Dear Arjan,

Thanks for the analysis of past work on the estimation of soil water
limits.  I have seen your Florence poster and I think that you have done
some very thorough analysis.

>Not unimportant is that the articles related to Ritchie1987 have severe
>errors in the equations, making that soils that differ significantly from
>those that were used for deriving the method, get wrong estimates for the
>soil-water parameters. This not necessarily means that the code as used for
>DSSAT has those errors also, but it is likely. I brought this up with Joe
>Ritchie a few months ago, and he acknowledged that his article equations
>were wrong (my compliments, Joe!).

A great scientist is not afraid to admit error.  The problem with the
earlier work was that it was a multiple regression analysis with too many
variables, plus their higher order terms.  It was too specific to the data
set that was used in the analysis.  It worked well for those data but could
yield negative water contents in other soils, as we show in our 1999 paper.
 This is why we simplified and we generalized the algorithm in the 1999
paper.  Simplicity is good if simplicity is what you want.  One of the
simplest equations in history is E=mc^2.  We do not offer it as a cure-all
solution but as an approximation when field data are lacking.  If
simplicity is not what you want, then you can look into more complicated
alternatives.  When comparing models, one should weigh the cost of
obtaining the extra data against the cost of the extra accuracy.  I do not
mean to disagree with you in any way, I just try to put the subject into a
wider perspective than that of comparing R squares and RMSEs.

Regards,

Aris Gerakis
[log in to unmask]
http://nowlin.css.msu.edu/aris