Bay Area weather is largely influenced by the synoptic state of the atmosphere. Lying on the western edge of mid-latitude North America, the region typically experiences westerly air flow from the Pacific Ocean and channeling through the Bay Area, whose delta provides a gap in the Coast Ranges which parallel the Pacific Coast. Synoptic conditions generally determine the strength and penetration of this marine layer flow and thus the degree of dispersion. Trough patterns aligned along the Pacific coast tend to increase ventilation and improve air quality, while high pressure ridges, both onshore and offshore, tend to reduce dispersion and allow tropospheric pollutants to accumulate to harmful levels in various locations.
Coarsely dividing Bay Area meteorology into only four classes provides a conceptual model of local air quality by identifying two high pressure conditions with distinct transport characteristics conducive to ozone buildup. However, using so few atmospheric states is rather limiting, as evidenced by the fact that there remains significant variability in ozone levels within each cluster. For example, two high pressure clusters contain almost all of the exceedance days, yet also have many days with considerably lower ozone levels. Here, we seek to expand the scope of the wind field clustering to identify additional atmospheric states, beyond the simple four-pattern model proposed previously, to further explain the effect of meteorology on Bay Area ozone levels.
An obvious method by which cluster analysis results can be refined is by forcing the clustering algorithm to output a solution with more clusters. There are two major disadvantages to this approach, however. First, as the number of clusters grows larger, their population sizes must decrease, resulting in decreased sample sizes for the statistical calculations. Second, a clustering algorithm cannot be guaranteed to identify new atmospheric states simply because the user queries for a larger number of such patterns. Previously, we found that four clusters capture distinct diurnal cycle patterns for surface air flow with an associated ozone response. Increasing the number of clusters did not produce additional discernable atmospheric conditions, and thus some other measure is required to further explain Bay Area ozone levels.
Because the cluster labels are generated continuously in time (for 122 consecutive summer days of 8 years), the time series nature of the cluster solution can be used to provide interpretation beyond considering the cluster patterns themselves, which are essentially static in time at the synoptic scale. Assignment of a particular cluster label to some day indicates the presence of a certain synoptic feature (ridge or trough) at some general location (onshore or offshore), however the evolution of the synoptic state, or the trajectories in time of the synoptic features, are not captured by the clusters themselves. Considering the sequence of cluster labels, however, can identify key atmospheric transitions that have important implications for Bay Area air quality. Additionally, the sequence of cluster labels can indicate certain periods of time for which the static cluster patterns do not adequately capture the actual prevailing meteorology, revealing new types of atmospheric activity that may affect local air quality.
Binomial statistics with small sample size corrections are used to model the probability of a transition occurring (moving forward in time) between each pair of clusters. Estimates for the true transition probabilities are compared to the null hypothesis that all state transitions are equally likely. Two major types of events are determined from such hypothesis testing: events occurring more often than would be expected by chance, termed the “favored” transitions, and the “disfavored” transitions occurring less frequency than by chance.
Recurring atmospheric phenomena drive the favored transitions. For example, cluster #4 is frequently followed by #1. This situation represents a high pressure ridge forming over the Pacific Ocean then moving from west to east over the Bay Area, producing some of the highest ozone levels on record and typically resulting in multiday NAAQS exceedance periods. While both clusters #1 and #4 experience generally poor air quality, the worst air quality is experienced on the days of transition between these two states.
The presence of disfavored transitions in the cluster solution suggests a trajectory of the synoptic state that is not physically realistic. For example, the transition of cluster #1 to #4 (reverse direction as above) would appear to correspond to an onshore high pressure system migrating westward to the Pacific Ocean, which would be anomalous. Instead, occurrences of such disfavored transitions in the cluster labels are taken to indicate inconsistencies in the cluster solution, not physically anomalous atmospheric states, and may indicate a type of event that is not revealed in the static four-cluster solution. For example, the transition of #1 to #4 typically captures conditions with an offshore low pressure cell in addition to the typical high pressure ridge.
Modeling cluster transition probabilities using binomial statistics provides further insight from the results of a previous cluster analysis of Bay Area wind field patterns. The presentation will elaborate on the physical nature of the identified atmospheric transitions and their implications for air quality in the Bay Area.