ABOUT extrAIM
In Layman’s terms, extrAIM aims to address an elephant-in-the-room type of challenge that regards the uncertainty (and mismatch, compared to in-situ measurements) of satellite-derived hydroclimatic products, which we all know it is there, yet we almost always choose to neglect. In particular, extrAIM through an optimum combination of explainable AI techniques and multivariate probabilistic methods will address the challenge of quantifying the uncertainty of satellite-based precipitation products, and eventually adjust them to account for the probabilistic behavior of extremes.
A new era...
The advent of new technologies as well as the recent, and game-changing, algorithmic advances in Information and Communication Technologies (ICT) and remote sensing, along with the multi-decade efforts on space exploitation and exploration has opened new frontiers in monitoring earth observations (EO) from space. Underpinning that we are transitioning in new era of satellite-data-driven EO systems (e.g., earth’s digital twins driven with satellite EO data – see for instance the DestinE initiative of ECMWF, ESA and EUMETSAT).
The challenge...
Such satellite-based products address several issues encountered when using gauge-based observations (e.g., limited spatiotemporal coverage and latency), yet they are not free of limitations. In comparison with gauge-based data, the QEs obtained by such datasets typically exhibit significant differences, especially in extremes, uncertainty, and bias, which arguably hinders their wide adoption by engineers and scientists. In this respect, developing innovative methods able to model and quantify the uncertainty of satellite-based hydroclimatic products, and eventually adjusting the products accordingly, becomes urgent.
Our vision...
The development of such methods and approaches would arguably improve the QEs provided per se (minimizing error), the reproduction of extremes’ behavior, the understanding of errors and error patterns, and would, as such, increase end-users confidence and trust on the quality and potential utility of satellite-based hydroclimatic products.
Our mission...
Develop a general-purpose framework capable of modelling and quantifying the uncertainty of satellite-based products. This will be done by combining both probabilistic and AI methods (with a focus on explainability) for multiple datasets integration, as well as uncertainty modelling. extrAIM will develop a first-of-its-kind, satellite-based, low-latency, uncertainty-aware precipitation dataset for the Mediterranean region, adjusted to account for the extremes’ probabilistic behavior.
The extrAIM approach
An important aspect to ensure the future-proofness, user-onboarding, and longevity of satellite-based hydroclimatic products is to build confidence and trust on data products of this type, which are known that when compared with in-situ measurements exhibit significant differences, especially in extremes, uncertainty, and bias.
To achieve this goal, the extrAIM’s vision aligns with the famous quote of Richard Feynman how suggested that “when we know that we actually do live in uncertainty, then we ought to admit it”. Paraphrasing this quote to better reflect the objectives of extrAIM we would argue that “when we know that our data/models are uncertain then we ought to admit it, and quantify it”.
This idea is the core of extrAIM, which though the combination of explainable AI methods, for the merging of multiple satellite precipitation products, with multivariate statistical learning and Bayesian modelling methods, for uncertainty quantification, will provide a novel modelling framework for such purposes, showcased through a large-scale proof-of-concept application that regards the generation of uncertainty-ware satellite-based precipitation product for the Mediterranean region.
To achieve this goal, the extrAIM’s vision aligns with the famous quote of Richard Feynman how suggested that “when we know that we actually do live in uncertainty, then we ought to admit it”. Paraphrasing this quote to better reflect the objectives of extrAIM we would argue that “when we know that our data/models are uncertain then we ought to admit it, and quantify it”.
This idea is the core of extrAIM, which though the combination of explainable AI methods, for the merging of multiple satellite precipitation products, with multivariate statistical learning and Bayesian modelling methods, for uncertainty quantification, will provide a novel modelling framework for such purposes, showcased through a large-scale proof-of-concept application that regards the generation of uncertainty-ware satellite-based precipitation product for the Mediterranean region.
