TidalFlats_Classifier

Purpose

Enables you to detect tidal flats with just two functions!

A user-friendly package that enables you to implement the approach of Murray et al. 2018 using the Google-Earth-Engine API in Python!

A pretrained Classifier will be used → no need to collect training data / train a model.\

Meant to support the conservation of migratory birds that rely on tidal flats as resting places.

Contents

The package contains of two functions, one example script and the classifier itself:

• download_classifier.py: Downloads a trained Random Forest into your system.

• run_classifier.py: After specifying the area of interest (aoi) and the start and end year of the analysis, it performs the classification.

• example_script.py: Shows a simple example workflow.

Requirements

• Google Earth Engine account

• Google Cloud project ID (for GEE authentication)

• Python >= 3.13

Installation & Setup

The package can be installed using:\

!pip install git+https://github.com/milesberberich/TidalFlat_Classifier.git

from tidalflats_classifier import download_classifier, run_classifier

To use Google-Earth-Engine, run:

import ee

ee.Authenticate()

ee.Initialize(project="your_example_project") 

Example

After installation and authentication of GEE, this is the only code required to run the model and download the results.

classifier = download_classifier()


result = run_classifier(path_to_aoi="/home/example_areaofinterest.shp", start_year="2016", end_year="2019", classifier=classifier)


url = result.getDownloadURL({'scale': 30,'fileFormat': 'GeoTIFF'})

print(url)

The classification can then be downloaded using the link given by the code.\

The start_year and end_year need to span a three-year interval.

Output

Using the link provided by the script, you can download a .tiff file with the classification.\

The classification will be saved like:

0 = "Other" (mostly land and vegetated areas)\

1 = "Water"

2 = "Tidal Flat"

The classification has a spatial resolution of 30m and a temporal resolution of three years. It uses EPSG:4326.

Reminder: Depending on the software used to visualize the result, the class "Other" (0) might be set as a NoData-Value.

Methodology

The classification uses an approach based on Murray et al. 2018.

The full workflow used to train the model is documented in https://github.com/GebTorte/WWFTidalFlats.

Most of the 56 parameters are indices and metrics derived from Landsat data. Furthermore auxiliary data like NOAA/NGDC/ETOPO1 and JRC/GSW1_4/GlobalSurfaceWater were used.

The Random Forest-Classifier was trained using the training data of Murray et al. 2018.

The model was trained using the code originally used in the paper.

Accuracy

Overall Accuracy = 95%

Tidal Flat Precision = 88,40%

Tidal Flat Recall = 96.97%

Tidal Flat F1-Score = 92.48%

The classifier is highly reliable and performs well. However, the model occasionally overpredicts tidal flats.

Scope

The purpose of this package is to provide an easy-to-use tool for conservationist to:

• locate current tidal flat habitats

• quantify habitat loss over time

• identify hotspots for conservation

All of that can be done without training data, computational resources or extensive programming knowledge.\

It can be part of a larger GEE workflow or just used as a standalone tool. The main purpose of this package is to support the conservation of tidal flats, especially for migratory bird.

Contributions and Context

The model was created as a joint effort by Rosemary Jones, Simon Sacher, Jule Pfeiffer, and Miles Berberich: https://github.com/GebTorte/WWFTidalFlats\

It was created as part of the class "EO in ecology" supervised by Dr. Wegmann for the WWF.