Event Classification Methods Comparison and Validation

As a means of validation for the methods we use for identifying NAM surge events, a comparison between those methods of Fuller and Stensrud (2000) and our methods will be shown.

To classify a surge event as 'valid,' it's occurrence must be concurrent with or proceed within two days a day where the accumulated daily precipitation for the Phoenix, Tucson and Yuma sites exceeds 0.13 inches, i.e., 80th percentile. Though there exist a potential for heavy rainfall at only one station to cause the threshold to be exceeded, using a validation method that depends on spatially-disperse precipitation readings lends itself to determine a procedure of surge classification that better identifies surges that produce widespread precipitation events.


Since, Fuller and Stensrud (2000) identified surge events in July and August, events that our methods found in June and September for these years will be ignored for the validation.


Results for the period of July-August for 1980-1984:

Rainy days above 80th percentile: 36

Events:
Fuller and Stensrud (2000) = 35
Current study = 20

Surge events with rain event within 2 days, i.e., valid events:
Fuller and Stensrud (2000) = 24 (69%)
Current study = 16 (80%)


Though Fuller and Stensrud's methods identified more valid surge events, our methods identified valid surge events with a greater precision. By using the methods of Fuller and Stensrud (2000) one would identify more valid surge events during this period, but they would also risk identifying an increased percentage of invalid surge events, a.ka. false-alarm surges, 31% compared 20% using our current methods.


-jamie

Surge Classification

There has been two main issues that I have seen with what has been done in the literature to extract surge events from the data:

(1) Only using data from one or two stations that could, at times, produce 'false surge' results due to their proximity to the GoC
(2) Not using precipitation as a classification variable

To avoid (1) we are using a spatially and temporally continuous dataset over a broad region that is tied to the NAM system. But, as of yet, I have not done much about (2). In my opinion, using dewpoint as a criteria for a surge event and not precipitation seems to be missing the point of finding this events, i.e., forecasting precipitation for the arid southwestern US. To this end I have decided to try some new avenues for classifying these events. Below I will show what work I have done in regards to DWP, IWVF and PRCP as variables and new thresholds.

To get an understanding of how dynamic an indicator each of my proposed variables truly is, I calculated the coefficient of variation for each. The coefficient of variation is a measure of dispersion of a variable and is dependent on the standard deviation divided by the mean. The usefulness of this measure is that it provides a means to compare different variables with different units. The coefficient of variation for DWP, IWVF and PRCP were 0.0166, 0.7232, and 1.5035, respectively. Values of 1 or greater are an indicator of a highly dispersed variables. Therefore, PRCP is our most dynamic indicator. Though not exceeding 1, IWVF does show signs of being changeable. Notably, DWP shows very little to no variability. Fig 1 is a composite of the DWP anomaly field in association with days where PRCP meets or exceeds 1-sigma over its mean (more on this criteria later). The darkest blue (red) is related to an anomaly of 0 (4) degrees Celsius above the mean. DWP shows some activity, but, interestingly, this activity is mainly located in extreme southern Nevada and not the core region normally associated with NAM activity in the US. I want to point out that where most previous studies have relied on data from, i.e., Yuma, actually shows little to no change during these events.

Fig 1: DWP Anomalies where dark blue (red) corresponds to 0 (4) degrees Celsius anomalies.

So, from this, I have come up with the following criteria for classification:

Event = IWVF meets or exceeds 1-sigma over mean on the same day that PRCP meets or exceeds 1-sigma over mean.

Some of this stems from conversations with Dr. Cordero about statistics (or what the BeeGee's might call "Stat Talkin' ") and how to find physically meaningful characteristics through statistical analysis. Initially, I was looking for events of 2-sigma above mean, but this did not provide results that I felt were very realistic, i.e., very few events. DWP never actually exceeds 2-sigma in our study period. Here our some of the numbers to go along with this criteria:

# of Events (1980 - 2007) = 172
~ 6.14 per year
~ 1.54 per month

I believe, and also through talks with Mike, that 1.5 events per month is a more realistic climatological number of surge events, not 3.

Here are some of the intra-seasonality of the numbers:
June: 3
July: 53
August: 64
September: 52

Fig 2 is a composite of IWVF anomalies, mean fields of Meridional IWVF, and PRCP anomalies associated with these events.

Fig 2: Composites for IWVF anomaly (top), mean Meridional IWVF (middle) and PRCP anomaly (bottom) fields in relation to events. For Meridional IWVF dark blue (red) corresponds to 0 (20), and for PRCP anomalies dark blue (red) corresponds to 0 (6) mm per day anomalies.

Fig 3 are the results for July of 1986 compared to those of Fuller and Stensrud (2000), their Fig 2. Their events are highlighted in blue (W = weak and S = strong events) and mine are in red. Units for IWVF (red, dashed line) and PRCP (blue, solid line) are on the left (green, dotted line) and DWP is on the right.

I was not able to put this image on the site because I updated it with PPT. Here is a link: http://www.met.sjsu.edu/~favors/Research/July1986_IWVFPRCPDWP.ppt

EOF of 500 hPa Heights: Comparing July/August to mid-August/September

Below will be a look at the variability due to seasonality in the EOF components. This look is based on the fact that through my event classification scheme the preponderance of events occur later in the year, late August through September, then seen in previous works, July through August.

Figure 1: Principal Component 1 for both seasons

Figure 2: Principal Component 2 for both seasons
Figure 3: Variance of the first 10 components for both seasons

With PC #1 a well defined trough (lower heights region) is present over central Alberta that stretches southward into southern California. But, with a shift in the time frame from July/August to mid-August/September we see this feature become much more well defined. This makes physical sense (we are moving closer to a time of year where more mid-latitude 'action' is found further south) and it also falls in line with our findings of trough interactions being important components or drivers of surge events in the late Summer.

Comparing the first to the second principal component shows that the July/August season is picking up more of a trough signature than that of the later season. This trough feature is associated with roughly 20% of the variance for mid-Aug/September and roughly 14% for July/August. It is reassuring to see that this trough feature is the leading component for the mid-Aug/September season in that it lends more foundation to our findings of the importance of these features.

The one issue I am having qualms with is the absence of the monsoon ridge over the Four Corners region. I believe we could be seeing some indication of this feature in PC #1 for July/August, but it appears very weak in nature. The problem could be that we are simply looking slightly too high in the atmosphere to pick out a strong signal with this feature, or perhaps my domain is slightly too large to pick out what might be considered a subtle feature in context of the overall synoptic and planetary patterns.

-jamie

500 hPa Height Pattern Composite

Here is a composite of the 500 hPa height pattern associated with strong events. Though we were questioning whether the 500 hPa height anomalies were showing weakening of the monsoonal ridge or an actual trough, it appears quite clear from the actual height patterns that the anomaly feature is indeed associated with a strong trough moving in from the Pacific.

Ridging, though relatively weak in nature, does appear to be in place slightly east of the Four Corners (i.e., Four Corners' High) on Day -4 and -3 to the event. At this time we are already seeing signs of a deepening trough off Northern California. As we progress closer to the event day the trough deepens dramatically, nearly explosively, and progrades into regions where it could potentially affect the characteristics and development of the NAM system. The actual height patterns and the anomaly fields display a strong spatial relationship to one another, and thus, I feel confident say that the anomalies we are seeing are associated with a trough feature.

There are still quite a few questions regarding the origin of this trough feature though. More on that to come soon.

Same Bat-channel...Same Bat-Time.

-jamie

Large Scale Composites for Northward NAM Surge Events

Following the same outline as laid-out in the most recent post regarding composite analysis, I will now show composites for large scale fields in relation to defined surge events.

Weak Surge

Weak Surge: Non-common with Strong Surges

Again, plots are shown in three day prior through 3 days after a surge event (middle column).

For these figures the top row is precipitation anomalies (0 to 5 mm/day), the second is 200 hPa isotachs (50 to 65 mph= Light Blue, 66 to 80 mph = Yellow), the third is 500 hPa height anomalies (-120 to 120 meters, blues = negative anomalies, yellows = positive anomalies) and the fourth row is 850 to 500 hPa lapse rate anomalies (-5 to 5 degrees Celsius, same color conventions as height anomalies).

Though I showed both for comparison, I will limit my discussion to only the non-common weak event plot since I feel it is more indicative of the true weak events. Isotachs show relatively zonal flow. From the 500 hPa height anomalies we notice pronounced ridging (positive anomalies) over the northern Pacific, and from event day on we notice troughing (negative anomalies) over the western United States. The ridging over the northern Pacific could potentially be a related to a slightly northwestward shifted North Pacific High. The ridging followed by troughing over the western United States is most likely related to surface warming then cooling (associated with precipitation) ahead and behind, respectively, of the surge event.

850 to 500 lapse rate anomalies show a clear, and persistent pattern throughout the composite. A couplet of positive and negative anomalies appear adjacent to one another over northern Mexico during the preceding days till the day after the event. This feature I am proposing is associated with some voriticty feature, either an easterly wave or some smaller scale vorticity lobe (TUTT?), that is causing upward and downward motions in the atmosphere. The upward (downward) motions are causing a stabilizing (destabilizing) of the atmosphere, i.e., negative (positive) lapse rate anomalies.

The patterns seen in these figures fit well into the classical model of how a 'typical' NAM surge event should happen.

And now onto the interesting stuff...

Strong Surge

First, let's point out what appears to be similar. From the 850-500 hPa lapse rate anomalies we are still seeing the stabilizing of the atmosphere post surge. From the 500 hPa height anomalies we see the increased heights (surface warming) over the western United States associated with some thermal ridging. From that point on things seem different...very different.

200 hPa isotachs are now less zonal and are showing some signs of a trough progressing inland from the Pacific Ocean. Most striking off all are the results shown on the 500 hPa height anomalies that stand out from previous examples. The composite shows a dramatic and well-defined area of negative height anomalies (up to -120 meters!) moving in off the west coast. The feature is most certainly an indication of a strong trough feature, the exact type of extratropical control feature we have been hoping to see with our research. By comparing the precipitation anomalies with the 500 hPa height anomalies you can see that the precipitation stays just downwind adjacent to the incoming trough. Clearly, the trough feature is playing some role in progressing the NAM surge event system.

Here are some of my thoughts so far on how things are starting to look:
(1) Weak events, as we have defined them, might end up being the ones tied to the tropical, e.g., easterly wave, controls.
(2) Strong events where precipitation is seen further north into the Great Basin regions, i.e., Andy's work, might be the ones intimately tied to the extratropics.

As you can tell, I am more than a little excited by the results from the strong events.

-jamie

Peak Day of IWVF for JJAS

To provide a prospective on the time-series of IWVF over the JJAS 'season,' a figure showing the day of maximum IWVF (daily average for 28 years) is shown.
Days are broken down as follows:
June: 1-30
July: 31-61
August: 62-92
September:93-122

The data shows that for the region most associated with NAM surge precipitation, i.e., AZ, western NM, southern UT and NV, and southwestern CO, the peak occurence of IWVF typically falls later in the season between mid-August and mid-September. Our data has been pointing towards this fact, but it is still striking to see how much of the region is dominated by later-season IWVF than most previous research has accounted for. What I am specifically revering to is that most NAM surge research is limited to only July and August analysis, thereby ignoring the vast majority of IWVF events, and most likely surge events, that are taking place.

To back up this point, draw your attention to the Yuma, AZ location (extreme southwestern AZ; Data point where most previous research bases 'surge event' definitions on). The peak for this region appears around day 70 (early August). So, if data analysis were limited to this region a season of only July and August observations might make sense. But, as data points out over the larger domain, this is merely an artifact of where you gather data, not actually physical manifestations for the region.

There are some swearly data areas where very early season peaks are shown directly adjacent to very late season peaks, e.g., northern Mexico. I believe this is an indication of regions with little variability in IWVF amounts during the period analyzed, therefore, resulting in drastic changes in date but not necesarrily amounts.

-jamie

Composites of Anomalies

I believe there were some issues inherent to the way I was previously defining two day events in the IWVF field. For example, in a situation where there were four days straight above 90th percentile, then my code would have registered 3 events. To combat this flaw I went through and defined a two day event as:

2 consecutive days above 90th percentile
for a zone where event day - 1 must not meet criteria

This appears to have helped out very well, and thus, I will show new and improved (and shiny, I might add) results.

Note: Since MATLAB disagrees with you about how large certain subplots should be when you start adding in colorbars, I will simply state data ranges here:

DWP Anomalies: -2:5
IWV Anomalies: -3:8
Precip Anomalies: 0:5

These restrictions are based on observed values in the data and which value range best described the data.

(1) Event for SAZ Zone
The figure includes IWVF and anomalies of Dewpoint, Precipitable Water and Precipitation. Anomalies also helped to weed-out patterns present in these fields that were not easily decernded from the basic field values. We clear see a northward flux of dewpoint, precipitable water and precipitation in relation to IWVF events. Precipitable water shows the most concurrent signature with IWVF, which should be expected as these fields are based on similar data. We can now see the slightly delayed, by one or two days, flux of precipitation and dewpoint into the AZ, NM and UT domains. As described more fully in a previous post, this is a feature we expect to see. So, things are looking good.

(2) Event for SAZ and NAZ Zones

Features are similar in this figure as with (1) except that we are seeing a stronger signal in all fields compared to (1). This should be expected since we are further refining our events to cases where IWVF moved more northward, thus a strong flux.

(3) All Zones (1 Day Criteria)

So, the final figure is for an event that shows up in all three zones on a given day. This event is not restricted to 2 consecutive days criteria as with above examples. Most of the signatures appear before event day for this criteria. What we are pretty much looking at in this 'event day' is the height of activity, i.e., furthest northward flux. If we were defining events of 2 consecutive days for SAZ and NAZ then it would probably show up on Day-1 here. Again, signals are very amplified here, as we should expect.

Well, I am pretty to call events defined. So, Events = Defined:

Weak Event = 2 consecutive days of IWVF at or above 90th percentile for SAZ
Moderate Event = 2 consecutive days of IWVF at or above 90th percentile for SAZ and NAZ
Strong Event = 1 day of IWVF at or above 90th percentile for SAZ, NAZ and SUT

-jamie