Simulation of long-term spatio-temporal environmental dynamics using a unified benchmark of neighbor augmenting, LSTM and graph attention models
Kaveh Karimadini, Sharareh Pourebrahim · 2026
Read Simulation of long-term spatio-temporal environmental dynamics using a unified benchmark of neighbor augmenting, LSTM and graph attention models by Kaveh Karimadini, Sharareh Pourebrahim (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/PMC13518308.txt.
Description
Kaveh Karimadini and Sharareh Pourebrahim run a fair fight between forecasting models for environmental change in a 2026 benchmark study. Long-range prediction across space and time is hard, and flashy graph-attention networks are not automatically best. Every contender trained and tested through one shared pipeline on annual satellite data from 2000 to 2023, holding 2024 fully unseen, and scored on the usual error measures.
A neighbourhood-augmented LSTM with helper variables won, scoring an R-squared near 0.95 and staying stable. Its cellular-automata style window over the surrounding pixels did the heavy lifting. A graph-attention model with a feed-forward block placed second; stripping parts made rivals wobble or degrade. Vegetation and water measures predicted most readily. The authors' engineering verdict: strong time modelling plus good side information beats fancy spatial attention for this job.
Catalog record
| CREATOR | Kaveh Karimadini, Sharareh Pourebrahim |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 94194 |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |