Dear Derek,

I don't know the details of your data and observation methods, but it seems to
me that starting with the re-parameterization technique would be very similar
to embarking on a fishing expedition: you might catch some fish, but still you
won't know anything about the system. In my opinion, a comprehensive review of
the model parameters requires detailed soil and crop monitoring data.

With what you have, perhaps you could first split your base of remotely sensed
data into two sets.
The first set could be used for a limited model calibration/adaptation. The
second set could be set aside as independent data set for testing
(validation).
Then you could use the model as it is and compare the results with the data of
the first set.
It seems to me that if you find considerable differences, you should try to
understand the reasons.
I imagine that potentially there are very many possible reasons for
discrepancies:
- How good are your input data: weather, crop management, soils (didn't you
mention in a previous mail that your soil data are very generalised?)
- How good  is the relation between remotely sensed LAI estimates and field
LAI?
- To which extent does the field situation compare to your idealised modelled
world (e.g. locust plagues, hail storms and soil salinity might have great
impact on LAI but are not taken into account in the model).

As long as uncertainties in the above have great impact on the model results,
there would be little gain in changing crop parameters. Once you are
reasonably confident with your input data, then you could check if there are
model parameters with a large uncertainty and to which the model is very
sensitive. In most cases, changing parameters (within realistic limits) will
principally affect trends. You could change e.g. initial LAI to obtain a
reasonable 'average fit' between remotely sensed LAI over time and modelled
LAI.

If this is OK, and your aim is to see if combining a model with remotely
sensed data improves your final yield estimates, then I would continue by
simply using the remotely sensed LAI estimates as input  to adjust the
calculated values while running the model, without touching the other
parameters. If large discrepancies tend to occur in particular fields or
particular years, it would be necessary to enter into the details of these
particular fields or years.

I hope this makes some sense.
Best regards,

Maurits van den Berg

- - - - - - - - - - - - - - Original Message - - - - - - - - - - - - - -
Hello,
Thank you all for your comments which were very helpful and
informative.  The reason I asked the question is because I am attempting to
incorporate infrequent (1-3 scenes per growing season) remotely sensed data
(Landsat 7 ETM+) into CERES-Wheat to see if it improves estimates of soil
water content and grain yield.  There are two methods described in the
literature that I am considering to accomplish this objective.  Both
methods essentially fine the 'best' fit of the simulated growth curve to
the remotely sensed data.  One method, however, will only change the
initial value of LAI, and is only applicable in a growth model which is
sensitive to initial conditions of LAI, the literature terms this the
re-initiation technique.  The other method results in new parameter values
for all parameters used in the equation to estimate LAI, the equation to
estimate LAI is
         LAI = (PLA-SENLA)*Plants/10000
So the equation to determine PLA and SENLA and there parameters will also
be changed.  I have not figured out all the parameters involved yet but I
believe there will be many.  This technique, termed the re-parameterization
technique, is consequently much more difficult to implement
programmatically and will result in greater computation time than the
re-initiation technique which is relatively simple.  It seems to me that
the re-parameterization technique will account for more of the variability
in inputs than the re-initialization technique.  However, based on some of
your comments changing the initial value of LAI based on remotely sensed
estimates of LAI (assumming these are correct) does have some merit and
might help to improve CERES-Wheat estimates.  Maybe it is best to try the
simplest solution first.  Does anyone have any comments on this.  Thank you
for your help.

Derek McNamara

Derek McNamara
Graduate Research Assistant
Mountain Research Center, Montana State University
P.O. Box 173490, 106 AJM Johnson Hall
Bozeman, MT 59717-3490, USA
406-994-5073
[log in to unmask]

- - - - - - - - - - - - End of Original Message - - - - - - - - - - - -