Climate change threatens to cause serious harm to our planet and our communities with extreme environmental and societal events, like floods, famine, loss of habitat, increased poverty, among others. Thankfully, artificial intelligence (AI) is stepping up to the plate, and has a secret weapon: Data. This article will touch specifically on AI and data in the context of floods.
Traditionally, flood prediction has relied on historical observations used in combination with complex models, which has been valuable, but has proven to be a limited approach.
And here’s when AI comes in, as it is a data hungry beast that with access to a vast amount of information, is learning to predict floods with remarkable accuracy.
The data deluge we’re lucky to be living in is empowering AI for flood risk prediction in many ways. For example, satellite imagery provides a constantly updated picture of our planet, which AI can analyze to identify areas with unusual changes or patterns in water levels or vegetation, potentially signaling an upcoming flood.
But only satellites are key to achieve accurate flood predictions, meteorological data acts as weather whisperer. To predict floods, it is crucial to understand rainfall patterns, atmospheric pressure, and wind direction – so that AI can analyze and identify weather patterns that might trigger floods, which is also complemented with sensor data from rivers, dams, and other ground based sources that provide real-rime information on water levels. The latter help to understand flood peaks and inundation zones.
By digesting these diverse datasets, AI is learning to recognize complex patterns that humans are likely to miss due to their comprehensiveness and complexity. This allows for:
a) Earlier warnings: The more AI understands the intersection between weather, geography, and water flow, the sooner it can identify potential floodings. This lead time is crucial and precious as it may allow communities to evacuate and take precautions, minimizing damages and saving lives.
b) Hyper-local forecasts: Traditional flood prediction models often use a broad brush but with AI, high-resolution data can be used to predict flood risks at a much more granular level. This allows for targeted warnings and resource allocation to the most vulnerable areas.
c) A dynamic model: AI doesn’t sit back and relax. As more data becomes available, particularly real-time flood events and their impact, AI models can continuously learn and improve their predictive accuracy.
The potential benefits of AI are undeniable. By harnessing the power of data, AI is becoming a game-changer in the fight against floods and other disasters, helping us build a more resilient future.
But what are the challenges for AI to become more accurate and better at predicting disasters such as floods? Data access.
On the one hand, sharing information across borders and ensuring data quality are crucial for robust AI flood prediction models. The accuracy of AI models hinges on the quality of the data they are trained on. So quality assurance, reliance, consistency, and standardization across different sources is essential.
While data sharing may seem like an easy task, it is often that sharing data across borders is hampered by political and security concerns, for which partnerships and collaboration across borders are key.
Finally, another issue which may crook results is data bias. When data used to train an AI system is not representative of the reality it’s meant to model, due to incomplete data, biased sampling or other factors, we may have unrepresentative datasets. This shy quality assurance becomes so crucial.
What kind of data fuels AI for climate predictions?
AI thrives on a diverse data palette. Typical key data sources that empower AI for climate predictions are:
Earth Observation Data: Satellite images provide a constantly updated picture of our planet that AI can analyze to identify changes in factors like: a) sea levels: rising sea levels increase coastal flood risk, so, analyzing satellite altimetry data can tell us about sea level changes and predict potential inundation zones; b) snowpack: its levels inform us potential spring flooding events; and c) vegetation health: AI can identify areas at risk by analyzing vegetation health and indices which may be indicative of drought or flooding conditions.
Meteorological Data: weather datasets play a crucial role in flood prediction, especially the ones related to: a) precipitation, as rainfall patterns are a key factor for flood forecasting, and AI can make comparisons between real-time precipitation data and historical data, b) atmospheric pressure data, which may help us identify weather systems that might trigger floods, and c) wind direction, c) wind data, which AI can factor into its flood prediction models to understand whether wind is pushing water towards certain areas.
Ground-Based Data: Real-time information from ground-based sources provides crucial insights for flood forecasting. These include: a) river gauges, measuring water levels in real time to predict flood peaks and inundation zones; b) weather stations, which AI can combine satellite and meteorological data for improved predictions.
Social Media: Even social media may become a source of data, a rather unconventional one, but that can provide valuable insights during flooding events. For example is it possible to track the spread of floods and identify areas in need of immediate assistance.
The biggest tchallenge for AI to improve and further develop its climate prediction skills, is data access, quality, sharing and bias. Nonetheless, we believe that the greatest one of these is data access, as this is the lifeblood of AI based climate predictions. Therefore, open source reliable data is essential for it.
A lot of the climate data has been typically found within government agencies or private research institutions, limiting access to it to a reduced number of people. This has changed in the past few years, as open source data initiatives have increased and aim to make this data freely available to the general public and scientific community. This fosters collaboration, innovation, and ultimately, more accurate AI models.
The benefits of open source data are that it allows for independent verification and scrutiny of the data, improving transparency. Moreover, by making data readily available, researchers can reproduce and build upon each other’s work, which accelerates scientific progress and allows for democratization of knowledge by allowing a wide range of people of institutions to contribute and collaborate together to climate research and AI development.
Building a Collaborative Ecosystem: Sharing Data for a Sustainable Future
As we mentioned before, for data access and quality to make a better AI-powered climate prediction ecosystem, international collaboration and supportiveness are essential. The development of international agreements that facilitate the secure and responsible sharing of climate data across borders, is crucial, but it will require addressing the political and security we mentioned in part I of this article, to foster trust and collaboration.
Data quality control and standardization are also key elements of a healthy open source data ecosystem. Open-access platforms for storing, sharing, and accessing climate data will significantly improve data availability for AI research. Furthermore, it may enable the creation of tools for data control, standardization and verification.
This all sounds like a fairly feasible plan. But, how do we deal with data collection, access and quality in developing countries without the proper infrastructure and capacity to process data and develop technologies that may save lives? We will cover this in an later article.
