.050]Can be your, you know, nitrogen availability
[00:03:57.600]or maybe water driven because of the topography of the field
[00:04:01.230]and all those, right?
[00:04:02.850]So soil based mapping
[00:04:04.710]and the management decision is common, right?
[00:04:06.990]So, you know, this trace back to a couple of
[00:04:10.770]even decades ago that there's a concept about
[00:04:15.725]the management zone development and the delineation
[00:04:19.230]and the concept is that, you know,
[00:04:20.760]you do not manage the field as a whole piece.
[00:04:24.150]You always wanted to divide them up into small pieces.
[00:04:27.870]And within those pieces, you know, it's more uniform,
[00:04:30.810]and then you can have a certain type
[00:04:32.670]of management practices tied to it, right?
[00:04:35.610]So some, the management is on development,
[00:04:38.610]and then after that you have variable rate application.
[00:04:42.150]So essentially, when you have
[00:04:43.451]the sprayers or the weeding device going
[00:04:47.720]through the field, and you know as you determine
[00:04:50.670]where you are in those management zones,
[00:04:53.550]you can induce a different management practice, you know,
[00:04:56.580]be it, different pesticide rating
[00:05:00.690]or different gross regulator rating and all those.
[00:05:04.290]So this has been, you know,
[00:05:05.826]it's not new, right?
[00:05:07.050]So it has been precision agricultural for a long time.
[00:05:10.410]So, but soil is a important part of it.
[00:05:15.000]So in order for you to understand
[00:05:18.450]what is a soil variability
[00:05:19.920]and what is a variability
[00:05:21.180]that cause the yield limiting factor, so you need to do go
[00:05:24.870]and sample soils, right?
[00:05:26.310]So and precision Ag,
[00:05:28.530]commonly our recommendation is one sample per acre.
[00:05:32.010]Now, you know in the area in Nebraska,
[00:05:34.440]and you can think about a pivot is,
[00:05:36.960]I don't know, 160 acres, right?
[00:05:39.420]Then, well, that's not really just economically feasible
[00:05:42.270]for you to go in to collect, walkthrough the field
[00:05:45.120]and collect all those samples.
[00:05:46.230]And then you have to bring in
[00:05:47.760]and do the ana do the processing in the lab
[00:05:51.900]and understand those two tool analysis.
[00:05:55.230]And after that, you will able to kind of produce a map
[00:05:58.980]that is, shows the spatial variability of the soil property.
[00:06:01.830]And that is just a very, very tedious process.
[00:06:06.420]So, and there are growing demands of high quality,
[00:06:09.600]high density soil data, right?
[00:06:11.490]So we call bigger soil data.
[00:06:13.140]And that turns out to be really used for a lot
[00:06:15.510]of things these days, you know,
[00:06:17.430]including like climate modeling, soil carbon modeling,
[00:06:21.480]and digital soil mapping,
[00:06:24.390]precision agriculture, right?
[00:06:25.710]So, you know, those are the, the things
[00:06:27.300]that would be directly impacted
[00:06:30.570]by the availability of the soil data.
[00:06:33.090]Now, throwing a couple of pictures there, right?
[00:06:35.280]So, you know, just to kind of show you the variability,
[00:06:37.740]and this is a piece of field that I worked for my PhD.
[00:06:41.760]So I worked in this field for three years.
[00:06:43.680]It's a cotton field, right?
[00:06:44.820]So it's a standard pivoted down in Texas,
[00:06:47.520]and you can see, right?
[00:06:48.900]So on the left is a remote sensing image
[00:06:51.930]of the bio soil, right?
[00:06:52.820]And you can see where the wet part, where the dry part,
[00:06:56.430]and then in the middle is
[00:06:58.020]where we drive the electrical connectivity survey
[00:07:01.115]through the field.
[00:07:03.810]And then now you can see the matching pattern of that,
[00:07:07.500]you know, what's being sensed
[00:07:09.870]by the electrical connectivity
[00:07:11.640]and what's the water, the wet spot in the field, right?
[00:07:14.550]So the padding matching pretty well.
[00:07:17.190]And then on the right is the yield of cotton
[00:07:19.860]for that particular year, right?
[00:07:21.720]And you can see the wet field,
[00:07:23.940]wet part of the field leads to a low yield of fiber.
[00:07:29.670]And so this is a problem with this particular field,
[00:07:34.860]is it gets too much water in the center
[00:07:37.230]of the field that leads to a water logging,
[00:07:39.990]and cotton does not like that.
[00:07:41.880]So it gets to the lower yield.
[00:07:44.040]So this kind of show you the usefulness
[00:07:48.510]of this kind of technology remote sensing
[00:07:51.030]and the proximal soil sensing for interrogate
[00:07:54.450]what is really the year, the limiting factor
[00:07:56.490]for the particular case,
[00:07:58.020]and how you gonna hopefully kind of develop
[00:08:00.960]a management decisions to remediate that, right?
[00:08:04.560]And then I also, when I go through the older files,
[00:08:08.250]I find a picture of myself that
[00:08:09.990]that's the field that I'm working in.
[00:08:12.060]So this is about,
[00:08:13.230]if anybody can guess how many years this is
[00:08:16.470]just looking at the,
[00:08:17.790]using your machine learning brain to figure out.
[00:08:22.980]so this is 2008.
[00:08:25.500]So this is my final years of PhD.
[00:08:29.010]Yeah, okay.
[00:08:31.500]So now, like I said,
[00:08:34.920]so soil sensing is really tedious, right?
[00:08:37.050]So, and hopefully you don't need to do that.
[00:08:39.240]And there's a lot of ways to rapidly gather
[00:08:41.910]that information from the soil, froms the field, right?
[00:08:44.749]So the soil sensing the mapping technologies,
[00:08:49.170]so remote sensing and the imaging of bare soil
[00:08:51.780]is a certainly one way, right?
[00:08:53.460]Now, when I moved to Nebraska,
[00:08:55.110]that's the quick lesson that I learned is,
[00:08:57.600]you know, there's a lot of no-till here, unlike Texas.
[00:09:00.960]So in Texas, getting a bare soil image is really easy.
[00:09:04.380]You go there and you will get it it.
[00:09:05.730]In Nebraska, it's really hard.
[00:09:07.920]And that's a part of the reason, like, you know,
[00:09:09.273]that this line of research in my group's gonna take some
[00:09:12.630]time to go, is I just, you know,
[00:09:15.540]every go everywhere they say, well, there's a residual,
[00:09:17.730]so therefore you cannot see the bare soil.
[00:09:20.100]So therefore you can't assess the variability
[00:09:23.250]in that field, right?
[00:09:24.630]So, and then there is EMI,
[00:09:27.000]so that's electromagnetic induction,
[00:09:28.860]and then the ER survey, electrical resistivity.
[00:09:32.280]So that's the second one right here.
[00:09:35.280]So the second one right here, right?
[00:09:37.433]You can see, so that's the sensor
[00:09:39.530]of the EMI sensor.
[00:09:41.850]So they put it under in a bus tab tub, right?
[00:09:44.010]And then pulled by a truck.
[00:09:45.810]And the reason for this is
[00:09:47.190]because the EMI is very sensitive to metal pieces,
[00:09:50.880]so there can to be any metal pieces
[00:09:52.500]around it, like three meters up in the proximity.
[00:09:55.680]So what you do is you have have this long,
[00:09:57.960]long rope to pull it
[00:09:59.550]and this thing so that the carrier cannot be made of metal,
[00:10:04.080]so it needs to be wooden pieces or plastics, right?
[00:10:07.770]So this is that.
[00:10:09.300]And then there is a gamma ray survey.
[00:10:11.310]Essentially there is a radioactive components
[00:10:13.830]in the soil, right?
[00:10:15.000]And then those can be picked up by the gamma ray sensors.
[00:10:19.380]And those tells you a lot of about the parent material
[00:10:22.710]and you know, where the soil developed from.
[00:10:24.990]And then use that,
[00:10:25.993]you can sort of develop models to look at,
[00:10:30.180]you know, the things like the clay fraction
[00:10:32.460]or the sand fraction
[00:10:33.540]or the water holding capacity and those, right?
[00:10:35.880]Again, so that's not a direct measurement,
[00:10:37.860]that's kind of a proxy sensing.
[00:10:41.730]And then you have to rely on model
[00:10:44.127]and many of those models site specific
[00:10:47.850]for that mapping purpose.
[00:10:50.010]And then there is a company called
[00:10:51.815]the Veris Technologies, right?
[00:10:52.920]So I love that company.
[00:10:54.420]So they're not very far away from us.
[00:10:56.190]So it's a small business company in Kansas,
[00:11:01.740]and they develop a bunch of technologies, right?
[00:11:04.380]And, you know, really unique about
[00:11:06.270]that company is the technology they develop
[00:11:09.360]is also engaging, right?
[00:11:11.100]So it's not like remote sensing,
[00:11:12.900]you fly or like you know this,
[00:11:16.140]that you're standing the signals penetrating
[00:11:18.060]through the soil,
[00:11:19.230]but they are developing a bunch of technologies right here
[00:11:23.130]that it's actually touching the soil
[00:11:25.590]while taking the measurement.
[00:11:26.970]And I think that's outstanding.
[00:11:28.740]And I follow their development,
[00:11:30.373]kind of quite long over the decades.
[00:11:33.990]I think this is the pal of
[00:11:37.332]the, what is it?
[00:11:39.540]The EM sensors that, you know,
[00:11:41.310]to measure the electrical connectivity.
[00:11:43.470]And I think this is a contact type pH sensor
[00:11:46.443]that you make a soil mix in the field, right?
[00:11:50.040]They use robotic waste
[00:11:51.510]and the dips, the sensor to it for pH measurement.
[00:11:54.390]And they also develop a kind of sensors
[00:11:56.520]that's integrated to the culture that as you go
[00:11:58.950]through it can measure the signal.
[00:12:00.870]Definitely love that.
[00:12:02.550]So those are the technologies that we have
[00:12:05.040]to quickly get a soil variability information
[00:12:07.560]from the field, right?
[00:12:09.450]So, but the technology
[00:12:11.040]that I really work on is the spectroscopy, right?
[00:12:14.670]So is, you gather the soil from the field
[00:12:20.130]and you process them in the lab
[00:12:21.792]to grind them, to passing them two millimeters,
[00:12:25.980]seething them.
[00:12:26.970]And then you have this dry soil that's, you know,
[00:12:30.450]sort essentially doesn't have its innate aggregates,
[00:12:34.140]the aggregates state anymore,
[00:12:36.420]but you can put them in a scanner
[00:12:41.040]to measure its spectral data
[00:12:43.200]and then you can rely on modeling
[00:12:47.280]to quantitatively infer
[00:12:49.830]what is those soil properties, right?
[00:12:51.780]Because, you know, there's soil mainly is a mineral, right?
[00:12:55.020]And then there's organic anywhere from a 1% to 5%
[00:12:58.890]organic stuff in it.
[00:13:00.990]And then many of those components have an active
[00:13:07.200]spectral signature in those wave bands.
[00:13:09.960]And then you can utilizing those information to infer
[00:13:13.410]what is the soil property.
[00:13:16.710]And it turns out for some of those soil properties,
[00:13:19.650]it's really accurate.
[00:13:21.450]Things like organic matter, right?
[00:13:23.370]And clay content in the mineral fraction.
[00:13:27.720]And iron oxides is one type of mineral that
[00:13:33.030]it causes the color change and the carbonate, right?
[00:13:36.000]So inorganic carbon, calcium carbonate,
[00:13:40.320]those kind, they can be estimated very, very accurately.
[00:13:45.270]And then there's other soil properties, right?
[00:13:47.820]You know, they are,
[00:13:48.960]what do we call the secondary soil properties?
[00:13:51.300]And you see them in the literature that they can be modeled
[00:13:55.230]with various degrees of accuracy with the spectroscopy,
[00:13:59.820]things like a pH, right?
[00:14:01.230]So nutrients, MPK essentially head on exchange
[00:14:05.760]and then salinity, right?
[00:14:07.020]So, and that this is almost like you are using spectroscopy
[00:14:12.930]as a pedal transfer function, right?
[00:14:14.640]So, you don't really have the first principle
[00:14:17.820]to measure those properties,
[00:14:19.530]but those properties are correlated with,
[00:14:21.720]you know the secondary properties are correlated
[00:14:24.180]with the first primary property.
[00:14:26.100]You can use a spectroscopy to predict the primary property.
[00:14:28.770]So there's a secondary correlation goes in there
[00:14:31.380]that allows you for a quick rapid detection, right?
[00:14:35.070]So a nice thing about this, you know,
[00:14:37.197]the whole spectroscopy thing in the lab is that it is all,
[00:14:41.370]it is actually, you know,
[00:14:42.690]all those wavelengths are aligned quite well
[00:14:45.330]with the satellite remote sensing system.
[00:14:47.460]So, you know, the satellites, the optical imaging sensors
[00:14:50.670]are flying in the sky.
[00:14:52.320]So they operate the same band as many of
[00:14:55.320]the lab equipment that we are using.
[00:14:57.870]So that you can see, right?
[00:14:59.190]You know, in the lab you are measuring dry ground soil,
[00:15:03.600]but remote sensing, look if you are lucky,
[00:15:06.060]get the bare soil image.
[00:15:07.620]You got a surface level aggregated
[00:15:10.500]with a different level of moisture and the residuals
[00:15:14.880]that you can measure.
[00:15:16.260]And it seems like there's a possibility that, you know,
[00:15:18.450]you can make the connections between the two so that,
[00:15:20.587]you know, what you do in the lab can be translated to
[00:15:23.850]what happens in the field.
[00:15:25.710]Okay, and so this is the vision ion
[00:15:29.760]machine that we use, right?
[00:15:30.821]You can see it's really simple.
[00:15:32.640]You've got the dry ground soil,
[00:15:34.200]you put them in the Petri dish,
[00:15:35.400]and then you measure them optically three seconds,
[00:15:37.801]you get the spectral, right?
[00:15:39.240]And then it really relies on the model to tell you
[00:15:42.210]what it is, and you have to have a model to do that.
[00:15:46.500]And so this is the mid-infrared machine that's available.
[00:15:53.340]And right here that's the EM spectrum
[00:15:56.520]that I usually kind of when I teach
[00:15:58.200]the undergraduate students,
[00:15:59.460]in which I'm not gonna be able this here.
[00:16:04.260]So that's the spectroscopy, right?
[00:16:06.180]You know, just to kind of give you an idea
[00:16:07.680]how accurate that can be.
[00:16:09.510]Okay, so on the top right here.
[00:16:11.850]So that's what you see, you know,
[00:16:13.680]that's the vision IR spectrum, okay?
[00:16:16.980]And that's going from visible range to the near infrared,
[00:16:20.880]and this is the nanometer.
[00:16:23.760]That's the unit.
[00:16:25.110]And so we using that,
[00:16:27.090]you can see, you can build the models to predict
[00:16:30.810]organic carbon, okay?
[00:16:32.160]So this organic carbon range is from zero to 5%
[00:16:35.422]for this particular set
[00:16:37.230]and the clay content from like 5% to roughly 65,
[00:16:41.040]70%, right?
[00:16:43.500]And then the pH value from five
[00:16:45.900]to nine, eight or nine.
[00:16:49.050]And so this is the accuracy that we can get
[00:16:52.050]for the vision IR, okay?
[00:16:54.510]And moving down,
[00:16:55.587]and that's the MIR, so that's middle infrared.
[00:16:59.040]And you know, the same thing,
[00:17:00.870]it's just operates at a different wavelengths.
[00:17:03.420]You scan the soil pretty much the same way
[00:17:06.090]and it's even more accurate, right?
[00:17:08.190]And so you can see, you can almost pinpoint
[00:17:10.380]where the organic carbon is,
[00:17:12.769]where is the clay and where's the pH?
[00:17:15.150]And you know, this is a kind of really,
[00:17:18.360]there's a interest to develop
[00:17:19.713]this kind of technologies, right?
[00:17:21.570]Because of the accuracy for things like carbon and clay,
[00:17:26.070]that you can develop technologies for precision agriculture
[00:17:29.280]and you can develop technologies
[00:17:31.710]for carbon trading, right?
[00:17:33.000]So the carbon market, so.
[00:17:38.940]Alright, so spectroscopy and machine learning, right?
[00:17:42.360]So correlations between spectra
[00:17:46.650]and soil properties are established
[00:17:49.560]via empirical modeling, right?
[00:17:51.480]So, that's the key for this whole research
[00:17:56.070]is you have to have a model.
[00:17:57.960]And where a model is from,
[00:17:59.280]well model is coming from your training set, right?
[00:18:02.040]So you have to have a set that you measure the spectra
[00:18:06.060]but you also have the lab ground tooth measurement,
[00:18:09.000]and you match the two and you figure out
[00:18:10.710]what's the quantitative relationship between the two.
[00:18:13.290]And that's your model.
[00:18:14.123]Once you have the model,
[00:18:15.240]you can always apply that model
[00:18:16.650]to the new sample you have, right?
[00:18:19.710]So, and if you go to the literature,
[00:18:21.900]there's just the explosive amount of information
[00:18:24.810]out there about machine learning, right?
[00:18:26.580]It's not just in soil spectroscopy,
[00:18:28.980]it's in everywhere, right?
[00:18:30.060]So, and they are proliferating, right?
[00:18:34.260]And so partially is the squares
[00:18:36.390]and you know, that's really the standards that we are using.
[00:18:40.020]But you know we also see things like
[00:18:42.270]the supportive vector regression, right?
[00:18:45.660]Random forest,
[00:18:47.010]they are slightly different in their statistical principle.
[00:18:50.220]And the neural networks has really take off recent years
[00:18:54.330]and because of the spectral data, so
[00:18:56.610]that a particular data format.
[00:18:58.590]So there is other more, even more recent modeling approach
[00:19:02.850]like 1D-CNN, right?
[00:19:04.280]So it's 1D, one dimensional convolutional neural network
[00:19:08.340]and a recurrent neural network.
[00:19:09.870]It's almost like a time sequence data,
[00:19:11.610]but this time it's a spectra data,
[00:19:13.890]it's a 1D it's slightly different
[00:19:15.660]form your image data which is 2D.
[00:19:18.180]So now machine learning models are useful,
[00:19:22.350]they gets bigger and less interpretable, right?
[00:19:25.740]And sometimes it's heavily criticized.
[00:19:27.870]So I have one slides to talk about that.
[00:19:35.460]So, right, so who are the supporters
[00:19:37.500]and who are the opponents, right?
[00:19:39.000]So, as you can see,
[00:19:41.700]it is substantially reduced the cost
[00:19:43.740]and time of soil analysis.
[00:19:45.420]There's no question about it.
[00:19:46.530]So, if you set up the system, right?
[00:19:49.650]And you know, if you say,
[00:19:50.820]I'm gonna have a soil lab that is just dedicated
[00:19:55.620]to surface let's say a region in Nebraska, right?
[00:19:59.580]And then, and your model really is
[00:20:01.950]developed in a way that fits the soil in this region.
[00:20:06.120]And then you can do this really cheaply, right?
[00:20:07.530]And then you can you can probably do,
[00:20:13.560]you know, like five, $10 a sample, right?
[00:20:16.890]So, and it can give you a lot of things.
[00:20:18.840]Not just one soil property,
[00:20:20.910]it can give you a bunch of them, right?
[00:20:22.950]So, and it's a case that you can see,
[00:20:25.680]this whole business can go virtual
[00:20:27.870]or network of distributed spectroscopy-based soil
[00:20:31.830]analytical labs, right?
[00:20:34.170]So that's really powerful
[00:20:35.820]because, you know, if you look at
[00:20:37.380]where the science go, right?
[00:20:38.910]Doing a lot of like, you know,
[00:20:40.560]coordinated large scale research
[00:20:43.050]and you know, I do research here in Nebraska,
[00:20:45.960]someone else is, the other part of the project
[00:20:48.420]is doing it in Europe or in Africa, right?
[00:20:51.390]And they have their own local soil labs.
[00:20:55.020]But if when you down the analysis
[00:20:57.570]and when you're trying to pull those data back on,
[00:21:00.330]so one missing piece that we don't know is
[00:21:02.930]if the World lab here do the analysis,
[00:21:05.910]is there systematically different from
[00:21:08.040]what they do in Africa in Europe?
[00:21:10.830]And there's not so much comparison
[00:21:13.770]or the analysis in that regard, right?
[00:21:15.870]So we don't know, essentially they may turns out to be very,
[00:21:18.810]very consistent, but they may not be right?
[00:21:21.450]So, but this virtual type of soil analysis
[00:21:25.680]is gonna alleviate that problem
[00:21:28.890]because it becomes,
[00:21:29.790]or the model becomes shared
[00:21:30.998]because it's not something that you do locally.
[00:21:34.530]The model is transparent to everybody,
[00:21:37.080]to all the participating labs.
[00:21:39.690]And you can see it real time
[00:21:42.030]and if everybody using the same model real time,
[00:21:45.390]and then you are guaranteed
[00:21:46.530]to give you the same same reading.
[00:21:48.570]So you can remove a lot of
[00:21:50.250]those inter laboratory variance, right?
[00:21:55.950]In your dataset,
[00:21:56.970]and this is particularly important for larger scale study.
[00:22:00.090]You have a collaboration with somebody in Europe
[00:22:02.430]and you are studying carbon sequestration
[00:22:04.170]at a global scale, right?
[00:22:05.820]So if your data is biased,
[00:22:08.910]your team's data is biased,
[00:22:10.470]then it can get really problematic, right?
[00:22:15.030]So and so those are the supporting arguments, right?
[00:22:19.080]For a virtual soil laboratory.
[00:22:22.590]But the opponents really is really,
[00:22:24.900]soil spectrum models are not based on the first principle.
[00:22:27.850]As I said, many of the soil properties are not
[00:22:31.710]and accuracy therefore cannot be insured.
[00:22:34.500]So model, essentially model is a soil dependent.
[00:22:37.770]And the model is location dependent.
[00:22:40.260]A model developed in Nebraska cannot be transferred
[00:22:44.640]to somebody in Europe or you know, Africa, right?
[00:22:49.470]And so therefore the cost benefit of the approach
[00:22:52.530]probably is harder to materialize, right?
[00:22:55.110]So, but those are sort of the competing
[00:22:57.780]arguments about the approach, right?
[00:23:00.660]But regardless, so I plug in picture here,
[00:23:04.080]so, FAO really likes the idea.
[00:23:07.230]They sort of kind of trying to push it in that way,
[00:23:09.390]particularly for some
[00:23:10.950]of the precision agricultural practices in the
[00:23:14.730]developing countries, right?
[00:23:16.200]So therefore they want us to,
[00:23:17.610]so that's the sort of like the part of the training efforts
[00:23:20.040]that FAO are trying to organize is,
[00:23:23.137]let's develop some training materials to
[00:23:28.020]get those out to the developing countries
[00:23:30.210]so that they know there are cheaper ways
[00:23:32.550]for the soil analysis
[00:23:33.930]and you can do this for the fertility management,
[00:23:37.050]for better knowing your soil resources and all those.
[00:23:39.930]So that's I'm very proud of that
[00:23:42.180]being able to plug in that efforts
[00:23:44.580]and I know that this is the first one
[00:23:47.580]and we're working on the second one
[00:23:49.350]that is more focused on sort of like machine learning stuff.
[00:23:54.720]All right, so now we said that,
[00:23:59.572]the study is really kind of very lab focused
[00:24:02.340]and you bring the soil in,
[00:24:04.440]you do the scanning and do the model development, right?
[00:24:08.490]But one of the things that we trying to do
[00:24:11.130]is to cooperate with the USDA-NRCS.
[00:24:16.867]So USDA-NRCS has its international soil survey
[00:24:22.170]headquarter here in Lincoln, Nebraska.
[00:24:24.510]And with that soil survey division,
[00:24:27.150]they have a national soil analysis laboratory.
[00:24:30.660]So many of you probably know it's KSSL,
[00:24:33.240]and you know, they have over 30,000
[00:24:36.739]of physical samples being archived
[00:24:39.210]in their warehouse, right?
[00:24:40.680]So that's in the federal building.
[00:24:42.180]They recently moved,
[00:24:43.860]I think it's in the process.
[00:24:45.720]But you look at all the physical samples they have, right?
[00:24:49.350]And we did a tally recently,
[00:24:51.600]so this go all the way back to 1950, right?
[00:24:56.490]And that's that's a tremendous amount of sample
[00:25:00.480]and also a tremendous assets for a lot
[00:25:02.520]of scientific questions, right?
[00:25:04.020]So if you wanted to, for example,
[00:25:05.520]if you want to study microplastic contamination,
[00:25:08.460]if you want to see is it the same
[00:25:10.410]50 years ago versus now,
[00:25:12.030]we go and find the 1950s
[00:25:14.190]and do the analysis using the modern approach, right?
[00:25:17.130]And then you may be able to pull out
[00:25:19.530]a pretty good paper out of it,
[00:25:21.630]but you know, that's a side.
[00:25:24.480]But so we are trying to do
[00:25:26.880]is to collaborate with them
[00:25:28.980]to get all those samples scanned, right?
[00:25:31.893]With the vision IR, with the MIR,
[00:25:34.290]and then we can have this huge data set
[00:25:36.360]that covers the entire United States.
[00:25:39.780]And among this 300,000,
[00:25:41.400]we also have some foreign samples.
[00:25:45.000]And so really the idea is,
[00:25:48.900]can we use this larger scale soil spectral library
[00:25:54.150]to support many different applications, right?
[00:25:57.090]The main drive of it is soil health,
[00:26:00.960]soil carbon sequestration.
[00:26:02.400]Because, there are some studies about spectroscopy
[00:26:05.700]can also infer some of the soil health properties,
[00:26:09.810]Again, a transfer function type of concept, right?
[00:26:13.410]And then digital soil mapping
[00:26:15.360]and soil crop and the climate modeling.
[00:26:18.120]But that also kind of brings the challenge, right?
[00:26:20.460]Because you know, the question really is right,
[00:26:22.853]are you gonna develop a super large model,
[00:26:25.350]like a larger language model
[00:26:26.880]and then you can apply that to
[00:26:28.837]thousands of different applications?
[00:26:31.380]Or do we need to really just if the application
[00:26:34.470]is about Nebraska,
[00:26:35.790]do we just kind of develop a local model
[00:26:37.650]from the Nebraska soils,
[00:26:39.660]hopefully that you're gonna have a better accuracy
[00:26:41.910]and a better inference power in that regard, right?
[00:26:45.930]So as you can see,
[00:26:47.190]that's the distribution of the soil samples
[00:26:49.380]and each dot is a panel
[00:26:51.060]and then got multiple samples come out of it.
[00:26:54.390]And then this is a modeling of,
[00:26:56.760]and this paper is actually a couple of years ago,
[00:27:00.750]so we can do so much better now.
[00:27:02.400]And this is a national scale data,
[00:27:03.870]and you can see the carbon.
[00:27:05.250]So organic carbon of 60%,
[00:27:08.100]that is not really like mineral soil.
[00:27:09.990]You're getting a like organic stuff, right?
[00:27:13.530]So, but you can see the range
[00:27:15.540]and we're putting together a large model,
[00:27:17.730]you know, it predicts pretty well
[00:27:21.630]from the totality perspective, okay?
[00:27:26.550]So this is the one thing that we do is
[00:27:28.050]to go geographically wider, right?
[00:27:29.980]So that we can get the spectroscopy national wide.
[00:27:35.040]And then the other thing that my lab do is go deeper, right?
[00:27:38.640]Because I'm an engineer,
[00:27:40.050]I like to develop things.
[00:27:41.670]So every time that I have a problem,
[00:27:45.540]I would like to look at the solutions of the hardware.
[00:27:48.240]Can I develop some instrumentation for it
[00:27:50.460]or can I program for it, right?
[00:27:52.560]So this is what we do
[00:27:53.700]because all the soil mapping technology
[00:27:58.980]that I introduced earlier,
[00:28:02.520]it only measures the surface soil, right?
[00:28:05.167]You know, like EM it does penetrates,
[00:28:08.250]but it penetrates to like one meter,
[00:28:10.140]but it gives you like an integrated measurement, right?
[00:28:13.380]You got one reading from that measurement.
[00:28:16.950]See, that does not give you
[00:28:17.970]layering information of the soil.
[00:28:19.800]So what we are trying to do is,
[00:28:21.870]because this sort of technology can be used in the field,
[00:28:26.820]you can move them out of the lab
[00:28:28.680]and you can get them smaller, right?
[00:28:31.560]Can we develop a penetrometer
[00:28:34.770]that integrated that sensing mechanism
[00:28:37.440]and then then pushed to the soil to see what happens, right?
[00:28:39.900]You can measure the reflectance out of the wall
[00:28:45.950]as the penetrometer goes deep, right?
[00:28:48.540]So this really is a very, very long journey, right?
[00:28:51.570]Starts in 2011 when I was at Texas ANM
[00:28:55.020]and unfortunately, right?
[00:28:56.130]So our patent is still with Texas ANM, okay?
[00:28:58.857]And somebody 10 years later gonna pick it up.
[00:29:03.480]So, 2021, right?
[00:29:05.430]So that's really kind of the concept.
[00:29:07.110]And 2023 gets the better.
[00:29:09.540]And 2025 is when I came to UNL our second year,
[00:29:12.750]I decided well, we should start to redo this, right?
[00:29:15.330]So that's the one that we developed
[00:29:17.730]and then that's when we testing
[00:29:19.790]in the field 2017, 2018,
[00:29:23.220]and this, it's quite a complex engineering system.
[00:29:25.990]We're trying to put a lot
[00:29:27.180]of things in it, like loader cells.
[00:29:28.962]The purpose of a loader cell
[00:29:31.020]is it gives you some sort of penetration push
[00:29:33.825]so that my hope, our hope is
[00:29:36.647]to get the bulk density out of it, right?
[00:29:38.910]So, you know as you go down
[00:29:40.620]and I touch that part a little bit later on, right?
[00:29:44.700]And then I put it 202X
[00:29:47.075]and that's our idea, right?
[00:29:49.350]So you have a handheld device,
[00:29:51.210]you go out and then get all the instrumentation
[00:29:54.267]and the AI and the cloud computing in the back.
[00:29:57.000]And as you push, you go to the location,
[00:29:59.430]there's no need to dig the soil samples out.
[00:30:02.190]There's no need to call the samples
[00:30:04.410]and bring them to the lab for the analysis.
[00:30:06.810]You know, all the models are gonna,
[00:30:07.950]you take the measurement,
[00:30:09.300]models are gonna do the prediction for you
[00:30:11.520]right there, right?
[00:30:12.390]So that's the idea.
[00:30:16.620]So we wanted to go deep for the deeper soil analysis
[00:30:20.790]to give you the layered information
[00:30:23.250]regarding soil property at the sampling location.
[00:30:26.700]So here are some kinda sort of,
[00:30:30.570]we do some tests of the system.
[00:30:32.340]We do have a prototype, right?
[00:30:34.350]You know, you can see the bulk density
[00:30:36.720]and the dots, then you section the soil
[00:30:39.600]when you pull the core out,
[00:30:41.820]and then you take the lab measurement
[00:30:43.860]and then the red curve is what we have
[00:30:46.583]with the VisNIR penetrometer system,
[00:30:48.570]because it is continuous,
[00:30:50.160]it's quite high in resolution as you go down, right?
[00:30:53.490]You can see it's not consistent
[00:30:57.570]with the lab measurement a whole lot.
[00:30:59.370]However, though what I'm happy to see
[00:31:02.250]is it does capture the trend.
[00:31:04.260]And as you can see, the bulk density goes high
[00:31:07.590]and the clay obviously you can see this is a surface layer,
[00:31:11.520]and then when you hit the B horizon,
[00:31:13.590]all of a sudden you have increasing
[00:31:15.390]in the clay content, right?
[00:31:17.100]And then the carbon is always kind of this nice
[00:31:21.060]going down as the depth is going up.
[00:31:27.780]So that's that.
[00:31:28.920]And then, so one of the challenge that we have
[00:31:32.070]as we develop it is really the soil moisture content
[00:31:35.610]and the aggregation state, right?
[00:31:37.410]And like I said, our model,
[00:31:39.420]the model we have is based on the dry ground
[00:31:42.060]soils in the lab.
[00:31:43.710]And when we get the sensor out to the field,
[00:31:47.310]we are measuring wet soils in its natural aggregated states,
[00:31:51.930]and we have to deal with it, right?
[00:31:54.210]So now, you know, here's another side, right?
[00:31:57.090]So if you go out to talk to somebody
[00:31:59.310]and anybody tell you they get a soil sensor
[00:32:01.620]that can measure soil properties right there in the field,
[00:32:04.620]you know, you probably need to have a second thought
[00:32:06.690]because, you know, they are getting
[00:32:08.730]the soil sensor measurement with something.
[00:32:11.610]So the ground truth is the dry ground soil
[00:32:15.180]measured in the lab, right?
[00:32:16.680]How are you going to deal with that very moisture content
[00:32:19.470]because the water is going to
[00:32:21.900]throw off your sensor readings, right?
[00:32:24.480]And soil moisture is very all the time,
[00:32:30.000]and that's the problem.
[00:32:31.470]You know, it's unpredictable.
[00:32:32.850]When you go today,
[00:32:34.350]it's gonna be at one soil moisture content,
[00:32:36.690]and you go the other time
[00:32:38.460]it's gonna be different soil moisture content, right?
[00:32:41.910]So, you know here are some examples, right?
[00:32:44.550]When we scan them in the field, right?
[00:32:46.800]You can see here show two samples.
[00:32:48.990]So the solid line is one sample,
[00:32:51.690]dash lines the other sample.
[00:32:53.430]But those different spectra
[00:32:55.350]is at the different soil moisture content, right?
[00:32:58.148]So you can see they change quite a bit
[00:33:01.350]as the soil gets drier.
[00:33:03.420]And so our challenge is really kind of
[00:33:06.810]very different from others,
[00:33:08.370]is others is they're trying to develop sensors
[00:33:10.350]to measure the soil moisture content, right?
[00:33:12.210]Soil moisture sensors.
[00:33:13.560]Ours is we don't want to have the soil,
[00:33:16.110]we wanted to reduce the effects of the soil moisture sensor
[00:33:20.070]to our modeling, right?
[00:33:21.510]It turns out that there are algorithms
[00:33:24.150]that allows us to do that, right?
[00:33:26.141]So, again I'm not gonna go into the details.
[00:33:30.990]So there are statistical ways that you can
[00:33:33.437]do the modeling to kind of remove or minimize
[00:33:36.660]the effects for moisture on your model
[00:33:39.330]so that you can actually grab the model,
[00:33:41.640]train from the lab
[00:33:43.050]and apply directly to the spectral you get from the field
[00:33:46.920]with a varying moisture content, okay?
[00:33:49.380]Now this is, and you can see it here, right?
[00:33:51.810]So you know, if you do not using that,
[00:33:53.940]you want to direct apply,
[00:33:55.500]you know, it's a mess, it's everywhere the points, right?
[00:33:57.930]But if you use that statistical analysis
[00:34:00.630]of algorithm to do the correction,
[00:34:02.670]you can actually improve the accuracy quite a bit, right?
[00:34:08.310]So here's the message, right?
[00:34:09.480]So, you're putting all those together, right?
[00:34:11.430]And that's what I have been trying to do
[00:34:14.210]in the lab for so long, right?
[00:34:15.870]Is you have a visit NIR penetrometer,
[00:34:18.360]that's your hardware system.
[00:34:19.890]It can be handheld,
[00:34:21.210]can be attached to a hydraulic probe
[00:34:22.830]or somebody later on
[00:34:24.510]you can develop a robots to carry it around,
[00:34:26.790]right, driving the field.
[00:34:30.180]And you have a spectral library that is national scale
[00:34:34.830]or maybe local scale.
[00:34:36.060]However, because it's all virtual, you know,
[00:34:37.740]you can be very flexible how to define in your software
[00:34:41.250]and you have a moisture correction
[00:34:42.930]or you know, sort of like the aggregation correction
[00:34:46.260]algorithm and all those three pieces together.
[00:34:49.080]Hopefully we can get a low cost institute,
[00:34:51.540]deep layer soil sensing system, okay?
[00:34:54.540]So that's what we are trying to work on over the years.
[00:34:59.070]Now, so it turns out, right,
[00:35:00.570]so, I show a slides about the carbon,
[00:35:03.360]so it's so lucky that organic carbon
[00:35:05.730]is just one of the soil properties that can be picked up
[00:35:09.600]by the spectroscopy very accurately, right?
[00:35:12.900]So therefore this can be used
[00:35:14.910]for potentially soil carpet MMRV, right?
[00:35:18.030]So MRV stands for measuring, reporting
[00:35:20.550]and verification through direct measurement.
[00:35:23.010]So a lot of the talk and discussion right now
[00:35:26.157]and some of the practices is to use model, right?
[00:35:29.607]So, and you know like model can be accurate.
[00:35:32.550]The model can also be very, very, very inaccurate, right?
[00:35:35.640]So therefore, there's nothing that can be replaced
[00:35:38.460]with direct measurement.
[00:35:41.370]If you want truly want to be serious about it,
[00:35:44.430]and you know that what you're receiving
[00:35:46.140]or what you're paying is actually get stored in the soil,
[00:35:50.250]then you have the direct measurement
[00:35:56.010]is your smoking gun evidence
[00:35:57.510]that you actually did it, right?
[00:35:59.520]So, therefore that's what we are trying to do.
[00:36:03.565]And you know, I show,
[00:36:06.030]we test our vision on our system,
[00:36:07.800]and this is supported by the ARPA-E,
[00:36:09.540]one of the ARPA-E project that we have,
[00:36:12.330]and a collaboration with Soil Health Institute.
[00:36:15.090]And so, you know we tested in the field,
[00:36:18.180]and so you can see the axis is measured
[00:36:21.977]of the soil organic carbon stock, okay?
[00:36:24.380]So this is not concentration anymore,
[00:36:26.850]this is a stock in terms of megagrams per hectare, okay?
[00:36:30.513]And each of those dots is a core
[00:36:33.540]that it kind integrates the two certain depths.
[00:36:37.380]And you can see our bias is really low, okay?
[00:36:39.810]And our score is a pretty satisfactory
[00:36:43.260]just using our penetrometer system, right?
[00:36:47.700]You know, one of the nice thing is,
[00:36:50.220]it accounting deep layer soil, organic carbon, right?
[00:36:54.060]And then it measures both carbon concentration
[00:36:57.840]in terms of percent
[00:36:59.430]and also the bulk density in terms of like
[00:37:02.730]you need those two to figure out what is the stock, okay?
[00:37:06.600]And we do have sort of like a collaborator
[00:37:11.040]from the ARPA-E project that they really like it,
[00:37:13.680]but they added their own idea to it, right?
[00:37:15.930]So that the company's yard stick
[00:37:17.940]and they are very, very successful these days.
[00:37:20.250]So they got a serious a investment
[00:37:22.230]and I just see like the company grow like crazy, right?
[00:37:26.700]But you know they license the technology
[00:37:29.880]that I mentioned earlier that belong to Texas ANM University
[00:37:34.830]and they do further development.
[00:37:38.130]And their idea is like,
[00:37:40.110]because our problem is,
[00:37:41.807]if you get hot soil, if you trying to push it,
[00:37:44.403]it's really hard, right?
[00:37:45.600]You can't do it.
[00:37:46.770]So their idea is they add a rotation mechanism.
[00:37:50.580]So essentially that becomes a drill, right?
[00:37:52.950]And how do you get the bulk density?
[00:37:55.140]Well they think that you can get the bulk density
[00:37:57.390]from the torque, right?
[00:37:59.280]So they are gathering data,
[00:38:00.420]they're doing the testing to be able to show that, right?
[00:38:03.210]So our way to get the bulk density is through the push,
[00:38:06.360]through the load cell measurement,
[00:38:07.890]and they have the motor
[00:38:09.870]and the motor can get the electrical current
[00:38:15.090]that drives the motor
[00:38:16.110]and you can get that torqueinformation
[00:38:17.550]to get the talking information
[00:38:19.260]and hopefully it can get you some bulk density.
[00:38:22.350]But I'll show you the video.
[00:38:23.520]So this one is actually can play, and this is a pretty cool.
[00:38:48.930]So you go from the top right?
[00:38:50.880]And then you go down, so there's a sensing head
[00:38:53.130]and those are also the one that in the middle,
[00:38:55.830]the white boxes all the instrumentation in it, right?
[00:38:59.307]And the hope is that you just drill that down
[00:39:02.220]to I think this one is maybe 45 centimeters
[00:39:05.940]and you are able to just get
[00:39:07.350]that carbon stock from the surface to the 45 centimeters
[00:39:11.250]through one push, right?
[00:39:12.540]You never need to pull out the soil core,
[00:39:14.940]you never need to segment it,
[00:39:16.320]send that to the lab for the separate carbon concentration
[00:39:20.307]and the bulk density measurement, okay?
[00:39:22.650]So that's we'll see, right?
[00:39:24.810]So they got a lot of money
[00:39:26.280]and you know, I'm really hoping
[00:39:28.080]that they can be successful to get that done, okay.
[00:39:32.520]Alright, so this is my last slides, right?
[00:39:34.980]So we have...
[00:39:36.210]So you know I am really lucky that
[00:39:40.470]being able to do this,
[00:39:41.790]keep this line of research in my lab, right?
[00:39:44.370]So it has been getting funding from USDA and RCS.
[00:39:47.793]USDA-NIFA and the most recent one is a ARPA-E, right?
[00:39:51.900]And also be able to train our graduate student
[00:39:55.710]and publish papers, right?
[00:39:57.180]You know, so these are some of the examples of the papers
[00:40:00.660]that we published over the time.
[00:40:03.630]And you know, arguably some of those journals
[00:40:07.230]are really top soil science journals, right?
[00:40:09.360]So, and it's a very nice,
[00:40:11.580]and you know, I think it's three lessons
[00:40:14.940]that I have learned from this journey, right?
[00:40:18.030]So number one is, when I come to here, right?
[00:40:20.810]So I sort of talk to my postdoc advisor.
[00:40:24.393]I say, "Well, I really want to do this,
[00:40:26.670]would you allow me to do it?" Right.
[00:40:28.410]Well, she said, "Yeah, I'm not.
[00:40:30.390]I probably can't do it without an engineer,
[00:40:33.060]go ahead and do your exploration."
[00:40:35.430]And I so therefore kind of second year here,
[00:40:37.588]I kind of starts to do this research
[00:40:40.200]and after that I thin we're still very good friends,
[00:40:44.220]you know, I'm still kinda really close relationship with her
[00:40:47.430]and I think those kind of support,
[00:40:50.190]having support even after graduation
[00:40:52.950]with your PhD mentors and the postdoc mentors
[00:40:56.850]probably can lead to some nice things, right?
[00:41:00.990]And the other is the second thing I want to say
[00:41:05.460]is really if you look at around the country,
[00:41:10.260]there's very few labs actually do this kind of research.
[00:41:13.800]So therefore there's not much competition,
[00:41:17.580]but there's not much excitement either.
[00:41:19.800]Like, you know, if I go to somebody say, you know,
[00:41:21.660]I'm doing soil spectroscopy
[00:41:23.160]and I'm doing, you know, for the
[00:41:26.400]precision agricultural research, I wanna say,
[00:41:28.320]well, yeah, that's nice.
[00:41:29.550]And then that's the end of the conversation, right?
[00:41:31.680]So this is a much less excitement
[00:41:36.120]than some of the other research I'm doing.
[00:41:38.100]But I'm very, I'm, I think I just like it
[00:41:40.047]and I think it's really not,
[00:41:44.490]it's a pity if I give it up
[00:41:45.810]because I started like as a postdoc
[00:41:47.880]and then I wanted to kind of
[00:41:50.100]to make more development onto it, right?
[00:41:53.910]So that's the second thing.
[00:41:55.110]So, you know, kind of I think that to most
[00:41:57.240]of the graduate student, the postdoc here is,
[00:41:59.970]kind of stick to you kind of instinction.
[00:42:04.350]Like, if you think somebody, something is truly valuable
[00:42:07.350]and something is truly you liked,
[00:42:09.116]go pursue it, right?
[00:42:10.800]And the third thing is what I learned from,
[00:42:12.930]you know, yard stick, right?
[00:42:14.100]So, Yardstick come,
[00:42:16.110]we actually pick it up from Yardstick
[00:42:18.630]along the way was up ARPA-E.
[00:42:20.760]So it was a really small company
[00:42:23.340]like two, three years ago, right?
[00:42:25.740]But is the lesson I've learned
[00:42:27.060]is how can I see it.
[00:42:28.770]Is their CEO really runs the company so successfully.
[00:42:33.300]And you know, part of it is I think that,
[00:42:35.469]we are Nebraska we are sometimes humble,
[00:42:38.550]but that CEO is go around the country,
[00:42:41.760]talk about the technology,
[00:42:43.080]even the technology is not existing yet, right?
[00:42:45.780]It gets the people excited
[00:42:47.010]and a successful landlord series,
[00:42:50.347]I don't know whatever it is, right?
[00:42:51.180]So the venture capitalist investment
[00:42:53.790]and I think that's something
[00:42:56.970]sometimes we just needs to be a little bit more bold
[00:43:02.670]to talk about our science
[00:43:04.140]and being able to look at some of those
[00:43:08.310]alternative foundings such as a private,
[00:43:10.500]like the venture capitalist, right.
[00:43:12.853]You know get maybe talk to SpaceX for example,
[00:43:15.750]get Elon Musk excited.
[00:43:17.490]You know, those kind of thing is the lesson
[00:43:20.310]that I have learned during this journey.
[00:43:22.920]Okay, so with that, I'm gonna stop here
[00:43:26.040]and I will entertain some questions you have.
[00:43:34.650]Yeah, how come all.
[00:43:40.729]The soil organic matter when you look at the...
[00:43:45.401]Thank you.
[00:43:46.507]The soil organic matter when you look at the database,
[00:43:50.100]do they have the same site collected over time
[00:43:53.370]or the, those 300,000 samples?
[00:43:56.250]Yes, so they are,
[00:43:57.540]so the way this National Soil analytical lab works is they
[00:44:01.620]work on like all the other UNRCS labs
[00:44:07.710]distribute around the country.
[00:44:10.110]So when they have a project,
[00:44:11.820]they would get a soil samples,
[00:44:14.280]they will send that to the KSSL for the analysis.
[00:44:18.148]So, the lab data is very consistent.
[00:44:21.330]So that is the protocol that they have used
[00:44:24.750]for all the samples.
[00:44:26.370]How variable is it from,
[00:44:27.930]I mean, is it changing significantly from
[00:44:31.491]every two years, five years,
[00:44:34.230]or of course it would depend on the type of soil,
[00:44:37.440]but you know, for the soils
[00:44:38.760]that do show enough organic matter is,
[00:44:43.139]where's the value point?
[00:44:45.147]You know, if you sample more than that,
[00:44:47.580]like we change.
[00:44:48.413]Yeah, we don't know yet, so we never.
[00:44:50.310]So, if I understand your question correctly,
[00:44:53.370]nobody ever kind of go into that database to look at...
[00:44:57.660]Like if you happen to have soil samples collect
[00:45:00.570]from the close area and compare,
[00:45:05.190]you know, what happens in 2000s and 1990s,
[00:45:08.050]like a 10 year period,
[00:45:09.900]is there organic carbon increase or decrease?
[00:45:12.780]You know, I guess we can do that analysis
[00:45:14.460]by looking at the database more carefully.
[00:45:17.430]Nobody has ever done that yet.
[00:45:19.590]Thanks.
[00:45:20.461]Yeah.
[00:45:24.900]Yeah, Daniel,
[00:45:31.290]Couple of questions.
[00:45:33.210]So just out of curiosity with your depth measurements,
[00:45:39.240]where does the percentage
[00:45:43.200]of soil carbon start dropping off?
[00:45:45.690]Do, do you have any feeling for--
[00:45:47.310]In terms of depth?
[00:45:48.390]Yeah.
[00:45:49.530]Well, it really depends on the type of soil, right?
[00:45:53.010]So usually what we see is, you know,
[00:45:55.854]the top soil usually have the highest
[00:45:58.650]carbon concentration, right?
[00:46:00.477]For the lowest bulk density.
[00:46:02.430]But when it gets deeper, like go to the B horizon
[00:46:05.490]and then that's where, you know, you increase the clay.
[00:46:08.010]But you know, the, and part of it is a lot
[00:46:10.397]of the crop roots doesn't grow that deep.
[00:46:13.680]So therefore, so that doesn't really have a whole lot
[00:46:16.560]of organic carbon accumulation.
[00:46:18.480]So like below one meter or just depends on the soils.
[00:46:22.560]No, it won't to be one meter.
[00:46:24.090]It was like 30 centimeters, 12 inches.
[00:46:27.520]Okay. Yeah.
[00:46:28.740]And then, you know, with this Yardstick,
[00:46:31.530]there going after this carbon market I assume,
[00:46:35.310]how is all that,
[00:46:36.143]you know, you get a little bit of a view into it
[00:46:38.820]from your work with them
[00:46:40.380]and is that really taking off
[00:46:44.070]or is it a kind of a flash in the pan type?
[00:46:47.940]Well, depending on the kind of a funding that they get,
[00:46:53.670]I think they are taking off.
[00:46:55.710]I think they are trying to,
[00:46:57.180]so they're trying to publishing,
[00:46:58.800]so they have test their the one
[00:47:02.310]that I show is already like the third generation hardware.
[00:47:06.300]So, and this is from last year, you know,
[00:47:09.780]they are probably continue keep improving it
[00:47:12.870]and as they have been hiring people,
[00:47:16.230]working, hiring local people to test the probe
[00:47:22.230]and their model.
[00:47:23.130]So, I don't really know the latest development.
[00:47:26.940]More in terms of the carbon market
[00:47:29.010]of people buying credits and things like this in general.
[00:47:31.950]Is it really benefiting the Nebraska?
[00:47:35.251]I don't know.
[00:47:36.256]I certainly hope so.
[00:47:37.770]You know, I think that that is a really good business model
[00:47:41.460]in my opinion to improve growers income, right?
[00:47:47.271]You know, you can argue, well,
[00:47:49.740]you can't just indefinitely store
[00:47:51.630]of carbon in your land, right?
[00:47:53.040]So there will be, you know,
[00:47:54.510]you since the agriculture and its cultivation
[00:47:59.040]from it's a prestine states,
[00:48:00.360]you kind of lose some carbon,
[00:48:02.250]but if you're trying to replenish it,
[00:48:05.100]you're gonna have some kind of top threshold value
[00:48:07.770]you can do.
[00:48:08.603]And that's really your value and your profit.
[00:48:11.550]But you know I think,
[00:48:12.870]I haven't really looked at what is that value, right?
[00:48:16.410]But I think that's gonna be quite a bit of profits there.
[00:48:22.440]But at the same time you market that in a way
[00:48:26.610]that this is it's just like organic produce, right?
[00:48:31.110]So this is climate smart produce
[00:48:33.930]or this is more sort of carbon zero
[00:48:38.550]produce that you can add a premier to it
[00:48:40.737]for to market it, yeah.
[00:48:45.540]Yep, great seminar.
[00:48:49.260]Learned a lot.
[00:48:56.400]I was interviewed by a reporter from Washington DC
[00:49:00.210]through the other day about the pusher Biden administration
[00:49:05.670]to sequester carbon with cover crops
[00:49:07.500]and conservation tillage, all that stuff, right?
[00:49:10.530]So the question is this, you know,
[00:49:13.200]we have been planting cover crops
[00:49:15.120]and not only my group, but others, right?
[00:49:17.280]So what we have found in some cases is that
[00:49:22.230]we cannot really catch differences
[00:49:26.340]between a field with cover crops
[00:49:29.130]and field without cover crops, especially in the short term,
[00:49:32.640]let's say three years, five years, and some cases even
[00:49:35.550]after 10 years under the rotations we have.
[00:49:39.840]So my question
[00:49:40.860]or any comments, I mean, I think is this is work
[00:49:43.260]and progress, right?
[00:49:44.460]So any thoughts on the
[00:49:48.780]sensitivity of this method?
[00:49:51.570]Because we have been playing
[00:49:53.190]with soil samples, right?
[00:49:54.780]But we don't see any differences.
[00:49:56.040]But with this catch, you know, small differences Yeah,
[00:49:59.610]that'd be better than the conventional method.
[00:50:02.520]Yeah, thanks, thanks for the question.
[00:50:04.320]You know, we do.
[00:50:05.434]We thought a lot about the sensitivity of the instrument
[00:50:10.800]and how that sensitivity compare
[00:50:12.630]to the lab-based measurement, right?
[00:50:16.290]I think, you know some of it is,
[00:50:18.210]like you said, you wait long enough so that you actually
[00:50:22.470]build a difference so that you can detect it.
[00:50:24.600]And that's one strategy.
[00:50:28.830]And then the other strategy is you can strategically
[00:50:36.720]place your sample,
[00:50:38.160]because if you talk about the field scale, right?
[00:50:40.327]You know, the organic carbon changes a lot
[00:50:43.080]and there's a availability by itself.
[00:50:45.360]So that availability is going to deter
[00:50:48.270]your ability to detect a change.
[00:50:50.490]So, you know, the part of the,
[00:50:51.690]what we are thinking is,
[00:50:53.400]can we use existing things like remote sensing imagery
[00:50:56.460]or you know, this other spatial data available
[00:51:00.270]to be able to find this field,
[00:51:02.670]get just to four or five more uniform field.
[00:51:06.450]And in each uniform field then you can do a carbon detection
[00:51:11.130]and then for the entire field you do a weighted area,
[00:51:13.590]weighted average to do that.
[00:51:17.460]And so that's the second one,
[00:51:19.290]if that makes sense.
[00:51:20.760]And then the third one is,
[00:51:22.410]we can do our hope really is,
[00:51:24.945]if you do the sampling,
[00:51:26.790]you can only do two sample, three sample, five sample.
[00:51:30.090]Our hope is that we can do many, many samples
[00:51:33.240]and that's surely is gonna increase the statistical power
[00:51:36.930]when you do that comparison.
[00:51:39.150]You know, the whole you know,
[00:51:40.590]so some of the analysis to look at is the break,
[00:51:43.200]like the break even point, right?
[00:51:45.180]You know, if sampling takes you $100 per sample,
[00:51:49.920]but if our prob only takes $5 per sample,
[00:51:52.950]then you can potentially do 20 times more sampling, right?
[00:51:56.430]And then when you run your statistical analysis
[00:51:58.980]with the same amount of cost.
[00:52:00.862]Yeah.
[00:52:01.695]So hopefully the statistical analysis,
[00:52:03.030]you will be able to do it better.
[00:52:04.822]So this is applied for most of those,
[00:52:07.560]you know, rapid sensing instrumentation sensor development
[00:52:12.600]is that, you know, many rapid measurement
[00:52:15.900]is probably better than one or two accurate measurement,
[00:52:19.410]particularly for a heterogeneous sample set, right?
[00:52:24.180]So that's essentially I...
[00:52:27.240]Somebody's laughing there, right?
[00:52:29.610]Yeah, so hopefully.
[00:52:31.230]Yeah.
[00:52:33.480]Thank you.
[00:52:34.313]Yeah, the last Christian, you showed them this slide,
[00:52:37.020]I think it was before this.
[00:52:38.580]Can we look at this again?
[00:52:40.110]That was very, very appealing.
[00:52:41.550]This one.
[00:52:43.830]I mean this is, in my world, in my small world,
[00:52:48.240]this R square 78% and then 82%.
[00:52:52.890]That's marvelous.
[00:52:54.780]Yeah.
[00:52:55.613]Problem solved.
[00:52:56.446]So we can use, now this spectroscopy
[00:52:59.430]or whatever that tool that guy is using, let's buy.
[00:53:01.950]I mean that's 78 and 82 is great.
[00:53:05.040]In my case, whenever I get 40% or 50%, I'm excited.
[00:53:08.940]Because I look at those, you know, basic--
[00:53:10.500]Let's talk about it.
[00:53:12.030]Let's get connected after the info.
[00:53:14.250]So we start a company.
[00:53:15.610]Yeah, so I mean, I'm wondering if this is that good.
[00:53:20.070]I mean we cannot not use it.
[00:53:22.140]We controlled it very, so this is a really research,
[00:53:24.360]we control it very well.
[00:53:26.190]So we go in,
[00:53:27.900]so this setup is.
[00:53:29.940]We actually, we don't use this probe.
[00:53:32.280]We are using, the one that we have gets a hook up
[00:53:34.670]to the hydraulic probe.
[00:53:36.480]So we push that in and we do multiple pushes
[00:53:39.330]and we average it to denoise it.
[00:53:41.575]And then we segment the soil sample really very carefully,
[00:53:46.740]like to do the comparison.
[00:53:50.010]So you know, those kind of thing.
[00:53:51.330]And it is, so for example, you know,
[00:53:54.420]here with the one field it's about like 27, 28 soil course.
[00:54:00.717]And it take us like three days to do it
[00:54:03.750]in that field for the data collection.
[00:54:05.730]But our idea isn't really in the reality,
[00:54:08.220]we would like to do 28 locations like three hours.
[00:54:13.680]That's our goal by one person, yeah.
[00:54:22.691]Great talk and a lot of great resources
[00:54:26.250]I didn't know about.
[00:54:27.840]I have a question about,
[00:54:28.920]I'm interested in nitrogen availability.
[00:54:32.670]Well, especially in natural soil.
[00:54:34.980]Now those are heavily industrialized prevalence.
[00:54:39.510]So I heard that people talk about the nitrogen
[00:54:43.620]availability in natural soil shows some sort
[00:54:46.680]of a longitudinal or elevational pattern.
[00:54:50.130]Okay.
[00:54:51.000]So that makes sense.
[00:54:52.230]'cause a kinda elevation
[00:54:54.390]that you have less vegetation probably is less
[00:54:58.382]mineralization process, well less nitrogen available
[00:55:03.120]in natural soil.
[00:55:04.650]So do you have any data
[00:55:06.540]to support this global nitrogen or vulnerability pattern?
[00:55:13.410]Yeah, so, well a lot of the study
[00:55:16.950]that we do is on the crop land.
[00:55:18.960]So it is of nitrogen being applied on those ground.
[00:55:26.010]So, you know, one way that I can think of is, you know,
[00:55:28.800]the NRCS database we have,
[00:55:34.800]one of the nice thing about the NRCS database
[00:55:36.970]is we do know where those samples are from.
[00:55:39.540]So they all have coordinates
[00:55:42.210]and many of those projects the nature isn't agriculture.
[00:55:47.280]So therefore there will be some native soil being captured.
[00:55:51.330]And then, so they usually have the physical sample
[00:55:54.480]large enough that you can go in
[00:55:57.630]and extract some and do the re-analysis.
[00:56:00.780]So if you are interested, you know,
[00:56:02.250]we can certainly try to see if that's gonna help you or not.
[00:56:05.790]Right, go dating back to 1950, okay.
[00:56:10.800]No, it's the US, it's just the US.
[00:56:13.590]So there are some international soils,
[00:56:15.690]but you know, the percentage is probably
[00:56:18.030]just like less than 1%, yeah,
[00:56:25.740]It is an interesting presentation
[00:56:28.440]because it sounds like the presentations would've been made
[00:56:33.810]40 years ago when we were starting to work
[00:56:35.730]with NMR spectroscopy with mortgage samples.
[00:56:40.320]And I guess just a couple comments that I might make
[00:56:43.260]with it is, is first off,
[00:56:47.130]your models are gonna be no better than the quality
[00:56:50.340]of your lab measurements on your samples that you do take.
[00:56:54.060]Right?
[00:56:55.290]You know, garbage in, garbage out
[00:56:57.030]is what's gonna happen with those.
[00:56:58.680]And so you have to really be careful
[00:57:01.290]with duplication, triplication,
[00:57:03.000]and making sure that the values
[00:57:05.700]that you're putting into that model with your initial
[00:57:10.260]developments are really critical there.
[00:57:11.820]And I guess my one main question for you is,
[00:57:15.720]have you tied into the history of those developments
[00:57:21.870]to kind of help guide some of your own work here
[00:57:25.590]in going to soils?
[00:57:26.940]Soils are gonna be much more complicated.
[00:57:28.680]But you know, with the forge area there,
[00:57:32.310]you know, we've had to move
[00:57:33.810]and make special ones for different types of forages.
[00:57:37.380]Right.
[00:57:38.213]But it's reached to the point where
[00:57:40.740]we're putting it on mailers
[00:57:43.350]getting moisture and getting various quality measurements
[00:57:46.350]in the field as it's going along.
[00:57:48.630]Right, yeah.
[00:57:50.460]Thanks for that comment.
[00:57:51.630]You know, I am aware of those development from the forage
[00:57:54.600]and USDA-IS lab many, many years ago,
[00:57:57.630]they kind of do the pioneering work.
[00:57:59.670]And in one of my, so yes, it's true.
[00:58:02.520]You know, it's the soil spectroscopy largely exists
[00:58:06.030]in the research world, right?
[00:58:08.150]Is the folks trying to push to putting the practition.
[00:58:11.100]And I think that's still, you know,
[00:58:12.900]there's a lot of resistance or the questions
[00:58:15.330]suspicion about it.
[00:58:17.280]And I have one of my PhD students,
[00:58:20.460]you know, one thing that I,
[00:58:21.990]one task I ask her to do that's probably become in her
[00:58:25.620]first thesis chapter is really going into the literature
[00:58:29.520]to ask the question why soil spectroscopy
[00:58:34.080]hasn't really been put into practition
[00:58:37.530]like forage or other biological samples yet.
[00:58:40.950]You know, part of it might be just because,
[00:58:43.050]you know, like you said,
[00:58:43.883]the soil is really complex, right?
[00:58:46.200]And you can't develop a uniform or in a consistent model
[00:58:50.940]good enough for the application.
[00:58:53.520]And part of it is, we see the good results
[00:58:58.020]with organic carbon or clay,
[00:59:00.000]but there are other properties.
[00:59:01.320]It's not really good enough yet.
[00:59:03.120]Like P and K particular, right?
[00:59:05.690]So the interest in guiding the nutrient applications.
[00:59:10.110]So, but yeah, so your point are well taken
[00:59:13.114]and thank you.
[00:59:14.773]And we see the same thing with forges that coming up
[00:59:17.370]with specific minerals is almost in many cases,
[00:59:22.110]rather than really a good correlation.
[00:59:24.480]And has to be really well tied into
[00:59:27.270]some of the organic components
[00:59:29.850]in order to get a good correlation.
[00:59:31.590]Yes, thank you.
[00:59:33.630]Here coming.
[00:59:34.658]So this will be the last encouraging question.
[00:59:36.690]There are a couple of technical questions online.
[00:59:38.701]One minute.
[00:59:39.900]Okay.
[00:59:42.270]This is probably kind of a complex question.
[00:59:45.840]Let's assume that I'm from John Deer
[00:59:48.390]and I have contacts with a lot of farmers.
[00:59:51.570]Some of them are selling carbon credits.
[00:59:56.130]So what have you got that would be of interest to John Deere
[01:00:00.510]that farmers would be willing to pay for?
[01:00:07.200]Yeah, well yeah,
[01:00:08.685]that is a very complex question.
[01:00:11.550]So now I'll try my best, right?
[01:00:17.345]I operate in the research world, right?
[01:00:19.170]So therefore, you know, I don't really think too much about
[01:00:22.350]those, you know, if I started the,
[01:00:24.180]if I start the business right,
[01:00:25.620]how you gonna make the cash flow?
[01:00:28.260]How you gonna try to make the people pay me?
[01:00:30.630]Or, you know, how you're gonna,
[01:00:32.430]the companies gonna be profitable.
[01:00:35.340]You know, I think that you can look at
[01:00:38.100]some of the different, so other Yardstick, for example,
[01:00:41.730]I think their business model is,
[01:00:45.600]I think they're not gonna sell it in hardware.
[01:00:49.470]They're not going to selling
[01:00:51.390]anything they're gonna just keep the company,
[01:00:53.880]they're gonna do the service, right?
[01:00:55.650]You know, they're gonna charge.
[01:00:57.540]So they're gonna go out and they are trying
[01:00:59.310]to develop the system such that the operating cost is so low
[01:01:03.450]that you can, I charge you $10
[01:01:08.100]for every acre of carbon gets measured or credit credit.
[01:01:13.138]And then they're gonna do, you know,
[01:01:14.490]things like a third party verification
[01:01:16.800]through direct measurement.
[01:01:18.750]And if that cost is lower than
[01:01:22.080]what the growers can get from the carbon market,
[01:01:25.470]they are probably willing to pay it
[01:01:26.820]and the company can still make a profit.
[01:01:29.520]But I don't know if that answers your question and,
[01:01:32.490]but yeah, I don't really consider a whole lot about that.
[01:01:39.570]All right, let's thank Professor Ge.
[01:01:42.166](audience clapping)

The screen size you are trying to search captions on is too small!

You can always jump over to MediaHub and check it out there.

Comments

0 Comments

Rapid Soil Sensing for Precision Agriculture and Carbon Market

Description

Searchable Transcript

Comments icon comment

Related Channels

Comments