Home Archived October 29, 2018
(i)

South Florida Information Access (SOFIA)


geer > 2008 > hindcasting water levels for eden gaging stations, 2000-2006

Hindcasting Water Levels for EDEN Gaging Stations, 2000-2006

Presented July 30, 2008 at the GEER Conference in Naples, FL

A PDF version of this talk, suitable for printing, is available for download (2.6 MB). Please note that you will need the free Adobe Acrobat Reader in order to view this file.

Paul Conrads, USGS South Carolina Water Science Center
Ed Roehl, Advanced Data Mining


What's on tap?

  • Methods development in WCA3a
  • Application to EDEN
  • Quiet desperation
  • The fix
  • Results and discussion


Problem : How to Estimate Water Depths at Ungaged Sites

  • Dataset –WCA 3a
    • Water-level from 3 sites
    • Water-depth data from 17 sites
    • EDEN grid and vegetation attributes
    • % prairie
    • % sawgrass
    • % slough
    • % upland
    • UTM North
    • UTM South
map showing Water Conservation Area 3A sites
[larger image]


Approach

  • Two step ANN model
    • First step: estimate mean water-depths using static model –“spatially interpolating” ANN scheme
    • Second step: estimate water-depths variability using dynamic variables


Two-step Model

illustration of two-step model input and final prediction equation
[larger image]


Static Model Results

graph of measured versus predicted static model water level difference
[larger image]
  • Each “step” represents a different site
  • Model able to generalize water level difference but not the variability


Dynamic Model

  • 5 “index” stations (red dots)
  • Combination of static and dynamic data
  • 5 validation stations (green dots)
map of Water Conservation Area 3A showing index and validation stations
[larger image]


Final Model Results

plot of measured and predicted water depths for Site W8
Figure 5. Plot showing measured (light trace) and predicted (dark trace) water depths for Site W8. Predictions are not continuous due to missing data for one or more of the index stations. [larger image]


Hindcasting 25 EDEN Sites

  • Existing EDEN sites : 7 years (61,400 data points)
  • New EDEN sites : 4 -12 months of data (925-8,760 data points)
map showing locations of Everglades Depth Estimation Network gaging stations in southern Florida
Figure 1. Locations of Everglades Depth Estimation Network (EDEN) gaging stations in southern Florida (modified from Pearlstine and others, 2007). [larger image]


Approach: Similar 2-Step Model

illustration of similar 2-step model input and estimate at a new Everglades Depth Estimation Network site
[larger image]


Approach: Similar 2-Step Model

  • Separate models for each area
  • Number of potential inputs sites reduced using dynamic clustering
  • Hindcast example –W2
  • Input Sites
    • Sites 63, 64, 65, 3ASW3, and 3AWS
map showing Water Conservation Area 3A sites, highlighting input sites, and hindcast example site W2
[larger image]


Spatially Interpolating ANN Model Stacked Dataset

illustration of spatially interpolating Artificial Neural Network model stacked dataset including five input sites
[larger image]


Static Model Results

plot of measured and simulated static model water level results for 5 input sites
[larger image]


Dynamic Model Results

plot of measured and simulated dynamic model residual water level results for 5 input sites
[larger image]


Initial Water Level Estimate - W2 “Quiet Desperation”

plot of initial water level estimate for site W2, static model results, dynamic model results, and measured data
[larger image]
What is going on?
Data for W2 not used to train models


Possible Causes

  • Datum confusion
  • Gage offset
  • W2 located lower edge of area covered by input sites
  • Limited information content of static variables
  • Did not use difference from a standard signal (decorrelated dynamic variables)


Third-step Model: Error correction

illustration of three-step modeling approach to make final water-level estimates at a station
Figure 7. Three-step modeling approach to make final water-level estimates at a station. [larger image]


Error Correction Model Results

plot of site W2 water level initial estimate, measured values, and residuals of error correction model results
[larger image]
plot of error model prediction of water level and residuals
[larger image]


Final Hindcast

plot of final hindcast water level results showing site 64 measured values, site W2 measured values, and site W2 record extension
[larger image]


Summary Statistics for Hindcasting Models

Table 4. Summary statistics for water-level estimates for new Everglades Depth Estimation Network (EDEN) stations.

[n, number of data values; R, Pearson coefficient; R2, coefficient of determination; Min, minimum; ft, feet; Max, maximum; ME, mean error; RMSE, root mean square error; PME, percent model error]

[Please note: This table is from Open File Report 2007-1350, and the figures referred to in this table are station location maps featured in the report.]

Site n R R2 Data range ME RMSE PME
Min, in ft Max, in ft
Water Conservation Area 1 (fig. 15)
North_CA1 41,721 0.943 0.889 13.82 16.38 -0.046 0.134 5.2
South_CA1 43,409 0.991 0.983 12.85 15.90 0.008 0.088 2.9
Water Conservation Area 2 (fig. 17)
EDEN11 2,637 0.950 0.902 10.97 13.19 -0.054 0.044 2.0
EDEN13 1,968 0.955 0.912 7.07 7.69 -0.018 0.006 1.0
Water Conservation Area 3A (fig. 20)
3A-5 5,684 0.998 0.995 8.15 10.10 0.002 0.001 0.1
EDEN4 2,399 0.999 0.998 6.96 10.30 0.001 0.002 0.1
EDEN5 1,653 0.999 0.999 8.01 10.20 0.004 0.001 0.1
EDEN8 2,442 0.999 0.999 6.79 9.19 0.001 0.001 0.0
EDEN9 925 0.999 0.998 7.69 10.55 -0.006 0.004 0.1
EDEN12 7,648 0.999 0.998 6.84 9.59 -0.002 0.001 0.0
EDEN14 969 0.906 0.821 8.66 9.59 0.011 0.010 1.1
W2 7,648 0.998 0.995 6.86 9.42 0.008 0.003 0.1
W5 7,648 0.999 0.998 6.84 9.59 -0.001 0.001 0.0
W11 7,628 0.999 0.998 7.08 10.08 -0.012 0.002 0.1
W14 7,628 0.998 0.997 7.00 10.00 -0.013 0.003 0.1
W15 3,859 0.999 0.998 7.47 9.67 -0.001 0.001 0.0
W18 7,628 0.998 0.996 7.92 10.21 -0.000 0.002 0.1
Water Conservation Area 3B (fig. 26)
TI-8 5,684 0.996 0.993 4.59 6.16 0.002 0.001 0.1
TI-9 5,684 0.996 0.992 5.29 6.42 0.009 0.001 0.1
EDEN7 2,419 0.998 0.996 5.24 6.96 0.008 0.001 0.1
EDEN10 2,419 0.995 0.990 5.06 6.32 0.007 0.002 0.1
Big Cypress National Preserve (fig. 29)
EDEN1 3,864 0.960 0.921 7.13 7.82 0.018 0.003 0.4
EDEN6 1,591 0.984 0.968 8.76 10.73 0.030 0.007 0.4
Everglades National Park (fig. 31)
EDEN3 5,294 0.989 0.978 0.07 1.75 -0.042 0.006 0.4
Met1 1,238 0.994 0.989 5.23 5.76 0.000 0.000 0.0


Summary

  • Estimation of water depth at ungaged sites
    • ANNs able to accurately predict water depths at ungaged sites
    • Use of static and dynamic variable produce a multi-variate “kreiging” of water depths
    • Methodology will be used to hindcast “new” network stations


Questions

Paul Conrads

USGS-South Carolina Water Science Center

pconrads@usgs.gov

image of Open-File Report 2007-1350 cover



| Disclaimer | Privacy Statement | Accessibility |

U.S. Department of the Interior, U.S. Geological Survey
This page is: http://sofia.usgs.gov/geer/2008/talks/GEER08_Conrads_Hindcast/GEER08_hindcast_talk.html
Comments and suggestions? Contact: Heather Henkel - Webmaster
Last updated: 04 September, 2013 @ 02:07 PM(TJE)