Você está na página 1de 4

Geog 169 Costa Rica Group

Anna Hubel
March 7, 2017

1. Investigating the association between the two tree height measures, LVIS and
GLAS

A scatterplot of the LVIS tree height measurement against the GLAS tree height measurement is included
below. A linear smoother is also included in order to visualize the trend.
Association between GLAS and LVIS Tree Height Measurements
LVIS Tree Height Measurement (m)

40

20

0
0 20 40 60
GLAS Tree Height Measurement (m)
Because there is a visible positive trend in these data, the next step is to formally fit a linear model and see
if a linear model is appropriate. Diagnostic plots below check model assumptions that:
the trend is linear
the errors are approximately normally distributed
the variance is constant across the fitted values

1
Residuals vs Fitted Normal QQ

Standardized residuals
256 256
20 2
Residuals

0 0

20 2
60 60
40 24 4 24
10 20 30 40 2 0 2
Fitted values Theoretical Quantiles
ScaleLocation Residuals vs Leverage
Standardized residuals

Standardized Residuals
2.0 24
256 60 2
1.5

0
1.0

2 101 196
0.5

0.0 4 24
10 20 30 40 0.00 0.01 0.02
Fitted values Leverage

The diagnostic plots above check the assumptions of the linear model fit to the data. The residuals vs. fitted
plot shows that there is little to no trend in the residuals of the model across the fitted values, so there is no
clear violation of the linear trend of the model. This means that a linear trend is appropriate. The Normal
Q-Q plot shows only minimal deviation from normality, meaning that the assumption of normality of the
errors is not violated. The Scale-Location plot shows little to no trend in variance across the model, meaning
that the assumption of constant variance is not violated, either. Overall, the only deviations in these model
diagnostics appear to be because of a couple points with high leverage, which means the linear model is valid
for these data.
Now that it is clear that the model is valid, we can show that the slope of the trend is significant and quantify
the strength of the association with an R2 value.

Call:
lm(formula = LVIS.H_Tree ~ GLAS.H_Tree, data = htree)

Residuals:
Min 1Q Median 3Q Max
-39.072 -8.069 0.469 6.972 29.884

Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 9.86331 1.27480 7.737 7.19e-14 ***
GLAS.H_Tree 0.56103 0.04567 12.285 < 2e-16 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Residual standard error: 10.26 on 434 degrees of freedom

2
Multiple R-squared: 0.258, Adjusted R-squared: 0.2563
F-statistic: 150.9 on 1 and 434 DF, p-value: < 2.2e-16
The extremely low p-value of GLAS.H_Tree in the model indicates that the slope of the trend is highly
significant. There is also an R-squared value of 0.258, meaning that about 25% of the variation in the LVIS
tree height measurement can be explained by the GLAS tree height measurement. Since there is a correlation
here, it may be appropriate to calculate the mean between these two tree height measurements and use it for
comparison with other variables.

2. Association between the mean tree height measurement and elevation

A scatterplot of the mean tree height measurement vs. elevation is included below.
Tree Height Measurement vs. Elevation
50
Mean Tree Height Measurement (m)

40

30

20

10

0 500 1000 1500 2000 2500


Elevation (m)
Again, the next step is to fit a linear model and check the model assumptions with diagnostic plots:

3
Residuals vs Fitted Normal QQ

Standardized residuals
20 2

10 1
Residuals

0 0

10 1

20 2
377
339
324 324377
339
20.0 22.5 25.0 27.5 2 0 2
Fitted values Theoretical Quantiles
ScaleLocation Residuals vs Leverage
Standardized residuals

Standardized Residuals
324
339
377
1.5 2

1
1.0
0

0.5 1

2
377
339
324
0.0
20.0 22.5 25.0 27.5 0.0000 0.0025 0.0050 0.0075
Fitted values Leverage

Again, none of the assumptions of the linear model appear to be violated. There are no trends in the Residuals
vs. Fitted or Scale-Location plots that would indicate violations of the assumptions of linearity or constant
variance, and there is minimal deviation in the Normal Q-Q plot, so the assumption of normally distributed
errors holds as well. Therefore, the linear model is valid for these data.
summary(modelMHTreeelev)

##
## Call:
## lm(formula = Mean.H_Tree ~ Elevation, data = htree)
##
## Residuals:
## Min 1Q Median 3Q Max
## -23.930 -5.861 1.407 6.407 21.605
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 29.2231496 0.7600627 38.448 < 2e-16 ***
## Elevation -0.0037228 0.0005454 -6.826 2.95e-11 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 9.37 on 434 degrees of freedom
## Multiple R-squared: 0.09696, Adjusted R-squared: 0.09488
## F-statistic: 46.6 on 1 and 434 DF, p-value: 2.946e-11
Because the p-value for Elevation is so low, this model is also highly significant. However, the association is
not particularly strong, as reflected in the R2 value of 0.09696.

Você também pode gostar