Datasets:
patrickbdevaney
/
mia9

Dataset card Data Studio Files
xet
 Community
1
text
stringlengths
0
6.44k
Caccia and Boyer (2005) performed an objective classification analysis on FIU’s
water quality data set from 25 sites collected monthly during 1994–2003. Water quality
parameters used for classification included: TN, TON, DIN, NH4, NO3, NO2, TP, SRP,
TN:TP ratio, TOC, DO, CHLa, turbidity, salinity and temperature. This spatial analysis
extended from North Biscayne Bay and to Card Sound but excluded Barnes and
Manatee Bay. They applied Principal Components, retained the scores and input mean
29
and standard deviations of scores to K-means cluster analysis. They defined five zones
having similar water quality characteristics (Fig 3.1), where a robust DIN gradient was
determinant in the spatial clustering from alongshore to offshore in the main Bay. This
gradient results from the dominating influence of freshwater input from canals which
drain the South Dade agricultural area, Black Point Landfill, and sewage treatment
plant. Their South Bay zone (i.e. Card Sound) was high in dissolved organic
constituents and low in inorganic nutrients, reflecting the little urban development and
the extended mangrove wetlands of its watershed. Lowest nutrient concentrations
occurred in the Main Bay, the area most influenced by water exchange with the Atlantic
Ocean. Caccia and Boyer (2005) concluded that water quality in Biscayne Bay is highly
dependent on the land use in the watershed and on the exchange with oceanic waters.
Hunt & Todt (2006) subdivided Biscayne Bay waters into five geographic
domains based on direct clustering analysis of a set of stations (SAS PROC CLUSTER,
Ward method-Euclidean distance) combining temperature and salinity data from FIU
and DERM programs. Their geographic domains were only based on12 quarterly data
of surface samples (<0.5m). Their cluster analysis explained 81 percent of variability in
the data and the distribution pattern is in general similar to that of Caccia and Boyer
(2005). In both approaches, the characteristics of North Biscayne Bay extend just south
of Rickenbacker Causeway (dark blue dots in Fig 3.1) suggesting an active connection
and exchange. In both classifications an east-west gradient is evident in Central
Biscayne Bay, driven by salinity and nutrient differences, in turn caused by mixing of
oceanic and freshwaters. Card Sound appears as a distinct class in both subdivisions,
defining an additional north-sound gradient.
Factor Analysis
FIU monitoring program began in Manatee and Barnes Sound in 1991 as part of
their Florida Bay program and in Biscayne Bay (including Card Sound) in 1993. Ten of
the initial sites in Biscayne Bay were discontinued and ten new sites were incorporated
after an assessment by the SFWMD and FIU in 1996. Several of the new sites were
located out in North Biscayne Bay. Additionally, reassessment of the monitoring
program in September 2007 dropped 6 additional sites. Finally, the period between
30
June-1996 to September-2007 contains the largest number of stations simultaneously
sampled. Hence, we selected that period and those 21 stations distributed throughout
the bay for our segmentation. To differentiate data from this time interval from the
overall POR, we have called it Selected Period of Record (SPOR).
Figure 3.1: a) Caccia and Boyer (2005) subdivision of Biscayne Bay waters into 5
zones, North Bay, Alongshore, Inshore, Mid Bay and South Bay; and b) Classification of
Biscayne Bay waters by Hunt and Todt (2006) into five geographic domains, using
salinity and temperature from DERM and FIU data sets combined.
The SPOR included fourteen biogeochemical parameters, TOC, TN, TON, TP,
SRP, CHLA, NOx, NO2
-
, NO3
-
, NH4
+
, DO, salinity, turbidity, and temperature, for 4,941
samples and a total of 73,368 analytical determinations. After considering: 1) the high
correlation between TN and TON, and NOx and NO3
-
; 2) the high proportion of nondetects for SRP; and 3) the constancy of temperature across the bay, we included only
ten biogeochemical variables in the model: TOC, TN, TP, CHLA, NOx, NO2-
, NH4
+
,
salinity, turbidity, and DO. The number of determinations was now 39,078.
Northern BB
Intermediate BB
Offshore BB
West Central BB
Southern BB
31
For the segmentation we used similar factor analysis and clustering techniques
as those described by Briceño and Boyer (2009) for Florida Bay. Briefly, raw data was
transformed into z-scores and used as input for factor analysis with VARIMAX rotation
of axes (Minitab 16®
statistical software) to obtain five Principal Components (PC). We
did not apply the Kaiser criterion or the scree test rigorously (Kaiser 1960) for the
selection of PCs, but the magnitude of our retained eigenvalues was above 0.9 and the
contribution to the accounted variance of the five selected PCs was above 10%. These
PCs are linked to specific water quality parameters controlling their structure (Appendix
8). PC1 is mostly a positive function of inorganic N species, NO2
-
, NOx and NH4
+
and is
negatively correlated with salinity. PC2 is linked to TN and TOC perhaps representing
inland contributions, and it is also negatively correlated to salinity. PC3 is strongly
dependent on TP and CHLa, perhaps representing a driver:response pair. PC4
depends upon CHLa and turbidity, perhaps representing a system response. Finally,
PC5 is a function of DO. In total, these five PC account for 79.2% of the total variance.
There is a striking similarity between our PCs and those PCs derived by Caccia
and Boyer (2005) for the whole Biscayne Bay for a shorter period of record. This
similarity in results suggests a highly robust and long-lasting relationship among WQ
variables in Biscayne Bay.
Hierarchical Clustering
The scores from the five PCs were retained and their descriptive statistics
calculated for each sampling station. After trying different combinations of statistics,
grouping was performed with a combination of two parametric statistics (mean and
standard deviation) and two non-parametric statistics (median and MAD) of the scores
for each sampling site as input for hierarchical clustering routines. Our hierarchical
clustering metrics were Euclidean distances and linkage was Ward minimum variance
(Minitab 16®