Академический Документы
Профессиональный Документы
Культура Документы
Brandon Bergenroth, Jay Rineer, Breda Munoz and William Cooter (RTI)
3040 Cornwallis Road P.O. Box 12194 Research Triangle Park, North Carolina, USA 27709
Phone 919-316-3537 e-mail bbergenroth@rti.org
1
Statistical Trend Analysis for STORET DATA
2
New STORET Tools (Services) Simplify Pulling
Data for Trend Analysis
3
Seasonal Kendall tests a common tool to help
confirm apparent trend patterns
Seasonal Kendall tests favored by the USGS, EPA
ORD, and many university researchers
Valuable where parameter show variability related to
seasonal changes in temperature or changes in flows
Can accommodate some degree of censored
observations (below detection limits or missing values)
4
Trend analysis functions/modules similar to
ESTREND (USGS) and Kendall (S-PLUS)
already implemented in the open source R.
5
R is supported by EPA through EMAP and
through initiatives such as NCEAs CADDIS
6
R-based Trend Analysis using STORET
river/stream station data
7
Non Parametric Statistic Tests
Non parametric statistic tests refer to the collection of statistic tests that
do not require any assumption on the distribution of the data. They are
also known in the statistic literature as distribution free tests and
distribution independent tests.
They are often used in hypothesis testing (e.g. existence of trends) and
therefore considered as confirmatory data analysis tools.
8
MannKendall
Let:
x1 ,..., x n be a sequence of measurements over time, to test the null hypothesis,
H 0 : x1 ,..., xn come from a population where the random variables are independent and identically distributed,
H 1 : x1 ,..., x n follow a monotonic (e.g. increasing or decreasing) trend over time.
n 1 n
The Mann-Kendall test statistic is calculated as S sgn( x j xwhere
k)
k 1 j k 1
1 if x j x k 0
sgn( x j x k ) 0 if x j x k 0
1 if x x 0
j k
E S 0
p
n( n 1)(2n 5) t j (t j 1)(2t j 5)
j 1
Var S if ties
18
n(n 1)(2n 5)
no ties
18
where p is the number of tied groups in the data set and is the number of data points in the jth tied group.
tj
9
MannKendall
A positive value of S indicates that there is an upward (increasing) trend (e.g. observations increase with
time).
If S is significantly different from zero, then based on the data H 0 can be rejected at a pre-selected
significance level and the existence of a monotonic trend can be accepted.
x xk 0 x xk 0
Note that S is a count of the number of times j for j k, more than j .
The maximum value of S (called it D) occurs when 1x x 2 ... x n.
n(n 1) p n(n 1)
S t j (t j 1) if ties
2 2
Kendalls tau is defined as tau D where D j 1
n(n 1)
no ties
2
10
MannKendall
The distribution of Kendalls tau can be easily obtained from the distribution of S.
A positive value of tau indicates that there is an upward (increasing) trend (e.g. observations increase with
time).
If tau is significantly different from zero (e.g. value less than 0.05 at the 5% significance level or less than
0.01 at the 1% significance level), then based on the data, H 0 can be rejected at a pre-selected
significance level (e.g. alpha = 5%) and the existence of a monotonic trend can be accepted.
Note that the test only allows the software user to conclude about the existence not about the magnitude of
the trend.
11
Getting Results
12
Visualizing Results
http://code.google.com/apis/kml/documentation
13
Visualizing Results
14
Visualizing Results
15
Report Results
http://iaspub.epa.gov/storpubl/storet_wme_pkg.Display_Station?p_station_id=SP-64&p_org_id=MWRD
16
Report Results
http://iaspub.epa.gov/stormoda/DW_resultcriteria_station
http://iaspub.epa.gov/webservices/StoretResultService/index.html?operation=getResults
17
Kendall Trend Analysis for pH
18
Kendall Trend Analysis for Temperature
19
Kendall Trend Analysis for Dissolved Oxygen
20
Kendall Trend Analysis for Total Suspended Solids
21
Kendall Trend Analysis for Turbidity
22
Kendall Trend Analysis for Cadmium
23
Kendall Trend Analysis for Zinc
24
Indexing STORET stations to the NHD can
help increase sophistication of trend
analyses
Group sites relative to
upstream NPDES
discharges
Group using Horton-
Strahler stream orders
Group in terms of
landcover patterns
using NHDPlus LU/LC
raster data
25
Indexing and combining station results
26
Next Steps
Additional work on pre-processing STORET station data to
focus attention on stations with enough data density to support
trend analyses
Develop a data mart of R trend analysis results including
saved images of scatter plots over time from R
Consider ways trend analyses can support either pro-active
study of anti-degradation effects [Goal: detect degradation trend
early on and consider management steps to avoid winding up
with additional 303(d) lists]
Also use trend analyses as a tool to document incremental
progress in meeting targets established under WQ Standards or
the TMDL program
27