Evaluation of Model Error by Lidar Measurements during Various Atmospheric Phenomena Observed in Complex Terrain

Complex terrain generates regular wind-flow systems in the lower troposphere that recur with some frequency. When they occur often enough over a season or over an annual cycle, aspects of their behavior may show up in averages over those time intervals. Characterizing model errors for repetitive phenomena may be an important aspect of diagnosis of model problems (indicator of where the problems lie) and improvement of model wind speed simulations.

In this paper, the high-resolution measurements from scanning Doppler lidars deployed to three sites in the Columbia River Gorge region are used to validate model skill in forecasting wind speeds and estimate seasonal, annual model error as well for periods of frequent meteorological phenomena observed in this area. When such recurrent phenomena occur often enough, aspects of their behavior may show up in seasonal and annual-mean statistics. Large error differences were found for cold (winter, fall) and warm (spring, summer) seasons. In late fall and winter, a high frequency of weak winds due to frequent cold-pool occurrences contributes to higher relative frequency of weak winds in the annual distribution than seen at other times of the year. During spring, summer, and early fall the mean westerly wind-speed profiles for mostly nocturnal hours are stronger than daytime profile speeds by several m s^-1 due to frequent warm-season occurrences of diurnal marine-intrusion flows at night. Such flows result from the strong day/night heating-cooling cycle in the arid Columbia Basin.

These error characteristics also were evident in the annual averages. The Wasco bias vertical profile for the year indicated a positive bias of 0.5 m s^-1 through and above the rotor layer, reflecting the wintertime overpredictions at the site, given the small negative biases in summer. The 1 m s^-1 positive annual bias seen in the forecast lead-time plots at Wasco was half the 2 m s^-1 cold-season bias. The annual mean Arlington and Boardman profiles show a low bias of 1 m s^-1 despite small high biases in winter, demonstrating that the low bias was due to model performance during the warm-season. The largest errors during the warm season were due to the diurnal wind systems, so the annual low biases are reflective of errors in the model’s handling of these frequently occurring flows.

Model systematic inability to faithfully represent frequently occurring phenomena can affect annually averaged wind-speed distributions and calculated annual wind power. Annual statistics and the ability of NWP models to accurately simulate them are important for wind energy operations. Examples of error in annual wind power and annual energy production (AEP) statistics calculated for several virtual turbines of different hub-height and blade length will be presented.