How can AI help us adapt to climatic extremes?

IEC Tech

article here

Artificial intelligence (AI) can be used to overcome some of the worse effects of extreme weather. International standards pave the way for its safe and responsible use.

Even climate change deniers seem to be changing their tune. In the face of persistent evidence, not least when floods and wild fires impact the lives of people the world over, the impact of severe weather episodes, which have been multiplying at an increasingly rapid rate, is getting more and more difficult to ignore.

As in other fields, from medical to transport, AI is being explored to help with their mitigation. According to UN Secretary-General António Guterres, a third of the world’s population, mainly in the least developed countries and small island developing states, is not covered by early warning systems (EWS). He announced that the UN would spearhead a new universal EWS with AI at its core.

With water scarcity and rising temperatures threatening food resources in all countries, scientists are also using AI to fast-track the development of new types of crops capable of withstanding climate extremes. In May, a UK-based plant genome editing lab said it was constructing a synthetic potato chromosome from scratch – “a world-first achievement that could transform agriculture”, it claimed. “If successful, the research will unlock new ways to make crops more resilient and productive… securing global food supplies amid the growing challenges of climate change.”

Harvesting data for weather resistant crops

There is a role for machine learning (ML) to help in better matching crops with environmental conditions, adopting growing techniques that boost their resilience and even developing new breeds of weather- and pest-resilient plants.

A number of these are outlined in a paper published by Nigeria’s Obafemi Awolowo University. For instance, AI-driven crop models which incorporate weather predictions and long-range climate data “can simulate how a crop will respond to future climate scenarios, allowing breeders to identify the most resilient varieties for specific conditions”.

AI can supercharge measurement of the physical attributes of crops. This process, called phenotyping, “traditionally required manual labour and was therefore limited by the availability of human resources”, according to the research. Data from drones and sensors fed into AI models can enable “automated and large-scale phenotyping, collecting detailed data on plant morphology, growth patterns and responses to environmental stress”. The speed  of AI systems allows more time for the farmer to find a better response.

Identifying plant genes

Academics and scientists are very active in using ML to identify plant genes which exhibit greater resilience to heat, moisture and disease. When genomic selection is sped through an algorithm, it can help predict the breeding value of plants and ensure that resources are used effectively, the Nigeria research outlines.

Researchers at Wageningen University in the Netherlands are unpacking the genome sequence of mustard with a view to building an AI model that could be used to identify the non-coding ribonucleic acid (RNA) across plant species. Non-coding RNA plays an important role in the development of organisms, for example by activating genes, according to project lead Michael Shon.

“By understanding the non-coding regions of plant genomes, breeders can develop crops with specific desirable traits, such as making crops more resilient to climate change,” he explains. “The potential impact could be huge.”

Early warning systems helped by AI

After the tragic loss of multiple lives when a flash flood raised the level of the Guadalupe River in Texas by eight metres in just 45 minutes, questions were asked about the performance of the region’s EWS. Weather forecasters had warned of heavy rainfall, and alerts were issued to mobile phones, but cellular coverage was patchy, and there was no county-wide EWS in place, with the cost of implementation blamed.

The need for better EWS was deemed urgent by the World Meteorological Organization (WMO) when it told COP27 in 2022 that the number of recorded disasters had increased five-fold “driven in part by human-induced climate change and more extreme weather”. Three years on and challenges remain in “hazard forecasting, risk communication and decision-making”, according to researchers at the Max Planck Institute for Biogeochemistry in Germany. They are exploring whether AI models, integrated with meteorological and geospatial data, could improve outcomes.

Since flood forecasting “suffers from many unresolved processes in hydrological modelling”, they suggest using ML to tighten predictions. They point out that wildfire risk prediction which links earth observation data with anthropogenic and meteorological information would equally benefit from ML.

Rubbish in, rubbish out

Regardless of whether AI is used or not, the results of any predictive model are only as good as the input. Where the data is weak, forecasts will be weak too. A project led by the University of Oxford with the UN World Food Programme and the IGAD Climate Prediction and Applications Centres claims to have developed a first-of-its-kind hybrid approach to generate more accurate forecasts without the need for expensive local infrastructure.

“We use AI models to fill in the traditional physical forecasting models,” explained researcher Shruti Nath. “This allows better, more accurate representation of reality, particularly for regions that don't have such good observations, like Africa.”

The USD 3,1 billion Early Warnings for All (EW4All) initiative led by the UN, due to come into force by the end of 2027, aims to ensure universal protection from hazardous environmental events.

Central to its effort is the collection, sharing and access to high quality data at a global scale with ITU tasked to improve warning dissemination and communication. At the heart of its efforts is AI. For example, AI can be “leveraged to contextualize disaster risk information comprehensively to facilitate well-informed and equitable decision-making”.

ITU says AI has the “capability to enhance the capacity for detection, observations, monitoring, analysis and forecasting of hazards”. AI will further “optimize information delivery with personalized ‘client’ profiles, ensuring timely, comprehensible and actionable insights”.

Standards are key

In order to be effective both at scale and at a local level, various technologies, ranging from satellite and plane imaging to IoT sensors and long-term meteorological modelling, ideally need to be standardized, transparent and equitable. The work of the IEC, often in concert with organizations like ITU and ISO, delivers essential work to ensure compatibility among diverse technologies and data sources and between multiple stakeholders.

A key set of standards is developed by JTC 1/SC 42, the joint technical committee formed between ISO and the IEC to produce standards for AI. It has for instance published ISO/IEC 5259-1 on the data quality for analytics and ML.

Standards for sensors are published by IEC TC 47, while anything linking sensors to IoT is standardized by another joint technical committee between ISO and IEC, JTC 1/SC 41. Among its publications is ISO/IEC 20005, covering sensor networks.

A new IEC and ISO joint technical committee, JTC 3, on quantum technologies is planning to help standardize such technologies and coordinate with the work of relevant committees. It is hoped that processing data using quantum tech will exponentially increase the capacity for computation and the speed at which calculations and analysis are performed.