To make sure that generation and load can be balanced at any moment, the required dispatchable generation capacity needs to be scheduled. For that reason we need an estimate of the expected residual load, i.e. the likely load and the contribution from vRES generation. Forecasts are uncertain and subject to deviations. This has to be taken into account in scheduling and allocation of balancing ancillary services.

Forecast errors mean that the system as a whole will not operate at the intended economic dispatch. The utilisation of reserves increases and this introduces extra costs to generation.
If forecast errors are much larger than foreseen, insufficient reserves may be available to make up for the deviation between schedule and real power capability of power plants. In this case system balance and, hence, security of supply may be compromised.
As a consequence, there is a relationship between the forecast error to be expected and the volume of reserves to allocate. Both values vary as the weather varies.

Specific meteorological conditions make it more difficult to produce good forecasts. Challenging conditions in the case of PV are dust, snow or fog.
In case of windpower, steep gradients caused by approaching storm fronts introduce major errors. Looking at the forecast for the day, the generation profile might be quite correct. But when the storm arrives one hour later than expected, in this very hour, the error may be huge.
In both cases, a high spatial concentration of installed capacity in areas with favourable resources makes the effect more problematic.

Forecast errors and the resulting imbalances, more precisely their compensation, causes costs. For that reason, good forecasts offer economic benefits from the very beginning of vRES development. Starting from phase 2 , they are one of the inevitable preconditions of reliable and secure system operation.
Also with dispatchable plants only, system operators always had to be prepared for forecast errors: if a power plant unexpectedly trips, for example due to a technical problem, the associated sudden deficit has to be compensated as well. The likelihood and the value of the error in such a case is different than scheduling vRES. However, because reserves are allocated based on likelihood and not for a specific purpose or event, the required amounts do not necessarily grow with vRES capacity.

Planning
Expected forecast errors are one factor influencing the necessary amount of balancing reserves. Hence, calculation methods assessing reserve needs have to take forecast errors into account.
Very large vRES projects, like offshore wind farms of several hundreds of MW, have to be treated the same as a thermal power plant, what concerns unexpected outage and calculating impact on the reserves needed in the system.
Appropriate regulation is essential for effective allocation of reserves and setting the right incentives for minimising deviations, including those resulting from forecast errors.

Operation
Forecasting tools for vRES generation improved a lot during the last decades. There are professional service providers with relevant experience in all parts of the world. System operators developed their own tools as well and integrated them in their operational management systems. Ideally, the volume of reserves is scheduled like power plant dispatch, depending on the forecasted vRES generation AND its expected accuracy.
Note 1: forecasts for larger areas generally will be better than for individual sites and for longer averaging intervals. An illustrative example: the output of a single rooftop PV system in a 5 minute time window depends on the individual clouds passing by. These cannot be forecasted seriously. An estimate for the average cloud coverage in one hour and in a region with many small rooftop PV systems can be realistically provided. As the differences between all houses balance out, the regional forecast will tend to be more robust. Here is an important policy choice:

• do you contract an independent entity providing forecasts for a country or region; or
• do you make project operators responsible for providing their forecasts and combine them later at a higher aggregation level?

In the latter case, the individual errors might or might not smooth out. Generally, the project operators have less means to improve their individual forecasts in particular in the critical cases with steep changes. In turn they have better knowledge of specific factors influencing forecast accuracy, like maintenance outages and site conditions affecting the output.
Note 2: a particular challenge arises when PV net metering and residential storage are combined. Residential customers will aim to minimise their individual residual load. Their 'schedule' becomes less predictable for network operators, but even their suppliers. Forecast errors for these users may increase substantially.