Your issue is that nls returns "singular gradient", but I believe the real
trouble is that
if or ifelse are not in the derivatives table. My own (experimental, but in
R-forge)
package nlmrt has nlxb to mirror nls but be much more aggressive in finding a
solution,
and it gives an error msg that ifelse is not in the derivatives table. It may
be that nls
has an alternative for getting numerical derivatives (perhaps Doug Bates or
Saikat DebRoy
could comment).
It may be better to convert to a function and use Nelder-Mead in optim, or
better, nmk
from dfoptim which is more up to date. I'm uniquely positioned to say this, as
I wrote the
version of Nelder-Mead in optim and I think it's getting long in the tooth,
though it
still works remarkably well. One of the routines in minqa may be better for
speed, but
possibly less stable.
Code:
wfn<-function(par, Cfl, Tsoil, Pw){
K<-par[1]
a<-par[2]
b<-par[3]
c<-par[4]
z<-par[5]
m<-length(Pw)
res<-rep(NA,m)
for (i in 1:m){
tmp<-1
if (Pw[i] <= z) tmp<-exp(-a*Tsoil[i])
ff<-K*tmp*exp(-b*Pw[i])+c
res[i]<-Cfl[i]-ff
}
as.numeric(crossprod(res))
}
a1<-optim(st,wfn, control=list(trace=1), Cfl=Cfl, Tsoil=Tsoil, Pw=Pw)
(which converges pretty fast)
Best, JN
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
