patrickbdevaney/mia9 · Datasets at Hugging Face

text stringlengths 0 6.44k
in situ temperature measurements (Figure 2) collected at three locations (NDBC buoys)
along the Florida Keys and Biscayne Bay (Figure 1). The NDBC SST in Fowey Rock,
at the vicinity of Biscayne Bay (Buoy FWYF1), followed a clear seasonal variation from
January 2005 to May 2007 (Figure 2a) with a mean level of 26.6 ◦C (Figure 2b). The
satellite-derived timeseries followed a similar distribution revealing a very high correlation
(Rp: Pearson Correlation Coefficient) and a small Root Mean Square Error (RMSE) with
the field measurements (Rp = 0.97 and RMSE = 0.68 ◦C; Figure 2b); the respective mean
value (26.42 ◦C) is slightly smaller than the one measured directly at the sea surface. The
agreement between the two products is better at Buoy MLRF1, located at the Key Largo
coastal area (Figure 1), where both timeseries showed a similar increasing trend (Figure 2c)
and a linear regression very close to the x = y identity line (Figure 2d). The Pearson
correlation coefficient and the RMSE confirm the very good performance of the satellitederived data (Rp = 0.98 and RMSE = 0.50 ◦C; Figure 2d). The correlation is even higher at the
Water 2022, 14, 3840 7 of 28
lower Keys (Buoy SANF1; Figure 1) during 2005, when the Pearson coefficient is Rp = 0.99
and the RMSE is less than 0.5 ◦C (Figure 2f). The satellite-derived SST data are capable to
describe all short-term peaks and lows that exceed the typical range of the seasonal cycle;
the extremely high values in August 2005 (>31 ◦C) and the large drop of approximately
2
◦C at the end of the same month (28 ◦C) are apparent at both field and satellite timeseries
(Figure 2e). Thus, we conclude that the high-resolution satellite-derived SST daily fields
are suitable to investigate not only the open sea and shelf SST variability [44] but also the
temperature distribution and the formation of MHWs over the South Florida coastal region.
Water 2022, 14, x FOR PEER REVIEW 8 of 31
Figure 2. (a) Daily evolution of Sea Surface Temperature (SST; °C) derived from the satellite data
(black line) and the NDBC Stations (red dots) (a) FWYF1 (coastal area of Miami), (c) MLRF1 (UpperKeys), and (e) SANF1 (Lower-Keys). The respective scatter diagrams (b,d,f) between the two
timeseries for each case are also shown. The linear regressions (trends and timeseries comparison),
the mean values, the Pearson correlation coefficients (Rp), the tests of statistically significant correlation (pvalue), and the Root Mean Square Errors (RMSEs) for all cases are presented.
3.2. Interannual Variability of SST
The mean annual SST levels over the entire study domain revealed an increasing
trend during the 1981–2021 period, following the respective air temperature interannual
trend (Figure 3a). In addition to the Mann–Kendall method that identifies the trend, we
employ the Sen’s slope [54] to characterize the magnitude of the change. We found that
both trends are characterized by positive Sen’s Slopes of 0.19 °C/decade and coefficient of
determination R2 = 0.46 for SST, and 0.21 °C/decade and R2 = 0.42 for air temperature. The
increasing trends are statistically significant (pvalue < 0.01: statistical significance level 99%).
ThSSTldd263°Ci1982dhdthllf27°C40Figure 2. (a) Daily evolution of Sea Surface Temperature (SST; ◦C) derived from the satellite data
(black line) and the NDBC Stations (red dots) (a) FWYF1 (coastal area of Miami), (c) MLRF1 (UpperKeys), and (e) SANF1 (Lower-Keys). The respective scatter diagrams (b,d,f) between the two timeseries for each case are also shown. The linear regressions (trends and timeseries comparison), the
mean values, the Pearson correlation coefficients (Rp), the tests of statistically significant correlation
(pvalue), and the Root Mean Square Errors (RMSEs) for all cases are presented.
3.2. Interannual Variability of SST
The mean annual SST levels over the entire study domain revealed an increasing
trend during the 1981–2021 period, following the respective air temperature interannual
Water 2022, 14, 3840 8 of 28
trend (Figure 3a). In addition to the Mann–Kendall method that identifies the trend, we
employ the Sen’s slope [54] to characterize the magnitude of the change. We found that
both trends are characterized by positive Sen’s Slopes of 0.19 ◦C/decade and coefficient
of determination R2 = 0.46 for SST, and 0.21 ◦C/decade and R2 = 0.42 for air temperature.
The increasing trends are statistically significant (pvalue < 0.01: statistical significance level
99%). The SST values ranged around 26.3 ◦C in 1982 and reached the mean level of 27 ◦C
40 years later, in 2021. The highest peaks were observed after 2015 and especially during
2019–2020, when the mean annual SST over the entire region reached the level of 27.4 ◦C.
On the contrary, the lowest annual level was observed in 1984 (<26 ◦C). Despite the general
positive trend, a period of relatively low SST that flattened the linear trend occurred
between 2004 and 2013, following a similar stagnation period in the air temperature levels,
which revealed a significant drop in 2010 (<24 ◦C). A respective increasing trend was
also computed for the 99th SST percentiles, that represent the highest values of each year
(Figure 3b). The 99th percentile of SST revealed a steeper trend over the 40-year period
characterized by a higher statistical significance (pvalue < 0.01) Sen’s Slope (0.21 ◦C/decade;
R
2 = 0.35) in comparison to the mean values. The beginning and the end of the period
show a mean difference of approximately 1.5 ◦C. The highest and lowest 99th percentiles
of SST were also computed for 2019 and 1984, respectively. The increasing trend that
was computed for the minimum SST levels was milder than the mean and maximum
values, while the positive Sen’s Slope was smaller (0.05 ◦C/decade) and not statistically
significant (pvalue > 0.01). A year of very distinctive behavior was 2010, when although
the colder air conditions prevailed (<24 ◦C; Figure 3a) and the lowest minimum SST levels
also occurred (Figure 3c), the 99th percentile was relatively high (30.7◦
; Figure 3b) resulting
in the largest annual variance among all years (>9 ◦C; Figure 3d). According to Soto et al.
(2011) findings, 2010 can be characterized as a year of high risk on coral losses. The cold
January of 2010 (Colella et al., 2012) affected the water temperature levels and reduced
the mean annual levels, but very high SST levels also occurred during the summer period,
increasing the 99th percentile annual variance. It is concluded that the observed general
increasing trend is mainly related to the summer maximum values and less related to
increases during the winter periods. For most of the years, the variance of the annual
values ranges between 5 ◦C and 6 ◦C, with a very small increasing trend throughout the
entire period (0.05 ◦C/decade; Figure 3d). Even though the variance showed a small
increasing trend, indicating larger seasonal differences, the annual variance is relatively
small (<5 ◦C) during the last decade (2012–2021), when all winter and summer levels were
high, confirming the general warming of the ocean; the highest minimum temperatures
were observed during the same period (Figure 3c).
3.3. Spatial Variability of SST
The spatial variability of the seasonal 10th and 90th percentiles is presented in Figure 4.
The 10th percentile represents the 10% chance that the temperature fell below this threshold
during the study period; anything below the 10th percentile is considered unseasonably
cool. Very low temperature levels, associated with cold water events, may have fatal
consequences on coral communities [22] but may also reduce the occurrence frequency of
the MHW events during the winter months [12]. The 90th percentile describes the 10%
chance that SST was above this threshold, and anything above it is considered unseasonably
warm. The monthly 90th percentile was used as the temperature climatology (threshold)
for the MHW computation (see Section 3.4). The colder waters have been detected over the
entire WFS between January and March (<17 ◦C; Figure 4), while very low SST also occurred
along the western Florida coast in December. Over the same areas and months, the 90th
percentiles were relatively low (<24 ◦C) revealing their lowest values between the coastal
region of Tampa (28◦ N) and Fort Myers (26◦ N). The highest 10th and 90th percentile values
were computed during July-September for the entire study domain; especially the 10th
percentiles were homogenously distributed over all areas. The maximum 90th percentiles
were computed over the southern WFS (>31 ◦C), and especially along the northern coasts
Water 2022, 14, 3840 9 of 28
of the Florida Keys during the summer months and early fall. The high 90th percentiles are

in situ temperature measurements (Figure 2) collected at three locations (NDBC buoys)

along the Florida Keys and Biscayne Bay (Figure 1). The NDBC SST in Fowey Rock,

at the vicinity of Biscayne Bay (Buoy FWYF1), followed a clear seasonal variation from

January 2005 to May 2007 (Figure 2a) with a mean level of 26.6 ◦C (Figure 2b). The

satellite-derived timeseries followed a similar distribution revealing a very high correlation

(Rp: Pearson Correlation Coefficient) and a small Root Mean Square Error (RMSE) with

the field measurements (Rp = 0.97 and RMSE = 0.68 ◦C; Figure 2b); the respective mean

value (26.42 ◦C) is slightly smaller than the one measured directly at the sea surface. The

agreement between the two products is better at Buoy MLRF1, located at the Key Largo

coastal area (Figure 1), where both timeseries showed a similar increasing trend (Figure 2c)

and a linear regression very close to the x = y identity line (Figure 2d). The Pearson

correlation coefficient and the RMSE confirm the very good performance of the satellitederived data (Rp = 0.98 and RMSE = 0.50 ◦C; Figure 2d). The correlation is even higher at the

Water 2022, 14, 3840 7 of 28

lower Keys (Buoy SANF1; Figure 1) during 2005, when the Pearson coefficient is Rp = 0.99

and the RMSE is less than 0.5 ◦C (Figure 2f). The satellite-derived SST data are capable to

describe all short-term peaks and lows that exceed the typical range of the seasonal cycle;

the extremely high values in August 2005 (>31 ◦C) and the large drop of approximately

2

◦C at the end of the same month (28 ◦C) are apparent at both field and satellite timeseries

(Figure 2e). Thus, we conclude that the high-resolution satellite-derived SST daily fields

are suitable to investigate not only the open sea and shelf SST variability [44] but also the

temperature distribution and the formation of MHWs over the South Florida coastal region.

Water 2022, 14, x FOR PEER REVIEW 8 of 31

Figure 2. (a) Daily evolution of Sea Surface Temperature (SST; °C) derived from the satellite data

(black line) and the NDBC Stations (red dots) (a) FWYF1 (coastal area of Miami), (c) MLRF1 (UpperKeys), and (e) SANF1 (Lower-Keys). The respective scatter diagrams (b,d,f) between the two

timeseries for each case are also shown. The linear regressions (trends and timeseries comparison),

the mean values, the Pearson correlation coefficients (Rp), the tests of statistically significant correlation (pvalue), and the Root Mean Square Errors (RMSEs) for all cases are presented.

3.2. Interannual Variability of SST

The mean annual SST levels over the entire study domain revealed an increasing

trend during the 1981–2021 period, following the respective air temperature interannual

trend (Figure 3a). In addition to the Mann–Kendall method that identifies the trend, we

employ the Sen’s slope [54] to characterize the magnitude of the change. We found that

both trends are characterized by positive Sen’s Slopes of 0.19 °C/decade and coefficient of

determination R2 = 0.46 for SST, and 0.21 °C/decade and R2 = 0.42 for air temperature. The

increasing trends are statistically significant (pvalue < 0.01: statistical significance level 99%).

ThSSTldd263°Ci1982dhdthllf27°C40Figure 2. (a) Daily evolution of Sea Surface Temperature (SST; ◦C) derived from the satellite data

(black line) and the NDBC Stations (red dots) (a) FWYF1 (coastal area of Miami), (c) MLRF1 (UpperKeys), and (e) SANF1 (Lower-Keys). The respective scatter diagrams (b,d,f) between the two timeseries for each case are also shown. The linear regressions (trends and timeseries comparison), the

mean values, the Pearson correlation coefficients (Rp), the tests of statistically significant correlation

(pvalue), and the Root Mean Square Errors (RMSEs) for all cases are presented.

3.2. Interannual Variability of SST

The mean annual SST levels over the entire study domain revealed an increasing

trend during the 1981–2021 period, following the respective air temperature interannual

Water 2022, 14, 3840 8 of 28

trend (Figure 3a). In addition to the Mann–Kendall method that identifies the trend, we

employ the Sen’s slope [54] to characterize the magnitude of the change. We found that

both trends are characterized by positive Sen’s Slopes of 0.19 ◦C/decade and coefficient

of determination R2 = 0.46 for SST, and 0.21 ◦C/decade and R2 = 0.42 for air temperature.

The increasing trends are statistically significant (pvalue < 0.01: statistical significance level

99%). The SST values ranged around 26.3 ◦C in 1982 and reached the mean level of 27 ◦C

40 years later, in 2021. The highest peaks were observed after 2015 and especially during

2019–2020, when the mean annual SST over the entire region reached the level of 27.4 ◦C.

On the contrary, the lowest annual level was observed in 1984 (<26 ◦C). Despite the general

positive trend, a period of relatively low SST that flattened the linear trend occurred

between 2004 and 2013, following a similar stagnation period in the air temperature levels,

which revealed a significant drop in 2010 (<24 ◦C). A respective increasing trend was

also computed for the 99th SST percentiles, that represent the highest values of each year

(Figure 3b). The 99th percentile of SST revealed a steeper trend over the 40-year period

characterized by a higher statistical significance (pvalue < 0.01) Sen’s Slope (0.21 ◦C/decade;

R

2 = 0.35) in comparison to the mean values. The beginning and the end of the period

show a mean difference of approximately 1.5 ◦C. The highest and lowest 99th percentiles

of SST were also computed for 2019 and 1984, respectively. The increasing trend that

was computed for the minimum SST levels was milder than the mean and maximum

values, while the positive Sen’s Slope was smaller (0.05 ◦C/decade) and not statistically

significant (pvalue > 0.01). A year of very distinctive behavior was 2010, when although

the colder air conditions prevailed (<24 ◦C; Figure 3a) and the lowest minimum SST levels

also occurred (Figure 3c), the 99th percentile was relatively high (30.7◦

; Figure 3b) resulting

in the largest annual variance among all years (>9 ◦C; Figure 3d). According to Soto et al.

(2011) findings, 2010 can be characterized as a year of high risk on coral losses. The cold

January of 2010 (Colella et al., 2012) affected the water temperature levels and reduced

the mean annual levels, but very high SST levels also occurred during the summer period,

increasing the 99th percentile annual variance. It is concluded that the observed general

increasing trend is mainly related to the summer maximum values and less related to

increases during the winter periods. For most of the years, the variance of the annual

values ranges between 5 ◦C and 6 ◦C, with a very small increasing trend throughout the

entire period (0.05 ◦C/decade; Figure 3d). Even though the variance showed a small

increasing trend, indicating larger seasonal differences, the annual variance is relatively

small (<5 ◦C) during the last decade (2012–2021), when all winter and summer levels were

high, confirming the general warming of the ocean; the highest minimum temperatures

were observed during the same period (Figure 3c).

3.3. Spatial Variability of SST

The spatial variability of the seasonal 10th and 90th percentiles is presented in Figure 4.

The 10th percentile represents the 10% chance that the temperature fell below this threshold

during the study period; anything below the 10th percentile is considered unseasonably

cool. Very low temperature levels, associated with cold water events, may have fatal

consequences on coral communities [22] but may also reduce the occurrence frequency of

the MHW events during the winter months [12]. The 90th percentile describes the 10%

chance that SST was above this threshold, and anything above it is considered unseasonably

warm. The monthly 90th percentile was used as the temperature climatology (threshold)

for the MHW computation (see Section 3.4). The colder waters have been detected over the

entire WFS between January and March (<17 ◦C; Figure 4), while very low SST also occurred

along the western Florida coast in December. Over the same areas and months, the 90th

percentiles were relatively low (<24 ◦C) revealing their lowest values between the coastal

region of Tampa (28◦ N) and Fort Myers (26◦ N). The highest 10th and 90th percentile values

were computed during July-September for the entire study domain; especially the 10th

percentiles were homogenously distributed over all areas. The maximum 90th percentiles

were computed over the southern WFS (>31 ◦C), and especially along the northern coasts

Water 2022, 14, 3840 9 of 28

of the Florida Keys during the summer months and early fall. The high 90th percentiles are