Trustable AI generated Mapping (TAIM)

点击翻译按钮获取中文摘要

Artificial Intelligence and Machine Learning based mapping based on remote sensing data presents a significant opportunity within the environmental field, removing the need for manually digitising features in imagery and for mapping larger areas to a finer granularity than is possible using traditional techniques such as Geospatial Information System (GIS) based analysis. As with any new technology however, there are reservations to its use 'on the ground'. This scepticism is to some degree warranted, due to the lack of standardised methods for comparing AI algorithms to ground results in a methodical manner.To address this challenge, EOLAS and partner organisation Scotland's Rural College's (SRUC's) Trustable Credit scheme will define a framework and candidate standard which allows for direct comparison of AI / ML generated mapping algorithms. This includes the definition of standard classes to be used by algorithms depending on the application, and baseline sites mapped to a high degree of accuracy. Using these organisations with an active interest in the field can run these trial sites through their algorithms to determine key performance metrics, comparing their results to pre-ground truthed data. The initial use case will be the carbon credit markets, selected due to the high requirement for trust in the outputs, emerging best practice, and its position as a high growth market suitable for SME involvement.The key approach to the project will be the definition of a consortium with interests in AI derived data products and wider engagement to ensure that a common methodology is defined and agreed. Here existing schemes such as Trusted Credit will be leveraged, due to their established networks of interested parties. A large part of this project will be engagement with the geospatial community more widely, ensuring a collaborative approach aimed at overcoming a challenge faced by all operators: user trust in the technology. This consortium will, through working groups, explore the issue and present solutions for standardised quality metrics.Through engagement and standard definition we will mature the use of emerging AI technologies within the carbon credit use case, serving as an example for techniques for wider adoption within the environmental and geospatial sectors. By providing open and transparent methodologies for assessing the quality of AI / ML derived data products this project will increase the overall levels of trust in AI as a mechanism for mapping features in combination with remote sensing data, benefitting the wider community.