High Resolution Phragmites Australis Classification in Delaware Estuaries

This dataset provides a high resolution (1-m) land cover map for Estuarine wetlands in the State of Delaware in the United States of America during the summer of 2017. This dataset was created to identify populations of the invasive marsh species Phragmites australis.

Input data:

This classification is derived from National Agriculture Imagery Program (NAIP) 1-m aerial imagery captured in the State of Delaware during June of 2017. NAIP imagery includes a blue, green, red, and near infrared band. To improve classification accuracy, a Normalized Difference Vegetation Index (NDVI) was calculated from NAIP imagery using the near infrared and red bands. A principal component analysis (PCA) was used on the four NAIP bands and the NDVI band to create five new PCA bands. The five PCA bands were used as input into a random forest classification.

NDVI = (Near infrared - Red) / (Near infrared + Red)

Classification methods:

We classified the input data using a Random Forest classifier with 100 trees. Data was classified into three coded land cover classes:

1 - Phragmites

2 - Other Vegetation

3 - Open Water

1,050 land cover reference points were collected with 70% used to train and 30% to test the classifier.

Accuracy:

Measures of accuracy including overall accuracy and per class user’s (UA) and producer’s accuracy (PA) of the random forest classifier were calculated.

Overall accuracy: 95%

Kappa: .92

Phragmites: UA = 97% PA = 95%

Other vegetation: UA = 92% PA = 96%

Open water: UA = 100% PA = 95%

Code link:

The Google Earth Engine code used in this analysis is publicly available.

https://github.com/mattswalter/Phragmites_Classification

Data for download:

The following zipped file is available for download:

       1. Delaware_Phragmites_Classification.zip

Contains a GEOTIFF titled "Phrag_DE_5PC" with the classified image for 2017.