Yeşil Görünüm İndeksi (Green View Index)
- Customer
- Istanbul Metropolitan Municipality
- Project manager on the customer side
-
- Year of project completion
- 2024
- Project timeline
- January, 2024 - June, 2024
- Project scope
- 1200 man-hours
- Goals
-
The primary objective of the project is to measure urban greenery experienced by citizens in their daily lives using a scientific, objective, and human-centered metric, going beyond traditional methods. It aims not only to quantify green areas critical for human health and ecological balance but also to evaluate their accessibility and perceived quality.
In line with this goal, the project seeks to analyze street-level greenery (trees, shrubs, etc.) that cannot be fully captured through satellite imagery and to generate a Green View Index (GVI) score for each geographic point, ensuring repeatability and comparability. Using these data, the project seeks to develop dynamic, spatially detailed maps that depict the city’s green structure at neighborhood and street scales, enabling the assessment of urban green distribution throughout Istanbul. The ultimate goal is to provide local government units operating in fields such as urban planning, landscape architecture, and public health with a strategic data infrastructure to guide the prioritization of new green space investments. Additionally, the project aims to use this dataset as a “starting point” for the periodic monitoring of future urban interventions and the impacts of climate change on urban greenery. - Project Results
-
The project has yielded multifaceted outcomes, delivering both significant outputs and strategic benefits.
Significant outputs include a comprehensive geodatabase containing GVI scores calculated for millions of points across the city’s road network; an interactive digital city map that allows GVI data to be examined at the neighborhood and street level; and comparative analytical reports illustrating the distribution of greenery across districts and neighborhoods.The strategic benefits of these outputs are significant for institutional decision-making processes. Municipal planners can now answer the question, “Where should we plant trees?” using objective data that prioritize areas with the lowest GVI scores. This enables urban greening budgets to be allocated efficiently, targeting locations with the greatest need and potential impact, thereby optimizing the use of public resources.Thanks to its robust analytical infrastructure, the project is positioned as a low-cost, high-impact smart city application that can inspire similar initiatives both nationally and internationally.
The uniqueness of the project
The project’s uniqueness stems from its emphasis on “human-scale” and “experienced greenery,” rather than relying on traditional satellite images that assess urban green spaces from a “bird’s-eye” view. By using 360° panoramic street images, the project directly measures how much greenery an individual sees at eye level, representing a shift from the notion of the ‘’presence of green spaces’’ to that of ‘’visual exposure’’. This approach provides a more realistic depiction of urban life quality. Another uniqueness of the project is its cost-effective and sustainable methodology. It was developed entirely using open-source AI models and Python-based libraries, without the need for commercial software or consultancy services. By applying existing pre-trained models, high-accuracy results are achieved without incurring significant model training costs, making the project both highly economical and transparent. Furthermore, the methodology encompasses not only large parks but also all street greenery, providing a comprehensive and inclusive measure of urban greenery. The underlying technology and methodology are universal and can be easily adapted to any city with similar street-level panoramic imagery.- Used software
- The project was implemented entirely using open-source technologies and internal resources to minimize external dependency and ensure cost-effectiveness. The primary data source was Istanbul Metropolitan Municipality’s (IMM) high-resolution, geo-tagged 360° panoramic street-level imagery archive, which was processed with the support of GPUs. The entire development process was carried out using Python. During this process, open-source deep learning frameworks such as PyTorch and TensorFlow were used; OpenCV was used for image processing; and Python libraries like NumPy and Pandas were employed for analysis and spatial operations. Pre-trained, field-proven open-source semantic segmentation models (e.g., DeepLab, UNet), trained on datasets such as Cityscapes, were applied directly, removing the need for commercial software solutions.
- Difficulty of implementation
- The main challenge in implementing the project was processing Istanbul’s existing panoramic image data (Big Data) to generate the GVI. The analysis of this massive dataset required the establishment of optimized data processing pipelines to efficiently feed the data into the available GPU resources. Another significant technical challenge was selecting the most suitable model from the numerous open-source models, capable of responding with the highest accuracy to Istanbul’s complex urban landscape including narrow streets, dense shadows, and seasonal variations, given that training a model from scratch was not an option. This model selection was critical to minimize potential errors, such as the misclassification of non-vegetative objects (e.g., green awnings). Once the most accurate model was determined, all data production was standardized through this single model. To complete the analysis of the entire city within a reasonable timeframe and within the limits of the available in-source GPU resources, tasks had to be highly parallelized for maximum efficiency. Developing the project entirely with internal resources was challenging but resulted in a sustainable knowledge base and increased technical capability within the organization.
- Project Description
-
This project is an innovative, data-driven analytical system developed to understand the impact of urban environments on humans and to design more livable cities. The project measures not only the physical area of green spaces in a city but also how much residents "see" and "experience" these spaces in their daily lives.
The project’s workflow consists of four main steps. First, millions of high-resolution, geotagged 360° street-level panoramic images covering the entirety of Istanbul were prepared for analysis.Second, these images were processed using pre-trained open-source deep learning models. Without requiring retraining, these models classified each pixel in the images into categories such as vegetation (trees, shrubs, grass), road, and building.In the third step, the GVI score for each geographic point was generated by calculating the ratio of pixels classified as “vegetation” to the total number of pixels in each image based on the segmentation results.In the final step, the resulting millions of point-based GVI scores were transformed into a spatial database using geographic data processing libraries in Python, resulting in the creation of Istanbul’s “Green View Index Map.”
- Project geography
- The current geographical scope of the project encompasses the entire provincial boundaries under the jurisdiction of IMM. The analysis includes IMM’s panoramic imagery and the entire asphalt road network accessible by collection vehicles, covering main arteries, avenues, boulevards, and the majority of streets. This level of comprehensive coverage of publicly experienced urban spaces is pioneering in the field. The project’s vision is not limited to Istanbul. The underlying methodology and technology provide a framework that is scalable globally. The developed model and workflows can be readily transferred to other metropolitan municipalities in Türkiye, potentially leading to the creation of a national “Green View Index Atlas”. The open-source philosophy further provides a standart that can be adopted by any city worldwide with similar panoramic imagery data.
- Additional presentations:
- GREEN VIEW INDEX.png