Validation of an Arsenic Rejection Computer Model for RO/NF Membranes

Arsenic contamination in groundwater supplies has long been a problem, and membrane technology is among the efficient methods for its removal for providing safe drinking water.

In 2004, the EPA decreased the maximum contaminant level (MCL) for Arsenic from 50 ppb to only 10 ppb. This made compliance more challenging for many existing facilities, while also creating additional design considerations for future membrane plants.

Currently, no membrane projection software predicts arsenic rejection, and no membrane manufacturers guarantee arsenic concentrations in the finished water quality. The only way to predict arsenic rejection through membrane systems is therefore to perform piloting at the location, which can be both costly and time consuming. Some systems have been “overengineered” using ferric coagulation with filtration as pretreatment for reverse osmosis.

It is known that As(V) is better rejected by membranes than As(III), and that conversion of the latter to the higher oxidation state can be achieved by chlorination. It is therefore important to have the ability to predict membrane rejection of both oxidation states to determine whether chlorination is necessary.

Using the weak acid properties of Arsenous and Arsenic acid, AKA As(III) and As(V), a computer model for rejection by RO membranes was developed based on the pH dependent valences of the various weak acid species. The weak acid constants (pKa) were corrected for temperature and ionic strength using the Van’t Hoff equation, and a modified Debye-Huckel equation respectively.

Membranes from various manufacturers were tested to validate the models. These included high rejection membranes, low energy membranes, and Nanofiltration membranes. Standard manufacturer test solutions contaminated with known levels of As(III) or As(V) were applied at standard test temperature and pressure conditions in a cell test apparatus.

The testing at the membrane manufacturer QC test conditions allowed for easy normalization of the data to full scale systems for the referenced membrane models. A good correlation was found between the predicted rejection values for As(III) and As(V) as compared to the computer model.