Transcript New visual style - Great Lakes Commission

Utilization of Remotely Sensed Data for
Targeting and Evaluating Implementation of
Best Management Practices within the Wister
Lake Watershed, Oklahoma
Scott Stoodley, Brenda Berasi
AMEC Earth and Environmental
Phillip Busteed, Michael White, Daniel Storm
Oklahoma State University
Introduction
Wister Lake Watershed


Located in southeast Oklahoma and Arkansas
Eutrophic
 Listed as a Nutrient Limited Watershed (NWL)
 Listed on the State’s 303(d) list in 1998 for
nutrients, siltation, suspended solids, flow
alteration, taste and odor
 Oklahoma Conservation Commission (OCC) has
begun using satellite imagery with water quality
modelling to identify potential critical source areas
of pollutants
Objectives
Utilize Landsat TM (30 meter resolution) imagery
to:
1. Target Best Management Practices (BMPs)
 Map landcover types for use in SWAT modeling to
target critical source areas of phosphorus from
pasture systems
2. Evaluate BMPs
 Evaluate the effectiveness of the OCC’s Section
319 cost share program by analyzing vegetation
and pasture grazing intensity for the summer of
2000 through the summer of 2004
Study Area
Landsat TM Imagery
August/September 2004
Landcover Classification
Categories of interest:






Forest
Pasture, high biomass
Pasture, low biomass
Shrub / Range
Bare Soil
Clear-Cut






Urban, high density
Urban, low density
Rock Outcropping
Mining
Water
Clouds
Landcover Classification
Methods


Unsupervised classification
 ISODATA clustering algorithm
 Iterative process
 Established set of decision criteria for each
landcover category
 3x3 Majority spatial filter used on large macro
classes
Change detection conducted using categorical image
differencing
Landcover Classification
Methods

Ground-truth data
 Collected by OCC and Oklahoma State University
(OSU)
 GPS coordinates, description of landcover type,
digital photos
 2 Data sets


255 locations surveyed in July
and August 2005
100 locations surveyed in
November 2005
2004 Landcover
Classification
Land Cover Categories
Water
Forest
High Biomass Pasture
Low Biomass Pasture
Shrub / Range
Bare Soil
Clear-cut
Rock Outcropping
High Density Urban
Low Density Urban
Mining
Clouds
2000 Landcover
Classification
Land Cover Categories
Water
Forest
High Biomass Pasture
Low Biomass Pasture
Shrub / Range
Bare Soil
Clear-cut
Rock Outcropping
High Density Urban
Low Density Urban
Mining
Clouds
Change Detection
Land Cover Change from 2000 to 2004
Unchanged Water
Unchanged Forest
Unchanged Pasture
Unchanged Shrub/Range
Unchanged Bare Soil
Unchanged Urban
Forest → Clear-cut
Pasture → Forest
Bare Soil → Forest
Clouds
Other
Riparian Buffer Zone
Land Cover
Water
Forest
High Biomass Pasture
Low Biomass Pasture
Shrub / Range
Bare Soil
Clear-cut
Rock Outcropping
High Density Urban
Low Density Urban
Mining
Landcover Classification
Accuracy Assessment

100 sites selected:



Stratified random sampling design
Within ½ mile buffer of roads
Minimum of 10 sites per category
Ground Truth
Need for temporally coincident ground truth data
21 July 2005
Shrub / Range
02 November 2005
Pasture
Vegetation Analysis
Objective

Analyze vegetation density to evaluate BMP
effectiveness
Methods


Vegetation indices
 Normalized Difference Vegetation Index
 Normalized Difference Senescent Vegetation Index
 Fractional cover of green vegetation
 Fractional cover of senescent vegetation
Change detection based on percent difference from
baseline
Vegetation Analysis
Multi-temporal
analysis
31 Aug 2004
28 Jul 2003
27 Sep 2002
22 Jul 2001
20 Aug 2000
Vegetation Indices
Normalized difference
vegetation index
(NIR-Red)
NDVI = (NIR+Red)
Normalized difference
senescent
vegetation index
(SWIR-Red)
NDSVI = (SWIR+Red)
Vegetation Indices
Fractional cover of green vegetation
FCgv = (NDVI-NDVIsoil)/(NDVIveg-NDVIsoil)
Fractional cover of senescent vegetation
FCsv = (NDSVI-NDSVIsoil)/(NDSVIveg-NDVSIsoil)

NDVIsoil and NDVIveg are the NDVI values of an area of only bare
soil and an area of total vegetation, respectively, as derived from
an image or measured in the field
Vegetation Analysis


All four indices computed for the pasture areas of
the Oklahoma portion of the watershed
Change depicted as percent departure from 2000
Vegetation Analysis
Vegetation Analysis
Vegetation Analysis
Evaluation of BMP effectiveness:





Mean FCgc and FCsv computed for 255 individual cost
share fields
Computed ratio of the mean FC for each cost-share
field to that of all other pastures within a given year
Ratio2000– Ratio2004 computed for each field
∑ Ratio2000– Ratio2004 for all fields that had
implemented the same combination of BMPs
Area weighted average was also computed for the
sum of the differences
Evaluation of BMPs


Some BMPs, such as a fence only, resulted in an overall increase in FCgv and
FCsv of the cost share fields relative to all of the non-cost share fields for
2004 compared to 2000.
Others, such as the combination of lime and fertilizer, resulted in an overall
increase in FCgv and an overall decrease in FCsv relative to all of the noncost share fields for 2004 compared to 2000.
Conclusions:
Land cover classification



The land cover classification indicates a
watershed dominated by forest, followed by
pasture.
The change detection reveals a dynamic
environment in which nearly 19% of the
Oklahoma portion of the watershed underwent a
change in land cover from 2000 to 2004.
Overall accuracy of 2004 land cover classification
was 93.0%
Conclusions:
Landcover Classification


Classification within the riparian zone offers a
qualitative targeting method to spatially locate
high risk landcover types.
When combined with estimates of nonpoint
source loadings through SWAT modeling, the
landcover classification provides a mechanism by
which to proactively identify and target areas
that are likely contributing to water quality
degradation.
Conclusions:
Vegetation Analysis



No clear trend was found in overall FC for the
cost share fields relative to the non-cost share
fields in 2004 compared to 2000.
Some BMPs resulted in increased FC, whereas
others resulted in decreased FC.
The overall effect of BMP implementation on
vegetation density may be obscured by changes
in land-use management decisions.
Conclusions:
Vegetation Analysis


By gathering data on the land management
practices that had been implemented between
2000 and 2004, a better understanding of the
effectiveness of the BMPs may be attained.
The OCC should consider the possibility of
placing management restrictions in conjunction
with Section 319 funding for BMP
implementation.
For More Information:
Brenda Berasi
AMEC Earth & Environmental
[email protected]
(978) 692-9090