Proceedings of the Workshop on Remote Sensing for Agriculture in the 21st CenturySession 4: How will the Data be Analyzed and What are the Priority Science Issues for the Next Decade? |
|
Introduction and Topics for Discussion
LB: I work with the databases and instruments in the field with
the agronomists. I’m an electrical engineer. More recently,
most of the work has been involved in data analysis with Professor Landgrebe
-- analysis of image data, multi-spectral data, high dimensional data.
I was invited because of the work I’ve done in agriculture back in the
‘70s and ‘80s, until agriculture was a bad word at NASA, but you can say
it now.
LB: This is from the standpoint of things that NASA [could do].
The [recommendations will be presented in a] paper to NASA to [identify]
where they can help out. There’s a big push from the public sector
now, but there may still be some issues [that need some help and can’t
be done by the public sector].
1. Primary science issues that prevent algorithms from becoming operational
in remote sensing analysis.
2. Are the science issues consistent with the needs of the end user?
3. What choices need to be made for spatial resolution and temporal
trade-offs?
4. What are the most probable products, and how are they going to be
used?
Data Products
LB: I was hoping to have a lot of end users in here. What
are the questions that the data products are to answer? I can see
where a user might have a question but might not know what the data product
is. P: From the discussion this morning that there were two
levels of questions being asked:
1. One was within-the-field questions where you’re going to need some
very high spatial resolution data and probably high spectral resolution.
For within-field, the spatial resolution is one meter or better.
2. The other category of products would be for large areas where you’re
spectral resolution issues may be the same, but you’re spatial resolution
issues would be very different. The large area resolution is probably 10-30
meters or larger. 250 meters would probably be adequate to tell agriculture
if you’re trying to identify fields or crops.
R: I heard a lot of talk this morning about the number of times a satellite has to be able to give you data, but with some crops it’s not as important as others. The same for resolution -- I think fields and row crops wouldn’t have the same demands as orchards and vineyards. If they’re looking for diseased crops or something like that, they want to catch it. LB: For vineyards, do you want one meter resolution or will 5 meters [do]? R: One meter would do it. Because you want to be able to see an individual [plant]. LB: Each plant is worth thousands of dollars. P: Most of what you see is ... ground for irrigation systems... certain header...send the irrigation person out there to ... the problem. Tell them what row, what number, etc.
P: One of the areas we’ve worked with is biomass estimates. Particularly for grazing areas and looking at carrying capacities and dealing with grasslands. That has a lot of applications. I see the efforts interested in it because of fire control. Soon the BLM and other large ranchers will be interested in it as a way of knowing the carrying capacities that would have all the ... If you go then to the whole area of commodities that we’re skirting this morning on a worldwide basis that come up on applications on who’s doing what, etc. I think it’s an area that I don’t believe we’ve felt we had a whole lot of success in terms of tonnage per acre, but it’s an area that we’re certainly interested in. It’s an area that people like to have numbers and often. It lends itself to remote sensing, however do we have the tool, not as much for identification, but for dealing with the texture, and translating that texture into a quantity of matter? -- Whether it’s the fodder or just weeds?
Potential Data from Satellites for Managing
Vineyards
Monitor Yields
P: For us, we’re wanting to do and be able to monitor our
yields... because of the value of land and using land as a resource and
keep it sustainable. I think it’s better to go in and replant so
that will get your yields up in areas where you have diseased vines.
All those things add up. Pretty soon you have an acre in a ten acre
lot that really doesn’t function.
Water Stress
P: [Also] water stress problems -- anything that doesn’t allow
the vine to work at its optimum. The goal is uniformity in a vineyard.
It doesn’t have to be high productivity, it can be high quality and lower
yields. [Uniformity is] going to make a difference when you harvest.
Differences between Spatial Distributing of Quality
P: We’re sampling the vineyard. We’re finding that
there’s sections of the field that taste better than others. What’s
that related to? If we look at GPS information and we find out that
it’s a certain soil type, slope, nutrition, soil moisture or vines with
a certain amount of canopy, etc., that gives us what we want. Whether
or not that’s doable, I don’t know. I’m even thinking of using a
GIS in a winery in the future for collecting data [so that it can be] analyzed
and compared. We’re using ArchView 3.
Balance
P: A grapevine has a certain capacity to produce fruit.
If it overcrops itself, the next year it will balance itself by having
less fruit. The idea is to find out what the balance is. A
lot of research has shown that shoots in a vertical train system[?] [can
grow] 46 inches long. You can get enough leaf area to support that.
I’d assume that other crops have similar for vascular production or quality.
LB: When you say balance, I can think of factors that probably go
into the balance of a vine, like soils, slope, the aspect etc. These
are all spatial factors that can become layers of a GIS.
Soil Moisture and Vertical Profile
P: We’re going to use [soil moisture and vertical profiles]
to assist us in our irrigation. When we go out and get information
from the neutron probe, we have key data bites around that probe site that
take rate of growth, development of the cluster... the different rates
of growth at various heights. All those things are going into the
GIS along with analysis from NASA -- nutrients, potassium of the vine.
Soil Maps
P: We’re also looking at doing more soil studies, because
the soils maps are horrible. LB: At the time, what the RS has was
adequate but now what they call inclusions, you need to know exactly where
those inclusions are. R: Yes and it makes a big difference
if you have half an acre of vines that you only get 6-8 tons/acre.
You’re losing 3 tons right there. [We want to] get an overall picture of
the vineyard on a newly acquired property or property you’re looking at.
Then be more site specific on where holes are dug and how we look at the
soils in conjunction with using the existing soils maps. LB:
Do you see that once you get your soil maps at the level you need it, then
probably that’s not going to change a lot? R: Yes, now we’re just
collecting data. Like was mentioned this morning, the growers want
immediate gratification and a problem solved. They don’t want to
spend a lot of money until you can prove to them that they’re going to
get something out of it.
Management Purposes
P: Oftentimes what I see in vineyard, there’s lot of holes
that are missing -- vines that die and aren’t replaced or they’re replaced
but not allowed to come up to fruition. And from a management standpoint,
if you have drip irrigation like we do, we have 5-6 irrigation people going
out and checking emitters. We don’t have time to do it ourselves.
Our latest infrareds that we took showed lots of places the emitters had
come out of the drip hose and excessive water was going out in that area.
Just for managerial standpoint. LB: That was brought up this
morning -- the double check. R: Exactly, the “cheat sheet”
theory.
High Soil pH
P: For us as a company, we’re in a warmer area for growing
grapes in California... high pH in grapes and wines. That’s
one of our big concerns, because that causes us lots of problems down the
road. With high pH in wine, you get less color, less fruit character,
etc. It’s better to have a lower, normal pH.
Nutrient Deficiencies
P: If the grower can show nutrient deficient areas in the
field from remote sensing, they can change it. I was interested in
seeing the tractor. I guess they’re using it for row crops especially
in the Midwest to look at what the yields are on a per meter basis.
I think it would be interesting to do the same sort of thing for a vineyard.
LB: They actually have a GPS system on a combine with a little antennae.
Most of them have a yield monitor and a moisture monitor for the grain.
As the grain is hitting the pressure plate, the more grain that hits it,
the more pressures and the higher the yields. Then you can set it
to sample every so often -- every second or two seconds. The location
of the sample is stored in a database and you end up with a map to generate
your rain moisture and your yields. P: I can see that we could
use that information in a vineyard, maybe have a pressure that goes down
the row. Then when we go out to prune, we know exactly where the
yield... instead of doing it one by one. At what point, there’s overkill,
I don’t know.
LB: I’m surprised how a lot of people, even some of the smaller
300 acre farms are using this. I’m surprised because I don’t know
what the payoff is. One person had a case where one area was high
in yield one year and low the next. The difference was moisture.
One year was wet. Different soils hold the moisture differently.
Basically they find that these yield maps relate to the soil maps more
than anything else. Plus they do see the effects of over-spraying
from the previous year. R: Well we have rolling hills, so we
have the pressure of crop emitters, so just being able to look at that...
How to Analyze the Data
LB: At the local grower level, you don’t have to start by doing
a classification of what’s there. I want to know about the variations within
the field. So that’s a different area of analysis than what’s been
going on historically in the last 20 years in remote sensing. Then
there’s developing information for short-term reaction or for long-term
trends like the moisture level for irrigation (short-term) and yield tracking
over years (long-term). The order to plant fields or harvest them
or the order of applications. The border effects -- this can get into litigation
because you can see where you’re neighbor over-sprayed and it had an effect
on your yields. The regional level was discussed this morning --
production estimates, where to put inventories. A lot of the tractor
companies [want to know] when [the crop] is ready to be harvested, and
they will make sure their harvesting equipment is at the dealers then.
Then land used as inputs to other local governments.
Raw Data versus Processed Data
P: Are you going to deliver data or a higher end product? If
you’re interested in the data, you’re going to generate your own processes
from your GIS. Another way of doing it (I think they’re both legitimate)
is to just deliver the product. Somebody else would make sure the
data is all on the right coordinate system. [Inaudible comment] LB: Mike
McDonald, in his original analysis [the growers] wanted their people at
TRW to give them a report that tells them to go to this spot, they didn’t
want the image. But Kevin Spry disagreed and said there was a market
for the image -- maybe it’s some interpretation on it. I would guess,
if I were a company delivering something, I’d still want to deliver the
image. I would think the end-user would feel much more comfortable
about being able to look at that and going to look at the field and doing
some verification. [Inaudible response] P: One grower may not
know what you’re looking for... In our case, it’s better for use
to get the image and then go out and... R: When you bring the data
down, most growers probably wouldn’t know initially how to deal with that
data. So you have an agriculture consultant who receives the data,
does the analysis on it and delivers a product. It could be an image,
not a raw image, but a second or third order image that ultimately you
could do that data processing without the ag consultant before it gets
to the actual user. So the question not only becomes what are the
standards needed, but also what standard products you might be able to
define if such a thing is possible. The space systems that we have
now do very little data processing on board, but they certainly could.
You could envision a system where a satellite was collecting data, doing
on-board processing and then delivering a data product every two or three
days right to your dish. You could use it as you wish -- you could
feed it into a GIS or just use it as an image. P: What would
the data products say other than show an image or maybe there are standard
things that have been ground-verified enough that that image could be analyzed...
LB: What the high dimensional data is used for -- that the 200 band
hyperspectral data is used for geology? P: Yes.
Need for Ground Verification
P: There’s certain relationships that apparently remote imaging,
you know with some degree of certainty that that’s what’s going to be occurring
on the ground. LB: Generally that’s vegetation. R:
Right, but right now, I don’t think you can take a look at different colors
and say exactly what disease it is or if it’s water stress. You still
need ground verification. But maybe that product could identify your
options. Here’s the point you go to, you might have water stress,
disease. The pattern looks like filotra[?]
Use in GIS
LB: For within-field, I think you’re always looking for variation,
whether it be water stress or canopy cover, canopy health. There’s
some variable that you want to look at. You could have one product,
you could have a series of products that would be delivered to you and
don’t even need to be colored if you’re looking at one variable.
It could be grayscale. You could feed that into a GIS to give you
more spatial information. Overlay it with a road map or whatever
so you could more easily identify where you are or show the location of
known ground locations. Then use it for whatever value it has.
One thing brought up in the Farm Bureau report, the need for the “expert
systems” or those type of things developed that helps between the data
and the end user -- some model that helps interpret the ... P: What I meant
was that in my trying to visualize what you’re talking about is with the
products...I was thinking more in terms of things that would have to be
verified. For instance, in forestry there’s a certain amount of information
that you can identify if it’s a tree ... signature. LB: Under certain
conditions, but even those have to be verified. P: ultimately...to
be able to get some kind of signature. LB: The ultimate is
the Star Trek sensor, the instrument. Take the instrument, point
it, it tells you everything about it. [Laughter]
Primary Science Issues
Spatial resolution
LB: What crop type needs one resolution versus other?
The smaller resolution you go, then you add a whole lot of variability.
Sometimes that becomes “noise”. You’ve got to integrate over two
or three rows and sample it over and look at the variation from that as
opposed to looking at the plant level, because you’re looking at shadows.
Textural information
LB: If you do get that small, the research issue is how
to use that as information about a canopy. We talked about improving
the resolution of soil maps. Also DEMs -- a lot are not adequate
from a local grower/producer level.
Spectral Data
LB: There’s always the [spectral] resolution, how many
and the coverage. We’ve talked about optical, but Agrometrics uses
the thermal and gets a lot out of it.
Microwave
Temporal Issues
LB: How often, and the timing -- should it be done in the
morning, midday, night? Then issues related to that include registration
and developing the GIS -- remote sensing data at the GIS, or to a projection,
or remote sensing data to remote sensing data. It makes a nice image
using cubic convolution, but there’s some people that don’t realize that
when you do that and an averaging, you destroy the data in a way that you
can never recover from. In particular, the second order statistics
may not now be valid. Sometimes it’s better to just take your standard,
and then how good does the registration have to be?
Calibration
LB: When is it needed or not needed? Particularly
with atmosphere calibration. If you do an atmosphere calibration
to 10% accuracy and you’re collecting the data to 12 bits of precision
which is far less than 1%, you’ve made the data worse than before. I really
believe there are times when you don’t want to do the atmospheric calibration
depending on the scale. We have an example of this where the ground
is flat and you don’t have variation of the amount of water vapor across
the 2 kilometer area.
Technology transfer
LB: Lots of questions, but you don’t know if they can help.
How do we transfer this?
? The cost of the analysis -- you can get a whole lot more information,
but then the cost will be too high.
? The band width needed -- do you give the information to the local
grower directly or just some regional person who does it and then feeds
it back. How can to get the data to the users? Can you do it
across the 14.4 or 28.8 modem?
Ground instruments
LB: I don’t know how much NASA wants to do with this, but
[ground instruments] are also part of remote sensing but not from a satellite
standpoint. There’s a lot that’s needed within these applicators
that vary the rate as you go across the field.
Coordinate Systems and Standards
P: Is there a standard coordinate system that everyone’s
using for the data? Because when I go to different sources for aerial
photography, it seems that everyone’s got a different coordinate systems.
We’re putting all of our stuff in State Clean 2. Is that something
that I should be doing? R: It’s a common problem. There
are no standards really. R: So someone might have some great
data that you’d love to get, but you can’t do it because first you have
to go through the expense of translating the data and all this other stuff.
R: It’s something you have to ... You have GIS programs that
have the ability to translate into different coordinate systems as a matter
of just touching it up. R: But you’re talking with an end user
that doesn’t have that ... We use ArchView, maybe the newer version
will have that capability. LB: There are different projections
for different reasons. When you look at global bases, there’s two
or three that they like to use there. [Inaudible response] P: The
Teomid[?] Center is using staple area projection, but one of the problems
with UTM in this area is that we split two zones. Some people reconfigure
the data so that it fits into ..., because they don’t want to have to worry
about the TMN level. Others says that with UTM there’s not going
to be equal area or accurate measurements. R: ...corporations.
They suggest putting it in StateView 2 ... but is there something better?
Is there going to be standardized coordinate systems?
LB: I think it’s going to depend upon the scale that one is working
with. You want to be careful that if you start with this projection
and then go to a different data with a different projection, pretty soon
you’re losing data, because you keep resampling. It’s much better
if you can start with the original and go to this one or that one.
R: But you’ve if got different agencies and different companies using
different coordinate systems, then you do have to go nag over here and
nag over there.
LB: Have you read the thing in here from the International
Farm Bureau? One key thing they [advocated] was standards as far as precision
farming data. They saw that as an impediment right now. Of
course, that’s something we shouldn’t be surprised by. They did say
companies need to get together so they can plug this instrument from this
one to this instrument for there and plug it together.
P: ...If you get a small area here in the Sierra, and I have
5 different surveys with 5 different ways of describing it with the same
series. At times, different basic ... description. You try
to set up a database for GIS, and use your judgment to draw out which attributes
you want. LB: Now put that into the SOTER -- that’s supposed
to be a new international classification for soils. A standard they’re
trying to set up that different.[Group ends and leaves to join Susan’s
group]