Artificial Intelligence (AI) plays an increasingly significant role in promoting environmental sustainability by processing large volumes of satellite images for real-time monitoring of the territory and utilizing Machine Learning (ML) to model non-linear relationships between data. This article presents four innovative projects demonstrating AI's importance in achieving environmental sustainability goals. In collaboration with the startup Latitudo 40, two tools have been developed. The first tool supports sustainable land planning by monitoring land use and built-up areas. The second tool provides accurate monitoring of carbon sequestration by green infrastructure, which is essential for balancing industrial emissions. A third project, in collaboration with Eni S.p.A, involves the development of an ML-based platform for the valorization of waste biomass in the production of biofuels. The platform suggests optimal pathways for converting biomass into biofuels, promoting more sustainable energy sources and optimizing environmental resources. The fourth project concerns building an AI system using intelligent cameras to detect fires even at great distances and recognize acts of waste spilling. This project aims to prevent and mitigate the environmental impacts of these events. These four projects showcase AI's potential to promote environmental sustainability and address global challenges.
Artificial Intelligence (AI) can be crucial in promoting environmental sustainability by proposing innovative solutions to address global challenges and supporting responsible management of natural resources. This article aims to present the work in AI applied to environmental sustainability, highlighting four innovative projects developed in collaboration with prominent partners.
The article is structured into four sections. The first section discusses the work carried out in the field of AI for sustainable land planning through the generation of impervious maps. The second section analyzes the AI project aimed at supporting the monitoring of carbon storage by plants. The third section explores the ongoing development work for an AI-based decision support system for selecting the process of valorization of waste biomass for the production of biofuels. Eni S.p.A. partially supports the latter project, while Latitudo-40 supports the former two.
The fourth project concerns building an AI system using intelligent cameras to detect fires even at great distances and recognize acts of waste spilling. This project aims to prevent and mitigate the environmental impacts of these events.
Monitoring the evolution of impervious surfaces can help bon absorption on a global scale, thereby facilitating the plan urban development more sustainably and resiliently, control of losses due to natural disasters or human activsuch as promoting the use of permeable materials for ities. Fig. 2 illustrates the carbon absorption before and pavements or planning green areas and public parks in after the wildfire in the WWF nature reserve in Naples dense urban areas. in 2017. End-users can access a tool that generates an
In [
In collaboration with Eni S.p.A., a project is underway to develop a platform aimed at valorizing waste biomass in the production of biofuels. The main objective is to create a machine learning-based platform to identify the most suitable process for obtaining high biofuel yields and the desired chemical properties from waste biomass. A decision support system based on machine learning models and data-driven approaches is being developed, rather than physically based models. This choice is motivated by the complexity of biomass conversion processes, which are often dificult to describe in detail using equations.
At the same time, artificial intelligence is particularly effective in capturing nonlinear relationships directly from data.
The planned platform will be able to analyze waste biomass from the agro-industrial chain and suggest the most suitable one among the possible transformation processes for producing a biofuel with desired properties in terms of yield and chemical composition. The data used to train the decision support system will be collected from various chemical processes in scientific literature.
A Natural Language Processing pipeline has been set up to retrieve relevant articles related to a specific process of interest and automatically extract the necessary data to support the data collection process. The platform will implement the principles of industrial symbiosis, intended as a production system that favours and optimizes the exchange and sharing of material resources or energy flows between diferent production chains. The results of the machine learning models system will be validated with the support of expert chemical engineers in bio-refining.
By applying artificial intelligence and computer vision techniques, we can assess the amounts of CO2 absorbed by forest areas globally. These goals are achieved by utilizing multispectral images from the Sentinel-2 mission of the Copernicus program and surface biomass data from ESA’s Climate Change Initiative Biomass project. This approach is vital for the United Nations REDD+ program, which requires estimating carbon sinks for each country. It also aligns with the United Nations Sustainable Development Goal 15, which promotes the protection of terrestrial ecosystems and sustainable forest management.
The project detailed in [
of interest by sending one or more detection notifications to appropriate management centres. The currently available systems described in the literature are based AI applied to image interpretation can play a fundamen- on methodologies that still need to be fully developed, tal role in safeguarding and monitoring an environment. and their performance limits their difusion on a large Thanks to the development and scientific evolution of scale. They can be improved above all considering their deep learning, some problems that have always been con- sensitivity, i.e. their ability to detect fires even at great sidered unsolvable can be tackled with adequate deep distances and recognize acts of waste spilling without architecture. In this case, methods for the real-time de- being confused by the behavioural dynamics that can be tection of events that can cause significant damage to detected. agriculture and the environment, such as forest and crop Therefore, the project proposal arises from these needs ifres and waste spills in unpredictable areas that create just described and aims to design and implement an inharmful leachates for the surrounding crops. The Ital- novative deep system capable of bridging the limits deian state forestry corps declares that from 1970 to today, scribed by limiting the occurrences of false positives, 12% of the forests have been destroyed by about 5,000 which represent a further limit to the large-scale difuifres a year. The Anti-Mafia Investigation Directorate re- sion of these systems. The aim of the research is the ports that 14,000 tons of waste have been spilt, estimating design and testing of deep neural networks for the detecabout 8,000 eco-crimes. tion of fires and illegal waste spills, also re-identifying
Therefore, a possible solution would be to cover the the perpetrator of this crime (classifying the colour of territory with intelligent cameras (self-suficient from the clothes, identifying the gender and verifying the presa computational point of view or at most accompanied ence of the bag and hat). These networks must overcome by a small-sized embedded system) with video-analysis the limitations present in the literature, i.e. test on small algorithms on board capable of detecting such events datasets that are not representative of the real conditions and decrease in "in-the-wild" performance. The research includes the definition of innovative architectures of neural networks, the definition of training methodologies and experimental validation procedures in real contexts, and comparing them with existing situations. Optimizing the system for real-time execution of onboard cameras with limited computing resources will also be addressed.