Title: Part B: Surface analysis
1Chapter 6
- Part B Surface analysis gridding and
interpolation
2Surface analysis - Gridding Interpolation
- Gridding (a form of systematic interpolation)
- to convert a sample of data points to a complete
coverage (set of values) for the study region - to convert from one level of data resolution or
orientation to another (resampling) - to convert from one representation of a
continuous surface to another, e.g. TIN to grid
or contour to grid - Quality of process
- Input data quality/accuracy
- Data density
- Data distribution
- Spatial variability of data
- Model applied typically weighted average
3Surface analysis - Gridding Interpolation
- Some other key issues
- Validity of continuity assumption/handling of
breaklines/stratifications and missing data - Validity of weighted average model
- How to choose weights?
- How many points to include in summation?
- Shape and size of neighbourhood region
- Selection of model and parameters
- Evaluation of quality
- Availability of related datasets
4Surface analysis - Gridding Interpolation
- Comparison of interpolation methods
Method Speed Type
Inverse distance weighting (IDW) Fast Exact, unless smoothing factor specified
Natural neighbour Fast Exact
Nearest-neighbour Fast Exact
Kriging -Geostatistical models (stochastic) Slow/Medium Exact if no nugget (assumed measurement error)
Conditional simulation Slow Exact
Radial basis Slow/Medium Exact if no smoothing value specified
Modified Shepard Fast Exact, unless smoothing factor specified
Triangulation with linear interpolation Fast Exact
Triangulation with spline interpolation Fast Exact
Profiling Fast Exact
Minimum curvature Medium Exact/Smoothing
Spline functions Fast Exact (smoothing possible)
Local polynomial Fast Smoothing
Polynomial regression Fast Smoothing
Moving average Fast Smoothing
Topogrid/Topo to Raster Slow/Medium Not specified
5Surface analysis - Gridding Interpolation
- Contour generation from grids
- Simple linear interpolation
- Smoothing splines or iterative thresholding
6Surface analysis - Gridding Interpolation
- Deterministic interpolation
7Surface analysis - Gridding Interpolation
- Deterministic interpolation IDW
- General model
- IDW version
- The values are computed for all or selected i
(e.g. 12) - The value for alpha is typically 1 or 2
- The division by the sum of weighted distances
ensures that the weights add up to 1
8Surface analysis - Gridding Interpolation
- Deterministic interpolation IDW
Simple IDW calculations for point 5,5
each data point contributes
Grid point (5,5)
Calculation of weights using alpha1 and alpha2
9Surface analysis - Gridding Interpolation
- Sample data for main examples (x,y,z)
10Surface analysis - Gridding Interpolation
- Deterministic interpolation IDW
IDW grid alpha2, no smoothing
Source data
11Surface analysis - Gridding Interpolation
- IDW
- All grid points in rectangle are assigned a value
- Sample points are exactly fitted
- Surface looks rather peaky with odd
hollows/bulls eyes - affected by point
locations, alpha and grid resolution - How do we know what values to choose - number of
neighbours, alpha, grid res., break lines? - Can we judge the goodness of fit of the
resultant surface? - Generally no, but could test the fit by (a)
additional sampling (b) comparison with other
sources of data (e.g. aerial photographs) (c)
computational procedures
12Surface analysis - Gridding Interpolation
- Deterministic interpolation Nearest neighbour
- Simple assignment on nearest neighbour basis
- Uniform within zones, stepped between zones
13Surface analysis - Gridding Interpolation
- Deterministic interpolation Natural neighbour
- Assumes each existing point, i, has a natural
sphere of influence its Voronoi polygon with
area Ai - Assumes a newly added point, p, (e.g. grid point)
captures part of these natural areas for itself.
These captured parts have areas Aip - Weights are based on initial and captured Voronoi
regions
14Surface analysis - Gridding Interpolation
- Deterministic interpolation Natural neighbour
Nat N grid extent limited to convex hull
Source data
15Surface analysis - Gridding Interpolation
- Deterministic interpolation Radial basis/splines
- A large family of interpolation functions
- Exact (or smoothing)
- Exhibit well-defined characteristics e.g.
properties of smoothness and fit - Thin plate splines widely used in Earth Sciences
- Core model is a form of weighted transformed
distances
bias or offset value
z-value at point p
Radial basis function, for radius ri
Weight for ith point
16Surface analysis - Gridding Interpolation
- Deterministic interpolation Radial basis/splines
- Computation
- Compute matrix D of all interpoint distances
- Apply radial basis function to all elements of
D?? - For grid point p, compute distance to all data
points, as a vector r - Apply radial basis function to all elements of
r?? - Form augmented matrix equation shown and solve by
inversion
17Surface analysis - Gridding Interpolation
- Deterministic interpolation Thin Plate Spline
Using 12 point neighbourhoods
Using all 62 data points
18Surface analysis - Gridding Interpolation
- Deterministic interpolation Modified Shepard
- Hybrid IDW-like system
- Typically uses local IDW or quadratic fit plus
longer range IDW with separate parameter
Example using 8 local points for quadratic fit
and 16 for longer range IDW
19Surface analysis - Gridding Interpolation
- Deterministic interpolation
- Triangulation with linear interpolation (i.e. 2D
linear model)
Linear grid extent limited to convex hull
Source data
20Surface analysis - Gridding Interpolation
- Deterministic interpolation Minimum curvature
- Seeks smoothed elastic membrane fit to surface
- Fit to residuals after linear regression of
original dataset - Linear component then added back
21Surface analysis - Gridding Interpolation
- Deterministic interpolation Local polynomial
- Fits a local polynomial (e.g. quadratic) to each
grid point, using window (e.g. circle) of given
size or number of points - Smoothing
Example local quadratic distance-weighted OLS,
with 12 points per fit
22Surface analysis - Gridding Interpolation
- Deterministic interpolation some other methods
- Triangulation with non-linear interpolation -
Clough-Tocher bi-cubic patches fitted to every
triangle, such that edge and vertex joins are
smooth - Bi-linear suitable for dense uniform datasets,
e.g. grids with some missing cell values simple
weighted averaging of directly or indirectly
neighbouring cells/points - Profiling input dataset is contour vectors.
Typically interpolation is linear between closest
pair of contours (8-point) - Moving average uses a moving circular or
elliptical window (similar to time series
analysis) to obtain simple local average - Topogrid conversion of contour vectors to
hydrologically consistent DEM. Includes
enforcement of local drainage, lake boundaries,
ridges and stream lines
23Surface analysis - Gridding Interpolation
- Geostatistical interpolation
24Surface analysis - Gridding Interpolation
- Geostatistical interpolation Core concepts
- Geostatistics definition
- models and methods for data observed at a
discrete set of locations, such that the observed
value, zi, is either a direct measurement of, or
is statistically related to, the value of an
underlying continuous spatial phenomenon, F(x,y),
at the corresponding sampled location (xi,yi)
within some spatial region A. (Diggle et al)
25Surface analysis - Gridding Interpolation
- Geostatistical interpolation Core concepts
- Addresses questions such as
- how many points are needed to compute the local
average? - what size, orientation and shape of neighbourhood
should be chosen? - what model and weights should be used to compute
the local average? - what errors (uncertainties) are associated with
interpolated values?
26Surface analysis - Gridding Interpolation
- Geostatistical interpolation Variograms
- Analyse the observed variation in data values by
distance bands using a spatial autocorrelation-lik
e measure, ? - Typically, analyse all pairs of observations that
lie within specific distance bands - Compute the average values of ? for each band
- Plot these averages against distance
- Fit an experimental curve to the observed pattern
27Surface analysis - Gridding Interpolation
- Geostatistical interpolation Variograms
- Analyse the observed variation in data values by
distance bands using a spatial autocorrelation-lik
e measure, ? - Semivariance measure is most often used
- Bands have width ?. N(h) is the number of pairs
in the band with mid-point distance h
28Surface analysis - Gridding Interpolation
- Geostatistical interpolation Zinc data (x,y,z)
- Burrough McDonnell (1998) App.3 98 soil samples
(zzinc ppm)
29Surface analysis - Gridding Interpolation
- Geostatistical interpolation Variograms
Smallest observed separation
Fitted curve
sill
Average semivariance for band 4
C1C0C (structural variance)
Range, A0
C0 Nugget
Lag (distance band)
model
30Surface analysis - Gridding Interpolation
- Geostatistical interpolation Variogram models
Model Formula Notes
Nugget effect Simple constant. May be added to all models. Models with a nugget will not be exact
Linear No sill. Often used in combination with other functions. May be used as a ramp, with a constant sill value set at a range, a
Exponential Exp() k is a constant, often k1 or k3. Useful when there is a larger nugget and slow rise to the sill
Spherical Sph() Useful when the nugget effect is important but small. Given as the default model in some packages.
31Surface analysis - Gridding Interpolation
- Geostatistical interpolation Core concepts
- Geostatistical methods assume a general form of
model which is a mix of - (a) a deterministic model m(x,y)
- (b) a regionalised statistical variation from
m(x,y) - (c) a random noise (Normal error) component
(actually two components which we cannot
generally separate) - (b) is chosen by analysing the semivariance, ?
32Surface analysis - Gridding Interpolation
- Geostatistical interpolation Further key
concepts - Sample size
- Support
- Declustering
- Thresholding
- Stationarity
- Transformation
- Anisotropy
- Madograms and Rodograms
- Periodograms and Fourier analysis
33Surface analysis - Gridding Interpolation
- Geostatistical interpolation Kriging
- Examine the data (z values) for spatial trends
and Normality and transform variables if
necessary, e.g. loge(z) or loge(z1) - Compute the experimental variogram and fit a
suitable model IFF the pattern warrants this
(i.e. the data is not just noise) - Check the model by cross validation and examine
size of mean squared deviation - Decide on method to use Kriging or Conditional
simulation - Generate a grid from the selected model and plot
- Compare the results to (a) the input data points
and (b) other sources of data/information
34Surface analysis - Gridding Interpolation
- Geostatistical interpolation Kriging
- Ordinary Kriging basic procedure is exactly the
same as for radial basis functions, but with
function applied being the modelled variogram - Computation
- Compute matrix D of all interpoint distances
- Apply variogram function to all elements of D??
- For grid point p, compute distance to all data
points, as a vector r - Apply variogram function to all elements of r??
- Form augmented matrix equation shown and solve by
inversion
35Surface analysis - Gridding Interpolation
- Geostatistical interpolation Ordinary Kriging
(OK) - Estimated value at p is then
- Estimated variance at p is
- where m is the Lagrangian multiplier and the ci
are the modelled semivariance values (the ?
values)
36Surface analysis - Gridding Interpolation
- Geostatistical interpolation Ordinary Kriging
Zinc data Predicted values Zinc data Estimated standard deviation
37Surface analysis - Gridding Interpolation
- Geostatistical interpolation Kriging
- Goodness of fit methods
- Residual plots, standard deviation plots
- Examination for artefacts
- Simple cross-validation
- Jack-knifing
- Re-sampling
- Examination of independent datasets
- Stratification
38Surface analysis - Gridding Interpolation
- Geostatistical interpolation Kriging other
models - Universal Kriging Kriging with a trend
- Indicator Kriging thresholding
- Stratified Kriging regionalised approach
- Co-Kriging use of associated data
39Surface analysis - Gridding Interpolation
- Geostatistical interpolation Conditional
simulation (Gaussian sequential method) - Similar to OK approach. 3 core steps
- Analyse the data/transform as necessary, fit
model variogram, define grid to use (possibly
multi-level) - Randomly select a grid node to visit (e.g. apply
a random walk process) as execute step 3 - Apply OK estimation at node based on local
observed data points. Take estimated value and
apply Normal random process with mean as estimate
and variance as estimated variance. Return to
step 2 - This process is then repeated until all nodes
have been visited. Then iterated M times
40Surface analysis - Gridding Interpolation
- Geostatistical interpolation Conditional
simulation, M100 iterations
Zinc data Predicted values Zinc data Estimated standard deviation