C2B2 Symphony Raster Analytics

A web application and Python toolkit for marine spatial data analytics, built for the Mistra C2B2 Hackathon. This project simplifies the analysis of geospatial raster data (GeoTIFFs) from the Symphony ecosystem model.

Our tool allows users to perform two key analyses:

• Compare Layers: Directly measure the statistical relationship between any two layers. For example, you can quantify how shipping noise correlates with porpoise density.

• Predict Ecological Impact: Use machine learning to predict an ecosystem layer from multiple human pressure and environmental drivers. For instance, you can model fish populations based on factors like water temperature, nutrient levels, and fishing intensity.

These powerful tools are accessible through both a simple, user-friendly web interface and advanced command-line scripts.

Features

• Web Application: A user-friendly web interface to run "Compare" and "Predict" analyses without using the command line.
• symph-compare: A command-line tool to compute pairwise Pearson/Spearman correlations between any two raster layers, generating an interactive scatter plot and a correlation heatmap.
• symph-predict: A command-line tool to predict a target raster from multiple predictor layers using a suite of high-performance models (Random Forest, GBM, XGBoost, Neural Networks).

Screenshots

Model Performance Metrics

Comprehensive model evaluation metrics including R², RMSE, and MAE for multiple algorithms

Parity Plot Analysis

Predicted vs. actual values visualization showing model accuracy across the full range of data

Correlation Heatmap & Distribution Analysis

Feature correlation matrix and target variable distribution for understanding data relationships

Installation

1. Clone the repository

git clone https://github.com/SuyashMullick/layer-correlation-symphony
cd layer-correlation-symphony

2. Set up a virtual environment

python -m venv .venv
source .venv/bin/activate   # On macOS/Linux
# OR
.venv\Scripts\activate      # On Windows

3. Install dependencies

pip install -r requirements.txt

Data Setup

The web application is pre-loaded with sample data. To use your own data, follow these steps.

1. Download datasets

Download the 2018 Symphony ecosystem and pressure layers from the Swedish Agency for Marine and Water Management:

• Bottniska viken
• Västerhavet
• Östersjön

Unzip these files into data/aligned/ (or any other folder). The web application will automatically find and list all .tif files located in the data/aligned directory.

💡 Note: The scripts assume all input rasters are aligned to the same grid (CRS, transform, and dimensions). You may need to pre-process your data using GIS software like QGIS or GDAL.

Usage

🌐 Web Application UI

This is the simplest way to run analyses. The web application provides a graphical user interface for the compare and predict tools.

1. Start the Web Server

From the root directory of the project, run:

python app.py

2. Open the Application

Open your web browser and navigate to http://127.0.0.1:5000.

3. Navigating the Application

The Home Page

The main page provides two choices: "Compare Layers" for simple two-layer correlation, and "Predict Ecological Impact" for more complex predictive modeling.

The Compare Page

This page is for a direct, one-to-one comparison between two raster layers. You can either select pre-loaded files from the dropdown menus or upload your own GeoTIFF files. Once you have selected two layers, click the "Compare Layers" button to start the analysis.

The Predict Page

This page allows you to predict a target variable using multiple predictor (or "driver") layers. Upload one target file and one or more predictor files. You can add more predictor file slots by clicking the "+ Add Another Predictor" button. Once your files are selected, click "Run Prediction" to train the models and generate results.

Viewing Results

After submitting a job from either the Compare or Predict page, a loading screen will appear. Once the analysis is complete, you will be redirected to the results page, which displays key metrics in a table and shows detailed, interactive diagnostic plots.

🔹 Command-Line Interface (Advanced)

For batch processing or more advanced control, you can use the command-line scripts directly.

1. Compare Two Layers

Compare two raster layers pixel-wise to measure linear (Pearson) and rank (Spearman) correlations.

Usage

python -m cli.symph-compare 
  --a <path_to_layer_a.tif> 
  --b <path_to_layer_b.tif> 
  --out <output_folder> 
  [--nodata <nodata_values>]

Example

python -m cli.symph-compare
  --a data/aligned/naturvarden-.och-belastningar-ostersjon-2018/National_eco_E_2018/01Porpoise_Baltic.tif
  --b data/aligned/naturvarden-.och-belastningar-ostersjon-2018/National_press_E_2018/32Nitrogen_Background.tif
  --out out/compare/porpoise_vs_nitrogen
  --nodata "-9999"

Outputs

out/compare/porpoise_vs_nitrogen/
 ├─ metrics.csv                # Pearson r, Spearman rho, and pixel count
 ├─ metrics.json               # Metrics in JSON format
 ├─ scatter.html               # Interactive scatter plot
 └─ predictor_heatmap.png      # Correlation heatmap image

💡 For a full list of options, run python -m cli.symph-compare -h

2. Predict Target from Multiple Layers

Train a model to predict one target raster from several predictor rasters.

Usage

python -m cli.symph-predict 
  --target <target.tif> 
  --predictors <x1.tif> <x2.tif> ... 
  --out <output_dir> 
  [--sample <n>] [--test_size <ratio>] 
  [--model {rf,gbm,xgb,nn,auto}] 
  [--remove_outliers] [--transform_y {none,log1p}]

Example

export BASE="data/aligned/naturvarden-och-belastningar-bottniska-viken-2018"  # On macOS/Linux
# OR
$BASE = "data/aligned/naturvarden-och-belastningar-bottniska-viken-2018"      # On Windows

python -m cli.symph-predict
  --target "$BASE/National_eco_N_2018/01Porpoise_Baltic.tif"
  --predictors
    "$BASE/National_eco_N_2018/20sill_lognorm_v2.tif"
    "$BASE/National_eco_N_2018/21skarpsill_lognorm_v2.tif"
    "$BASE/National_eco_N_2018/19torsk_lognorm_v2.tif"
    "$BASE/National_press_N_2018/17Noise_2000Hz_Shipping_20181122.tif"
    "$BASE/National_press_N_2018/18Boating.tif"
  --out "out/predict/N_porpoise_prey_noise_boating"
  --sample 200000
  --transform_y log1p
  --model rf

Outputs

out/predict/<name>/
├── feature_importance.csv
├── metrics.csv
├── parity.html
├── parity.png
├── predictor_heatmap.png
├── residuals.html
└── residuals.png

💡 For a full list of options, run python -m cli.symph-predict -h

---

📁 Repository Structure

layer-correlation-symphony/
├─ app.py                       # Main Flask web application
├─ cli/
│  ├─ symph-compare.py         # CLI for correlation analysis
│  └─ symph-predict.py         # CLI for predictive mapping
├─ src/                         # Core analytics and visualization modules
│  ├─ analysis/
│  ├─ geo/
│  └─ viz/
├─ static/                      # CSS and generated result files
├─ templates/                   # HTML templates for the web app
├─ data/
│  └─ aligned/                 # Location for input GeoTIFF layers
├─ out/                         # Default folder for CLI outputs
├─ screenshots/                 # Application screenshots
└─ requirements.txt             # Dependencies

👥 Credits

Developed for the Mistra C2B2 Hackathon #1 (2025) by:

• Younus Mashoor - younusmashoor@gmail.com
• Suyash Mullick - suyash.mullick@gmail.com
• Hamza Zia - hamza.zia@lnu.se
• Abdullah Saeed - as228aw@student.lnu.se

Special Thanks To

• Amilcar Soares Junior
• Rafael Messias Martins

For their guidance and support throughout the project.

📜 License

MIT License — free to use, modify, and extend with attribution.