Computational sustainability

Computational sustainability is an emerging field that attempts to balance societal, economic, and environmental resources for the future well-being of humanity using methods from mathematics, computer science, and information science fields. Sustainability in this context refers to the world's ability to sustain biological, social, and environmental systems in the long term. Using the power of computers to process large quantities of information, decision making algorithms allocate resources based on real-time information. Applications advanced by this field are widespread across various areas. For example, artificial intelligence and machine learning techniques are created to promote long-term biodiversity conservation and species protection. Smart grids implement renewable resources and storage capabilities to control the production and expenditure of energy. Intelligent transportation system technologies can analyze road conditions and relay information to drivers so they can make smarter, more environmentally-beneficial decisions based on real-time traffic information.

History and motivations
The field of computational sustainability has been motivated by Our Common Future, a 1987 report from the World Commission on Environment and Development about the future of humanity. More recently, computational sustainability research has also been driven by the United Nation's sustainable development goals, a set of 17 goals for the sustainability of human economic, social, and environmental well-being world-wide. Researchers in computational sustainability have primarily focused on addressing problems in areas related to the environment (e.g., biodiversity conservation), sustainable energy infrastructure and natural resources, and societal aspects (e.g., global hunger crises). The computational aspects of computational sustainability leverage techniques from mathematics and computer science, in the areas of artificial intelligence, machine learning, algorithms, game theory, mechanism design, information science, optimization (including combinatorial optimization), dynamical systems, and multi-agent systems.

While the formal emergence of computational sustainability is often traced back to the years 2008 and 2009, marked by the initiation of an NSF-funded award, and specific conferences and workshops, the exploration of computational methods to tackle environmental and societal sustainability issues predates this period. The use of statistical and mathematical models for sustainability-related problems has a long history, paralleling the evolution of computing technology itself. A notable example is the early attempts at climate modeling, which were constrained by the limited computing resources available at the time, necessitating simplified models.

In the realm of artificial intelligence, particularly within machine learning, the 1990s saw research efforts addressing ecological modeling and wastewater management, among other sustainability issues. This work continued into the 2000s, supported by groups like the "Machine Learning for the Environment" working group established by the National Center for Ecological Analysis and Synthesis in 2006. Research on optimization to aid sustainability challenges, such as designing wildlife reserves, can be traced back to the 1980s.

The early 2000s also witnessed a growing concern over the environmental impact of computing technology itself, with green information and communications technology (ICT) gaining attention among ICT companies. This interest extended beyond the immediate environmental effects of computing to consider second-order and higher-order impacts, such as the potential of ICT to reduce the carbon footprint of air travel through online conferencing or to optimize delivery routes to lower CO2 emissions. International policy efforts, particularly by the Organization for Economic Cooperation and Development (OECD), have since focused on a framework recognizing these multi-tiered effects of ICT, a focus that continues today.

Before the OECD's 2008 conference, mathematicians proposed using their expertise to combat climate change, signaling a growing recognition of the research community's role in sustainability. This period also saw the establishment of the Institute for Computational Sustainability in 2008 and the launch of the International Conference on Computational Sustainability in 2009, pivotal moments that significantly advanced the field. The inclusion of sustainability themes in major AI conferences further integrated sustainability into the broader computing and scientific discourse.

The field of computational sustainability has continued to expand, with significant initiatives like the Sustainability-focused Expeditions in Computing award to the University of Minnesota in 2010, aiming to advance climate understanding through data mining and visualization. The establishment of sustainability-related tracks and awards at various conferences, along with targeted funding by organizations like the NSF, underscores the growing importance of computing in addressing sustainability challenges.

Biodiversity and conservation
Biodiversity conservation focuses primarily on preserving the diversity of species, sustainable utilization of species and ecosystems, and maintaining life-supporting systems and essential ecological processes. Conservation of species is an important sustainability goal to prevent biodiversity loss. As urbanization is expanding across the globe, it threatens wildlife in and around cities. An effort towards conservation has included the creation of wildlife corridors that are used to connect wildlife populations that have become isolated from man-made habitat fragmentation. Building these wildlife corridors is a challenge due to barriers between habitats and property owners (Zellmer, Goto). Moving species to connect core conservation areas through corridors results in an optimization problem. This is where technology can help, not only in optimizing corridors but in terms of helping with cost-benefit analysis. Moreover, artificial intelligence serves as a tool in the ongoing battle against biodiversity loss and illegal activities such as poaching. In recent years there has been significant research on wildlife monitoring strategies to better understand patterns and enhance security to combat poaching (Gomes). This integration of AI into wildlife conservation efforts represents a significant step forward in collective efforts to safeguard and protect natural ecosystems.

United Nations' Sustainability Development Goals
The United Nations lists seventeen different Sustainable Development Goals (SDGs) to protect the planet, all of which are important in different ways. Sustainable Development Goal 14 emphasizes protecting life under water. Sustainable Development Goal 15 references protecting life on land. While technology has historically favored profitable sectors, its potential to revolutionize environmental sustainability, particularly in wildlife conservation, remains largely untapped. By examining challenges, contributions, and potential contributions presented by technological advancements in achieving Sustainable Development Goals 14 and 15, computational innovations can be harnessed to protect life under water and on land.

The application of machine learning techniques to address challenges in fire prediction and management in Alaska's boreal forests. Studies have underscored the importance of adapting existing fire management strategies to the evolving fire landscape, especially considering the impact of climate change on fire frequencies. By incorporating diverse variables such as topography, vegetation, and meteorological factors, the research aligns with the computational sustainability paradigm, which seeks to leverage computational models for sustainable environmental practices.

There is one novel machine learning framework for fire prediction, which represents a significant contribution to computational sustainability in the field of environmental monitoring. The model, centered on the identification of specific ignitions likely to lead to large fires, provides a more straightforward and interpretable alternative to existing, more complex prediction models. The emphasis on two key variables, vapour pressure deficit (VPD) and spruce fraction, reflects the paper's commitment to practical and actionable computational approaches in environmental assessment. The assessment of how active fire management influences fire regimes highlights the role of human intervention in shaping environmental outcomes, illustrating the potential of computational sustainability for informed decision-making in environmental monitoring and assessment.

Species distribution modeling
Computational sustainability researchers have advanced techniques to combat the biodiversity loss facing the world during the current sixth extinction. Researchers have created computational methods for geospatially mapping the distribution, migration patterns, and wildlife corridors of species, which enable scientists to quantify conservation efforts and recommend effective policies.

Renewable and sustainable energy and materials
Using "affordable and clean energy" is one of the seventeen Sustainable Development Goals (SDG) worldwide (Gomes et al., 2019).

The Sun, as the only planet-hosting star in the solar system, can provide clean and renewable energy to meet the demands of increasing populations. Unlike fossil fuels, the Sun does not generate pollutants or greenhouse gases. Therefore, relying on solar energy can reduce carbon footprints, which abates global warming and ecosystem destruction. The Sun will live for approximately 5 billion years more, which serves as a long-term and stable energy source. If humans can extract and convert energy efficiently, both the environment and the economy can benefit, contributing to sustainability.

However, renewable energy, including wind and solar energy, is non-dispatchable. Humans cannot control these energy sources or predict energy production in advance. If using renewable energies, scientists need to seek different sources for compensation, which usually links back to fossil fuels that are considered unsustainable. Alternatively, people can store the energy from these renewable sources for the difference, which can be expensive.

Scientists should consider different factors when designing the storage system, including frequency regulation, energy shifting, peak shifting, and backup power (Gomes et al., 2019). Deciding whether to utilize diverse energy sources or store energy to prepare for unexpected situations will be hard for scientists. The approaches for each strategy are complicated. Scientists have turned this scenario into an optimization problem that involves the three "broad sustainability themes"—simulation, machine learning, and citizen science (Gomes et al., 2019).

Climate change and renewable energy interrelate with each other. Renewable energy sources such as the Sun and wind highly depend on the climate. On a cloudy day, people will acquire less solar energy due to shielding. The UV index will also impact solar energy production. Reversely, using renewable energy on large scales benefits the environment, reducing global warming and extreme weather. Therefore, constructing an accurate climate model and predicting the weather for renewable energy production becomes essential.

In Jones' article, he explores the usage of artificial intelligence (AI) in simulating the climate (2018). The major problems of using computers for climate modeling arise from a lack of details and slow simulation. Different computing approaches can also result in different and inaccurate results. For example, one model predicts that the temperature can increase more than three times than the other model if the carbon dioxide level doubles in the atmosphere. Therefore, scientists incorporate machine learning frameworks into the existing climate models. This combination enables the computer to efficiently discern more unnoticeable details than traditional computers, even with slight uncertainties and deviations, to give accurate simulations and predictions (Jones, 2018). Simultaneously, machine learning techniques, including normalizing flows, can infer long-term patterns and behaviors from data from a short period.

People can take advantage of running small-scale simulations that are more efficient for predictions, especially with characteristic models. For instance, the information on how clouds evolve in a few miles region over a short period will be sufficient for "Cloud Brain", a deep learning code to infer climate change due to increasing carbon dioxide emissions. Then, the framework can figure out the climate model on a large scale and over long periods. This model is more efficient than traditional high-resolution simulations yet gives similar and realistic results (Jones, 2018). Normalizing flows also performed similar functions as the "Cloud Brain". After inputting initial and final conditions into the neural network, the algorithm can figure out a chain of transformations. The given conditions generally come from a short period, while the chain can be universal for long-term scenarios to infer and predict.

However, developing these machine learning techniques to predict the physical world is still challenging. Machine learning functions "intuitively" and may not follow the rules in the world. When predicting and establishing the climate model, AI cannot consider different factors in physics, including gravity and temperature gradient, for efficiency. Lack of rules in the framework can lead to unrealistic results. These frameworks can be inflexible and do not adapt to a new and diverse environment. "Cloud Brain" cannot accurately predict when the temperature is high (Jones, 2018). Like the "black-box function" in SMART-Invest (Gomes et al., 2019), these machine-learning techniques obtain little transparency. People struggle to recognize and comprehend the models (Jones, 2018). In normalizing flows, learning the exact bijective transformations takes extra effort, and few packages have the functions to express each transformation explicitly. Some particular transformations can disobey the physics laws, but scientists have no way to identify and fix the issue. Therefore, training the model comprehensively with appropriate supervision of physics laws becomes necessary. However, heliophysics can be complex, and scientists are uncertain about the nuclear fusion process inside the Sun. In such cases, no physics equation is established to describe the energy conversion process, which affects the amount of solar energy humans can extract. Without a "rulebook", machine learning is the optimal approach to figure out the pattern and correlation (Jones, 2018). When implementing normalizing flows in solar energy and heliophysics, some degree of freedom needs to be allowed for the neural network to discover patterns in the unknown solar physics regimes.

Spatial planning
Spatial planning refers to the methods and approaches used by the public sector to influence the distribution of people and activities in spaces of various scales. It encompasses a broad spectrum of activities related to the use and management of land and public spaces, aiming to ensure sustainable development and to improve the built and natural environments.

Spatial planning covers a wide range of concerns including urban, suburban, and rural development, land use, transportation systems, infrastructure planning, and environmental protection. It aims to coordinate the various aspects of policy and regulation over land use, housing, public amenities, and transport infrastructure, ensuring that these elements work together to promote economic development, environmental sustainability, and quality of life for communities in all types of areas.

This term is often used in a European context and can be seen as an integrated approach that looks beyond traditional urban planning to address the needs and development strategies of a wider range of environments. It involves strategic decision-making to guide the future development and spatial organization of land use in a way that is efficient, sustainable, and equitable.

Transportation
Intelligent transportation systems (ITS) seek to improve safety and travel times while minimizing greenhouse gas emissions for all travelers, though focusing mainly on drivers. ITS has two systems: one for data collection/relaying, and another for data processing. Data collection can be achieved with video cameras over busy areas, sensors that detect various pieces from location of certain vehicles to infrastructure that is breaking down, and even drivers who notice an accident and use a mobile app, like Waze, to report its whereabouts.

Advanced public transportation systems (APTS) aim to make public transportation more efficient and convenient for its riders. Electronic payment methods allow users to add money to their smart cards at stations and online. APTS relay information to transit facilities about current vehicle locations to give riders expected wait times on screens at stations and directly to customers' smart phones Advanced Traffic Management Systems (ATMS) collect information using cameras and other sensors that gather information regarding how congested roads are. Ramp meters regulate the number of cars entering highways to limit backups. Traffic signals use algorithms to optimize travel times depending on the number of cars on the road. Electronic highway signs relay information regarding travel times, detours, and accidents that may affect drivers ability to reach their destination.

With the rise of consumer connectivity, less infrastructure is needed for these ITS to make informed decisions. Google Maps uses smartphone crowdsourcing to get information about real-time traffic conditions allowing motorists to make decisions based on toll roads, travel times, and overall distance traveled. Cars communicate with their manufacturers to remotely install software updates when new features are added or bugs are being patched. Tesla Motors even uses these updates to increase their cars efficiency and performance. These connections give ITS a means to accurately collect information and even relay that information to drivers with no other infrastructure needed.

Future ITS systems will aid in car communication with not just the infrastructure, but with other cars as well.

Utilities
The electrical grid was designed to send consumers electricity from electricity generators for a monthly fee based on usage. Homeowners are installing solar panels and large batteries to store the power created by these panels. A smart grid is being created to accommodate the new energy sources. Rather than just electricity being sent to a household to be consumed by the various appliances in the home, electricity can flow in either direction. Additional sensors along the grid will improve information collection and decreased downtime during power outages. These sensors can also relay information directly to consumers about how much energy they're using and what the costs will be.

Active information gathering
Another way that computational strategies are used is in active information gathering. The use of technology to measure tons of information and sort through them is a powerful tool in many fields of study. For example, NASA uses satellites to get SAR (synthetic aperture radar) data in order to map the surface of the earth. They are able to perform active data collection of visible, near-infrared, and short-wave-infrared portions of the electromagnetic spectrum using their satellites. These findings can help to identify deforestation and rising sea levels and help predict future changes to different ecosystems based on the wavelengths and polarization of the radar. NASA has made this data publicly available beginning with the European Space Agency's (ESA) Sentinel-1a in 2014.

Vision and learning
Computer vision and machine learning play a crucial role in advancing computational sustainability, offering innovative solutions to complex environmental challenges. By harnessing the power of these technologies, researchers and practitioners are able to analyze vast amounts of data, extract meaningful patterns, and develop sustainable strategies for managing natural resources and ecosystems.

Wildlife conservation
Computer vision is used to monitor and track endangered species, such as tracking the movements of animals in their natural habitats or identifying individual animals for population studies. For example, camera traps equipped with computer vision algorithms can automatically detect and identify species, allowing researchers to study their behaviors without disturbing them. Machine learning algorithms can analyze these data to understand animal behavior, habitat preferences, and population dynamics, aiding in conservation efforts. This is helpful in assessing the effectiveness of conservation measures and identify areas in need of protection.

Environmental monitoring
Remote sensing technologies combined with machine learning can monitor air and water quality, detecting pollutants and assessing environmental health. For example, satellite imagery can be used to monitor algal blooms in water bodies, which can be harmful to aquatic life and human health. Computer vision techniques can analyze satellite imagery to detect deforestation and illegal logging activities. By identifying areas at risk, conservationists and authorities can take action to protect forests and biodiversity.

Sustainable agriculture
Computer vision is used to monitor crop health, detecting diseases and nutrient deficiencies early. For example, drones equipped with multispectral cameras can capture images of crops, which are then analyzed using machine learning algorithms to identify health issues. Machine learning algorithms can analyze data from sensors and drones to optimize resource allocation in agriculture. By providing insights into soil health, moisture levels, and crop growth, these algorithms help farmers make informed decisions to improve productivity and sustainability.

Climate change mitigation
Machine learning models can analyze historical climate data to predict future climate patterns. This information is crucial for developing strategies to mitigate the impacts of climate change, such as planning for extreme weather events and sea level rise. Computer vision techniques can be used to monitor renewable energy sources, such as solar panels and wind turbines. By analyzing data on energy production and environmental conditions, these techniques help optimize the use of renewable energy and reduce reliance on fossil fuels.

Data-driven decision making
Computer vision and machine learning enable data-driven decision-making in sustainability efforts. By analyzing large datasets, researchers can identify trends, predict outcomes, and make informed choices to conserve natural resources and protect the environment.

Efficiency and accuracy
These technologies improve the efficiency and accuracy of environmental monitoring and management. They can process data faster and more accurately than traditional methods, enabling timely interventions to prevent environmental degradation.

Conservation impact
By enabling more precise monitoring and analysis, computer vision and machine learning enhance conservation efforts, helping to protect endangered species, preserve biodiversity, and mitigate the effects of climate change.

Sustainable development
Insights from computer vision and machine learning contribute to the development of sustainable practices in agriculture, forestry, and other industries. By optimizing resource use and minimizing environmental impact, these technologies support long-term sustainability.

Agent-based modeling
Agent-Based Modeling (ABM) has many applications across a variety of domains within computational sustainability. In wildlife conservation and ecosystem management, ABM simulates animal behaviors and interactions in ecosystems. This displays the impacts of habitat destruction or climate change on biodiversity. In sustainable agriculture, ABM assesses how farmers decide on crop selection, land use, and the adoption of sustainable practices. Urban planning benefits from ABM by traffic simulation and pedestrian patterns to potentially optimize public transportation systems, reduce carbon emissions, and improve urban life quality. These are some examples of ABM applications that provide a powerful tool for exploring sustainable solutions and help stakeholders make informed decisions for a sustainable future. NetLogo is one of the leading and most popular ABM softwares. It allows for researchers to design, develop, and implement complex ABMs in a manner that is still accessible for those without programming backgrounds. Due to this, it has widespread applicability and use in educational settings that can help students develop a broader understanding of sustainability issues. The base version of NetLogo comes with many sample models, this includes 7 models under the folder of Earth Science. These models tackle various sustainability issues ranging from the impact of carbon dioxide emissions on climate change to the percolation of oil through permeable soils during oil spills.

Mobile apps
Mobile applications are increasingly being used in biodiversity monitoring and conservation citizen science projects. These apps allow volunteers to easily record and share species observations, photos and other ecological data directly from the field using their smartphones. By harnessing the power of mobile technology and an active citizen community, these projects can gather large amounts of valuable biodiversity data across a variety of settings in a cost-effective way, compared to traditional survey methods conducted by professional scientists alone.

Some popular examples of mobile apps for biodiversity monitoring include iNaturalist, eBird, and Merlin. iNaturalist allows users to record observations, share them with fellow naturalists, and contribute to biodiversity science by sharing findings with scientific data repositories.eBird, managed by the Cornell Lab of Ornithology, enables birdwatchers to enter their sightings and access tools that make birding more rewarding, such as managing lists, photos, and audio recordings, and seeing real-time species distribution maps. Merlin, also from the Cornell Lab, helps users identify bird species through AI-powered visual recognition and question-based filtering, and contributes sightings to the eBird database. These apps showcase effective design practices that have enabled them to gather significant biodiversity data through public participation.