You should have performance curves for a single cell based on air and water flow/properties. Now increase the flow through a single remaining cell by the flow lost through the single cell divided by the number of remaining cells, and hopefully you're still within the boundaries of prediction.