Machine Intelligence is Helping Tackle Environmental Crisis

The frequency and intensity of extreme weather events are constantly increasing because of human-made climate change and they hit vulnerable communities in developing countries disproportionately hard. The population of developed countries is also suffering from human activities harmful for the environment, such as burning fossil fuels for power and transportation: 90 % of the worldwide population is exposed to air pollutants that exceed World Health Organization (WHO) air quality guidelines [2]. In fact, dirty air kills around 7 million people each year! In addition, it causes long-term health problems, such as asthma, and affects children’s cognitive development. It is not only awful for our health but also stresses our social and economic systems. According to the World Bank, health damage due to air pollution costs global society around $5.7 trillion every year. That is equivalent to approximately 5% of the worldwide gross domestic product (GDP) [3].

 

Even though these challenges might overwhelm us and seem unsolvable, we have a new tool which will help us protect the planet – Machine Intelligence (MI). Below we consider several examples of how MI is already helping tackle climate change and its impact.

 

​

MI supports environmental, climate, and weather research

 

In order to be able to fight climate change, we first need a very good understanding of all the related phenomena. Therefore, there is an enormous amount of research in this field including the development of various climate models. These are complex mathematical models designed by meteorologists, geophysicists, and climate scientists to simulate and understand historical climate development and to predict future climate changes. Indeed, there are dozens of climate models accepted by the scientific community [4]. However, these models rely on different assumptions so that predictions can vary a lot for long-term forecasts depending on which model is used. One way to deal with this is to simply average predictions, but this approach is very unprecise. Here is where MI comes into play: Claire Monteleoni, a computer science professor at the University of Colorado, has used Machine Learning (ML) to create a weighted average of various climate models [5]. By learning the strengths and weaknesses of each single model, this ML algorithm gives each model a particular weight with respect to the final output. As a result, this approach considerably improves climate forecasts compared to conventional methods.

 

MI is not only applicable in fundamental climate research but is also used by climate scientists to tackle practical problems. For example, in 2016, researchers reported the first application of Deep Learning algorithms to identify extreme weather events such as tropical cyclones, atmospheric rivers, and weather fronts [6]. This research has shown that intelligent machines can predict extreme weather events with precision like that of human experts. Since machines, compared to humans, can process enormous amounts of data within a fraction of a second and can work 24 hours a day, they can be applied to observe and monitor weather all over the world in real time and to alert human experts if an extreme weather event is likely to happen. Thus, such algorithms can be applied as a part of early warning systems for extreme weather events and can help save thousands of human lives.

​

​

Real-time tracking of air pollution

 

One of the major sources of climate-damaging CO2-emissions is large fossil fuel power plants. While in the majority of developed countries, emissions of power plants are constantly measured and observed, there are a lot of communities on our planet that cannot afford expensive CO2-emission monitoring systems. In addition, many power companies worldwide try to shroud their actual pollution in secrecy in order to prevent bad publicity or escape governmental regulations (see e.g., [7]). As a result, the actual emissions remain unknown, transparency towards the public is not met, and governmental actions are based on incorrect assumptions. MI can be used to better this situation. For example, Carbon Tracker Initiative is using Artificial Intelligence (AI) and satellite images to measure CO2-emissions from all large power plants worldwide [8], [9]. In this initiative, an AI algorithm detects and quantifies multiple indicators of power plant emissions including visible smoke as well as infrared signature of smokestacks and cooling water intakes. Emission data collected in this way creates public transparency regarding actual stress to the environment and can help authorities adapt their climate strategies and actions in a data-driven manner.

 

Air pollution due to burning fossil fuels is not only responsible for current climate change but is also harmful for our health. Especially in cities with strong industry and a lot of cars, polluted air is an urgent issue. However, most cities fail to implement effective measures to tackle air pollution, mainly because of the lack of reliable data: conventional air quality monitoring stations are very expensive and difficult to install because their large size requires planning permission from the local authorities. As result, air quality is measured only in a few locations and is not representative of a city’s air quality as a whole. To solve this problem, a German startup, Breeze Technologies, has developed low-cost, small-scale air quality sensors that can easily gather a lot of data across the whole city [10]. Breeze Technologies uses AI for automatic calibration of air quality sensors in order to increase data reliability and accuracy. They also utilize AI to predict which sensors are likely to fail, allowing timely repair of faulty hardware or to push-out software fixes remotely. In this way, MI can help provide authorities with the data needed to better mitigate air pollution in the cities.

​

​

Saving energy using MI

 

In order to slow down climate change, we have to not only reduce emissions that are harmful for the climate but also have to consume much less energy than we are currently doing. In fact, this problem is extremely urgent since global energy consumption is expected to grow by nearly 50% by 2050 (with respect to consumption level in 2018) [11] and most of the current energy is still produced by burning fossil fuels [12]. MI can also help us tackle this problem.

 

For example, a German cross-industry initiative, WindNODE, which focuses on smart cities of the future, is using an ML algorithm as part of a research showcase dealing with controlling the heating consumption in a Berlin district [13]. Their smart, self-learning home manager system can optimize energy consumption in accordance with weather and tenant’s habits. When the tenant leaves her home, the system automatically adjusts the heat supply and lowers the temperature in the rooms. Just before the tenant returns, the system gradually heats the home. Thereby, the ML algorithm uses not only the inside temperature as input data but also the outside temperature and further external factors such as the weather forecast. This system is capable of autonomously learning and can determine how much energy is required to heat the home depending on the actual weather at any time. It was shown that this prototypical system is able to save about 20-25% of the energy than previously used.

 

Another initiative dedicated to reducing energy consumption is running in the Landshut University of Applied Sciences and aims to significantly lower energy demand through intelligent networking [14]. To this end, researchers install various sensors and control devices in large buildings such as hotels, indoor swimming pools, factories, and office buildings. The collected data are fed into an ML algorithm which is connected to the existing building’s energy-efficiency management system. ML component analyzes collected data, learns how to minimize energy consumption, and steers the operation of the energy-efficiency system of the building accordingly. In this way, a holistic and intelligent energy-optimization system is created. With this approach, researchers aim to reduce energy consumption of large operations buildings by more than 50 percent! If the current research phase is successful, the developed model/system might become a blueprint for intelligent and resource-saving energy management systems of the near future.

​

​

Deforestation

 

Another environmental problem we are currently facing is deforestation and forest degradation, especially of the rain forest. Indeed, deforestation contributes up to 17% to the total anthropogenic CO2-emissions [15] and is one of the major causes of global warming. Thus, reducing deforestation should be one of the priorities in the fight against climate change. Also, here MI can provide a significant help.

 

Before one can take actions against deforestation and degradation, one should know where these phenomena appear. There is a lot of data which can help to identify regions in danger, e.g., government records, commercial satellite imagery, and publicly available data sets. However, due to enormous amounts of data and large forest areas, it is very difficult for humans to process all the available data and to identify problematic regions. This is where an AI-driven solution from OpenSurface comes into play [16]. This open-source platform combines data from various sources, including satellite imagery, and is able to detect changes in the forest, monitor, and forecast deforestation and degradation in each observed region. This information can be then used by governments, companies, and researchers to define and prioritize preventive actions.