Urban Land-Cover Classification for Mesoscale Atmospheric Modeling
Post on 05-Jan-2016
20 Views
Preview:
DESCRIPTION
Transcript
Urban Land-Cover Classification for Mesoscale Atmospheric Modeling
Alexandre Leroux, M.Sc., Ing.
• Canadian Meteorological CentreEnvironment Canada’s National Center for data assimilation and numerical weather prediction, climate and air quality modeling
• Environmental Emergency Response DivisionProvides highly specialized support to environmental emergencies including atmospheric dispersion and trajectory modeling
Context
• High resolution atmospheric numerical models require detailed characterisation of the Earth’s surface to drive sophisticated surface parametrisation schemes. This requirement is even more important for complex urban environments
Objectives
• Goal:Provide an urban land-cover database for North-American cities for mesoscale atmospheric modeling, specifically, for the Town Energy Balance scheme (TEB)
• Mean:- Approach #1 (snapshot overview)
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
Input data processing and analysis
• Satellite urban land-cover classification:Mid-resolution unsupervised classification of Landsat-7 and ASTER data
• Building height appraisal:– SRTM-DEM for elevation at top of features (e.g. trees,
buildings)– CDED1 (Canada) and NED (USA) for soil elevation– The subtraction evaluates the building height
Satellite approach results
• Computed statistics and fractions are feed to the decision tree
• Main results:– 12 new urban classes generated at 60m– +/- 5 vegetation classes
• Processing and analysis: ~ 1 week / urban area• Results over OkC, Mtl and Van are satisfactory
Oklahoma City, 60 mOklahoma City, 60 m
Montreal, 60 m (detail)Montreal, 60 m (detail)
Vancouver, 60 m (detail)Vancouver, 60 m (detail)
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
Scripted Spatial Data Processing
Complete automation:• Automated dataset identification• Read/write multiple formats, including CMC custom
formats• On-the-fly reprojection and datum management• Different spatial resolution / scale management• Spatial data cropping, subtraction (cookie cutting),
buffering, rasterizing, SQL queries on attributes, multiple layer flattening (merge down), basic spatial queries, LUT value attribution and much more…
• Makes use of GDAL and OGR open C libraries
Results
• Results for Montreal and Vancouver– Raster output at 5m spatial resolution, generates rater
data with 10,000 x 12,000 pixels (50 x 60 km, Toronto)
• Other processed cities– Calgary, Edmonton, Halifax, Ottawa, Quebec, Regina,
Toronto, Victoria, Winnipeg
• The methodology, processing, analysis and results are well documented
TEB classes
• 44 ‘final’ aggregated classes– Buildings (18 classes)
• 1D & 2D, height, use (i.e. 24/7, industrial-commercial)
– Residential areas, divided by population density (5 classes)
– Roads and transportation network (6 classes)
– Industrial and other constructions (5 classes)
• e.g. tanks, towers, chimneys
– Mixed covers (3 classes)
– Natural covers (7 classes)
Population density Population density classes, Montrealclasses, Montreal
1 km
Population density classes, Population density classes, VancouverVancouver
1 km
1 km
18 building classes, 18 building classes, Downtown MontrealDowntown Montreal
1 km
18 building classes, 18 building classes, Downtown VancouverDowntown Vancouver
Transportation network, Transportation network, VancouverVancouver
1 km
Detail of Montreal,Detail of Montreal,
Scaled-down, 44 classesScaled-down, 44 classes
1 km
1 km
Detail of Vancouver,Detail of Vancouver,
Scaled-down, 44 classesScaled-down, 44 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– NTDB 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…
Urban canyon modeling: linking mesoscale models to CFD models at the urban scale
top related