You are on page 1of 12

Predicting Forest

Composition
using LandSat 7
Predictive Modeling Pre-Processing Presentation
J. Shannon
K. Colbert
Project Goal
Quantify the distribution of Black
Ash across St. Louis County,
Minnesota

● St. Louis County, Minnesota


● Black ash tree
● Threatened by the invasive
species Emerald Ash Borer

https://www.arborday.org/trees/health/pests/images/figure-emerald-ash-borer-1.jpg
Data Sources and Descriptions

● Forest Inventory Data


○ 5876 Plots
○ September 2012 - March 2018
○ Basal area of black ash
● 120 Predictors from Landsat imagery
○ Raw data
■ Visual, Infrared, RGB
○ Derived measures from raw values
Plot the Data
Plot the Data
Plot the Data
Plot the Data

Correlation matrix plot of all predictors Correlation matrix plot of predictors from March
Missing Data
Skewness & Transformation
Imputation and Predictor Reduction

● KNN Imputation, k=5


● Center and scale
● Remove collinear predictors
❌ PCA

✔ Correlation cutoff (0.95)


Outliers
Conclusion
Pre-Processing Steps Data Spending

● Remove missing observations ● 80 training set /20 testing set


● Box Cox transformation ○ Random selection
○ 17 → 83 symmetrical ● Resampling
● Removed 22 highly correlated ○ 10 fold Cross Validation
predictors
● Spatial-sign transformation to
remove outliers

You might also like