Proceedings of the Workshop on Remote Sensing for Agriculture in the 21st CenturySession 5&6: How Can Agricultural and/or Ecological Models be Combined with Remote Sensing Datasets to Develop New Sustainable Agriculture or Other Environmental Resource Monitoring Programs? |
|
Introduction Uses
of Remote Sensing Defining
Sustainable Agriculture and Approach Justification
of Remote Sensing Agricultural
and Ecological Models Summary
of TopicsIntroduction
MZ: Goals of this discussion:
1. What are the issues of sustainable agriculture?
2. What are the issues of the environmental monitoring programs?
3. What are the data requirements of existing models?
Representatives from the remote sensing industry can tell us what remote sensing can offer in terms of acquiring these datasets for use with the models.
Uses of Remote Sensing
Determination of Evapotranspiration
P: [Remote sensing] hasn’t been used very widely in California, because
[the effort to estimate evapotranspiration (ET)] failed miserably in the
1980’s and turned a lot of people off. I think it is possible to use remote
sensing for ET, if you take the right approach. One of the biggest
problems is determining the percent of canopy cover. The crop coefficient
that we use to estimate ET from a reference value is based on what percentage
of land is covered by plant material. The problem with determining the
percentage of canopy cover with photographs is shadows. We can’t
differentiate the plant material from the shadows, so we have to trace
them by hand and figure out the percentage. That’s an ideal [application]
for remote sensing. In California, there are 230 major crops planted at
different times. Each farmer could go out and estimate the canopy
cover but a lot of people want to be more precise than that. If you
could do a fly-over once a week to determine the percent of ground cover
[that increases with growth], and [combine it with ground-based weather
data], you can get the crop coefficient that relates the reference value
to the ET problem for one particular field. One of the weaknesses
with remote sensing is taking one measurement during the day and estimating
ET from that. ET fluctuates in morning vs. afternoon, when the wind
picks up, if clouds roll in, or wind direction changes -- you need continuous
measurements. R: There have been a lot of studies that have
related the vegetation indices developed from spectral reflectance to index
percentage of ground cover. You’re probably not going to get anything
much better out of remote sensing. R: I’ve never had anybody approach
me about the availability of that information.
MZ: We found that you need to have the ground cover as your base so you can relate the indices and the ground observations -- to look at the relationship. Otherwise, the remote sensing data is just a picture. We use the ground data relating to the penheld[?]. That sensor has 18 degrees of view at 1 meter above the canopy with 50 centimeters in a circle. We’ve estimated the percent of cover, then taken measurements on the spectral. We expect that the spectral would reflect on the percentage, so we separate the green, woody, etc. and go through regression and other indices. We’ve learned by doing this, but the problem is who’s going to go out and take samples? R: That’s why you need a fly-over that just gives them the results.
P: Craig Weagon in the USDA did a study a few years ago, where he collected vegetation index data and agronomic data from a number of the major agriculture crops at various locations all across the U.S. and developed relationships between the vegetation indices and things like leaf area index, active radiation, etc. He was able to show that there were consistent relationships, assuming that you made observations under similar conditions, for different places around the country, and that you could in fact extend to some international areas. People don’t get consistent relationships because they’re looking with different sensors or at different heights, different times of the day, etc. There is evidence that there is some consistency in the vegetation index and agronomic relationships. R: I think percent cover or ... fraction of a rate of absorption is a pretty good relationship too.
Approaches to Determining Evapotranspiration by Remote Sensing
P: The guys from agronomics in Arizona say they get a lot more
than crop coefficients, in fact they get exactly the information the farmer
wants as an end result without having to go through the crop coefficient.
They can find a better discipline measure of temperature and determine
water demand. R: I’m saying the missing link for improving
ET models is to know the percent ground cover. R: It all depends
on the model though. Your technique essentially estimates the potential
water demand of the crop and modifies that by a crop coefficient.
Another approach is simply to use remote sensing and weather sense data
to actually estimate the energy balance of the crop and directly [estimate]
the ET. It can be done with an airborne spectral thermal imager [reference
to Susan Moran’s work in the southwest]. R: Maybe because a lot of
the assumptions that are used in that instrumentation aren’t valid for
California crops, because you have too rough a surface.
Soil Mapping
P: Before we go into the EPA, I think ... others are going to
review that information quicker than we will for lending purposes, real
estate market values, etc.-- that would frighten me as a grower.
We’ve targeted the EPA and other government regulatory agencies as the
bad guys, and in fact I think that [?] is a lot closer to home than the
EPA. Because that’s land values. R: One of the markets
in the Columbia[?] Basin in Washington rent fields, and there’d be a field
with awful soil patterns. We asked [farmers] how they knew which fields
to buy. They said they used USGS maps and soil conservation survey
productivity numbers, and they just have their favorite fields. They
had gotten some really bad fields. When we asked if they wanted some
information on the best fields, they were hesitant about it. They
said there wasn’t too much risk. When we asked the insurance guys,
would they like to track the growers and whether they got good fields and
just not produced in those fields, or whether they got bad fields that
would be a higher insurance risk for those fields? The insurance
guys weren’t interested, because then they’d have to assess a true risk
for the location to the true grower. That would put a lot of growers
out of business. If they just use the Soil Conservation Service (SCS)
data, then they set a price and just pro-rate it across everybody.
Soil Characteristics
P: The growers get a single soil sample at a represented location in
the field. They would like a better soil model, but they don’t like
soilection[?]. The soilection process involves tromping through the
field and costs them $1500. Use remote sensing before and during
the season to try to find the optimal locations for one or several soil
samples if you have major soil groupings in your field. The goal
was to make the single sample soil model more efficient by understanding
the difference within the field and then using our spatial implement to
make some corrections. That was a successful application. R:
The problem with using remote sensing to map the soils is that the soils
are based on the parent material, so you need a profile and remote sensing
only covers the surface. R: That’s not correct. Agricultural
soils are tilled so there is no soil structure in most agricultural soils.
There isn’t much of a profile. MZ: It also depends on the crop
too, if you talk about the roots’ impact.
Soil Types
P: One of the first people who did site specific technology analyzed
their data and asked what the fundamental underlying principle that governs
agriculture? There is a fundamental relationship between soil type
and yield. That should be the fundamental testable assumption but
when they tested it they found no relationship. So they called in
the USDA to micromap their soil types - no relationship. They had
the USDA to do 30-foot boring so they could look at underlying soil structure
- no relationship. It is a fundamental principle of Midwestern agriculture
that has no relationship. R: It doesn’t matter what soil type it is, it
depends on how much N2 is in the root zone.
P: Let me make one point that deals with 1 meter resolution and how much resolution you have in the application system. [Let’s say] I take one soil sample and base a uniform application on it, because I don’t have variable root technology. You design your fertilizer application on that one value. In the second case, if you go out and sample your field on a one meter grid and then determine what the average application should be to meet some objective, the uniform application you would apply would be much different than in the first case where you took one sample. The reason is because the processors are not linear. That is not well known from the people who are marketing this information. Most people think if you add up all the little pieces, you get the average but in fact when you run it through the processes that are not linear, you can’t just add them up 1 to 1. So there is a huge amount of utility in having high resolution information in just making the decision as to what the uniform application should be, not only in addition to the ability to do precision farming. Those differences are not trivial in a lot of cases, in irrigation they’re not.
P: As an ecosystems modeler, another need I’ve found is tell me something about the soil. The soil maps from SCS are just not useful at this scale, there’s just a minimal amount of information. I want to know soil texture, depth, hydraulic conductivity, etc. R: The problem is translating a field survey for your purposes, correlating that with the national standards of soil types. The SCS does not have and will not have the resources to map the intensity that’s needed for a particular practitioner. It boils down to having someone that is an expert in soil taxonomy, genesis and classification to work for you and provide what you need to move ahead. R: You’ll never get that [soil] information from remotely sensed data. What you could get from the remotely sensed data is the characterization in real time of the soil moisture content at the surface. If you get some good radar data, you get some indication of the depth of the rooting zone. Long-wave radar may show us some things in the future.
P: Did we talk about soil moisture? It would be interesting for both rain-fed and irrigated systems. In one, you’re looking at what your plant population would be and in irrigated systems you’re looking at your water management plan for the year.
MZ: One of the people in this room has done soil texture classification from remote sensing spectral that works well. A paper just recently published in remote sensing looked at soil texture.
P: I’d like to see more soil moisture data, but I also think there’s room in between what the soil taxonomist puts on a map and remote sensing images. Bringing in geomorphic knowledge or the concept of a catina[?] on a landscape, for instance, to use these ideas and concepts along with digital elevation models from remote sensing. To interpolate over the landscape some important characters -- soil depth increases, etc. These pieces of common knowledge are not applied in a manner that could be to create use. There’s an immediate product that could be put together to reveal great answers.
P: The short and dirty answer to that is if the NRCS had the money to do it, they already would have done it. Their budget has been cut over the last 30 years and prevented them from starting that kind of work. If the precision farming becomes the way of the future, it is going to be equal employment opportunity for consulting soil scientists, because it’s going to take a well trained soil scientist to give you the high quality base map of the soil series on any particular farm for that precision work. There’s a number of ways of getting at that data, none of which are remotely sensed. For instance, you could have several mechanisms on a tractor pulling a shank through the soils that could reveal soil texture, because it’s related to the amount of energy it takes to pull that shank through the soil. A better job is if you could deliver to the farmer a characterization of all the soils. That comes from looking at soil profiles on a gridded basis, which is a lot of expense. I expect that over time some surrogates will evolve from some recording circles that tell you what the energy requirements are as you move across the field. That’ll be indexed and have a heavier texture. Some of the soil characteristics are related to chemistry or percolation. Some of the soils in U.S. are very uniform, but the farther west you go or east past the Ohio Region, it gets very difficult with broken topography and changes in the parent material.
Application of Water Quality Issues to Remote
Sensing
Salinity/Drainage Problems
P: Those days are going to be limited, where things get socialized.
Another example of that is drainage issues in the San Joaquin Valley.
The water management district is scared if I released the map that shows
who the good and bad guys are in terms of upwelling and downwelling. We
have remote sensing data to add to it that reveals salinity levels that
are changeable over space and time. The [water management district]
is worried they’re going to have wars within the district. They say
they’re worried what the State Quality Control Board will do about it,
when in fact they’re worried in and amongst themselves whose going to pay
the piper to meet the regulatory guidelines. It’s really easy to
blame one group when it’s a complex issue with a lot of players involved.
Pollution of Rivers
P: I’ve been exposed to a similar situation with rivers.
The Delaware River was an absolute cesspool back in the 1960s. I
was involved with a variety of people who wanted to model a section [of
the river] to improve it. Somebody up the river was causing his problem.
Then a government agency came in and a whole sequence of people wanted
to get a model. We did enough of these to realize we weren’t going
to get a simple answer. There was an assemblage of regions of all
the groups that got together and worked on building a model from the pollution
standpoint. Instead of pointing fingers, they realized everyone would
need to make changes to reduce pollution. When the information became available
to the people involved, then they got together and worked on it -- that’s
what remote sensing has the potential of doing in a lot of these areas.
The next step [for the salinity problem in the San Joaquin Valley] is to
bring the people in to start working on improving in it.
Remote Sensing to Evaluate Irrigation
P: Remote sensing [could be used] for system evaluation of surface
irrigation. One of the problems is having the manpower to go out
and evaluate surface irrigation, which is probably the biggest limitation.
70% of CA [farms] have surface irrigation, but they don’t know their efficiencies
in terms of advanced recession.
Defining Sustainable Agriculture and Approach
P: Did anyone define sustainable agriculture? R:
Sustainable agriculture is a concept of maximizing agriculture production
while minimizing environmental impact, plant diversity, economic diversity
and jobs creation etc. MZ: Sustainable agriculture may not
be maximization if you want to minimize environmental impact. If
they are put together, it has to be optimization. R: The war over
words is not as important as taking a systems approach to understanding
how the system reacts to different perpetrations. The system is so complex
with so much variability that we probably can’t model it to the nth degree.
At least we can measure to see what a given perpetration results in.
Applying Environmental Issues to Sustainable Agriculture
P: Who’s going to use this and what is it for? Are these
for decision support systems or environmental compliance? I can see
the agriculture side of sustainable agriculture but I don’t know where
we’re looking at the environmental side. Who is the end user?
What are they using this for? I realize we’re trying to be inter-disciplinary
by bringing in the environmental sciences, but when we talk about environmental
crop monitoring, where is that going to go? To the farmer to make
a decision in sustainable agriculture or is that some other area?
P: A fellow from Kansas is doing these maps of the U.S. showing the greenness -- “the green maps.” Now I can see, are my forests being stressed or are they healthy? They’re using it for agriculture because that’s where the money is. When you use it for the environmental side, who is it [for]? The EPA? What are they looking for and what do they think is worth $X to drive the research? R: I think the ground level of people in production would like to know if these models being used to monitor the environment are going to come back to the farmer and say, “Now we have information that shows your farm damages the environment.”
Justification of Remote Sensing
P: I would like to hear some arguments why the USDA should sit
down with NASA and say these are the databases that are critical to the
Department of Agriculture, to the Extension Service, to the producers in
the United States. [We need to tell NASA which databases] are worth
fighting for and committing $4-6 millions/year and are worth a $35 million
launch of a light satellite or a $75 million launch of a larger satellite.
R: How much are you at the USDA paying for predictive work at this
time? R: Just for the commodity forecast, we pay $3 million/year
just for hard copy photographs. The declassified data that comes
in from remote sensing has a bill of $33 million.
P: It’s very hard to put your hand on all the benefits because in some cases it takes 20 years to get the investment going. R: If we’re looking 50 years into the future, we want to make a good case for both the commercial side and the research side that there is an urgent need for databases that are critical to the century model for predicting changes in grassland ecosystems, for example. R: There are very few thermal sensors. R: We can’t answer the question, until we have the information and it is tried and used. So the data needs to be at a higher resolution.
Arguments against Remote Sensing
Existing Methodologies
P: What information do we need from remote sensing that we can’t
get some other way that is cheaper, faster, better, earlier in order to
move us into the next 10-50 years? My fundamental argument is that
we don’t need remote sensing. Maybe 30 years from now it will be
of more value. It’s not worth putting more taxpayers’ money into
another set of satellites to get data to agriculture to enhance models
of old systems. What I’m suggesting is that the biggest competitor to remote
sensing is the existing methodologies. Will the data from this workshop
tell NASA that they should be spending more money to create more databases
and more satellites, or are there other ways of getting this data?
I need to make the decision as to where the research money goes.
Difficulty in Finding Agricultural Problems that
haven’t been Addressed
P: The farmer does their commodities stuff, and they already
have their existing means, and that’s what we ran up against. The
growers are not helpless people lost -- they’re making a profit and running
their businesses well. It’s difficult to find a hole in their program.
The remote sensing opener is that the grower only has the 80% solution
because your management of your personnel in the field is not adequate.
That’s a reasonable market. But there aren’t any real large problems
we can help them with. There are some high value things, but they
are few and far between. For example, the late blight in potatoes
that can wipe out a crop in 48 hours. So they ask us, when the atmospheric
model indicates a high probability of late blight conditions, fly over
twice a day and find where the stress starts happening. That’s too
expensive for them and we can’t guarantee that the indicators will reveal
the stress. Even if they do catch it early, it may be 7 hours before
they can get a plane up to spray with fungicide. [Remote sensing]
is not a panacea.
Models for every crop have been validated
P: There’s basically a model for every river basin in the U.S.
including forested, savanna, grassland and cropland. We have a crop
model for every crop grown in the U.S. They’ve been validated and
are used throughout the world. They’re either physiologically based
or heat-unit based. They provide yield or leaf are cover from which
you can get an approximation of biomass production. Most of the models
help producers manage drought stress or provide projections for when to
fertilize, irrigate, apply pesticides and herbicides. All models
are an approximation of reality.
Arguments Supporting Remote Sensing
To assess regional concerns
(Benefits include less manpower needed and centralization of data processing).
P: If you’re looking to do some kind of assessment on a regional
scale, it’s very difficult to do that with ground-based techniques.
If you call that many people out to make observations and samples on an
area the size of a large watershed, for example, you start running into
problems of logistics. It would be [ideal to use an] existing satellite
system that could be adapted to provide the aerial coverage to serve as
a novel data input to assess things. Do you want to pay to have 10,000
people out in the field making observations, or do you want to have an
existing satellite observation? Not only do you have less people
out in the field but you can also centralize your data processing.
Existing methodologies aren’t effective enough
P: Observations can be collected more quickly and efficiently and be
worked on by a few people. Another example, I have a weather map
with observations. I can sit in one spot and know how hot it is every
five miles in a grid. I can run that data through these models and
say this side of the valley has high risk; this side has low risk.
Ideally if you could tell me which crop is being stressed by what organism,
but if not, just identify stressed crops [Remote Sensing to Determine Location
of Stressed Crops through Thermal Sensing]. It helps me differentiate
that field from other fields. I don’t have enough people in the field
to guarantee that the field that gets diseased gets looked at when I need
it to be looked at. Regional observations that are collected very
quickly and efficiently and send it back to a few people to work on [would
be ideal].
To provide more information at a higher resolution
P: The concept is that the farmer has more information (even
if we don’t know what information he needs). Is there a higher probability
that we’re going to have a more efficient system both for agriculture and
the environment than without it? When computers were introduced,
it took years and the equipment didn’t work at first, but the government
invested in them in the early stages. We’ve got a similar situation
here. At first the satellites may work for awhile, but there may
be new technology that appears later on that makes them obsolete.
A lot of models won’t work. Does having more information on a broader
basis across the world going to give us a better quality of life than without
it? There’s also a question of fairness with respect to the USDA.
Does everyone have all the information that’s around or do only the wealthy
receive it? R: The commercial decision is whether or not they’re
going to make money when the satellite goes up. R: We can’t answer
that question -- it’s a marketing question. P: Instead of looking
at technology seeking a problem solved, identify key problems that technology
may be able to solve? What kinds of problems are facing agriculture
in the future?
Use of Remote Sensing for Risk Assessment
MZ: I would use remote sensing for risk assessment. Since
I would work for companies, it’s already market-oriented. But I often
cannot use the county average for crops, because I don’t know where crops
grow.
Use of Remote Sensing to Study Problems in the
Oceans
P: The biggest [problem] people are looking at is the hypoxia[?]
problem in the Gulf of Mexico. In the middle of the Gulf of
Mexico, there is an area that develops, in the summer, {whole oxygen content[?]}
and the fish in the area die off. That’s just a theory of the cause.
The EPA has been taken to court by the Sierra Club, whose forced the implementation
of regulations. The Sierra Club’s smoking gun is nitrogen, which
is being applied by growers in search of higher yields. The hypoxia
problem is real and grew after the floods of 1993 and hasn’t subsided in
the intervening years. If research is going to be done, let’s make
sure its good science and not based on manipulations that may affect agriculture
adversely. Identifying the real problem would takes remote imaging,
because the problem manifests itself in the middle of the Gulf of Mexico
and would need to be systematically traced back through a very large watershed.
It could possibly affect some 27 states.
Agriculture Meeting Increased Population Demands
P: I would postulate that over the next 50 years, the pressures
of increased population will increase the demands of agriculture.
Remote sensing is a good way to look at things on a national or global
scale to protect and understand what’s happening. R: There are examples
in other countries of also using this technology appropriately. In
Australia, they implemented a hyperspectral[?] imaging program to try to
systematically start following components of the ecosystem. They
modeled microorganisms, chemicals, etc. to trace the mass flows from their
source. You can define an approach to solving a problem [with remote
sensing]. If the U.S. wants to maintain or increase the $50 billion foreign
trade in agriculture, then need to improve agriculture efficiency.
R: [Improving efficiency] is going to be done over a period of time with
a number of small improvements by many people based upon the information
they have.
Agricultural and Ecological Models
Need for Calibration Data to Increase Accuracy/Predictability
of Models
P: I don’t think we can determine what specific databases will be needed,
but we do know that with more information, people can do a better job at
managing their agriculture businesses. R: We have some idea of the
types of data needed for use in models. Very few of the agricultural
models in existence have calibration data. Remote sensing will probably
not add much to any one individual field, but field-based data could be
used to calibrate a model during the growing season. Developing models
for regional areas have problems with data collection. What they’ve
used in the past from remote sensing is the leaf area index that comes
from a vegetation index that comes from the satellite reflectance data.
Certain models such as crop stress can be used from thermal infrared data.
That capability has been demonstrated.
P: A model can indicate where a particular field is in the model
by identifying certain indicators. Two things that the model can
apply:
1. a spatial component to apply the model differentially via remote
sensing technology
2. the temporal aspect which indicates where a crop is in its growth.
We provided a product for the corn problem called the Initial Field
Productivity Base Map, which was basically an emergence map that gave us
a first cut at the productivity of that portion of the field at that time
for that particular crop. The farmer was interested in getting a
corn dryness map to determine harvesting time and increase his yield and
economics. Neither issue is critical to the potato farmer.
P: The two ways that remotely sensing data can really help to
improve the accuracy of an agricultural model is by
1. using “within season calibrations” where a model tells us
certain things will happen during the growing season and then to remotely
sense or observe the occurrences. Given the fact that models are
only an approximation of what happens in the real world, there not going
to be terribly accurate. But if you have these observations from
the field or remote sensing data, you can use that information to re-calibrate
the model to produce a new simulation that is more accurate.
2. The second way is 2) spatial variability. In the past, they would assume an average value of a field parameter that was representative of the plants in the field. They’d run the model once and come up with a yield. By using remotely sensed data and looking at the variation of the vegetation index across the field, you can classify your field into various portions that have vegetation indicators of growth and then run your model for each portion of the field and produce an aggregated yield estimate. If there’s a lot of variability, it would be more accurate than to do an average value. If it’s important to harvest crops at different times, remote sensing will also give you a means of being able to know where that’s going to occur.
P: What index do you need for the correction of these models? What tells you that you need to re-calibrate the model? R: You need some spectrally unique signature or visible queue and that is temporally unique. The first one that is detectable is emergence. R: Most of the agricultural models simulate leaf area index because that involves the absorption of light. Leaf area index can also be estimated through remote sensing. The model continuously gives you an estimate of leaf area index throughout the growing season. When you have remotely sensed observations, you have distinct points that you can compare on a given day to what the model and simulations says its going to be. If they’re not similar, you need to go back and re-parameterize the model so that you can bring the simulation in line with the observation.
P: I’m not completely convinced that the leaf area index is a reliable estimate. R: When people use the term biomass, you don’t see or sense biomass. R: They’re some work being done in three spectral areas -- 2 in the infrared and 1 in microwave -- that essentially correlate the response to the plant. The idea of using the remote sensing data to drive the model based upon measured remote data as against what you would do from the normal leaf -- they’re working to identify the correlation between the signals that they’re getting from overviews and remote sensing to the stage of the plant and using that to drive the models to determine if they can do a better job. They improved their regional prediction to a better level 9-13%.
Remote Sensing and Weather Models
P: Can remote sensing determine this weather station is applicable
to this area because it has the same slope (in order to predict frost,
etc.)? We know that there are differences in temperatures based on
slope, topography, location at the base of a hill etc. R: A GIS with an
elevation model and the national weather model. A satellite should
be able to determine the temperature. A thermal image of the ground right
at sunrise would tell you a lot about frost protection.
MZ: The Midwest only has flat topography, so you end up with problems trying to model that data. R: So the ADAHR data really isn’t going to be of value - it’s either going to be a ground based station or it’s going to be a component of data terrain and a lot of historical knowledge of the climate.
P: You can always argue on a problem like that. That problem was specific in San Diego where people wanted to put in some tropical [crops]. The problem was that there is a freeze every 1 in 10 years and you were wiped out. R: The citrus growers have fought the problem on the flat land, and the crop zone in Florida has moved south as a result. There hasn’t been a predictable model yet, except the night after the big freeze. We’re looking at a piece of the land bounded by thermal mediating bodies [of water] on both sides. They haven’t done a good job yet, whether you’re using ground based systems or remotely sensed systems. We need good models there and ways to maintain them.
The Evolution of Spatial Ecosystem Models and
Remote Sensing
P: The only way we can have spatial ecosystem models is by having the
complete spatial coverage that remote sensing can supply. In a sense,
because the evolution of ecosystem models is from point models and small
area models, the kinds of parameters and the way we conceptualized the
system evolved from looking at it from a human scale of plants, leaves,
etc. not at the scale that remote sensing has the ability to look at the
system. What does remote sensing have to tell us about the ground surface
that we haven’t even thought of to incorporate into our models because
we don’t conceptualize it that way? Are there some interesting and unknown
products from remote sensing that remote sensing researchers interfacing
with the model might provide and [lead to] changes in the way we do models?
There might be something inherent in remote sensing at a larger scale that’s
not a parameter we normally use right now. R: A lot of the energy flux
and biogeochemical cycling activity is already in the global change models.
It’s hard to really find the needs for this kind of information, because
there are other ways and most of them are at the ground or just above the
ground. R: For global change modeling, there are not imaging or sensing
systems that give us any good data on the emission of greenhouse gases
in the troposphere. Everything that’s up there is looking at it in
the stratosphere. All those are ground-based systems that are tunable
laser systems that are measuring emissions of greenhouse gas fluxes.
They are basically all ground-based systems.
Remote Imaging Scale
P: One of the problems we run into is that we’ve taken the models
and applied them at a space scale that is much larger than they were developed
at. As a result, we find that they don’t predict the way they thought
they would which shouldn’t be surprising at all. If we have higher
resolution data, they wouldn’t necessarily be able to predict an order
of magnitude better, but their predictability would improve because you
would capture more of the spirit under which the model was developed to
begin with. So you need the higher resolution.
P: With regard to scale, the producers want to talk about field-size scale and the [commercial industry] wants landscape scale. The models were developed at a scale in between. Some people won’t need the resolution that others do.
P: One of the reasons the existing crop models haven’t been used is that there isn’t money to use them. In the past, crop yield models have found the largest application in large area yield estimation because that’s where the money is. R: To say we’re going to get more accurate data and therefore a more accurate model is not linear. You don’t necessarily get a more accurate model. You can get a less accurate one because of your ability to measure the spatial extent of a potato field, for example. What accuracy do we need to have for geo-registration of our images to the actual correction on the ground so we don’t have an error in an acreage estimate from this field that is significant. If you’re 6 feet off, that’s a lot of potatoes. The more accuracy you get, the more significant the small errors in your remote sensing data. Because it’s difficult to get that accuracy, you may actually go backwards in your modeling. R: A lot of times we have to do things [in order to find out] they don’t work.
P: One scale of resolution may be suitable for one crop, but in the case of grapes, the scale is much smaller than that because vines have a higher value and a smaller surface area. What scale should we recommend? R: The gross scales are the commodities scale, the regional scale and the scale necessary for precision farming, which is currently defined by mechanical variable capabilities (50 feet or the width of a harvester). R: In five years, it will be a meter or less. So remotely sensed databases for agriculture should consider delivery of data at a meter or less. R: What I’ve heard is, the most important thing is getting the monitor on the harvest so we understand what it is, which would be at about a 3 meter scale. The data and mechanics necessary to get down to the foot isn’t going to work. R: It depends on the crop. In some cases you may need to look at each individual plant. I would think that in most cases 1 meter resolution would be good. If you’re going to do measurements on individual plants, you need to identify where you are and know that the soil sample is good for only one square foot.
MZ: What’s the wavelength we want to look at? R: You need to be able to follow nutrients and agriculture chemicals by species. That’s about 1000 channels across the UV to the thermal IR. R: It hasn’t been proven that it can be done yet. R: We already know we need more IR.
Need for Geo-Registration
When the flurry got started, the question was about the accuracy of
the GPS that’s needed for that. That gets back to the type of databases
we need. We’re going to need data to geo-register this precisely
to the field. We did that in Washington by going out with the different
GPS and finding something we considered to be a feature would stay there
a couple of years (house, bush) and we would have a center pivot.
There’s also features you wouldn’t want to take data on. But we would
take 4-5 for the field to get accuracy at 2-meter data. If you’re
going to do this for satellite operations, you’re going to need to get
them manually or get some dataset that is registered in a database to that
accuracy with which you can correlate your reference image. The cost
of an airplane doing this job comes down to near 0 because you have a video
camera and it doesn’t matter how the pilot flies. You don’t need
to have sophisticated navigation because you can take out all those errors
[from the remotely sensed geo-referenced map]. A geo-registered digital
database of U2 data down to one foot would be fabulous. In order
to do that you need a lot of GPS data. We did the soilection process
at 50 feet grids on a field, got an emergence map, soil color map (for
organic content) and there was zero correlation. In other fields
it worked.
Summary of Topics
Two different scales -- landscape scale and sub-field scale.
The discussion today focused on the sub-field scale, defined as 1 meter.
Issues that need to be addressed at sub-field scale:
1. Evapotranspiration expressed as percent canopy cover.
2. Crop stress as indicated in thermal radiation.
3. Within season calibration for field in order to attach what was
happening in the field to models
4. Soil characteristics within the field, as far as spectral resolution
goes at that scale (defined broadly as to what was needed -- identification
of plant nutrients and chemistry), texture, soil moisture in irrigated
and rain-fed systems
5. Digital geo-registered dataset for each year registered to a base
map-.
Issues that need to be addressed at landscape scale:
1. Water pollution and water distribution
2. Regional/Global food production
3. Soil moisture
We shouldn’t worry about definition of sustainable agriculture, but take a systems approach.
Approach to Sub-Field and Landscape Scale (Role
of Governments/NASA)
P: When looking at the sub-field scale, there isn’t much that
NASA can do, it has to come from the bottom up -- from the growers.
The solutions for growers would more appropriately be addressed by private
industry. I don’t see how NASA could effectively design a program
from the top down. Regional issues are a different question -- government
solutions are more likely because there is no commercial market for that
data. Data on the regional/global scale (like global warming) can
have a tremendous impact on agriculture and our ability to sustain an agriculture
system that can feed the population.
P: Agriculture has input from county, state and regional [government]. They do a lot of data gathering and regulation, especially in CA, that impacts the farmer. An example that comes to mind is the urbanization in the Central Valley. Cities and counties are digitizing their master plans (visions of development for the next 5-10 years) -- we can take all the master plans and see what the entire valley is going to incur. Urban area growth has a tremendous impact on agriculture lands around them. Not only are they converting land, but the relationship between them tends to change how the agriculture land is used. R: Their voice needs to be heard as to the input for agriculture, and we haven’t done that. R: You can replicate that over the entire U.S. Regional scale and the field scale has been ignored [in the past].
Environmental and Ecosystem Issues -- Forest,
Range, Woodlands, etc.
P: From my perspective, the landscape field scale modeling is
for forest, ranges lands and semi-arid shrub lands or woodlands.
From an ecosystem standpoint, what I want from remote sensing is percent
cover. Even though I’m doing a regional scale, I’d like to see it
in 1-meter resolutions so I can get information on the canopy roughness
and the canopy structure (including trees and shrubs). The structure
of vegetation has many non-linearities on how that percent cover scales
up depending on how it’s structured. We need to know that if we apply
big leaf models. We need to take the non-linearities into account.
R: Would you need 1-meter data for the entire region or could you
use samples? R: If I have a computer that can handle the data, I’d
take the entire region. Adding more pixels for the marginal modeling
cost is minimal, but the marginal computing cost is excessive. R:
You would have to consider the data cost as well.
P: When we started doing spatial simulations in the early days,
there were big problems as to how many pieces you need to actually do something.
We’d extend it to the capability we had, but usually we’d back off at some
point realizing that we didn’t really need that much information.
But you have to go out and actually look at all the data before you can
say that three meters is fine. There’s a value in being able to look
at it, but you may not want or need to look at [that fine scale] all the
time.
