Geography 3842:477 Advanced Geomatics

1
Geography 38/42477Advanced Geomatics

• Topic 5 Raster Data Orthorectification

2
The Raster Data Model

• Brief review of raster data structure/model
• Array of discrete grid cells
• Each grid layer represents a different theme
• Most appropriate when representing continuously
  distributed phenomena
• Used when conducting certain types of statistical
  analysis
• Problems with data redundancy
• Resolution dependent on cell size
• Rasterisation converts from vector to raster
  format

3
Types of Raster

• Raster datasets can be used to represent thematic
  (discrete)information or continuous rasters
  (floating point)
• Thematic rasters include categorical or ranked
  data as integers (e.g. landuse, soil type,
  suitability)
• Continuous rasters contain real numbers (e.g.
  soil pH, snow depth, temperature, etc.) AND images

4
Raster Images

• Raster datasets representing satellite imagery or
  aerial photography contain grid cell values that
  represent colour or spectral reflectance of the
  corresponding area on the Earths surface
• Image data can be
• single band (e.g. panchromatic)
• or contain multiple bands each representing
  reflectance in a different portion of EM spectrum

5
Raster Data

• Thematic raster datasets have an associated VAT
  (values attribute table)
• contain each unique integer value or class
• the number of pixels with that value
• any other attributes for that group of pixels
• Continuous raster datasets do not have a VAT
• store actual data values only as floating point
• no associated attribute information
• what attribute info would be stored anyway??

6
Georeferencing

• Raster data that have not been georeferenced are
  displayed in image space or file coordinates
• pixels displayed by row/col position in the file
• When a raster dataset is georeferenced, the array
  of grid cells is aligned to an existing real
  world projection and coordinate system
• the projected location of each grid cell with
  reference to that coordinate system is calculated

7
Georeferencing

• Raster data that are already projected (e.g.
  derived from projected source) can be
  georeferenced by creating a world file or
  defining the projection properties of the dataset
• Raster data that are not projected (e.g. derived
  from an unprojected source) are georeferenced by
  identifying a series of ground control points

8
Sources of GCPs

• GCPs are derived from
• projected data sources (raster or vector)
• map coordinates
• (d)GPS surveys
• GCPs link coordinate position of projected source
  to unprojected data
• From and To GCPs

9
Rectification

• GCPs used to derive polynomial equation
• Equation used to transforms file/image
  coordinates to real world coordinates
• Equation is a least squares fit similar to a
  simple linear regression equation

10
Rectification Error

• Single nth-order polynomial equation applied to
  all grid cells in data set
• So calculated and real coordinate of GCP are
  likely not exactly the same

11
RMS Error

• Root Mean Squared Error
• Amount of error between
• actual location (indicated by GCP)
• as compared to calculated location (derived by
  equation)
• RMS Error (xa-xd)2(ya-yd)2-1/2
• Mean RMS Error ? (xa-xd)2(ya-yd)2-1/2/n

12
Rectification

• Conceptually, square grid cells all represent and
  equivalent area of the Earths surface
• In reality, the shape and area represented by a
  grid cell on the Earths surface is not uniform
  due to distortion occurring as a result of the
  projection

13
Rectification

• Consequently, when raster data are georectified
  (projected) the output cell size and/or number of
  rows/columns may change
• Similarly, when raster data are transformed from
  one projection to another the shape and area
  represented in the real world changes and the
  cell size and/or number of row/columns may also
  change

14
(No Transcript)
15
Georeferencing

• A geometric transformation of a raster data set
• The geometry of grid cells are
• rotated
• skewed
• shifted
• and scaled to conform
• . . . to the projected coordinate system

16
Georeferencing

• Location of each grid cell changes as compared to
  the input grid coverage
• AND location of output grid cell does not
  correspond with any input cell value
• So, output grid cell values need to be
  recalculated
• This process is referred to as resampling

17
Georeferencing

• Geometric transformation consists of two fcns
• Establishing the new grid cell coordinates
• extent of output grid is determined by applying
  polynomial transformation to the bounding box of
  the input grid
• area is then divided into array of grid cells
  based on user specified cell size

18
Georeferencing

• Geometric transformation consists of two fcns
• Determining new output cell values
• identify location of output grid cell on the
  input grid
• calculate output cell value based on values of
  neighbouring grid cells on input grid
• three techniques typically used for calculating
  output cell values

19
Nearest Neighbour

• Location of closest grid cell on input grid is
  assigned to output grid cell
• Does not change cell values
• Used for discrete (integer) raster datasets
• Input cell values do not change

20
Bilinear Interpolation

• Uses value of four nearest grid cells on input
  coverage
• Calculates weighted mean based on linear distance
  from each of four cells
• Used for continuous raster datasets
• Not appropriate for discrete rasters
• Output cell values are real numbers

21
Cubic Convolution

• Similar to bilinear interpolation but considers
  16 adjacent neighbours
• Produces a smoother looking image
• Used for continuous raster datasets
• Not appropriate for discrete rasters
• Output cell values are real numbers
• Note that since cells values change, geometric
  corrections are often applied after images have
  been classified

22
(No Transcript)