2013 GISCO Track, Wildfire and Water: Utilizing LANDSAT imagery, GIS, and Statistical Models for...

Preview:

DESCRIPTION

The June 2012 High Park Fire burned over 87,000 acres of forest and 259 homes to the west of Fort Collins, CO. The fire has had dramatic impacts on forest ecosystems. Of particular concern are its effects on the Cache la Poudre watershed, as the Poudre River is one of the most important headwaters of the Colorado Front Range, providing important ecosystem and economic services before flowing into the South Platte, which in turn flows into the Missouri River. Within a week of the fire, the area received several days of torrential rains. This precipitation—in conjunction with steep riverbanks and the loss of vegetation by fire—caused soil and ash runoff to be deposited into the Poudre’s channel, resulting in a river of choking mud and black sludge. Monitoring the effects of this disaster is critical and requires establishing immediate baseline data to assess impacts over time. Utilizing LANDSAT imagery, GIS layers, and boosted regression trees modeling, the NASA DEVELOP team based at the North Central Climate Science Center at Colorado State University conducted an investigation into riparian, wetland and headwaters modeling within the Cache la Poudre watershed. These efforts produced a preliminary model of predicted wetlands across the watershed, which is currently being refined by field data collection and modeling within three elevation-based “life zones.” The ultimate goal of this ongoing project is to provide important spatial data for land managers and create a riparian and wetlands modeling methodology that can be reproduced throughout the intermountain west region.

Citation preview

Wildfire &

Wetlands

Mapping with Landsat,

GIS and Statistical

Models

Stephen Chignell Sky Skach

Colorado State University

A NASA DEVELOP Project

High Park Fire

• On June 9th, 2012, lightning ignited Roosevelt National Forest west of Fort Collins, Colorado.

• At the time of containment the fire had impacted over 87,000 acres and burned 259 homes.

• Days later, torrential downpours caused major flooding and erosion in the Poudre Canyon.

• Large amounts of soil and ash run-off were deposited into the Cache la Poudre River.

Monitoring effects of fires such as these on sensitive habitats like wetlands and riparian areas is critical, and there is urgent need for baseline data to assess change over time.

• The Poudre River runs

140 miles from its

headwaters in the

Rockies

• Drops 7,000 feet to

its confluence with

the South Platte

River, ultimately

flowing into the

Missouri River

• Annual flow of

280,000 acre feet

Cache la Poudre

Watershed

• One of the most important river systems on the Colorado Front Range:

– Drinking Water

– Agricultural Water

• Vulnerable to a range of environmental and anthropogenic stressors

Community Concerns

Importance of Wetlands

• Biogeochemical processes: • water cycling • carbon storage

• Ecosystem services: • water purification • flood protection • shoreline stabilization • groundwater recharge • stream-flow

maintenance • Biodiversity:

• home to a diverse array of flora and fauna

Wetland Mapping Status Percentage of basin mapped:

• Digital NWI Mapping: 61% • CPW Riparian Mapping: 70% • Potential Fen Mapping: 0%

Total area of NWI mapped wetlands: 30,145 acres Percent of total area: 4%

Colorado Natural Heritage Program, 2013: http://www.cnhp.colostate.edu/wetlandinventory/

• Provide a current map of wetlands in the Cache la Poudre

watershed for use by local land managers.

• Create a reproducible methodology for mapping wetlands

in other headwaters regions of the Intermountain West.

Project Objectives

Colorado Natural Heritage Program

Colorado State University

Geospatial Centroid at CSU

Natural Resource Ecology Laboratory at CSU

North Central Climate Science Center

USDA Forest Service Rocky Mount. Research Station

USGS Fort Collins Science Center

Project Partners

Map of Predicted Wetlands

Existing Wetland Data

Landsat 5 TM Ancillary GIS Data

Statistical Regression Model

Methodology

Landsat 5 Data

Image Acquisition • Landsat 5 Thematic Mapper • Path/Row: 34, 32 • Cloudless • Multiple years and months

Available Cloudless LANDSAT Data for Path 34, Row 32

June 2011 March 2010 March 2009

November 2010 January 2010 February 2007

September 2010 August 2009 December 2006

April 2010 May 2009 July 2003

Landsat Data Processing • Atmospheric Correction

Derived Environmental Variables • Tasseled Cap: Brightness,

Greenness, Wetness • NDVI, NDMI, NDWI, SAVI,

Edge Filter

Ancillary Data • NED Digital Elevation

Model (30 m) • Slope, Aspect, CTI,

Curvature

Normalized Difference Moisture Index (NDMI)

Map of Predicted Wetlands

Existing Wetland Data

Landsat 5 TM Ancillary GIS Data

Statistical Regression Model

Existing Wetlands Data

1. Exclude water bodies

2. Select only palustrine – Includes all inland, non-tidal,

wetlands which lack flowing water.

3. Exclude all polygons less than one hectare

4. Inverse -30 m buffer

5. Generate 150 random presence points within polygons – 30m distance between points

Presence Data Preparation

1. Buffer 60 m away from wetland polygons

2. Exclude buffered areas from all analysis.

3. Generate 150 random absence points from background

Background Data Preparation

Map of Predicted Wetlands

Existing Wetland Data

Landsat 5 TM Ancillary GIS Data

Statistical Regression Model

2. Merged Dataset Builder i.e. “Extract Multi Values

to Points” in ArcMap

USGS Software for Assisted

Habitat Modeling (SAHM)

1. Project, Aggregate, Resample, and Clip (PARC)

3. Covariate Correlation Matrix

Removal of highly

correlated predictor variables

Boosted Regression Trees Modeling

Split Wetland Presence Points

80 % Train

20 % Test

(Figure: Elith et al, 2008)

BRT Code within SAHM

Binary Map

Map of Predicted Wetlands

Existing Wetland Data

Landsat 5 TM Ancillary GIS Data

Statistical Regression Model

Initial Results

Refinement: Elevation Zones

Three models for three zones:

• Foothills and Plains (< 1800 m)

• Montane (1800 m – 3500 m)

• Alpine (> 3500 m)

Results

Source: Esri, DigitalGlobe, GeoEye, i-cubed, USDA, USGS, AEX, Getmapping, Aerogrid, IGN, IGP, swisstopo, and the GIS User Community

Plains and Foothills PCC: 72.5% AUC: 0.82 Kappa: 0.46

Source: Esri, DigitalGlobe, GeoEye, i-cubed, USDA, USGS, AEX, Getmapping, Aerogrid, IGN, IGP, swisstopo, and the GIS User Community

Montane PCC: 91.3% AUC: 0.99 Kappa: 0.99

Source: Esri, DigitalGlobe, GeoEye, i-cubed, USDA, USGS, AEX, Getmapping, Aerogrid, IGN, IGP, swisstopo, and the GIS User Community

Alpine PCC: 86.7% AUC: 0.97 Kappa: 0.73

Source: Esri, DigitalGlobe, GeoEye, i-cubed, USDA, USGS, AEX, Getmapping, Aerogrid, IGN, IGP, swisstopo, and the GIS User Community

Validation Statistics

Elev. Zone Lower Middle Upper

Correctly Classified 72.5 % 91.3 % 86.7 %

AUC 0.82 0.99 0.97

Kappa 0.46 0.99 0.73

Source: Esri, DigitalGlobe, GeoEye, i-cubed, USDA, USGS, AEX, Getmapping, Aerogrid, IGN, IGP, swisstopo, and the GIS User Community

Results and the High Park Fire Visual assessment of model against current Landsat 8 imagery (June 29, 2013)

Conclusions • Numerous unmapped wetlands in the CLP watershed,

including the High Park Fire burn area.

• Modeling within distinct elevation zones is a valuable strategy for refining wetland models.

• Modeling wetlands in developed areas continues to be a challenge for methodologies relying on moderate-resolution remotely sensed data but may be less important than less urbanized areas.

• Potential methodology to aid monitoring effects of wildfire and land use change in wetlands.

Future Work

• Field validation through wetland plant identification

• Refinement of model in developed regions

• Inclusion of forthcoming cloudless Landsat 8 imagery in model

• Application to other watersheds in the Intermountain West

Project Partners David Merritt, U.S. Forest Service

Jeremy Sueltenfuss, Colorado Natural Heritage Program The City of Fort Collins

Science Advisors

Paul Evangelista, Natural Resource Ecology Lab, CSU Jeff Morisette, USGS, North Central Climate Science Center

Melinda Laituri, Ecosystem Science and Sustainability, Geospatial Centroid, CSU Nicholas Young, Natural Resource Ecology Lab, CSU

NASA DEVELOP Past Contributors

Amy Birtwistle, CSU Brenda Kessenich, CU Boulder

Matthew Luizza, CSU Amber Weimer, CSU

Recommended