Posters

Abstract

In this contribution, we present a novel method for estimating MCP consistency uncertainty that is site-specific and allows for year-by-year assessment of potential consistency changes without prior knowledge related to the input long-term data sets. We accomplish this by drawing inferences about the drift of long-term data relative to the site through observation of the drift of multiple long-term data sets relative to each other. Site wind assessments are typically based on high-quality site measurement campaigns spanning from several months to several years (“short-term” data). A process called Measure-Correlate-Predict (“MCP”) is common for adjusting short-term measurements to a longer period of climate conditions (typically ~20 years). MCP utilizes one or more long-term reference data sets (“long-term”), each consisting of either measured data from an off-site reference station or modelled data from reanalysis or a mesoscale simulation. MCP adds value to a site wind assessment by effectively extending the period of measurement and thereby reducing the uncertainty related to the temporal variability of the wind resource. However, it adds two new components of uncertainty: the first is the uncertainty in the statistical fit of whichever MCP process is selected; the second is the consistency uncertainty. Consistency uncertainty describes the potential bias in using a statistical transformation calibrated to one period of time to make predictions about site conditions in another period. Such biases may arise from a variety of causes, including sensor degradation, terrain changes, changes to model inputs and structure, or changes in large-scale climate patterns. As a recent example, there is potential that the Covid pandemic may have temporarily influenced reanalysis data quality by impacting the availability of aircraft weather data. The causes of change may be varied and subtle but nonetheless accrue meaningfully over time. Traditionally, consistency uncertainty is difficult to estimate because of the unknown nature of the relationship between long-term data and site conditions in the periods for which short-term data does not exist. Empirical estimates may be assessed from experience at other sites, but such estimates lack specificity when assessing new locations, and typically offer little insight into optimal lengths of long-term data for achieving the lowest overall uncertainty on a site assessment. Our solution to this challenge is to utilize multiple MCP-corrected long-term datasets at a site (e.g., ERA5, MERRA2 and CFSR reanalysis datasets), and to draw conceptually from a step-by-step random-walk process with multiple particles moving chaotically from a single point of origin. We use relative changes in year-over-year adjusted wind speed averages between the long-term datasets to estimate the scale of the year-over-year changes in the relationships between each of the long-term datasets and actual site conditions. This methodology has allowed us to estimate the evolution of consistency uncertainty as long-term measurement periods are extended, and thereby derive optimal periods for performance of MCP to minimize the combined uncertainties of climate variability, MCP fit and long-term consistency. From this, AEP uncertainty may be reduced versus simply using the maximal extent of available long-term data.