ANALYSIS OF SIMILARITY FOR SCALING OUT, an example with the Andean region

1
ANALYSIS OF SIMILARITY FOR SCALING OUT, an
example with the Andean region
byJorge Rubiano Martha Otero German Lema Victor
Soto
2
Objective

• To identify similar areas outside of the pilot
  catchment in the Andes Eco-region BY
• Applying classification methodologies
• -Weight of Evidence and Logistic Regression
• Fast Cluster Analysis
• FOR future research extrapolation

3
Weight of Evidence and Logistic Regression

Log linear model for multivariate analysis to
obtain a probabilistic distribution of the
modeled events.

4
Weight of Evidence and Logistic Regression

Pilot catchments (training points) Colombia
Neusa, Miel, Cauca Ecuador El Angel,
Ambato Peru Piura,Alto Mayo, Jequetepeque,
Arequipa Bolivia Colomi. Explanatory
Variables Rainfall, elevation, length of growing
period, land use, roads, population
5
1. Pilot Catchments in Andean Eco-region
(training points)
2. Explanatory Variables (evidential themes)
3. Posterior Probability Map


6
Probabilistic Map All pilot catchments
Probabilistic Map e.g. one basin Miel
7
Fast Cluster Analysis
Uses an iterative Euclidian distance algorithm to
classify the space in clusters accordingly to the
attribute used to describe the area large data
sets The most predominant clusters in the
catchments are the clusters 4, 5 and 8.
Characterize by elevations higher than 2000
m.a.s.l and rainfall around 1000 mm.

8
e.g. Cauca Tributaries belongs to clusters 2, 4
and 8
Clusters in the Catchments
2
4
8
9
Discussion

• The two method presented differ in classification
  procedure and outputs but are complementary
• - the weight of evidence and logistic regression
  give an integrated measure (probability) of the
  similarity of any location with the training
  sites provided
• the fast clustering describes the characteristics
  of each variable in the produced clusters.