Different ways of fitting Van-Genuchten model on some experimental data - Part3

Finally, today is the day, I want to finish this topic. Sorry for the delay, I was somehow busy with writing a paper and I didn't have too much chance to finish this topic up.

In this post I would like to talk about what I did in my project. In my project I was supposed to find out whether there is significant difference in water retention curves of five treatments or not.

So basically, I had 45 set of water retention data (9 water content and their corresponding suction head) for 5 treatments with 3 replications in 3 harvesting time. For all of my sets I used one of those non-linear methods to fit Van-genuchten model and to find 45 alpha, n, theta_s and theta_r. Actually my design was split-plot in time and I didn't have hard job to do, so I used SAS to perform the analysis of variance for each of these 4 parameters.

There is just one quick tip for working with time dependent measurements. If you have measured your data repeatedly over the duration of the experiment you need to make sure that the residuals are not correlated. If you could recognize pattern in your residual you can be sure that you messed up somewhere. The residuals should follow the normal distribution and they should look ........ "belly" :-| I guess. If you could see that there is time correlation in your measurements you should take them into account by modeling the covariance structure of the residuals.

Here is the part of SAS code I used to analysis my data for Alpha :

Proc mixed data=HW7 Cl IC;

Class trt rep HarvTime;

Model Alphas= trt HarvTime trt*HarvTime / DDFM=SATTERTH;

REPEATED HarvTime / SUBJECT=rep(trt) TYPE=UN R RCORR;

LSMEANS trt HarvTime trt*HarvTime / PDIFF=CONTROLL CL;

contrast 'Two methods' trt 0 1 1 -1 -1;

contrast '3% VS 6%' trt 0 1 -1 1 -1;

ods output infocrit=un;

quit;

Good luck :)

Hamze