Proceedings of the Workshop on Remote Sensing for Agriculture in the 21st Century

October 23-25th, 1996 
 

Session 3:  How will the Data be Analyzed and What are the Priority Science Issues for the Next Decade?

Transcription Notes
 
| Ag 21 Agenda |
 
Moderated by Susan Moran
Session 3; October 23, 1996
West Room, Alumni Center
Note:  Interjections are in [brackets].
SM=Susan Moran; LB=Larry Biehl; P=Participant; R=Response

Introduction
Main Goal: Suggestions and comments to NASA about future research and development of remote sensing in agriculture.  Includes satellites and aircraft for precision farming and commodities.

Topics to Address:
1. Science issues
2. Product and Dataset Validation
3. Sensor and System Design - Delivery and Spatial Resolution
4. Data Products
Let’s make prioritized lists and have general discussions on these four topics.

What are the primary science issues that prevent tested and validated operational algorithms from becoming operational systems?
SM: Let’s say we do have an operational algorithm.  How do we get someone to accept it? R:  Ideally, when they proposed the study to do this algorithm, they’ve also identified the group that’s anticipating its use.  If they haven’t, it’s more of an academic exercise than a real one. Ideally, they would know who’d want to use it. SM: Agrometrics[?] [determined the] crop water stress index and is now flying it and selling it to farmers to do irrigation scheduling.  The private companies are paying for it -- that may be one way to interpret this question.

Need for a Yield Monitor
 P:  I work with a vineyard and a winery. What’s lacking there is a yield monitor. R:  A yield monitor is not from remote sensing. [Response muffled] They can with GPS, if you knew the amount of yield from each acre or each area in a block.  Once you did that, you’d try and find out why they are differences [between blocks].  Without that, the image may or may not be true, but with yield you know its true. R:  Biomass doesn’t always tell you what the yield will be.  In fact, in most cases it doesn’t.  SM:  I think a lot of the science issues of remote sensing can be converted to yield. R: With some degree of accuracy. R:  For grain crops we have a device yield monitor attached to the combine.  SM: If you can get yield at certain locations, [muffled parts] remote sensing could provide a supervised classification mode for something or in a relative yield mode, which isn’t absolute. R: One of the problems in using remote sensing to give you information about yield is that yield at the end of the season is an integration of what has happened over the entire season, whereas each remote sensing observation is a point in time.  It would be nice to have a way to use remotely sensed data and get an integration over time.  You would expect that would be better correlated with yield at the end of the season, rather than trying to rely on one given observation at some time during the growing season, where subsequent activity (hail storm etc.) might make it completely wrong.
 
Spatial and Temporal Resolution of Sensors
 P:  I’m not so concerned with having the yield be accurate.  The farmer’s trying to find his best block and [investigate why other blocks aren’t doing as well].  If you could move them all up to the same level... R: Perhaps use remote sensing to estimate bigger growth... R: Or uniformity. R: In order to do this, you’re talking about [muffled] a multi-temporal approach to the way we use remote sensing. R:  There’s some research being done but not enough, because the price of the data is too expensive.  The spatial resolution of the sensors often prohibits us from getting the temporal resolution that we need to be able to get yield.  If you’re using Landsat, you might get 3 or 4 images over the entire growing season.  If you’ve got cloud cover during that time, you’re out of luck.  So this is jumping down a bit [in the topics] but sensors that give us higher temporal resolution  -- it may need higher spatial resolution. R: You need to be able to compare data from one day to the next ... R:  It depends on the region of the world.  In my part of the world, the farms are 50,000 acres.  If you’re in the Midwest talking about wheat, corn, soybean, etc. yield is a lot different than in California, where farms are 50 acres.
 
Involvement of the Users
 P:  The other science issues that are always based on this [muffled] is involvement of the user. If you’ve got something that wasn’t developed to meet a particular demand, then the only way you can popularize it is to go through a regional pilot project where you involve all the growers.  There’s usually some government involvement (university, etc.) to actually demonstrate to the users how they might use something to minimize their costs and risk of using it, until they get educated.  It might be a process for determining if something works in the field or not.  [2-3 minute Muffled portion]
 
 P:  If you could take small different blocks and say, there’s ten different yield levels in there.  If you can take two and bring them up to a higher level, esp. in California with irrigated agriculture. R: As farmers go out and look at their lower yielding areas, they find that there’s not much they can do about raising that yield. It may be that the topsoil is most gone at that ridge and they’re better off going somewhere else.  The major benefit of some of our corn and soybean yield maps is that they find the areas that were higher yielding and apply more inputs at that place.

Protection of the Crop
 P:  [Another issues is] the idea of being able to protect certain situations in the field that the farmer can do something about -- whether it’s an insect infestation or a disease outbreak.  To find relationships between remotely sensed information and the onset of a problem where the farmer can attack through his management.  In other words, if he can see the outbreak of an aphid infestation in a cotton field that’s a localized hot spot, he could conceivably go in and treat that area and not have to treat the whole field and save a lot of money.
 
Reduction of Pesticide Use
 P:  With the environmental consciousness in California, people would like to see farmers apply as little pesticides as possible.  There’s a lot of pressure on farmers in California to decrease the use of pesticides.  Spot treatment of fields would help accomplish [that goal], decrease the cost of spraying for the farmer, and make the public happier.  SM:  What makes someone adopt remote sensing if they have this pressure that only allows them to do certain things?  They’re going to have to determine where to put their chemicals when they’re only allowed to put out a certain amount. R:  The number of regulations that are in the government for chemical use in CA is growing all the time, so it’s not getting easier for farmers to do things.

Lack of Research with Agronomists
 P:  I don’t think there are any high pressures.  I would disagree with the irrigation scheduling, I think most water scientists will too.  I think if operational algorithms means something that the grower can go out and use, then there’s nothing preventing them from being adopted (except that there hasn’t been any developed yet).  If I took a poll of all the growers in California that are using remote sensing and infrared to schedule their irrigation, I bet you’d [be able to count them on your fingers and toes].  The thing that strikes me is how little actual research has been done that has involved the actual agronomist in developing it.  There just hasn’t been any money for that. R: They’re saying experience in irrigation scheduling in Arizona using thermal infrared has worked there.  Also up in Washington, they did irrigation scheduling from flying thermal infrared. R:  That’s the exception.  That was years of research to develop that.  SM:  Absolutely, at this table there’s hundreds of years of research just sitting waiting for the same opportunity. R: But it also took a long time before there was a commercially available thermal scanner that you could fly and get the data to do it.  Now the devices are available.  [Muffled portion]
 
Sensors
 SM:  Could we say that one of the biggest impediments may be the sensors themselves? For example, a certain algorithm requires data every three days for 24-hour turn-around and 20 meter resolution.  All the requirements never seem to come together for farm management applications. R: There seems to be a discontinuity between research and design, R&D, and what the needs of the agriculture community are. R: None of the systems that have been launched to date have been designed to address a single focused application.  Spot was the first attempt at commercial application, but it was not focused at agriculture application, so your point is well taken. R:  If you want to address these issues, you have to see if there is a sufficient market to support a commercial enterprise that goes up and puts up a system. R: Some of the commercial companies that are starting to build satellites are putting bands on these sensors that are missing the most important bands that are needed for crop purposes -- middle infrared bands.  Some of the spectrometer work I’ve been doing is showing that in terms of predicting biomass and leaf area, the bands for Landsat are actually reducing the correlation between the spectral data and biomass and leaf area by as much as 50%, because of where they’ve positioned the cross-over regions and correlation.  They’re actually destroying the correlation, where if they were moved one direction or another, or just a bit narrower, they would greatly increase the correlations that we would get.  I think there is still some research that needs to be done on that.  R: You’re talking about their width, but their location isn’t too bad.  Those bands are that wide, because the technology that was used when they were first put up were continually widened until you had a good enough signal.  But certainly the commercial farms are just following on the heritage of that instrument.  I think the bands have been useful (although, there’s room for improvement).
 
 P:  Resource 21 is the only commercial company that has openly discussed their design and openly stated that their market is agriculture, and they do include the middle infrared bands in their design. [Muffled response] Initially, it was because geologists wanted to see relief and shadow, but it’s proven to be a good time because of cloud cover too, so there’s been a tradeoff between reduced shadow and cloud cover. R: Maybe the TRW that has 384 bands, they solved this problem and fought with agriculture ... [muffled] R: But those sensors are research and experimental, and they will not acquire the volume of data required to apply the data in operational sense.  To fulfill the applications that are being discussed here, there has not yet been a sensor or system directed towards those applications.  R: During the design phase, they had much finer resolution that expanded over time.  There’s probably a lot of agricultural applications that would find that 10-meter resolution is not much of an improvement over Spot. R: [Resource 21] will tell you they designed their resolution on the basis of their market research. R: They’re looking at the large volume crops rather than the intensive problems like we have here in California.
 
Cultural Acceptance of Remote Sensing Technology
 P:  If you’ve got thousands of dollars invested per acre, paying a little bit more for an image of that is almost an annoyance. R: The way to change [the cultural mindset] is to have some success stories out there where some respected people have used it and start to talk about it. R:  We need some pilot studies to test the process and develop the technique and see if its a possible tactic.  If it doesn’t save [the farmer] any money, then it doesn’t matter how good it is. R:  Another point is that often growers have ideas that we wouldn’t necessarily have thought of, so I think it’s very important to involve the grower for a variety of reasons.  [Muffled comments]
 
 P:  We need to connect the science results to commercial base applications for real end users.  [Muffled comment] I think that part is important, but I think the other thing is that NASA tends to pull back from something after 3-5 years.  In agriculture, they say it always take 15 years to get adoption of something.  Everyone’s watches the early adopters to see if they’re going to continue with it. I’m really interested in our better communication techniques that may help speed up agricultural [acceptance]. R: You go with what you’re sure of.  There’s already a lot of risk.  I’ve been dealing with my tenant [farmer in Nebraska] and talking about precision farming and use of remote sensing.  He and his son and farm 3400 acres.  They have seven different landlords, land in three different counties, and it’s all he can do to keep track of which piece of equipment is where.  He can’t add another layer of complexity onto his management system.  He needs somebody to comes along that can interpret [the data] for him and tell him “Do this and you’ll gain $X/acre.” R: Pilot studies are so potentially important, because they have the capability of removing a great deal of the risk aspect.  A regional pilot project, in which the farmers would have access, maybe funded in cooperation by NASA and the USDA. R: The senior management of NASA is gun-shy of staying anywhere too long because someone is [always pressuring them to do cutting edge research]. What happens is you need a longer cycle to do this technology transfer, and there’s a significant amount of real estate.  Now we see 5-8 firms looking at remote sensing, but we haven’t yet seen that happen.
 
 P:  There’s another issue in agriculture that extends the timeline -- it takes a year to grow a crop and typically it takes several years to actually discern an effect.  R: Maybe we should let other agencies carry out the pilot projects for longer periods of time.  I think that UC Extension is the perfect people for that.  Because the UCA scientists know the issues, and the extension person involved will also help with the distribution to make certain that you pick the good growers. [Muffled comment] [Response] You need somebody else in there as a gopher to make certain and from our operation, the most believable people are the extension. R: I’ve said for a long time that NASA needs the equivalent of an Agriculture Extension Service to take a stand on remote sensing. R:  Most of the spacecraft primarily are resident in aerospace and engineering departments. [Muffled comments]
 
 P:  There needs to be a partnership among agencies, like NASA and USDA working together for a number of years. R: If you want to include USDA, you should be talking to more than just the agriculture research service.  There’s more agencies within the USDA that fund agricultural research.  It used to be the Cooperative States Research Service.
 
 SM: We could put down spectral, temporal, radiometric, spatial, turn-around time. P: Bandwidth position.  SM:  There needs to be a strong emphasis on operational methods, not just proving a scientific concept -- applications. [Muffled comment]

Product and Dataset Validation
Who will provide the Data to the Farmers?
P: I don’t think the farmer himself is going to analyze remotely sensed data.  A dealer would supply that information in conjunction with a service.

P:  You usually work with the growers, but it’s often the crop consultants that end up actually implementing the [products], because they have an established relationship with the growers. R:  As we see more of the small farmers dropping out and the big companies coming in, then I can see the company running this farmland having someone hired to do image analysis, but I don’t see the “Mom and Pop” farmer doing this.  SM:  Maybe it’ll be the irrigation management district that hires a person on to provide information to people in the district. R:  It’s conceivable that we could have an university extension specialist in remote sensing. R: It could also be with a commercial company -- Farmland Industry is trying to hire a remote-sensing specialist right now.  R:  Most farmers in the Midwest would probably rely on other services -- crop services, coops, etc.  I’ve seen remote sensing companies bring it in with the coop and 2-3 people are GIS and remote sensing experts. [Rest of the response is muffled] R: My experience is that farmers don’t want to hear about the sensors or pixels, they just want the meaning behind the images.  You do have exceptions, and you’ll probably have more exceptions in the future.

P: Fifteen years ago in the Midwest, everyone fed information through the extension service to the farmer.  Now everybody is giving advice to the farmer and sometimes it’s conflicting.  The farmers have to integrate the information from all these different groups -- that’s what I call the information multipliers.  You need to think about who provides the information to that farmer and what impact they really have on [the farmer’s] management decisions. R: The problem with it is that if NASA starts dealing with the chemical dealer A but not the fertilizer dealer B, you going to have much more trouble than with an extension agent from the university. R: NASA just has to recognize that agriculture is really complex and it has so many different sides to it.  In the Midwest, there are farm cooperatives that sell the feed, seed fertilizer, etc. and I think these are possibilities that NASA should think about in addition to the extension service.  Mike Cook, is a distinguished professor in cooperatives at the University of Missouri says that a lot of cooperatives are going back to forming neighborhood groups, almost like the Coinnia[?] groups in churches today.  The purpose of these neighborhood groups is that farmers talk together and share information that they know.  Farmers learn best from other farmers and get their incentive to do things from other farmers. R: We’ve had someone come out from Australia who wants to form users groups for whatever product.  It’s very effective.  SM:  I think that relates to how the scientists take their science to the user.  The user gets in at the beginning, even at the conception of the idea to do the research.

SM: You’ve got two separate things: 1) data validation or data quality and 2) product validation. R: If I develop a model between remote sensing and crop yield, I could certainly validate that relationship in a scientific sense by doing a validation study which would look at a large number of fields.

Calibration of the sensor vs. uncalibrated data
 P:  There was a statement made this morning that satellites weren’t going up with calibration systems. R: The commercial system just wanted to sell pictures, but you can’t use [pictures] for modeling purposes if you don’t calibrate [the satellite]. R: Without calibration, you won’t be able to look at qualitative variations within a given image.  You won’t be able to compare multi-temporal [muffled]. R: Stoney laid out the commercial systems.  There’s Resource 21 and there’s everybody else.  Everybody else is basically taking a high resolution, panchromatic image and throwing in some multi-spectral bands, but I think they’re counting on their market being a high resolution panchromatic. R: GIS people are also using this as the backdrop and digitizing. R: None of these systems, with the exception of this new firm (AIE), were designed specifically with agriculture in mind.  Resource 21 and NASA has had an informal dialog.  We have told them that calibration is important, but they have some data processing methods in mind, where they feel they don’t need to do absolute calibration. I know they’re doing relative calibration detector to detector in their system -- if they don’t they end up with a striped image that’s not aesthetically pleasing.  I don’t know what they’re doing temporally.  They seemed to have gained a greater appreciation in their brief conversations with us [for calibrations], but they did not set out with calibrations as a high priority. R:  We’ve had a high priority for calibration. R:  I think the importance of calibration also depends on the application.  If you’re doing a single farmer’s field where you’re interested in the relative values of the data, then the calibration is less important.  I don’t think uncalibrated data is worthless. R: How could you compare, in a relative sense, seeds acquired on different days if not there’s some common calibration?  The digital counts on any part of the field could be different, simply because of radiance, characteristics, etc. R: There’s the absolute calibration of the signal and the variability associated with the signal.
 
Temporal and Spatial Variability of the Crop
 P:  I think the variability of the crop on a day to day basis as being more [of a problem].  The calibration is not at the sensor but at the crop’s response to the radiation.  The wind is blowing when it comes over.  Temporal and spatial variability. R: I’m speculating that maybe they’re tracking NDVI over the growing season and using that as an indicator.  They don’t really care about the absolute calibration, they only care about that their system is stable over the season, because they’re looking for trends and statistical correlations. R: One of the Resource 21 products that they delivered was a change image between the current image and previous image. R:  Change can be caused by atmospheric changes. R: NDVI, for example, hides a multitude of sins, and a multitude of information, but it’s utilized.
 
Summary of Larry Biehl’s Session
[Larry Biehl was the other breakout session coordinator.  His group joined with Susan’s group and he leads the rest of the discussion.]
LB:  I’m Larry Biehl.  I’m from Purdue and I’ve been involved with LARS remote sensing since the mid 1970’s, so I’ve been involved in the LACIE program and the field measurements, and the work with NASA, North Dakota, and our agronomy farm in Purdue and LACIE AgRISTARS.

[Everyone introduces themselves -- Steve Moss, USDA; Chris Johannsen, Director of LARS at Purdue; Kevin Price, University of Kansas; Marvin Brower, University of Minnesota; Susan Moran, USDA; {Two muffled introductions} Wes Wallender, UCD; Allen Hoop, San Diego State University; Mike McDonald, TRW.]

LB: In the other session, there were only two people, so I thought we’d go over what we discussed at that session. We had one person from Maze[?] and one person was a vineyard grower.  The first question I proposed was, what are the products we’re talking about?  We discussed that we were going to have analysis from within the fields to the large areas.  Different parts of analysis are going to be done.

The goal for the vineyard grower was for uniformity to get the best balance.  He had some specific needs that remote sensing may or may not be able to address.
1. Monitor yields.
2. Map diseased areas.
3. Locate water and nutrient stress.
4. Why taste varies across the vineyard, even though they’re all the same variety of grape.
5. Verification of field notes.
6. Help control of high pH in vines.
7. Biomass estimates and creating reports for carrying capacities.
8. Which coordinate systems are the best?  He’s tried to put the different GIS layers together from different sources, and everyone’s got it from a different projection, but he doesn’t know which is the best one or how to determine it.  We’re talking about technology transfer and education.  The standards comes out in our book from the Farm Bureau Association.

P:  At the farm level, wouldn’t he be more interested in the coordinate system that’s going to give him the detail that he wants?  LB:  The recommendation was that you want something that’s going to be [muffled] equal area.  Are there any standard products that we could define that several companies would produce?  Is that possible or not?  So that’s a quick summary of the issues discussed at the first section.

Sensor and System Design - Delivery and Spatial Resolution
LB:  What products need real-time delivery, which only need occasional, sporadic delivery?  P:  For commodities operations, you want occasional, weekly or bi-weekly evaluations, so you can get a handle on what commodities are happening.  The guy from agrometrics had a good point that (and we’ve seen this at TRW), if you have a problem related in the field to a pest vector, you may catch that on a weekly flight, but if you’re going to mediate that with an application of pesticide at that location, you want images every two days after that.  [At the field level], there’s a temporal requirement for certain applications where there is a threat to the field by pests or disease.  Once the problem has been eliminated, then you can go back to the weekly process. R:  If this gets to a timeline of acquisition, then it also gets to a timeline of delivery, and I think everyone wants the product within 24 hours.  SM:  I think there’s some applications at the field level where you don’t need the data and you don’t need it within 24 hours. R:  At the beginning of the year, you’re right.  LB:  It depends on the user.  At the other session, they said that better soil maps are needed at the field level.  But that is something that once you get it, it’s not going to change rapidly. R:  We did some test field where we flew the fields before the season, did a soil color evaluation, then flew the fields during the whole season.  We also did a gridded soil map of the area and then  the soil firm came up with the contours.  We looked at the emergence path which is our base map (called the Soil Productivity Base Layer).  You would expect that the areas emerging late to be deficient in some nutrients, and it didn’t seem like there were correlation with the gridded soil contours in that field.  LB:  There’s some research issues that could be in NASA’s domain to help out.

Hyperspectral.
 P:  I’m curious, what is hyperspectral data going to give us in agriculture?  Except a lot of data bands. R: I know that some of the attack has been used in the same approach to using geology.  We have our database that’s available on line...
 
Monostatic and Bistatic Radar.
 P:[Muffled] NASA ... with the ultimate goal to have a bistatic radar in space.  You have one satellite with radar and more than one with receivers.  The bistatic is one of the hottest topics with the DMV.  When you get to the shadow side, you’re getting bistatic optical because you’re getting the transparence -- you’re seeing the greenness that’s coming through the leaf. R: This is something that’s probably not going to pay off for 20 years.  We’re going to be in a situation where, for example, if you want to look at methane worldwide, use a bistatic radar.  You’ll get radar glitter from all the light around the world, as well as weapons.  We can do that with optical data.  We’re looking at wetlands, but we’re not able to look at them in equatorial regions because of cloud cover.  The same way I think we should use radar in the long term for agriculture because many of the crop growing areas in the world (the ones that aren’t irrigated) are often covered by clouds, simply because that’s what provides [the water].  It behooves us to in the long term, figure out what’s going on.  So monostatic and bistatic radar. R:  I was against using radar coming in the room, and now you’re swaying me. R: Bistatic is in the long term.  We can get pay off in maybe 10 years from monostatic. R:  What I’d like to see is “Wild” Bill’s cloud chart corrected for the growing season.  Then I would like to see the vignetting analysis.  Let’s take all of the crops in the Central Valley and find out how many of those are periodic in a bad way for the resolution of the sensors that are coming out.  I don’t think any of the satellite companies have thought of any of that stuff. R: I don’t think [satellite companies] realize the magnitude of the problem.
 
Data Processing Problems, Resolution and Moray Patterns
 P:  When you talk about the spatial resolution data, are you also including the data processing etc.?  You don’t want wall-to-wall 1-meter data.  Do you insert it at certain locations within the existing coarser spatial resolution? R: I don’t have an answer, but I think there’s a technological solution out there as well as an informational processing solution. R: Look at the total problem -- not just at the gathering of the data, also the analysis and everything else.  R: You said that grapes and raisins give you a moray pattern at 2 meters. It may be that those patterns are frequencies that are just going to hose some people. R:  What spatial resolution do you need to manage the field?  Do you need to see every grape arbor or see every artichoke?  There’s a real easy way to get rid of the moray pattern...  So you have a lower resolution, but you still have the information you need at that resolution.  Just because you can’t resolve every row, doesn’t mean that there isn’t information that spectral data ... R: The guy from the vineyard wanted to find the area around the vine.

End-Users and Data Products
End-Users
P:  We assumed an end user, and I think that’s correct for the majority of problems, but in the agribusiness community, there’s other users.  Is the grower for remote sensing purposes the user 70% of the time? 90%, 50%?  For example, I’ve had people come to me from Smith-Barney who are interested in customers decisions based on using satellites to predict anything from disaster assessment, frost prediction, etc. R: The answer depends on where you are.  In the upper Midwest where you have corn, soybean and wheat systems in these big commodity groups and companies -- there you would have very different end users.  In the West and Southeast, you have different crops that are less associated with big organizations.  SM:  The speakers this morning were from commercial companies and were concentrating on commodities as much as they want [to escape it].  They mentioned commodities, insurance companies and then they said they were trying to put together a decision support system for the farmer. R: There are crop scouts that drive all over the U.S. with maps, so we believe there is a market out there for more than just a field [grower] level. R: One person took one of our products and made $10,000 in one day, because he was about to bet on the future and figure out what was going to happen when the early freeze came through.  He was a meteorologist also, so he saw the freeze coming in and knew which crop was going to be wiped out because it was planted late and was able to use that information to his advantage.  R: The transportation industry wants to know where and when are the crops coming along, so they know when to move trains and trucks to move into the area.  If they send it to the wrong place and the train sits on the tracks for a week, it costs them a lot of money. R:  Most types of application of the remote sensing requirements for those [predictions] are completely different from precision agriculture. R: If you talk with the Cases[?] and the John Deeres, prior to harvest season, they want to know where the wheat is looking better than normal.  That’s the location where they start stockpiling their combines for their fields.  So there’s also that kind of information out there with real applications and real money. R: On the large scale, though, the USDA is going to be putting out these crop estimates.  The goal then is to project which way they’re going to change ahead of time.  Just guessing what’s going to be put on that USDA report and get it out to their customers, before the report comes out, so that everybody’s already done their positioning in the market. R: [International agriculture forecasts] Farmers want to know what to do with next years crop by December 31.  You’re looking at South America and other places that are raising the crops, how do you think it’s going to come out there? R:  By August 15, some companies have to know whether or not to release the contracts from South America.  If they make a mistake and release the contracts, it could cost them millions of dollars if there isn’t going to be enough seed produced here in America for the next year for growing seed.

SM:  Can we as a group see applications for remote sensing for anyone interested in agriculture? [Laughter] Yes we can. There seems to be a lot of applications and a lot of different uses R:  I had mass employees coming to me wanting to see the maps we’ve got because they play the future’s market. [Laughter] R: We have cotton producers in the San Joaquin Valley that have asked me about setting up a technique for estimating or predicting the cotton production in the valley just for that reason.  SM:  I can presume they could use it for harvest dates (farm management). R: A lot of information that can affect how much money they make is not really dependent upon actually how much their crop is producing, it’s [the farmer’s] production relative to what’s happening somewhere else. R: And they don’t trust a lot of the information that’s out there.  Government included.

How can Communication Technology be utilized to Provide [Products to Users]?
P:  One thing that struck me this morning is that we had someone driving 80 miles to the farthest region because they didn’t have computers? R: That’s right.  We were going to buy them computers.  Their kids will use it, they’ll keep it, get used to it and then buy the data.  Our distribution plan initially was hard copy, and with 24-hour service, it’s hard to get distribution through the mail, so you have to do it yourself.  The other option was to have modems at computers where you dial the information up and download it to their computer -- to have a Web site with [user codes], so you can get a picture of the field you want to look at any time of the day.  Then there was special dedicated communications lines, local private wireless networks. R: A lot of people (like 150,000 subscribers) to ETM[?] and broadcast partners that send the data by way of communications satellites to farmers that have satellite dishes and then they bring it in on a TV antennae and put it on a TV screen.  Data comes up on a minute by minute basis. R:  We had a great idea for The Farm Channel, so you get it on your regular TV, you contract for community access from 3:30 -4:00 in the morning.  The [farmer] can set his VCR and you have a host like a standard TV show.  Now you don’t want to let your neighbors know that you have late blight because they’ll sue the heck out of you if you don’t do something, but there’s a lot of pathways to get it out there.  We could just put these films up with a code number.

Research Issues Dealing with Communications Technology
P:  [You need] image standards, formats, and compression standards, if you’re going to put it over any communications line.  When it comes out the other side, that’s largely the commercial people who will figure that out, if you’re using jpeg compression, it’s going to be unjpegged at the other side.  If you go to Netscape and say, can you make a Java apple-it which will display all these things in a RGB format instead of a color look-up data thing.  They’ll handle all the communication for you, if you have a reasonable customer base for it. R:  And the sub-areas are okay, because I know where my brother farms there’s 4 phone companies in one county.  You can’t call five miles without being a long distant phone call. R: That’s a problem when you have a 15x15 image that coming down and your kids want to use the phone.  So our solution was that all that happens at night. R: Our solution in the other room was to have an intermediary or consultant or a coop person.  If your going to do that, have it at the coop office. R:  The coops are actually wanting this, because they want to provide a service to their users that says we’re doing something for [growers].  SM:  One experience we had related to this -- we flew a farmer’s farm and came back with the images.  The farmer needed the information immediately, so we’d process the images and the farmer would come in and videotape our presentation.  He’d take that tape over to his farm which was an hour away, and have his growers watch that videotape and he’d videotape their session, so we could hear the grower’s comments on our comments.  It worked quite well because the feedback went in both directions.

Raw vs. Processed Data
 SM:  We identified three levels [of data] that seem to have potential:  1) direct raw image to a farmer 2) final product such as apply pesticide to this part of the field today, or 3) there’s a value added organization or service support where a raw image goes to that agency and then goes to the farmer or user.  LB:  The other discussion came up with similar concepts.  The agrometric model, which also seems to work well for them, is “Here’s your finished enhanced product.”
 
Problems with Classifying Images
 P:  Providing a classified image to the grower is a mistake.  In our processes, image-to-image normalization and, day-to-day normalization is difficult.  When you use a classification process, it has to be and needs to be automated.  Otherwise it’s subjective and then there’s no continuity between day-to-day or even frame-to-frame sometimes.  Variation is caused by sensor calibration differences, by illumination differences, angles of view differences, atmospheric column characteristics etc.  That causes images taken over the same field to have differences in illumination for the exact same field taken 5 minutes apart.  Something taken at an offnater[?], which is the worst, or from a higher altitude, is going to have a different population.  If you use an automated algorithm, and you don’t normalize those things extremely well, the center point where you’re dividing spectral space up into classifications in N space, those boundaries shift.  A very slight change in a boundary can completely change the visual appearance of a classified image.  We found this out and [recognized the problem] that we’re not calibrated to it.  Use an automatic hypercube classifier that’s a function of X, Y position on the screen.  All the image-processing packages have them.  Those signatures weren’t stable enough, because what we wanted to do is find really find differences in the field, and they were within the noise of those lack of calibration from day to day.  If we use three things, all the information is lost and this variable effect is very strong.  If we use ten, the variation still throws it between classes, but if we color code the classes reasonably close, then visually it’s not so bad.  So we adopted a smooth continuous, maybe 200 different colors, which is a very small rate of going from the green to red in a nice color wedge.  It allowed us to do our normalization as best we could.  In presenting it, the binning[?] or classification of the images wasn’t done to such an extent that the minor variations made green turn to yellow.  It made green turn to a yellower green.  We’re basically pseudocoloring the image, but we’re doing it in an extremely fine pseudocolor pattern which overcomes some of the problems we have in this registration which is almost the same as not classifying it at all.
 
 P:  Don’t you have trouble selling as one class as one versus another, because of color differences? R: Exactly.  And we didn’t give them statistics on classes.  On the side we had a bar which said, the top 20% is very healthy vegetation, here is good vegetation, and it wasn’t a linear mapping.  We had a function that we looked at the spectral characteristics of the fields.  They would look at the field data from this week’s and last week’s, and the variation would not be an artificial function of classification assignments in this look-up table.  It’s an artifact of taking spectral classes and putting them in information classes. That’s the classic problem.  They don’t fit.  It’s very difficult to do that, because the visual boundaries are high contrast.  In this case, they kind of do [fit] because we’re doing NDVI, and that’s going to give us the robustness of an EIR signature, which is generally related to plant vigor. [Muffled Response]
 
 P:  If you drive down any row crop field and look at the sunny side you’ll see a bunch of dusty pants.  If you drive down the non-sunny side and look toward the sun, the plants will look darker and greener. R: It’s even more complicated than that.  If you look straight down at a soybean field, you’re looking at soil plus vegetation.  If you’re looking at it offnater, you’re looking at all vegetation. R: And that could be a better image. R:  With Landsat we looked at two images that overlapped sagebrush and looked at it.  [We found] major differences in between the brightness values. R: For aircraft, this gets us into debates.  Kevin Spry has the whole crew on oxygen at 14,000 feet.  He can only stay up there for 1/2 hour without oxygen, so everyone’s hooked up.  I said, “Why are you doing this?  You’re going to have this thick dirty atmospheric column in between you and the ground.”  It basically comes down to the way it’s lensed.  But what it does do is it gets a more vertical view on all of the rows if you don’t have complete canopy.  We’re flying at 5,000 feet, looking out of a 45 degree angle, and that’s the middle of the two adjacent fields, so we’re a little bit less than that from border to border and so that affects the spectral difference across that look with respect to the illumination.  [Muffled response] R: That might be something for NASA to fund, because that’s a bear of a problem to understand.  They flew the GER sensor for us, what data we looked at had this severe angle of illumination problem, and we had to come up with a function to correct it for each image we worked with.
 
 P:  How well will it solve the problem of trying to extract the soil component or vegetation component? R:  Vegetation versus bare dirt is one of the best signature differences around, but you have to process them both out, and you have to have multi-spectral to do that. R: It seems to me that we haven’t completely solved that problem in terms of being able to say, “This is definitely a soils component, this is a vegetation component.”  There’s still some bleeding between the components. R:  There’s also a complicated factor  -- you’ve got sunlit soil and shadows. R: If that’s a mixed pixel -- if all that happens sub-resolution, that’s even harder.  So the coarser your resolution you have, the more mixture you have. R:  That’s a good question -- how good are these mixture models at separating these components? [Muffled response]  If you have a canopy for one part [muffled].  If on the other hand you have four fields, and a pixel is a portion of each of those four fields, then unmix that situation, because it’s linear.  People have had global forest canopies, for example, with this in mind and they’ve shown that [muffled]. R: When I was working with PJ[?], the shadow effect was actually creating an inverse relationship between near infrared and the camera, because shadow with the sun angle was 2/3 larger than the canopy itself.  So the shadow had a bigger effect than the trees did.  It inversed the relationship until the tree reached 50% cover when the relationship flip-flopped. R:  If you were going to choose to do a classification on your image, are you going to wait until you’re sure you have 100% canopy cover?  If you don’t and image-mix pixels, you have to try to evaluate a mixed pixel of dirt, roughage, shadows and illuminated dirt, etc. so you have this mixed spectrum coming up at you.  You want to be able to say that the addition or subtraction of a certain near infrared component as a percentage of this total is a function of stress rather than of more bare ground.  So I would say don’t even think of classifying that image.  Leave it as raw until you have canopy cover, and now you have hopefully, all the same homogenous material.  Provide them as a raw image which shows you that pixel A is a reflectance in red, green and blue, etc.  Just interpret it as a photo.  Now the NDVI gave you some information in a ratio, but I imagine that that would also be an error too.

 LB:  What is the list of products?  The yield map, the NDVI?  Maybe we’re not at the stage to define that very well. R:  We came up with five products, and our motto was “Some will get shot down, and then the farmers will tell you what the products are.”  So each of these things [that are developed] will lead to a niche mini-business application.  My view is that there are hundreds of those little niches.  LB:  Even before the product, what are the questions that the producers/growers have?  Then we help define the products that can answer these questions. R:  The people are interested in what’s happening at the national level as well.  Not all of the agricultural people are going to be interested in what’s happening at the field level. R:  The reason we’re thinking about Internet is because we want an interactive system that brings information back so we can modify it or ideally trigger some discussion among the farmers so that multiplies and comes back to you.  That’s what’s wrong with the idea of the TV dish.  You can deliver data but they can’t give feedback. R: They call up a 1-800 number and talk to the people on the phone. R:  With the new Netscape version that’s out, you can download images or video clips, you can have a video teleconference with somebody on the other side -- audio and text at the same time, wait 6 months and those types of communications are going to get better.  We used this in a proposal to the Defense Mapping Agency, who has a similar data distribution problem -- they have billions of images, products and map/layer components and they want to put all of it on a web server in a distributed relational database so Mr. Marine in the foxhole can download an air target chart.  They’re making these large relational databases to patch these things over, and industries are going to keep pushing the communications technology.  As part of that, we set up a concept that there would be an operator [that’s on-line with you].  That’s driven by the commercial world now.  You could actually do a consultation over the Internet, given that there’s some infrastructure to get those lines in. R: One thing we need to realize is, if one channel does it to the precision that’s needed then you don’t need to spend a heck of a lot more on it.  Sometimes, there’s overkill on what’s needed.

Potential Research Areas
Crop Cycles over the Growing Season
 P:  I’m thinking in terms of research areas.  I think that a lot more work needs to be done in the area of [?] and how crops are changing over the growing season.  When do they green up and for how long?  When do they peak in their greenness?  How long does it take from the time they peak until they reach maturity and try to relate those things to yield and the way a crop is performing. R:  And also [relate] the critical times as designated by the emergence state and the greenness clicanic[?] closure time and the critical time when you need to look at that crop so it doesn’t get damaged.  For potatoes, the farmers are far less worried late in the season because the potatoes have finished bulking.  When they’re emerging sprouts, they want [know any stress] at that time, because the rest of the life of the plant is shortened. R:  You’re talking very high-resolution data [to read plants when they’re emerging]. R: Nothing I’ve seen has shown that multi-temporal analysis has been incorporated into something we’re using now on a common basis.  One reason is because we couldn’t afford the data, but with the kind of data that will be available in the future, the price of it begs to go back to some of the problems we started looking at.
 
Research in Crop Models
 P:  One of the possible solutions would be to use remotely sensed data in crop growth simulation models, because that provides the other tracking of the crop growth and where it is in the season based on climatic conditions.  We can continue to develop into those models and interface with remote sensing and crop simulation models. R:  Using a crop simulation model, maybe you can predict what the crop state should be and see what the correlations are between the simulation and what you’re getting from satellite imagery. R:  The model could also give you some idea of the canopy plant size.  That could [help ameliorate] the shadow problems.
 
Temporal Research and Analysis -- Crop Time vs. Calendar Time
 P:  One of the difficulties in the temporal analysis is you can do that by date, which is what we did week by week.  The guy would flip through a notebook or click through it on a computer (watch the crop mature).  That’s calendar time, but crop time needs to be overlain on that.  It’s October, but when did the crop get put in?  The only metrics that I’ve identified is emergence (which is variable across the field) and 100% canopy closure (which occurs at different times in different fields).  Beyond that I don’t think there’s any other time -- well flowering.  [For one emergence model, they use the emergence of a moth that only appears] after a sufficient number of degree-days and it’s correlated with one of the crops. R:  Fruit setting stages, somewhere in the growing period when the nitrogen distribution goes from the plant growth to the fruit development. R: That’s hard to do with remote sensing. R:  You’d have to make up a special system to do that.  Technology would have to increase, because when you go down to a very low altitude to increase, with a restricted field of view, your pixel ground speeds overcome in the re-ferentiated[?] camera or the integration time of the detectors, you get blurred. [Muffled Response]  There are two parameters plus emergence that define the shape of this curve.  I’m only assuming in this instance for corn and soybeans.  The power is that you can look at any one of these parameters and get all the information for the season.  [So you can analyze and compare fields].  SM:  I think an extension on that is to use the physical model to define and refine the model using remotely sensed data.  Then you can predict. [Muffled responses]  LB: It’s really trying to get the models to an operational effect basis.
 
Models Give Information on Canopy Structure
 P:  Another area involving the models that could provide some possible alleviation of this problem of the small canopy size is actually have the model tell you something about canopy architecture so that you can a radiometric model that can actually take into account shadows, etc.  You might have information from the grower about the size of the crop that you could use in the model to characterize the size of the canopy.  If you know your image acquisition time, you can at least do some type of estimation of sunlit ground, shadowed ground, and build that into a scene-reflected section.  It’s a lot more work... R: It’s getting really complicated, but the one thing you need to remember is that the bumblebee still flies.  No matter what angle you take the picture at, no matter what you do to that thing, whether film or digital, you still get a picture that is pretty useful to the grower.  It just limits the fancy high-tech stuff that you can do to it. R:  If you can find a way to extract that information you need out of [the image]. R:  The same information may be in two images, and we may not be able to get at it. R: So what you’re saying is a lot of times, you can see the differences, you just can’t explain it yet.
 
Determinant vs. Non-determinant crops
 P:  I would like a list of all the crops that are determinant and not determinant, so I wouldn’t waste any time thinking or talking about the possibilities of a yield measurement in sugar beets or potatoes (non-determinant crops).  Because right now, you can have a healthy plant on the top, and 1) a big sugar beet with no sugar content or 2) a small sugar beet with acceptable sugar content that’s too small.  That’s the problem -- you can’t just a book by it’s cover for certain plants.  I would like to focus on those, because there may be a way to cracking that puzzle or not trying to apply that type of technology to those. R:  The savings you may have from nitrogen management in sugar beets, is anywhere from $2200.  Sugar beets are very sensitive to nitrogen management.  That may be one of the advantage of precision farming over coops. R:  W e would have to study the non-visual signatures for that to see if we could get any clues to [non-determinant crops].
 
Proposal of Radar Use in Agriculture
 P:  You can use something like ground-penetration radar.  I’m not promoting doing that from remote sensing.  One of the values would be to potatoes.  There’s a system mounted on the back of a truck. R: You could look at the soil to find organic matter.  R: We tried to look at soil color and find a relationship to organic matter, but we ran into problems with wet soil vs. dry soil and couldn’t find a correlation.  I have worked radar for a long time, and I’m not a proponent of using radar in general for  agriculture.  We used everything from large regional mapping radar to high-resolution radar, multi-polarization multi-spectral.  We have a wavelength, what modifications can we use it for?  We had some [success using radar to determine the maturity of] rice crops in the Central Valley.  [However] a lot of times with row crops it’s difficult to differentiate row plants between each other in radar signatures.   You have to have pretty high resolution to do that.  It’s a very non-intuitive type of interpretation scale at high resolution from aircraft.  You’re also getting pings and bounces from everything else that is on the field and so it’s a very dirty data type in my mind. R: You could look at it in another way in that this type of radar image could be very useful in agriculture, because radar is probably the only surface that is possibly flat. I think radar is probably best used in agriculture, because we can use radar to look at crop soil moisture [muffled] and it works quite well in a temporal oasis[?].  I’m not saying it’s perfect, but radar should be included [in research].  R: If we’re looking at soil moisture, the cheapest solution to soil moisture in a field is a soil moisture block in a represented location in the field.  Maybe you can’t get into the center of a cotton field, but you’re not going to get much radar from it, either. R: There just hasn’t been much work done with multi-frequency, multi-pull, looking at different incidence angles all at the same time.  [In the past] it’s been one frequency, one polarization on a study.  [Research on radar has] been very fragmented and it’s hard to really pull together and get an understanding of what’s possible with radar. R:  We combined multi-polarization, and multi-frequency radar with multi-spectral data, and we found out that combining these two did not give better classification than just EL.  It did allow us to look at rice fields better, because of the resolution of the multi-spectral doesn’t quite pick up those plants, so there are applications for it.
 
Soilection and Remote Sensing vs. Ground-Based Sensor
 LB:  A comment was made that all related agriculture and remote sensing is still [based on] the research of ground-based systems.  P:  I was talking to Soil Tech that has their soilection process, which is on the Golden Soil Probe.  They’re using a single spectral band to try and define organic matter content on that [probe].  I told them to talk to anybody who has a hyperspectral sensor, plug that in the hand-held fiberoptics and stick it in the blade ... R: It won’t hold up on a tractor. R: Correct, but Lockheed can make a little tractor-built thing that will make it more rugged for a couple million.  But the idea being, if you have 256 channels and you’re looking at a controlled spectrum light source, you can do that too.  You’re going to get water, but I would think with hyperspectral that you should be able to get the chemical components or at least the organic matter. R: Soilection [muffled]visible to get the organic matter and there’s also an NRI based system to do the same thing, and they are working well.  But do all the other chemicals, there is much that can be done easily.  You don’t need the fiberoptics [on your tractor].  There are other ways of doing that. R: And you can’t [measure the organic matter in] the field with a tractor once the crops are in the ground. R:  That’s a question of how many times you’re going to go through the field.  If you have to go through the field every time, that also costs you because the operating costs [add up].
 
Recommendations
Start providing products.
 P: My perception of remote sensing is that it never quite connects with agriculture.  You never actually get something to the end user and work with them.  You do one tiny thing, and you keep stepping up the technology.  With the research, the end user gets left behind again, so we haven’t connected at all with the 100 farms that are using remote sensing data.  So now we’re talking about bi-static radar multi-spectral and multi-polarization, they won’t get the data [from all the research] for another 50 years.  [NASA] talks about the applications they want, but they don’t put any money into it.  I’d like to see them invest some really money into it over time, but if they get this [discussion], they might leap [onto putting up] a new radar [system].  The way to make this connection with agriculture [and remote sensing] is start providing [products] and keep working with [users] to improve [the products], rather than getting five steps ahead without filling in this back portion.

Fund pilot projects.
NASA could help fund pilot studies with emphasis on operational methods, applications and implementation, rather than studies to develop the science, per se.

Involve growers in pilot projects and research.
Growers should be involved in the pilot projects and at research plants, so the minute we get an idea or conception, we should get together to focus the direction of the research. R: They ought to be the ones to draw up with the questions.

Partnership between agencies.
 SM: There should be a partnership between the agencies including funding and conducting of research to continue it on for 15 years to allow for this reality.  Agriculture has a 15-year adoption rate, and NASA is often restricted to three-year funding. R: We really need to take a closer look at the easier adoption of things, and that with current communication systems, how could we cut that 15 years in half? R: What are the reasons that the yield monitors became so accepted in such a short period of time? R: They saw an application. R: The soil moisture probes that you just stick in, get the little number, went in right away because they saw the utility right away, they were easy to use and cheap.
 
LB:  Let’s come to some key points.
1. In pilot studies, it was important to direct [application] toward the user.
2. The temporal issue was very important, from the user’s standpoint.  It’s also to get him the information that he wants.
3. The spatial resolution is a third issue that could have impact in the near term.
4. Integrating radar.

Concentrate of economically important and geographically widespread problems.
P:  Agriculture covers such a broad spectrum.  In a conference lecture years ago on commercial radar, a lot of the applications that came out were mapping fault lines and topographic mapping that was one-time only.  The radar people said what are the applications that require images every few days?  They said there were no applications.  So we should come back and say there are these needs for something in a general sense. R: I’ve been hearing things about vineyards, lettuce and cabbage.  It that as important a topic economically or geographically as corn, soybeans, the winter wheat?  If [NASA] only has limited dollars, how do they decide how to place ...?   LB:  They may want to lean on the commercial companies.  They’ve gone through the market analysis. R: There’s two ways to look at those numbers.  One is total volume and you pick corn and soybeans.  But if you’re hitting a narrow swath with a satellite, and you’re only hitting so much real estate, then you look at how much are they netting per acre.  Then you’re into small stuff, vineyards and those things.  The high-resolution commercial people have done the market analysis, but they’ve only done it from their perspective. R:  That goes back to integrating the spatial resolution. R: That ought to be done. R:  They’re also looking at urban applications.

Role of Government in Agriculture
NASA help develop the applications and help create the market.
 P:  Is there a perception that the government (NASA) has a role to play in agriculture and remote sensing?  Private companies visit NASA headquarters regularly and say, “Remote sensing in agriculture, get out of it.  It’s a mature market, we’ve got it, we know what to do with it and you guys go do something else.” R: They say that because they’re starting to provide data, and they don’t want to compete with NASA.  This is one of things I remember from a commercial meeting held in Washington that they would really like to see. R: Part of agriculture is mandated by the progress to produce crop yield estimate.  [In the past] that was seen as an important part of the economy.  Of course a lot of private companies do that also. R:  That’s an important one, because there’s a few people who make major decisions that affect millions of dollars based on business by information.
 
 P: [Companies] can launch a stable sensor, they can take data, classify it, and tell the farmer what’s wrong with the field in the simplest way.  But as soon as they start doing that for a couple of years, the company is going to need to realize the need for follow-up on it.  SM:  In Dan’s talk this morning, his company was going to take all the space imagery and start value adding and start providing it to all the commodities and insurance companies and all these people in the format that they want it. R: I wasn’t impressed with the depth of his knowledge of remote sensing.  It was superficial. SM:  After two years of providing this data, they need someone to come in that says, “I know these groups and I’m going to take the data out and make it into the product that I know they want.”  Already, I think that has started. R: Maybe NASA does the initial work or the university, but at some point (year 3), [NASA should] take any firm who wants to come in or select additional people that will work for a few more years.  In the end, [private firms will] have to [supply the products].  They don’t want the government supplying it.  [Private firms] want to be included and I think that’s valid.  Just make it scheduled.
 
Technology transfer, pilot projects, application development.
R: I think here it’s important to distinguish between agriculture and agribusiness, because I suspect that remote sensing applications will be much more quickly adopted in agribusiness.  First because the technical issues are not quite as difficult and also the sums of money are a lot bigger and the number of players are a lot smaller.  If you look at actually applying a direct benefit to the individual farmers, that traditionally is where the government has had to play a role.  Even with the large farms of today, farmers are not large enough individually to actually affect the process. I think as a farmer they would look at all the commodity markets and if it works and there is perfect knowledge all over the world, all it’s going to do raise the cost.  Now they have to buy all this extra information.  To me, that’s a real negative.  Farmers are still regarded as being a small industry and less able to deal with these issues than an ADM or a commodities or some of these bigger players.  SM:  The second is actually an agency (cooperative extension) to work with whoever that industry or farmer is.