2020 was a devastating year for flooding across Africa. The Nile rose to its highest levels in half a century and Ethiopia and Sudan saw large areas swamped with water devastating farms and rural areas. Many African cities are ill prepared for natural disasters and with the continent’s urban population rising fast floods will increasingly devastate cities as well as rural areas.
As well as washing away crops, homes and livelihoods, floods can bring malaria and other water borne diseases to a weakened populace.
Climate change is making a hotter and wetter world with more unpredictable rainfall a perfect recipe for devastating floods.
The floods that shocked Germany in the summer of 2021 were proof that rich countries will not escape climate risks either. But wealth and technology can play a role in adapting to and measuring climate risk.
The World Observed
Across the world thousands of earth observation instruments, satellites, sensors and cameras are constantly collecting millions of points of data. This data includes temperature, greenhouse gas emissions, polar ice melt, wildlife statistics, forest cover (often using LIDAR which measures the density and carbon content of a forest) and lots of other useful information.
These critical indicators help capture the sadly declining health of the natural world and spiralling climate risks. But the enormous quantities of data collected by these instruments can appear overwhelming.
However, as the world is discovering machine learning techniques can be applied to the data, allowing organisations to sift through the information and turning into useful indicators.
Machine learning is a branch of artificial intelligence which uses data and algorithms to imitate human learning and therefore improve in efficiency over time.
Three Magic Ingredients
The three ingredients which make this possible are the collection of huge amounts of data via sensors, satellites etc, allied to powerful computing power which can handle the data and the application of machine learning systems which can improve the way in which they collection and interpret data.
Machine learning is a sub-division of artificial intelligence, and it means that systems can learn and improve the way in which they collect and interpret data independently.
Satellite data can used to track climate risks like drought, flood and deforestation which can be transformed into financial indicators with the help of machine learning helping firms monitor their risk profile.
Machine learning can be used to sift through the information provided by camera traps in wildlife reserves. Trailguard cameras use AI to spot different species and send an alert when the camera detects humans (or more importantly poachers) rather than animals. Within minutes of poachers entering one of Africa’s wildlife reserves wardens are alerted and can respond to the threat.
Climate is Data Problem
Machine learning techniques are being deployed in an ever-wider number of settings.
Google have created the Environmental Insights Explorer which monitors the carbon dioxide footprint of buildings and transport networks allowing users to calculate their carbon footprint.
Satellite technology can monitor large scale carbon emissions, while sensors collect energy use data in buildings which can be used to optimise heating and cooling systems.
Climate risks are now a stark reality and machine learning can be used to help the world adapt. This could be through using data to predict floods or weather changes more accurately. Machine learning can also be used to effectively forecast energy usage to optimise renewable usage.
What Does the Future Hold
Identifying the scale and intensity of climate risks is a major challenge for governments and companies. Banks and insurers want to understand the physical risks to their portfolios. In other words: how much and where will sea level rise, how and where will extreme weather and wildfires impact on their assets on a granular level.
This is a problem for machine learning – collecting billions of points of data about the land, climate and sea and using them to accurately predict the future.
For example, researchers from Montreal Institute for Learning Algorithms (MILA) simulated what would happen to homes in Canada after damage by intense storms and rising sea levels. The objective is to try and make the risks of a changing climate real to people and businesses and provide them with actionable information.
How Can Machine Learning Aid the Battle Against Climate Change
Machine learning (ML) is particularly well primed to help the battle against Climate Change. Below are just some of the many methods and sectors Machine Learning can be harnessed to help mitigate, measure and adapt to a fast changing climate.
Machine Learning (ML) can cut the leakage of methane through the monitoring of seepage data from pipe sensors and satellite imagery. ML can also be used to pinpoint where repairs are necessary in electrical infrastructure, cutting wastage and therefore increasing energy efficiency. However, ML could help the fossil fuel industry by making oil and gas more efficient and cheaper to extract and emit more fossil fuels.
ML can be used to model carbon emissions and energy mix helping planners to optimise the usage of renewable sources.
For countries without universal electricity ML can identify which electrification methods would be most suitable for a particular region. Normally this could require intensive resource heavy surveys. ML can use satellite imagery to speed the process up.
Transportation is the source of around a quarter of global greenhouse emissions and represents low hanging fruit in terms of decarbonization. Cars and planes are the usual villains in this sector, but cargo ships are a major source of emissions.
There are four ways in which to decarbonize the transport sector:
- Reducing transport activity
- Improving fuel efficiency
- Switching to Alternative fuels
- Moving to transport alternatives (car to train)
ML is in a prime position to drive these strategies. Firstly, ML can be used to collect large amounts of data about transport habits and patterns. For example, traffic can be monitored, and models created to forecast future demand which can help drivers and planners avoid congestion.
Predicting public transport usage and aeroplane take off times can make for more efficient, less energy using and time wasting travel. Similar techniques can be used for freight, using ML to consolidate trips to avoid empty trains/lorries and ships therefore driving efficiency and cutting the number of journeys.
Driverless cars driven by artificial intelligence are a particularly controversial transportation topic. In theory using driveless vehicles would be more efficient and safer (taking the quickest routes) and would free up time for humans.
However, this technology is unproven at scale and there remain many ethical concerns about the use of these vehicles, primarily who is responsible when things go wrong. The driver, car company or the programmer.
Electric Vehicles (EV) by contrast are now a familiar sight on the roads in many countries. ML and EV dovetail in climate friendly ways. Manufacturers can monitor EV and use ML to predict faults, battery management and usage. Over time as more drivers use EV more data will be collected on their usage. This can be used to improve performance by proactively spotting fault and identifying battery state degradation.
Cities and Buildings
Building, cities, towns and workplaces are a major source of carbon emissions. But they also offer some of the easiest fixes. Many modern state of the art buildings consume virtually none or no energy. Smart buildings using sensors and control systems can monitor energy usage identifying where more is required and using tech like windows with built in solar panels to collect energy for the building.
ML can be used to model energy consumption and in modern buildings with sensors optimise energy usage.
Energy efficiency also comes with major cost savings for occupants as well as the environment.
The challenge is that you cannot replace building stock very quickly. Buildings also vary widely in size, shape and usage so one size fit solutions will not work. This means that these solutions take a long time to implement.
Climate Forecasting and Modelling
The many satellites orbiting the world are constantly producing huge amounts of data about land use, weather and climate patterns of the world. This data can used by scientists to build ever more complex climate models. ML has been utilised to classify crop cover, pollutants and many other kinds of data.
Deep neural networks could be used to account for cloud cover, a major source of uncertainty in climate models. Clouds can block sunlight and trap heat but are difficult to account for in climate models. Deep neural networks (an extension of ML) can be used to simulate cloud behaviour which and learn over time to improve the accuracy. Putting more accurate cloud behaviour into climate models should make them more reliable.
ML can be used to make climate predictions at a local level. Knowing which areas are likely to flood or suffer more wildfires is valuable information for firms trying to map the climate risks to their assets.
For example, an agribusiness dependent on the land will be impacted by any shifts in long term rainfall, while flooding and extreme weather could also drastically alter the productivity of the land. Understanding and modelling future shifts in climate at a very granular level will help businesses adapt.
Forests and Farmland
Huge amounts of carbon cycle through the biomass of trees, soil, bogs and peatland. Thanks to unsustainable farming practices and deforestation along with intensive cattle and livestock farming around a quarter of all greenhouse gases are released through agriculture.
As the world heats tinder dry summers cause ever bigger forest fires and permafrost which release ever more greenhouse gases. Carbon release in agriculture is not only a major contributor to greenhouse emissions but also one of the most difficult to tackle given well established agricultural practices and growing demand for meat in much of the world.
ML along with satellite imagery can identify how much carbon is released from the ground and how much is held within forests and soil. This would make is easier to identify how to manage land to where regulations are being breached and to help governments avoid further carbon release.
ML can aid reforestation by ensuring that trees are planted efficiently – by locating planting sites and then analysing data about tree health and biodiversity. ML could also be utilised to predict the direction and speed of fires, allowing firefighters to decide where to fight and where to try and stop the spread of the fire.
Adaptation to a new Climate
A new era of climate change or Anthropocene will be one of rapid adaptation. Humans will have to quickly adapt to a new world of painful extremes. Navigating a world of devastating natural disasters, disappearing coastlines and blistering heat will mean a wholesale shift in how societies function. Current infrastructure and agricultural systems will buckle under the stresses of these changes and new ways of working and living will have to be rapidly improvised.
ML will be a critical part of mapping and planning this new world. ML has applications from predicting disasters, to designing and maintaining new infrastructure. From monitoring ecosystems to detecting carbon release from soils and forests.
ML is not a replacement for the political and economic shifts that are required for a zero-carbon economy, but they represent a potent tool which can aid the policy decisions which drive reductions in carbon emissions.
ML could also be used to measure and coordinate individuals and groups to push for action on decarbonisation. By identifying and predicting how individuals or groups will react to changes like carbon taxes, governments can assess their impact prior to implementation. This could help avoid protests and strikes that could hold up meaningful climate action.
ML could also assist some of the speculative technologies which have been proposed to tackle global heating. Carbon capture and storage would collect C02 direct from the air and store it in the ground. The technology has been trialled, but it is extremely costly and unproven at the scale required to make any difference. ML could help model and detect where the prime underground storage places for carbon might be.
Solar geoengineering is the idea that reducing solar radiation would cut temperatures on Earth and the impact of global heating. Proposals include cloud whitening, robot boats crossing the ocean, mirrors in space reflecting the sun. All of these proposals are speculative but as desperation about climate change grows the incentive and temptation to try these moonshot ideas will grow.
Machine Learning is far from a solution or silver bullet for global heating. Instead it represents a potent tool which can be deployed in many different approaches that can aid in the battle to mitigate and adapt to the climate era.
One thought on “The Machine Learning Revolution: How it can be used in the fight against Climate change”