Dud weather forecasts prove a headache for AEMO, particularly in biggest load centre

One of the key aspects of managing the electricity grid is knowing what supply is available, and what the demand is likely to be. Weather forecasts are a key component of both, but it looks like Australia’s Energy Market Operator has been getting some dud advice, particularly in winter, but also in heatwaves.

A new report quietly released on the AEMO website this week reveals that one of its weather forecasters was dumped last year because they weren’t up to scratch, but the error rate of the replacement and the other existing providers is still causing concern, especially when it is cold.

It turns out that one of the biggest headaches is in Penrith, in western Sydney, where the error rate is the highest. That’s a problem for AEMO because that area drives the biggest residential load in the country’s main grid.

Another problem area is Adelaide West, where AEMO reports that all its weather forecasting providers regularly over-forecast. “The systematic over-forecasting bias is evident for all providers in extreme cold temperatures,” it notes.

This matters because big errors in forecasts can have an impact on the grid, and cause a scramble to ensure enough capacity is available to meet demand.

One example was in Queensland in early July last year, when extremely cold conditions took the forecasters, and the market operator, by surprise. Variations in temperature forecasts, clouds and demand – the state recorded its highest winter demand of 8,716MW – created problems.

AEMO was caught out, and on both July 4 and July 5 it had to declare “lack of reserve” forecasts, and on July 5 had to trigger its emergency reserve trading mechanism and demand response (where customers are paid to reduce load) to ensure there were no outages.

This graph below illustrates the difference between demand predictions – based on weather forecasts – and the reality in Queensland on those days.

The AEMO report reveals some interesting foibles from the weather forecasters.

“Provider B” – the one brought in to replace the previously underperforming provider – emerged as the best performing forecaster across 4 hour, 24 hour, and 72 hour forecast timeframes.

But it also performed the worst in one hour time frames. Perhaps they forgot to look out the window, although that might not have changed much because it was the “early hours of the morning” where it was weakest.

Provider A – hitherto wobbly in the one hour forecasts – managed to lift its game in that department during last year’s winter, at least in comparison to the previous year. Provider C was consistent with the previous year.

AEMO says its response to this problem is to continue work with the providers to ensure that their tools are fit for purpose for energy forecasting, and to adopt a “probabilistic” approach to resource adequacy, and to include “uncertainty margins” in its reserve calculations.

It also wants to improve the observational network around the renewable energy zones that are being created across the grid to improve forecasting, and improve the tools assessing likely changes in output of variable renewable energy, particularly distributed solar PV.