In November 2023, GeoVille launched the Dunia on behalf of the European Space Agency (ESA) as an innovative technology designed to streamline the processing and distribution of Earth Observation data across Africa. By overcoming infrastructure challenges, Dunia offers a cloud-based computational platform that facilitates the direct interpretation of Copernicus data. This platform boasts low bandwidth consumption, compatibility with handheld mobile devices, and allocated usage allowances for eligible users, making it accessible to a wide range of individuals, from novice developers to seasoned experts.

Dunia empowers users to harness its user-friendly data dissemination and robust processing capabilities to enhance Earth Observation services tailored specifically for African needs, thus contributing to the continent’s sustainable development goals. 

Introduction to the Platform 

Creating on the Dunia platform follows a structured approach, ensuring that users of all skill levels can easily navigate and utilise its features to develop projects. The workflow, categorised into “discovery, build, and exchange” phases, guides developers through introductory steps, access to free tier services, utilising dataset search and download APIs, navigating the Sandbox environment, and understanding the nuances of scale-out processing across Africa.

Moreover, Dunia emphasises collaboration among developers to minimise redundancy and optimise problem-solving. Developers can seamlessly exchange insights with colleagues and clients through secure data sharing and real-time collaboration features, regardless of their location. The platform’s infrastructure supports various tools and processing capabilities, facilitating the sharing of scripts, products, and outputs through a dedicated marketplace. Furthermore, Dunia’s optimisation for bandwidth and mobile devices ensures that users can work efficiently on handheld devices, providing a seamless and immersive experience.

Features of the Dunia Platform

1. Datasets: The Dunia platform hosts a meticulously curated selection of remote sensing datasets tailored to provide valuable insights into Africa, catering to various project requirements. Users can explore datasets such as Copernicus Sentinel 1, 2, and 3, Copernicus Sentinel 5P, Landsat 5, 7, and 8, Copernicus DEM at 30m resolution, Envisat, which offers comprehensive data on land, oceans, ice, and the atmosphere, Soil Moisture and Ocean Salinity (SMOS), Global Administrative Boundaries, and the Global Database of Political Administrative Boundaries. These datasets are designed to meet diverse user needs and serve as a solid project foundation.
2. Wiki: The Dunia platform features a designated information hub labelled “Wiki,” the go-to resource for users to delve into FAQs, troubleshoot issues, and gain comprehensive knowledge about the platform. This repository hosts tutorials, quick-start guides, and reference materials covering all EO Africa Dunia service aspects, ensuring users can effortlessly initiate their journey without encountering any hurdles.
3. The Dunia Application Hub: The hub is a user-friendly environment accessible directly from your web browser. It consolidates the processing capabilities of EO toolboxes such as Sentinel Application Platform (SNAP), EnMAP-Box, GDAL, and Orfeo Toolbox, along with free EO data. Each user has a dedicated private workspace within this hub, enabling seamless access to processing tools and data resources.
4. The Map Browser: This tool on the Dunia platform provides users with an intuitive interface to explore and analyse geospatial data effortlessly. Integrated seamlessly with the datasets, this tool allows users to visualise various datasets and manipulate and analyse data layers directly within the map browser using the Application Hub. With its user-friendly design and integration with other platform features, the map browser facilitates streamlined exploration and interpretation of Earth observation data for diverse projects and research endeavours.
5. Dunia Sandbox: Creating projects with the Dunia Sandbox offers developers a convenient and accessible platform for research, development, and deploying services. Accessed directly from web browsers, the sandbox environment provides a range of free Earth Observation (EO) datasets and tools, empowering users to create customised solutions within their dedicated virtual space.

Users can follow this step-by-step guide to begin utilising the Dunia Sandbox. Successfully launching the sandbox initiates a user-defined virtual desktop environment allocated upon registration. Within this environment, developers can explore, experiment, and develop applications tailored to their needs, leveraging the available EO datasets and tools.

NOTE: This virtual desktop can launch specific tools, including GIS software like QGIS. Additionally, users can install Python libraries like tensorflow, open-cv, and so on that may not be pre-installed in the virtual environment. If a user believes a newly installed library could benefit other developers, they should contact the Innovation Lab Support Team. This will allow them to assess the potential benefits and make the library available for other users. Otherwise, users can install libraries within their environment for their use.

Exploring a Sample Project: Deep Learning for Advanced Image Classification in Earth Observation Over Africa 

This project aimed to leverage deep learning techniques for advanced image classification in Earth observation data over Africa. The implementation used Jupyter Notebook as the development environment and TensorFlow as the primary deep learning framework. The dataset utilised for training and validation was sourced from EuroSAT, a satellite image classification dataset featuring different land use and land cover classes.

However, a step-by-step guide to visualise the process for intending users has been developed to improve the understanding of the Dunia Sandbox and how to create projects using the platform. The workflow to creating a project involves the following key steps:

Step 1: Creating a New Notebook 

For this project, a new notebook was created within the virtual environment, and TensorFlow was selected as the software. TensorFlow is an open-source software library for machine learning and artificial intelligence tasks. While it offers versatility across various tasks, its primary focus lies in the training and inference of deep neural networks. To install TensorFlow into a virtual environment, you can simply execute the “pip install tensorflow“ command and run the programme. Additionally, all other essential libraries required for this task have been imported, and the results are displayed below.

The version of TensorFlow used in this project was determined by executing the command “tf.__version__” while the necessary variable was established utilising the dataset sourced from EuroSAT. This dataset, known for its diverse satellite images covering land use and land cover across Africa, provided the foundational data for the variable in question. Additionally, the utilisation of Jupyter Notebook underscores its significance as a popular environment for interactive computing, particularly in machine learning where TensorFlow, denoted by “tf” in the command, holds substantial importance. By combining these tools and resources, researchers and practitioners gain access to a robust framework for developing and experimenting with various machine learning models tailored to geospatial analysis and classification tasks.

Step 2: Data Collection and Preprocessing 

The EuroSAT dataset, containing thousands of satellite images across different African countries, was obtained and preprocessed to classify African landscapes. This preprocessing likely involved resizing, normalisation, and augmentation techniques to enhance the robustness of the model. The data was prepared for the model, and the input datasets were visualised with the command, as shown below.

Step 3: Training and Validation 

The model was trained using a portion of the preprocessed dataset while keeping aside another portion for validation. Training likely involved optimising the model parameters (such as learning rate, batch size, etc.) to minimise the loss function and improve classification accuracy. See the visualisation of the trained model below. 

Step 4: Evaluation and Result Analysis

After training, the model’s performance was evaluated using various metrics, including accuracy, precision, recall, and F1 score. A confusion matrix was also generated to visualise the model’s performance across different classes and identify misclassifications. 

The image classification results obtained from the model were thoroughly analysed, possibly focusing on areas of high accuracy and instances where the model struggled to classify images correctly. Insights gleaned from this analysis could inform further iterations of the model or provide valuable information for stakeholders involved in Earth observation and land use planning. The project’s result is shown below after training, analysing, and evaluating the dataset. 

Overall, this project demonstrated the efficacy of deep learning techniques, specifically TensorFlow-based CNNs, for advanced image classification tasks in Earth observation data over Africa. By leveraging state-of-the-art methodologies and datasets within the Dunia Sandbox, the project contributed to advancing remote sensing applications and provided valuable insights into classifying diverse landscapes across the continent.

Join the mission to combat Africa’s challenges using the power of Earth Observation (EO) data. From safeguarding the environment to managing disasters, the potential applications are limitless. With Dunia, you gain access to the tools and resources, including invaluable data from the Copernicus Sentinels, to build tailored solutions that address the specific needs of African communities. Explore the Dunia platform today and create innovative solutions to drive positive change in Africa and globally. Click here to start your journey on the Dunia platform.