I am working with estimates of vegetation height derived from radar data.
We have a nonlinear model to correct these estimates for errors associated
with viewing geometry. I am trying to estimate a single parameter in this
model while accounting for spatial (spherical structure) autocorrelation.
I'd also like to statistically test the influence of several vegetation
parameters. The gnls() function in the nlme library seems well-suited for
fitting this model, but I am having trouble getting it to converge, even
without the autocorrelation structure. Here is the model I'd like to fit:
th=eh*((1+theta/thetaref)/(theta/thetaref))^(1/n), where th=true height
(dependent variable); eh=estimated height (independent variable); theta is
local incidence angle (independent variable); thetaref is fixed; n is the
parameter to be estimated. My question is: is this parameterization
efficient for gnls()? I've gotten reasonable results by changing the
control settings, but also lots of warning messages.
Thanks,
Alan Swanson
[[alternative HTML version deleted]]
