Cluster-Segregate-Perturb (CSP): A Model-agnostic Explainability Pipeline for Spatiotemporal Land Surface Forecasting Models

Tushar Verma, Sudipan Saha

Keywords: Climate AI, Explainability, ConvLSTM, spatiotemporal analysis

Abstract: Satellite images are increasingly valuable for modeling regional climate change. Earth surface forecasting is one task that combines satellite imagery and meteorological data to understand how climate evolves over time. However, understanding the complex relationship between meteorological variables and land surface changes remains a challenge. Our paper introduces a pipeline that integrates principles from perturbation-based techniques like LIME and global explainability techniques methods like PDP, addressing the limitations of these techniques in high-dimensional spatiotemporal models. This pipeline facilitates analyses such as marginal sensitivity, correlation, and lag analysis, etc for complex land forecasting models. Using ConvLSTM for surface forecasting, we analyzed influence of variables like temperature, pressure, and precipitation on the NDVI of the surface predictions. Our study in EarthNet2021 Dataset (primarily consists of samples from the European Alps region, collected during the spring to fall seasons) revealed that precipitation had the greatest impact, followed by temperature, while pressure has little to no direct effect on NDVI. Additionally, interesting nonlinear correlations between meteorological variables and NDVI have been uncovered.

Primary Area: interpretability and explainable AI

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 9682