Poultry manure as Influent.

[vss0015@auburn.edu]

Greetings Dr. Jirka,

I work with 25 cm long and 15 cm dia. cylindrical columns. Our tracer experiments point towards high preferential flow in these undisturbed columns. We have amended poultry manure on the top of these columns. We know the trace metal (copper and zinc) conc. of the poultry in terms of total conc. and also of various fractions (soluble, exchangeable, residual, bound to carbonates, etc.).

Our column experiments were run for less than 5 hours. We have analyzed the leachate for trace metal concentrations. I am using a dual-porosity model with physical non-equilibrium and only kinetic sorption in the mobile region to model my data.

My concerns are:
1. As our experiment time is short and we have preferential flow. We won't have equilibrium sorption in the mobile region. The immobile region will have equilibrium sorption.

I took frac = (mobile water content/saturated water content), Alpha for trace metal was calculated using alpha for bromide and changing the diffusion coefficient in water, Frac_M =0, dispersivity, immobile water content, etc were taken from bromide tracer models (MIM model).

I only optimized Omega and Kd. Thus, I am considering linear sorption and my model works fine. But if I go for non-linear sorption (Freundlich) and try to optimize Beta (Freundlich co-efficient), My 95% C.F. becomes negative for 1 or more parameters.

I cannot go for batch experiments as our experiments are not long thus, adsorption coefficient values from those experiments would not represent our conditions. Moreover, as we have poultry at the top, as the conc. of influent (trace metal) changes with time we cannot use kinetics curves too as we don't have a constant conc. with time.

So, my only option is to optimize kd, beta, and omega. Is there any way, to get +ve C.F for all the optimized parameters?

Thank you.

With regards,
Vishawjot Singh Sandhu

[Jirka]

You should always work the simplest model (with the smallest number of parameters) that can explain your data, since if you use more complex models, you may be overfitting your data and get nonunique estimates of optimized parameters (large parameter errors and confidence intervals). If the linear model works, there is very little reason to try to fit a nonlinear model (with more parameters and thus more parameter uncertainty). J.

[vss0015@auburn.edu]

Thank you Dr. Jirka for your response. I will go for linear sorption and a less complex model then.

Have a good day sir.