coastal-landcover-classification

This repository contains code that was used to develop the first national scale coastal specific landcover classification for New Zealand using public earth observation satellite data. Google Earth Engine was used to develop composite imagery for 2019 from Sentinel-1 and Sentinel-2 sensors which was classified into nine landcover types using rule-based and supervised machine learning techniques in a Python workflow with RSGISLib.

Classes
Artificial surfaces
Bare rock
Dark sand
Gravel
Intertidal
Light sand
Supratidal sand
Vegetation
Water

Table of contents

1. Installation
2. Python modules
3. Jupyter notebooks
4. New Zealand coastal classification

Installation

Packages and dependencies handled are handled by conda

conda create --name coastal-classification

conda activate coastal-classification

conda install --file requirements.txt

python -m ipykernel install --user --name=coastal-classification

jupyter notebook

Python code

Code in this repository is contained in the coastal_landcover_classification package, which consists of two modules:

1. coastal_landcover_classification.composite handles the preprocessing and generation of annual composite imagery from all available images within a specified year.

• Filters imagery to area of interest and year.
• Applies preprocessing steps to both optical and SAR data.
• Derives statistical aggregations of vegetation and water based indices (NDVI, NDWI, MNDWI and AWEI).
• Downloads composite imagery locally or to Google Drive.

2. coastal_landcover_classification.classification contains the functions to classify composite imagery to provide an annual coastal specific landcover classfication.

• Applies a series of hierarchal rules using automated Otsu thresholding to identify water, intertidal and vegetation from multispectral composite imagery.
• Classifies remaining classes using a random forest machine learning classifier trained with a manually derived national training dataset, included in this repository, using the multi-spectral and SAR composites images.

Jupyter notebooks

A series of jupyter notebooks containing a working example of both steps are provided:

• Annual composite creation
• Classification example

New Zealand coastal classification

The classification output generated for the year 2019 is available to view as a Google Earth Engine App.