A method for assessing the impact of historical weather dataset size when doing long-term electricity system planning

With the current large-scale uptake of variable renewable electricity, the impact of interannual variation in power production from these electricity sources must be accounted for in long-term planning. The values used to characterise renewable production during the planning process are generally based on historical weather data. Production variability is influenced by the climate zone and resource quality at the location of production and by the geographical distribution of production locations. Additionally, the required level of accuracy in renewable production representation will depend on renewable penetration, particularly when doing modelling to establish electricity system reliability. Thus, there can be no one-size-fits-all recommendation for the number of years of weather data that needs to be included in electricity model input datasets. Any attempt at this in the literature is subject to several caveats. In this paper, it is hypothesised that analysing how well a sample of a given size represents the behaviour of the subsequent 20 years will give valuable insights into appropriate dataset size selection. A methodology is developed, using a South African case study, to produce prediction error bounds for annual mean production, seasonal mean production, and minimum and maximum production over one-, seven-, 30-day and annual periods for wind or PV production based on resource quality or climatic zones. These calculated error bounds are then combined to produce estimated production error bounds for a fleet of generators. The error bounds contextualise the impact of dataset size on production projection accuracy, to guide decision-making.