Research map / Probabilistic and scientific modeling / Dynamical systems
SIDE: Socially Informed Drought Estimation Toward Understanding Societal Impact Dynamics of Environmental Crisis
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes SIDE, a framework that uses social media and news text to jointly estimate US drought severity (DSCI) and its societal impact across 11 determinants, using LLM-assisted topic-determinant mapping and social-physical cross-attention.
Problem
Drought monitoring focuses on physical severity and neglects the diverse, time-varying societal impact of drought; estimating both jointly from social and news media is challenging due to implicit temporal dynamics and social-physical interdependence.
Approach
Quantify societal impact by BERTopic topic extraction from social and news text, then map topics to NIDIS societal impact determinants with LLM likelihood scoring; model severity and impact jointly with a social-physical cross-attention encoder-decoder.
Key results
- On California, SIDE reports MAE 30.07 and MSE 1823.20 versus the best baseline TimesNet (MAE 40.78, MSE 2026.89), a 26.25% MAE reduction and 10.05% MSE reduction.
- On Texas, SIDE reports MAE 48.08 and MSE 4369.98, versus iTransformer at MAE 50.50 and MSE 4693.32.
- Ablation removing social input raises CA MAE from 30.07 to 35.28 and TX MAE from 48.08 to 55.19.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add social input (ablation) | SIDE · California drought severity estimation | MAE: 35.28 -> 30.07 | Helped |
| add social input (ablation) | SIDE · Texas drought severity estimation | MAE: 55.19 -> 48.08 | Helped |
| add news input (ablation) | SIDE · California drought severity estimation | MAE: 33.47 -> 30.07 | Helped |
| add news input (ablation) | SIDE · Texas drought severity estimation | MAE: 51.97 -> 48.08 | Helped |
| add attention for integrating social and news inputs (ablation) | SIDE · California drought severity estimation | MAE: 33.96 -> 30.07 | Helped |
| add attention for integrating social and news inputs (ablation) | SIDE · Texas drought severity estimation | MAE: 50.33 -> 48.08 | Helped |
Surprising findings
- On Texas, the best baseline is iTransformer, while TimesNet performs poorly (MAE 116.06), indicating baseline instability across states.
Limitations
- Societal impact ground truth is itself computed by the same DSIQ pipeline, so no independent human annotation validates it.
- Location filtering relies on a list of entities because few posts carry geolocation.
- Evaluation is limited to two US states over about six years.
Experimental setup
Datasets: SocialDrought
Benchmarks: SocialDrought (CA) SocialDrought (TX)
Metrics: MAE MSE RMSE MFA
Baselines: iTransformer Koopa TimesNet One-Fits-All RWKV-TS
Builds on: BERTopic
Closest papers by meaning
- DroughtSet: Understanding Drought Through Spatial-Temporal Learning (AAAI 2025 · similarity 0.87)
- World Models Should Prioritize the Unification of Physical and Social Dynamics (NeurIPS 2025 · similarity 0.81)
- Fostering the Ecosystem of AI for Social Impact Requires Expanding and Strengthening Evaluation Standards (NeurIPS 2025 · similarity 0.81)
- Assessing Large Language Models on Climate Information (ICML 2024 · similarity 0.81)
- Holistically Evaluating the Environmental Impact of Creating Language Models (ICLR 2025 · similarity 0.80)
- A Statistical Framework for Algorithmic Collective Action with Multiple Collectives (NeurIPS 2026 · similarity 0.80)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.