Estimation - Geostatistical R Package

o What can be estimated ? Estimation. ➢ Punctual estimation. ➢ Block average estimation. Estimate the value Z. 0 at the nodes of. Geostatistics & RGeostats. 2.
2MB taille 59 téléchargements 471 vues
GeoEnv - July 2014

Estimation D. Renard N. Desassis

Geostatistics & RGeostats

1

Estimation o What can be estimated ? Punctual estimation Block average estimation

Estimate the value Z0 at the nodes of a regular grid Samples Estimate the average of the variable Z over the block

Geostatistics & RGeostats

2

Estimation o Linear estimation techniques Linear estimator: • Each estimation is obtained as a linear combination of the values measured at sample points

Several linear interpolation techniques: • Moving average • Inverse distance (closest point) • Inverse distance

Properties of the estimation: • Smoothness • Unbiasedness • Exact interpolation

Geostatistics & RGeostats

3

Interpolation o Illustration Exhaustive data set (reality) Irregular sampling used as data For each method, represent the estimation as a map and along a section >=3 2.7 2.4 2.1 1.8 1.5 1.2 0.9 0.6 0.3 0 -0.3 -0.6 -0.9 -1.2 -1.5 -1.8 -2.1 -2.4 -2.7 radius 33.3%

33.3% 37.0%

σ = 0.45 2

33.3% 25.7%

48.7% Geostatistics & RGeostats

26.0% 25.7%

σ 2 = 0.526

σ = 0.48

37.0%

2

50.0%

σ 2 = 0.537

50.0% 36

Cross-validation o Principle At each data point:

Zα*0 =

• Suppress the sample value • Estimate its value by Kriging • Compare real to estimated values

Statistics on: • Error:

• Normalized error

Geostatistics & RGeostats

λα Zα ∑ α α ≠

0

ε α = Zα* − Zα 0

ε αR =

Zα*0 − Zα

σα

37

Neighborhood o Principle Kriging considers available samples in the system When too many samples, kriging system becomes very large and may become difficult to invert (unstable, slow) Kriging weights of peripheral points are small: could they be neglected? Neighborhood: Unique: Take all data available Moving: Select the most appropriate subset of neighboring samples • By number • By maximum distance • By angular sector

Geostatistics & RGeostats

38