Apppublic

kavinda0126/coral-reef-bleaching-prediction

sourceHugging Faceupdated 6mo agoView on Hugging Face

0likes

App README

🪸 Coral Reef Bleaching Risk Predictor

A machine learning web application that predicts coral reef bleaching risk using four trained classification models. Built as part of an ML assignment at SLIIT.

Overview

Coral bleaching occurs when ocean temperatures rise above normal levels, causing corals to expel their symbiotic algae and turn white. This app takes real-world oceanographic measurements as input and predicts whether a reef site is at risk of bleaching.

Dataset: Global Coral Bleaching Database 1980–2020 (BCO-DMO)

Models

Model	Test Accuracy	F1 Score	ROC-AUC	Scaling
Logistic Regression	67.39%	0.6835	0.7338	Yes
Random Forest	90.45%	0.9080	0.9704	No
XGBoost	90.21%	0.9049	0.9683	No
SVM (RBF Kernel)	87.89%	0.8817	0.9285	Yes

All four models vote on the final prediction. The consensus is determined by the number of models predicting bleaching:

—4/4 votes → CRITICAL
—3/4 votes → HIGH RISK
—2/4 votes → MODERATE
—1/4 votes → LOW RISK
—0/4 votes → SAFE

Input Features

Feature	Description
`ClimSST`	Climatological Sea Surface Temperature (°C)
`Temperature_Mean`	Mean SST (°C)
`Temperature_Minimum`	Minimum SST (°C)
`Temperature_Maximum`	Maximum SST (°C)
`SSTA`	Sea Surface Temperature Anomaly (°C)
`SSTA_DHW`	SST Anomaly Degree Heating Weeks
`TSA`	Thermal Stress Anomaly (°C)
`TSA_DHW`	Thermal Stress Anomaly DHW
`TSA_DHW_Frequency`	Frequency of TSA DHW events
`Windspeed`	Wind speed (m/s)
`SSTA_Frequency`	Frequency of positive SSTA
`SSTA_Frequency_Standard_Deviation`	Std dev of SSTA frequency
`Turbidity_ct`	Turbidity count
`Turbidity`	Water turbidity
`Cyclone_Frequency`	Frequency of cyclone events
`Distance`	Distance to nearest land (km)
`Depth`	Reef depth (m)
`Latitude_Degrees`	Latitude
`Longitude_Degrees`	Longitude
`Date_Year`	Year of observation

Key bleaching risk indicators:

—TSA_DHW > 8 → severe bleaching expected
—Temperature_Mean > 30°C + SSTA > 1.5 → high risk
—Low turbidity + high DHW → maximum risk

App Features

Single Prediction

Adjust ocean condition sliders to get an instant prediction from all 4 models with a probability bar chart and consensus gauge.

Sample Reef Sites

Load pre-configured data from 10 real-world reef locations including the Great Barrier Reef, Maldives, Red Sea, and Caribbean.

Batch Prediction

Run all 10 sample sites through all 4 models simultaneously and view results as a probability heatmap.

Model Info

View loaded model performance metrics (accuracy, F1, ROC-AUC) and feature counts.

Project Structure

├── app.py                    # Gradio web application
├── requirements.txt          # Python dependencies
├── models/
│   ├── lr/                   # Logistic Regression artifacts
│   │   ├── lr_model.pkl
│   │   ├── lr_scaler.pkl
│   │   ├── lr_features.pkl
│   │   └── lr_metadata.json
│   ├── rf/                   # Random Forest artifacts
│   ├── xgb/                  # XGBoost artifacts
│   └── svm/                  # SVM artifacts
└── notebooks/                # Training notebooks

Local Setup

bash

pip install -r requirements.txt
python app.py

Tech Stack

—ML: scikit-learn, XGBoost
—UI: Gradio
—Data: pandas, numpy
—Visualization: matplotlib