Identification of Sea Breeze Recirculation and Its Effects on Ozone in Houston, TX, During DISCOVER‐AQ 2013

Abstract

In coastal environments, sea breeze recirculation has been found to be an important mesoscale meteorological phenomenon that causes high ozone episodes, yet the identification of this small-scale circulation pattern remains difficult. In this study, a new method was developed to automatically identify sea breeze recirculation in Houston, TX, by applying K-Means clustering algorithm to surface winds measurements at near-coast sites during the DISCOVER-AQ (Deriving Information on Surface Conditions from Column and Vertically Resolved Observations Relevant to Air Quality) field campaign period from August to October 2013. The key to the clustering algorithm is seven features derived from site-based surface winds on each day, including zonal (U) and meridional (V) winds in the morning and afternoon, 24-hr transport direction (θ) and the recirculation factor, which is the ratio of net transport distance (L) to wind run distance (S). For comparison, the same clustering was applied to San Antonio, TX, a noncoastal city yet within the synoptic-scale distance from Houston. Four clusters were obtained for each region, including three synoptic patterns common to both regions and one mesoscale pattern that differs by region, classified as Stagnation and Sea Breeze Cluster for San Antonio and Houston, respectively. The clustering outputs were verified by wind profiler data in Houston. By linking the wind clusters with surface and aircraft ozone measurements, we revealed a clear connection between circulation patterns and daily ozone variability showing that maximum daily average 8-hr (MDA8) ozone levels and spatial distributions differed by cluster type (e.g., the highest ozone found in the Stagnation/Sea Breeze Cluster and the lowest ozone in the Southerly Cluster). This automatable method of sea breeze identification we developed can be potentially applied to other coastal cities because it has low data requirement and no ad hoc location-specific adjustments.

Classification of synoptic and local-scale wind patterns using k-means clustering in a Tyrrhenian coastal area (Italy)

“…A decrease in S (58.5 km) is also observed. These values are in agreement with the results in Li et al (2020): although the regions of interest have different morphological and climatological characteristics, the wind-integral quantities have similar values for comparable patterns. To appreciate the relative influence of synoptic and local atmospheric circulation on the different clusters we identified, Fig.…”

“…Clustering is carried out by considering seven objective parameters, based on the wind-integral quantities proposed by Allwine and Whiteman (1994) and on the diagnostics suggested by Li et al (2020). The former allow the evaluation of the daily ventilation/stagnation conditions, while the latter provide useful parameters to identify the onset and the cessation of the SBR.…”

“…However, the above integral quantities are insufficient to fully describe the dynamics of coastal circulation and to discriminate between different weather regimes. Additional diagnostics are thus needed to better characterize the SBR and isolate it from the concomitant synoptic weather conditions (Li et al 2020). We therefore adopted the following list of relevant objective parameters: The first four parameters were selected on consideration of the typical hour range that comprises the onset and decline of the SBR in the study area (Di Bernardino et al 2021c), regardless of the seasonal trend in insolation.…”

“…In recent years, clustering algorithms have been widely used for meteorological and climatological studies, although the choice of the input parameters represents a critical constraint for successful algorithm setup (see the review by Huth et al 2008 for details about clustering techniques and meteorological applications). Among these, thanks to its simplicity of implementation and computation efficiency, the k-means algorithm is widely used for the classification of environmental data-e.g., observations from meteorological stations (Li et al 2020), evapotranspiration measurements (Ferreira et al 2019), and air quality parameters (Adame et al 2012)-as well as for wind energy resource assessments (Al-Shammari et al 2016) and rainfall prediction (Geetha and Nasira 2014).…”

“…The investigated domain also includes the city of Rome, which is the largest and most populous Italian city. Following the procedure proposed by Li et al (2020), the clustering procedure only requires wind speed and direction data at the surface, measured at coastal stations. The minor modifications that have been introduced to better account for the region's peculiarities are described in Sect.…”