Dosing forecasts from rainfall data: a simple planning model

The best forecast we have seen on a leachate site was three columns in a spreadsheet, maintained by a supervisor who had never taken a statistics course. It predicted polymer use for the coming week within 12 percent, which was better than the consultant model the group had paid for the year before.

Forecasting does not need to be clever. It needs to be close enough to order the right tonnage and staff the right shifts, and it needs to be updated often enough that it does not quietly drift away from reality.

Three inputs are enough to start

Rainfall, sump level and conductivity. Rainfall gives you the inflow; sump level gives you the current inventory and the trend; conductivity gives you the strength, which sets the dose rate. Everything else is refinement.

The arithmetic is short. Take the seven-day rainfall total and multiply by the effective catchment area and a runoff coefficient you have estimated from past events. That gives an inflow volume. Add it to the base flow, which is the volume arriving on dry days, and you have the week’s likely throughput. Multiply by the dose rate from the conductivity curve and you have polymer demand.

Calibrating the runoff coefficient from your own records

Pick three past rain events with good data: total rainfall, sump rise over the following 48 hours, and any tanker movements that removed volume. Back-calculate the coefficient for each. Most sites land between 0.2 and 0.5, and the variation between events tells you how much the cover and the drainage network matter.

Do this once and the model stops being theoretical. A coefficient derived from your own site beats a textbook value every time, because it already contains the cracks, the capped areas, the French drains and the neighbour’s runoff that nobody documented.

What the model cannot see

Three things will break a rainfall model, and it is worth naming them. Recirculation changes the base flow without changing the rain. A change in waste composition — a new industrial customer, a closed cell being capped — changes the strength without changing the volume. And a membrane plant coming online changes which stream reaches the pad at all.

The model should therefore be treated as a planning tool, not a dosing instruction. Its job is to answer “how many tonnes do I order” and “do I need a second operator on Thursday”, not “what dose do I run at 14:00”. The dose still comes from a test on the actual liquid.

Review the forecast against actual use every Monday. A persistent 15 percent over-prediction usually means the runoff coefficient is too high or the base flow includes recirculation. A persistent under-prediction usually means unmeasured inflow, often from a neighbouring property or a broken drain.

Keep the sheet short enough that a busy supervisor will still fill it in during a wet week. Three columns and five rows beat a forty-tab workbook that nobody opens after the first month, and a rough forecast that gets used every week is worth more than an accurate one that gets abandoned in February.

The model that stopped a stock-out

A site with a 90-day rainfall outlook started using the three-column sheet in autumn. In the first week of November the forecast called for 70 mm and 1.9 tonnes of polymer against 0.9 tonnes in stock. The supervisor ordered early, which felt excessive at the time because the sky was clear.

The rain arrived on the Thursday. By Sunday the site had processed 380 m³ more than a normal week and used 1.7 tonnes. The delivery landed on Friday morning. Without the order, the pad would have run dry on Saturday afternoon with two tankers already booked.

None of that required software. It required someone to look at the gauge, the sump and the store count on the same morning, and to write the number down. If you want a starting template, the leachate treatment guide has a weekly planning sheet that matches this structure.