# Noise Injection

## Introduction

Our data contains the trend we are looking for and some noise that should be based on a specific distribution. Since we cannot go out and sample our field sites over and over again, we can use the computer to inject different amounts of noise into our data to see what impact the noise might be having on our models.

## Noise Injection in R

The code below will inject noise into one of the columns of our data table, run the model, collect statistics, and then repeat the process. We also load the data each time through to loop so our errors do not accumulate on each successive run.

# Setup the vectors for the statistics of interest AICs=vector() # loop while loading the data, injecting noise, and creating the model Count=1 while (Count<20) # loop injecting different noise each time through the loop { # Read the data each time because we will be modifying it below TheData = read.csv("D:\\jimg\\Classes\\GSP 570\\GSP 570\\Cali_5Minute/DougFir_CA_Predict 4.csv") # Remove any NA (null) values TheData=na.omit(TheData) # Get the number of rows in the data to create a vector with random data NumRows=length(TheData$AnnualPrecip) # Add some random data TheData$AnnualPrecip=TheData$AnnualPrecip+rnorm(NumRows, mean=0, sd=10) # Create the model attach(TheData) # attach is required for the "predict.gam()" function TheModel=gam(Height~s(AnnualPrecip),data=TheData,gamma=1.4) detach("TheData") # Add the AIC for this model to our results AICValue=AIC(TheModel) AICs=c(AICs,AICValue) # Incrememnt our counter so we don't keep going forever Count=Count+1 } # Output stats on our model runs summary(AICs) plot(AICs) hist(AICs)