Urban Land-Cover Classification for Mesoscale Atmospheric Modeling

Alexandre Leroux

Objectives

• Goal:Provide an urban land-cover database for North-American cities for mesoscale atmospheric modeling, specifically, for the Town Energy Balance model (TEB).

• Mean:- Approach #1 (presented last year)

Satellite imagery and DEM analysis- Approach #2Vector data processing and DEM analysis

Satellite approach - Workflow

Satellite imagery unsupervised classification

Building height assessment

through SRTM-DEM minus

CDED1 or NEDP

rocessing and analysis

Statistics and fractions at a lower scale

Decision tree

Results readied for

atmospheric modeling

Satellite approach results

• 30 to 40 “simple elements” identified on satellite imagery at a 15-m spatial resolution– e.g. asphalt, concrete, roofs, water, trees, grass &

fields

• Results from the decision tree:– 12 new urban classes generated at 60m– +/- 5 vegetation classes associated to gengeo

• Processing and analysis: ~ 1 week / urban area

Oklahoma City, 60 mOklahoma City, 60 m

Montreal, 60 mMontreal, 60 m

(detail, zoom 2x)(detail, zoom 2x)

Vancouver, 60 mVancouver, 60 m

(detail, zoom 4x)(detail, zoom 4x)

Vector approach - Workflow

National Topographic Data Base

• Vector data with 110 thematic layers– e.g. water, vegetation, golf course, built-up areas,

buildings (points and polygons), roads, bridges, railway, etc

• Most layers with attributes– e.g. a road feature can be ‘highway’, ‘paved’,

‘underground’.

• A total of 2474 1:50,000 sheets covering Canada• Available internally within the federal government

Statistics Canada - 2001 Census Data

• Canada-wide coverage• Used to distinguish residential districts

– Population density calculated using this dataset– Includes the number of residences

• Available internally (license purchased by EC)

Statistics Canada – Population density

Topography and Height data

• SRTM-DEM– Top of features (e.g. buildings, vegetation)– Worldwide coverage and free– “Poor” spatial resolution (3 arc-second, ~90m)

• CDED1– Ground elevation– Canada-wide coverage and free– 1:50,000 (mtl: 16 x 23m)

• Subtraction to evaluated building height

“AutoTEB” Spatial Data Processing

• Automated dataset identification• Read/write multiple formats, including ‘.fstd’• On-the-fly reprojection and datum management• Different spatial resolution / scale management• Spatial data cropping, subtraction (cookie

cutting), buffering, rasterizing, SQL queries, multiple layer flattening (merge down), basic spatial queries, LUT value attribution and much more…

Results

• Some results for Montreal and Vancouver– Raster output at 5m spatial resolution, generates rater

data up to 10,000 x 12,000 pixels (Toronto)

• Other processed cities– Calgary, Edmonton, Halifax, Ottawa, Quebec, Regina,

Toronto, Victoria, Winnipeg (SRTM-DEM - CDED1 not yet processed for those cities)

• The methodology, processing, analysis and results are well documented

TEB classes

• 46 ‘final’ aggregated classes– Buildings (18 classes)

• 1D & 2D, height, use (i.e. 24/7, industrial-commercial)

– Residential areas, divided by population density– Roads and transportation network– Industrial and other constructions

• e.g. tanks, towers, chimneys

– Mixed covers– Natural covers

Population density Population density classes, Montrealclasses, Montreal

1 km

Population density classes, Population density classes, VancouverVancouver

1 km

1 km

1 km

Transportation network, Transportation network, VancouverVancouver

1 km

Detail of Montreal,Detail of Montreal,

Scaled-down, 46 classesScaled-down, 46 classes

1 km

1 km

Detail of Vancouver,Detail of Vancouver,

Scaled-down, 46 classesScaled-down, 46 classes

1 km

1 km

Main benefits

• Canada-wide applicability– Full data coverage – Approach directly applied anywhere over Canada

• Complete automation– Single command with only one input parameter– One optional exception: SRTM-DEM minus CDED1– Fast! From 3 min to 40 min for the whole processing

• Numerous other advantages identified…– No interpretation and reduced human intervention– Flexible approach, code developed reusable– Spatial resolution of the results

Main limitations

• Up-to-date data– BNDT data based on “old” aerial imagery: missing

some downtown buildings and suburbs

• Thematic representation– No layer corresponding to rural areas and parking lots– Almost no distinction in vegetation types

• Various other minor limitations identified…

The future of the vector approach

• Adaptation to CanVect and other datasets, potentially including US territory datasets

• Use of 3D building models required for CFD modeling within the vector approach

• Various other improvements envisioned…– TEB sensibility analysis to urban LULC databases– Scientific article to be written– much more…