A Quantitative Assessment of Uncertainty Reduction as a Function of Measurement Campaign Length Using Linear and Machine-Learning MCP Models

This study evaluates the impact of measurement campaign duration on wind resource characterization using three MCP (Measure–Correlate–Predict) models: Total Least Squares (TLS), Multiple Linear Regression (LR), and Quantile Gradient Boosting (GB). The analysis is based on data from 30 meteorological masts (nine primary and twenty-one secondary masts) installed worldwide across different terrains, with up to twenty-seven months of concurrent wind measurements between primary and secondary masts. Fixed campaign durations of 3, 4, 5, 6, 9, and 12 months were simulated using moving intervals to quantify the effect of measurement length on mean wind speed estimation. This working framework also serves to represent conditions typical of campaigns where LIDAR systems are used to complement meteorological mast deployments, as LIDAR units generally operate for shorter periods due to frequent relocation as part of broader measurement strategies. Wind speed estimation was assessed through metrics such as Mean Absolute Error (MAE), relative uncertainty, and monthly uncertainty reduction, taking into account terrain complexity and correlation coefficient (R2) between masts. Results indicate that extending the measurement period improves the accuracy and consistency of wind speed estimates, with significant reductions in uncertainty observed after six months. Across sites, the average monthly uncertainty reduction ranges from 0.13% to 0.41% of the mean wind speed per additional month of measurements, depending on terrain complexity and inter-mast correlation. Linear models (TLS and LR) consistently show better performance in terms of error and uncertainty reduction compared to GB. Based on an extensive and diverse MCP dataset covering multiple terrains and locations, this study provides empirically derived monthly uncertainty-reduction benchmarks for campaign-length optimisation under different site conditions, contributing to more reliable wind resource assessments and, consequently, energy yield estimates.