Applying 2022-01-12 shrubland models to all lidar coverages.
This report is a companion to Shrubland 0.0.1: Making It Happen. It describes the same models as that document.
The logistic ensemble model was used to predict shrubland in all areas with LiDAR-derived CHMs. Predictions were restricted to only pixels of vegetated LCPRI classes (excluding developed, barren, ice/snow, and water). The predicted probabilities were then classified using four thresholds, documented in the model report document, chosen to target certain values of specificity in order to improve positive predictive value.
These results pool training, validation, and test pixels with the remainder of the state. As only 00.02% of pixels were used in any of these roles, these results are for all practical purposes equal to out-of-bag accuracy.
Threshold: 0.416
Confusion Matrix and Statistics
1 0
1 1790234 20837811
0 317255 58932889
Accuracy : 0.7416
95% CI : (0.7415, 0.7417)
No Information Rate : 0.9743
P-Value [Acc > NIR] : 1
Kappa : 0.1025
Mcnemar's Test P-Value : <2e-16
Sensitivity : 0.84946
Specificity : 0.73878
Pos Pred Value : 0.07912
Neg Pred Value : 0.99465
Prevalence : 0.02574
Detection Rate : 0.02186
Detection Prevalence : 0.27636
Balanced Accuracy : 0.79412
'Positive' Class : 1
Threshold: 0.848
Confusion Matrix and Statistics
1 0
1 831553 3394254
0 1275936 76376446
Accuracy : 0.943
95% CI : (0.9429, 0.943)
No Information Rate : 0.9743
P-Value [Acc > NIR] : 1
Kappa : 0.2364
Mcnemar's Test P-Value : <2e-16
Sensitivity : 0.39457
Specificity : 0.95745
Pos Pred Value : 0.19678
Neg Pred Value : 0.98357
Prevalence : 0.02574
Detection Rate : 0.01016
Detection Prevalence : 0.05161
Balanced Accuracy : 0.67601
'Positive' Class : 1
Threshold: 0.882
Confusion Matrix and Statistics
1 0
1 513678 1565742
0 1593811 78204958
Accuracy : 0.9614
95% CI : (0.9614, 0.9615)
No Information Rate : 0.9743
P-Value [Acc > NIR] : 1
Kappa : 0.2256
Mcnemar's Test P-Value : <2e-16
Sensitivity : 0.243739
Specificity : 0.980372
Pos Pred Value : 0.247029
Neg Pred Value : 0.980027
Prevalence : 0.025739
Detection Rate : 0.006274
Detection Prevalence : 0.025397
Balanced Accuracy : 0.612056
'Positive' Class : 1
Threshold: 0.902
Confusion Matrix and Statistics
1 0
1 260243 626483
0 1847246 79144217
Accuracy : 0.9698
95% CI : (0.9698, 0.9698)
No Information Rate : 0.9743
P-Value [Acc > NIR] : 1
Kappa : 0.161
Mcnemar's Test P-Value : <2e-16
Sensitivity : 0.123485
Specificity : 0.992146
Pos Pred Value : 0.293488
Neg Pred Value : 0.977192
Prevalence : 0.025739
Detection Rate : 0.003178
Detection Prevalence : 0.010830
Balanced Accuracy : 0.557816
'Positive' Class : 1
Raster files can be downloaded from GitHub at this link.
Files ending in classified have been classified into four categories:
Files not ending in classified represent the binary shrubland maps (1 = shrub) using the threshold specified above.
The probability map is too large to share easily; I’ll be adding it to Labrador in the near future.
If you see mistakes or want to suggest changes, please create an issue on the source repository.
For attribution, please cite this work as
Mahoney (2022, Jan. 12). CAFRI Labs: Statewide Shrubland Prediction Accuracy. Retrieved from https://cafri-labs.github.io/acceptable-growing-stock/posts/statewide-shrubland-prediction-accuracy/
BibTeX citation
@misc{mahoney2022statewide,
author = {Mahoney, Mike},
title = {CAFRI Labs: Statewide Shrubland Prediction Accuracy},
url = {https://cafri-labs.github.io/acceptable-growing-stock/posts/statewide-shrubland-prediction-accuracy/},
year = {2022}
}