1. ATLANTIC MULTIDECADAL
OSCILLATION
ANDTHE ENHANCEMENT OF ENSO ON
PRECIPITATION PATTERNS IN SOUTH FLORIDA
Alannah Irwin
Mississippi State University Department of Geosciences
Applied Meteorology Program
June 21, 2016
3. Purpose
• Investigate Atlantic Multidecadal Oscillation (AMO) and potential
enhancements of El Niño Southern Oscillation (ENSO) on rainfall
•Research region = South Florida (SFL)
• 1985 – 2015
• Important to understand connection btwn precip extremes & climate
variability
• Provide forecasters with better tools for more accurate forecasts in long-term
• Seasonal Outlook
•Yearly Forecasts
• Etc.
4. Background - AMO
• SST variation in equatorial Atlantic
Ocean
• Less known climate oscillation
• Variations correlate w/regional
increases/decreases in precipitation
• Changes occur on much larger
timescale
• AMO warm phase in SE US
• Decrease in precip
• Increase in temp &TC activity
•AMO cold phase in SE US
• Increase in precip
• Decrease in temp &TC activity
Northeast Fisheries Science Center - http://nefsc.noaa.gov/publications/crd/crd0911/
5. Background - ENSO
• SST variation in equatorial
Pacific Ocean
• Most frequently studied pattern
• Warm phase – SE US
• Cool/wet if forms in winter
• Dry/drought in summer
• Cool phase – SE US
• Warm/dry if forms in winter
• Increase hurricane activity in
summer
• Strength depends on where it
forms (EPAC vs CPAC)
http://serc.carleton.edu/eet/pmel/part_3.html
8. Background – South Florida
• Research region limited to SFL
• Seasonality in rainfall
• Dry season – Nov 1 – Apr 30
• Wet season – May 1 – Oct 31
• Consists of the KOE – Kissimmee-Okeechobee-
Everglades Watershed
• Hurricane landfall important for water balance and
ecosystem vitality
• Understanding climate patterns can help better
prepare for short- and long-term forecasts
Evergladesrestoration.gov
9. Methodology – Climate Data
• Gather AMO & ENSO data from 1985 –
2015
• Must coincide with defined seasonality
• Separate into dry & wet season averages
• Dry season begins Nov of previous year
• Wet season begins May of year in question
• Classify as warm/cold &
weak/moderate/strong
• Negative values = cold
• Positive values = warm
• After entire analysis is done, calculate
correlation using weighted coefficients
Event
Strength
Weak Moderate Strong
AMO (+/-) 0-
0.1°C SST
anomaly
(+/-) 0.1-
0.2°C SST
anomaly
(+/-) 0.2+°C
SST anomaly
ENSO (+/-) 0-
0.5°C SST
anomaly
(+/-) 0.5-1°C
SST
anomaly
(+/-) 1°C SST
anomaly
Data from http://www.aoml.noaa.gov/phod/amo_fig.php for AMO definitions and
http://ggweather.com/enso/oni.htm for ENSO events.
10. Methodology – Defining the Research Region
• Boundaries and stations w/in the
South Florida Water Management
District (SFWMD)
• Include all of KOEWatershed
• Rainfall sites come from within
SFWMD
• Easy to eliminate outlier data from
northern regions
http://floridaswater.com/history/images/5_districts1.gif
11.
12. Methodology – Obtaining Precipitation Data
• Rain stations must have complete period of
record (POR)
• From or before 1984
• Tipping bucket rain gauges provided by
SFWMD
• 23 sites used
• Full dry season POR n = 181 (182 leap year)
• Full wet season POR n = 184
• Seasonal averages of rainfall calculated for
analysis
http://novalynx.com/store/pc/260-6011-260-6021-Rain-Gauges-p269.htm
13. Methodology – Analysis using Correlation
• Correlation of season averages of
rainfall and climate data calculated
in R
• Best to determine relationship
• Repeat analysis using weighted
average of AMO and ENSO
• Smooth out and removes bias
• Multiplied average by total number
of occurrence
• Hypothesis testing done to further
determine correlation
Season and
Strength
AMO Weighted
Coefficient
ENSO Weighted
Coefficient
Dry/Strong 6 9
Dry/Moderate 14 12
Dry/Weak 11 10
Wet/Strong 14 5
Wet/Moderate 10 9
Wet/Weak 7 17
14. Methodology – Analysis Using Hypothesis
Testing
• Plot histograms and bell curves
to determine normal
distribution
• Create test, null and alternative
hypotheses, and determine
accepted significance interval
(p-value)
• Three HypothesisTests
• Using p value, “reject” or “do
not reject” null
Test Null Hypothesis
(Ho)
Alternative
Hypothesis (Ha)
Significance
Combined
AMO/ENSO
Ho = Rainfall is not
reliant on both
AMO and ENSO
Ha = Rainfall is reliant
on both AMO and
ENSO
For p ≤ 0.05, we
reject the null
For p > 0.05, we
cannot reject the
null
AMO Ho: Rainfall is not
reliant solely on
AMO patterns
Ha: Rainfall is reliant
solely on AMO
patterns
For p ≤ 0.05, we
reject the null
For p > 0.05, we
cannot reject the
null
ENSO Ho: Rainfall is not
reliant solely on
ENSO patterns
Ha: Rainfall is reliant
solely on ENSO
patterns
For p ≤ 0.05, we
reject the null
For p > 0.05, we
cannot reject the
null
17. Results – ANOVA
Ho = Rainfall is
not reliant on
both AMO and
ENSO
Ha = Rainfall is
reliant on both
AMO and ENSO
For p ≤ 0.05, we
reject the null
For p > 0.05, we
cannot reject the
null
Station Wet Dry
S331 0.434 0.065
S20F 0.340 0.007
Miami 0.343 0.096
S140 0.910 0.165
Ft Laud 0.297 0.731
East Beach 0.628 0.009
East Shore 0.720 0.053
Belle Glade 0.599 0.017
LWD.L38M 0.443 0.062
KPBI 0.862 0.012
Marco 0.901 0.002
Cork 0.749 0.021
Alico 0.631 0.021
Clewiston FS 0.285 0.008
S133 0.675 0.003
Basinger 0.494 0.005
S80 0.684 0.927
Blue Goose 0.964 0.050
Sebring 0.820 0.000
Creek 0.463 0.000
Beeline 0.936 0.015
S65 0.267 0.000
S49 0.765 0.510
Combined AMO/ENSO
Station Wet Dry
S331 0.24 0.84
S20F 0.99 0.49
Miami 0.19 0.77
S140 0.64 0.78
Ft Laud 0.27 0.64
East Beach 0.68 0.95
East Shore 0.40 0.27
Belle Glade 0.68 0.52
LWD.L38M 0.97 0.68
KPBI 0.65 0.51
Marco 0.49 0.26
Cork 0.55 0.84
Alico 0.48 0.89
Clewiston FS 0.49 0.46
S133 0.91 0.97
Basinger 0.23 0.53
S80 0.78 0.81
Blue Goose 0.92 0.72
Sebring 0.44 0.21
Creek 0.67 0.72
Beeline 0.79 0.28
S65 0.31 0.12
S49 0.46 0.64
AMO
Ho = Rainfall is
not reliant solely
on AMO
patterns
Ha = Rainfall is
reliant solely on
AMO patterns
For p ≤ 0.05, we
reject the null
For p > 0.05, we
cannot reject the
null
Station Wet Dry
S331 0.81 0.021
S20F 0.06 0.002
Miami 0.41 0.028
S140 0.86 0.076
Ft Laud 0.24 0.552
East Beach 0.69 0.004
East Shore 0.61 0.069
Belle Glade 0.35 0.008
LWD.L38M 0.03 0.020
KPBI 0.67 0.004
Marco 0.46 0.006
Cork 0.49 0.010
Alico 0.63 0.007
Clewiston FS 0.06 0.004
S133 0.29 0.001
Basinger 0.55 0.003
S80 0.21 0.720
Blue Goose 0.96 0.021
Sebring 0.90 0.00003886
Creek 0.18 0.0001198
Beeline 0.81 0.006
S65 0.29 0.000394
S49 0.78 0.508
ENSO
Ho = Rainfall is
not reliant
solely on
ENSO patterns
Ha = Rainfall is
reliant solely
on ENSO
patterns
For p ≤ 0.05,
we reject the
null
For p > 0.05,
we cannot
reject the null
18. Conclusions
• Correlation showed that individually, the climate
patterns do not impact rainfall
• All three tests inconclusive for wet season rainfall
• Dry season rainfall patterns showed significance for
both combined AMO/ENSO and ENSO tests at most
sites
• Implies that AMO enhances dry season rainfall patterns in
South Florida
• Dry season rainfall patterns had no significance in
AMO test
• Many possible reasons – would need further
investigation
19. Thank you!! I will
answer any questions at
this time
Editor's Notes
Location of rain gauges for this research project.
The correlation was used to show that independently, the climate patterns do not have an effect on rainfall. If there was correlation, then there would be no reason to conduct the hypothesis testing in this project.
As expected, the analysis shows that there is virtually no correlation between the individual patterns and rainfall. Some of the ENSO values are close to 0.5, but this is not strong enough to indicate a direct correlation between solely this oscillation and rainfall. This is why hypothesis testing needed to be done.
It is important to note that one pitfall of normal distribution is it is extremely sensitive to the number of bins or breaks used. For this analysis, the bins have been set at 31 for the sample size of this research.
In order for the null to be rejected, the significance must fall within the range of the p-value designated in the hypothesis test. If the p-value does not fall in this range, the null cannot be rejected. The p-value determines occurring, i.e., at a p-value of 0.05, there is a 5% chance of the null hypothesis occurring. Values within this range suggest that the alternative hypothesis is more likely to occur, meaning that the null can be rejected
These tests were performed using the weighted AMO and ENSO values discussed in the methodology in order to reduce the skewness created by the difference in magnitude for the anomalies in each climate pattern.
The values in yellow show that the null hypothesis can be rejected. While this does not mean that the alternative hypothesis is accepted, it means that further investigation is needed to conclude if there is a combined effect of AMO and ENSO
Values in red do not fall within the rejection region, but they should be of interest since they are relatively close to our 0.05 p-value. Had I used a significance value of 0.10, these values would have fit into the study. But 0.05 is the most widely-accepted p-value in statistical analysis.
There are many factors as to why the results turned out this way:
1.) Site location – These sites are located in either rural or urban areas. While urbanization was not included in this analysis, the location could have an impact on the results
2.) POR – The period of record
3.) Lack of hurricane landfall in Florida since 2005; future research can divide data into pre- and post-2005 to see if hurricane deficit impacts rainfall patterns. This will imply an indirect link to AMO and ENSO
