SUPPLEMENTARY 1- SPATIAL AND TEMPORAL
PATTERNS OF ICE BREAKUP DATES IN MAINE LAKES
To assess whether the PC1 for the ice-out date of eight Maine lakes used in this
study is a good representation of the dominant temporal pattern of ice out date in Maine
lakes, we performed Principal component analysis (PCA) on the ice out date of sixteen
Maine lakes for the period between 1950-2010 with some years missing. The leading
principal component (PC1) retains 79% of the total variance in ice out date of the sixteen
lakes and therefore may be considered as the leading pattern of ice out date variability
(see table next page). Figure 1a shows the strong similarity between the time series of the
PC1 of the two sets of lakes and the correlation is 0.98 (p<0.05). Furthermore the
homogeneity of signs in the sixteen lakes indicates that there is coherence between these
lakes in their pattern of ice out dates, which was also observed in the PC1 of the eight
lakes (see figure 1b).
Figure 1. The time series and loadings for the first principal component (PC1) of lake ice
out dates in Maine for the period 1950-2010. For comparison the PC1 of ice out date for
two sets of lakes is shown: one set containing eight lakes with no missing ice out date
and another set having sixteen lakes with six years missing. (a) Time series of PC1 for
the first and second sets of lakes. The correlation between the two time series was 0.98
(p<0.05). (b) Lake loading of PC1 for the ice out date of the second set with sixteen
lakes.
Table 1a. Results of Principal component analysis (PCA) for ice-off dates of sixteen lakes
(with missing years)
Summary table on the principal components (PC1s)
Standard deviation
Proportion of Variance
Cumulative Proportion
PC1
29.087
0.791
0.792
PC2
11.048
0.114
0.905
PC3
4.948
0.023
0.928
PC4
4.144
0.016
0.944
Loadings of Lakes in PC1 and PC2
Lake
Mooselucmeguntic
Richardson
Rangeley
Aziscohos
Umbagog
Moosehead
Portage
Sebec
Kezar
West Grand
Auburn
Swan
Maranacook
China
Cobbosseecontee
Damariscotta
Loading of PC1
-0.197
-0.197
-0.198
-0.183
-0.181
-0.217
-0.149
-0.208
-0.208
-0.284
-0.287
-0.290
-0.288
-0.314
-0.311
-0.368
Loading of PC2
-0.351
-0.338
-0.329
-0.326
-0.296
-0.267
-0.251
-0.099
-0.046
0.070
0.171
0.176
0.206
0.222
0.227
0.325
PC5
3.611
0.012
0.956
PC6
3.154
0.009
0.966
PC7
2.745
0.007
0.973
PC8
2.595
0.006
0.979
SUPPLEMENTARY 2- INFLUENCE OF NORTHERN
HEMISPHERE SEA SURFACE TEMPERATURES ON
MAINE’S WINTER TEMPERATURE
For the sake of comprehensiveness, the PC1 time series of winter AFDD and
AMDD in Maine were correlated with sea surface temperatures and upper air (500mb)
geopotential heights to determine other oceanic-atmospheric circulation patterns that
influence winter temperature variability and in turn spring ice out date in Maine. The PC1
of winter AFDD and AMDD showed a significant correlation largely with sea surface
temperature anomalies along North Atlantic and Pacific and Tropical Atlantic (see figure
next page). This result indicates that teleconnection patterns such as Atlantic tripole
(ATI), Atlantic multidecadal oscillation (AMO) and Pacific decadal oscillation (PDO)
with sensitivity to sea surface anomalies in this region may shape inter-annual variability
of winter degree-days and spring ice off dates in Maine (see table next page). However
the mechanisms that underlie the mode of these sea surface temperature anomalies (and
their teleconnections), their persistence and seasonality are less understood. Thus, we
limit our focus to the understanding the season-ahead causal relationships between select
teleconnection patterns, suitable degree-day indices in the context of lake ice out.
Figure 1. The 1950-2010 composite winter sea surface temperature and 500mb geopotential height anomaly maps correlated against the time series of PC1 for winter AFDD (top)
and PC1 for winter AMDD in Maine (bottom). Correlation coefficients exceeding 0.25 are
significant at 95% confidence level.
6
Year
Year
PC1 of Winter AFDD
PC1 of Winter AMDD
PC1 of Spring AFDD
PC1 of Spring AMDD
Winter TNH Index
Winter NAO Index
Winter PNA Index
Winter PDO Index
Winter AMO index
Winter AT index
Winter NP index
Winter AO index
October SAI index
Lake Ice out dates
1
-0.15
-0.09
-0.07
0.24
0.14
0.44
0.35
0.4
-0.16
0.21
-0.2
0.3
0.16
-0.26
PC1 of
Winter
AFDD
PC1 of
Winter
AMDD
PC1 of
Spring
AFDD
PC1 of
Spring
AMDD
Winter
TNH
Index
Winter
NAO
Index
Winter
PNA
Index
Winter
PDO
Index
Winter
AMO
Index
Winter
AT
index
Winter
NP
Index
Winter
AO
index
October
SAI
index
Lake
ice out
dates
1
-0.5
0.2
0.26
0.38
0.06
0.1
0.16
-0.50
-0.41
0.13
0.01
-0.52
0.38
1
-0.14
-0.26
-0.24
0.44
-0.07
-0.06
0.11
0.17
-0.13
-0.31
0.56
-0.51
1
0.61
-0.17
-0.02
0.19
-0.03
0.08
0.01
-0.18
-0.2
-0.38
0.13
1
-0.07
0.19
0.29
0.26
-0.05
-0.01
-0.19
0.06
-0.47
-0.85
1
0.12
-0.11
-0.09
-0.15
-0.14
0.3
0.26
-0.56
-0.27
1
-0.05
-0.09
-0.14
-0.11
0.12
0.82
-0.79
0.21
1
0.74
0.05
0.17
-0.87
-0.21
0.72
0.29
1
0.01
0.12
-0.66
-0.24
0.5
0.24
1
0.85
-0.01
-0.11
-0.07
0.08
1
-0.1
-0.09
-0.27
-0.15
1
0.37
-0.57
0.14
1
-0.8
-0.1
1
-0.43
1
Table 1. Correlation between the first principal component (PC1) of lake ice out dates, winter and spring degree-days and winter
teleconnection patterns. Correlation coefficients in bold are significant at p<0.05.
7
SUPPLEMENTARY 3- ANALYZING THE EFFECT OF
CHANGING THE TELECONNECTION THRESHOLDS ON THE
RESULTS OF THIS STUDY
In this paper, analyses and observations on the influence of “strong” or “extreme” teleconnection
patterns on lake ice out dates/winter degree-days in Maine is based on the threshold of the
selected teleconnection patterns being in the upper and lower quartiles of the historical indices.
The selection of this threshold was made based on the limited sample size (n=61) and higher
thresholds would limit the number of subsamples that can be re-sampled. In our case, we have a
sample size of 15, wherein mean, median, and box-plot statistics can be meaningfully computed
and compared.
However, to explore if alterations in this threshold would result in a whole sale change in our
observations and conclusions about the efficacy of TNH and NAO phases in engendering early
spring ice breakup dates in Maine lakes, the significance of the shift in the median ice out date of
select eight lakes during different lower/upper percentiles (0.1, 0.15, 0.2, 0.25, 0.3) of
TNH/NAO phases were compared with the unconditional median ice out dates for randomly
chosen subsamples years in the study period (see figure on next page). It should be noted that the
size of the random sub-samples depends on the percentile chosen. For instance, for the
upper/lower terciles of TNH/NAO patterns, the subsample size for the random sub-sample is six
(0.1x61=6). It can be observed in figure 1 that indeed strongly negative TNH phases during
winter induces a shift towards earlier dates in the spring ice out date of lakes for all regions of
Maine although the shift in the median ice out dates for one or two lakes is not highly significant
(p<0.05) during the different percentiles. It can also be noted that the influence of strongly
8
positive NAO phases during winter in engendering earlier spring lake ice out dates is limited to
the coastal regions since in most cases the shift in the median ice out date of northern or southern
interior lakes is not significant (p<0.1). The basis for such observations is that the strength of
positive NAO phases in producing warmer winter degree-days is weak in the interior regions as
observed in the figure in appendix D.
9
Figure 1. Shifts in median ice out dates and its significance for eight selected lakes from 1950-2010
during (a) lower tercile TNH pattern (TNH<-0.9) and upper tercile of NAO (NAO>0.62) phases (b)
lower 15th percentile TNH pattern (TNH<-0.75) and upper 15th percentile NAO (NAO>0.4) phases (c)
lower 20th percentile TNH phases (TNH<-0.63) and upper 20th percentile NAO (NAO>0.29) phases (d)
lower quartile TNH indices (TNH<-0.47) and upper quartile NAO phases (NAO>0.2) (e) lower 30th
10
percentile TNH indices (TNH<-0.34) and upper quartile NAO phases (NAO>0.16). The variability band
for the median ice out date of each lake was constructed using the bootstrap method. Solid lines show
10th-90th percentile range while dotted lines indicated confidence interval at 5th-95th percentile range.
The filled triangles represent the median ice out date of each lake either during lower and upper tercile
TNH and NAO phases.