/GEE-App-Classification-Floods-Porto-Velho

WorCAP 2023

APLICAÇÃO DO GOOGLE EARTH ENGINE COM CLASSIFICAÇÃO DE INUNDAÇÕES DO MUNICÍPIO DE PORTO VELHO – RO (GOOGLE EARTH ENGINE APPLICATION WITH FLOOD CLASSIFICATION IN THE MUNICIPALITY OF PORTO VELHO – RO)

Status License

It was possible to classify the water class with RF and identification of floods in areas of interest, as per presents the Charter initiative map with media images spatial resolution (30 m) of the Landsat-8 satellite. In works future classification will be carried out with other algorithms pattern identification.

📝 Table of Contents

🧐 About

The flood events that occurred in 2014 in the municipality of Porto Velho, capital of the state of Rondônia, were analyzed in this study using flood maps generated by the initiative The International Charter Space and Major Disasters. Charter's products show where events occurred and were used in this work to perform sampling for a pattern recognition algorithm. In this context, 4 class labels were made, namely water, soil, vegetation and urban, to highlight only the areas with water accumulation. Data was separated for training and testing in the Random Forest (RF) machine learning algorithm. The processing of images from the Landsat-8 satellite, OLI sensor, was carried out using the Google Earth Engine (GEE) cloud platform, and aimed to analyze an image before the floods (August 2013), and another after the floods. (April 2014), in order to identify these targets. To verify the learning of the RF, images from 2013 and 2014 were classified, thus obtaining a difference between flooded areas and where the drainage network remained the same. When building the application in a GHG environment, it was chosen to visualize only the “water” class on the final map, before and after the events, aiming to highlight, in the application, only the places where there were floods. The application results show, on the left side of the sliding window, the classification of the “water” class of the image before the flood, in blue, and, on the right side, the classification of this class after the disaster. On the right side there is also the water classification (in white) from the image before the disaster, to highlight the flooded areas.

🏁 Getting Started

These instructions provide a working copy of the project for development and testing purposes. See Development in: GEE Platform Development, GEE Application. These are the project implementation notes.

Prerequisites

To use GEE you need to open an account using gmail at: https://earthengine.google.com

After creating the account, access the platform via the link: https://code.earthengine.google.com

🔧 Running the tests

The following code presents the metrics used to evaluate the RF according to the separate test samples:

//////////////////////////////////////////////////////////////
// Classificar o conjunto de testes
var testClassified = testSet.classify(classifier);

// Calcular a matriz de confusão
var confusionMatrix = ee.ConfusionMatrix(testClassified
  .errorMatrix({
    actual: 'Class', // Nome da propriedade de classe real no conjunto de teste
    predicted: 'classification' // Nome da propriedade de classificação no resultado da classificação
  }));

// Calcular as métricas de produtor e consumidor
var producerAccuracy = confusionMatrix.producersAccuracy(); // Produtor Accuracy
var consumerAccuracy = confusionMatrix.consumersAccuracy(); // Consumer Accuracy

print('Confusion Matrix:', confusionMatrix);
print('Overall Accuracy:', confusionMatrix.accuracy());
print('Producer Accuracy:', producerAccuracy);
print('Consumer Accuracy:', consumerAccuracy);

🎈 Usage

The GEE App can be used to identify flood areas, with remote sensing images classified before and after the event.

⛏️ Built Using

✍️ Authors

🎉 Acknowledgements