Research map / Probabilistic and scientific modeling / Molecules and proteins
iTrendRNN: An Interpretable Trend-Aware RNN for Meteorological Spatiotemporal Prediction
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 iTrendRNN, an interpretable RNN that models meteorological evolution as a hidden-state trend (proximity, attention-based sequential, and flow trends) fused adaptively, outperforming recent ConvRNN baselines on ECMWF temperature and humidity data.
Problem
Spatiotemporal prediction of meteorological elements (temperature, relative humidity) with existing neural models being overcomplicated and uninterpretable.
Approach
Formulate evolution as hidden-state increments ht+1 = ht + δt and estimate δt via three interpretable trends: proximity trend (last change), attention-based sequential trend (ATU over historical trend states, supervised by true trend), and flow trend (flow-aware spatial attention over neighborhood differences), fused by an SKNet-inspired spatially adaptive trend fusion.
Key results
- iTrendRNN reduces temperature MSE from 1.04 (best baseline TAU) to 0.87 and MAE from 0.70 to 0.63.
- On relative humidity, iTrendRNN achieves MSE 36.97 and MAE 4.17, vs. 39.57/4.28 for the best baseline MIM.
- Ablation: removing the sequential trend raises temperature MSE to 1.46, the largest drop among trend ablations; the SKNet-style fusion variant reaches 0.89 vs 0.87 for the proposed fusion.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add sequential trend (ATU-based) module (ablation) | iTrendRNN · Temperature and relative humidity | MSE: temperature 1.46 -> 0.87; humidity 45.82 -> 36.97 | Helped |
| add proximity trend module (ablation) | iTrendRNN · Temperature and relative humidity | MSE: temperature 0.98 -> 0.87; humidity 39.09 -> 36.97 | Helped |
| add flow trend (FAU-based) module (ablation) | iTrendRNN · Temperature and relative humidity | MSE: temperature 1.06 -> 0.87; humidity 40.17 -> 36.97 | Helped |
| replace trend fusion with average, convolution, or SKNet fusion (ablation) | iTrendRNN · Temperature and relative humidity | MSE: temperature 0.87 -> 0.89-0.94; humidity 36.97 -> 37.89-39.15 | Hurt |
| proposed iTrendRNN vs. strongest baselines (MIM/TAU) | iTrendRNN · Temperature and relative humidity | MSE: temperature 1.04 -> 0.87; humidity 39.57 -> 36.97 | Helped |
Surprising findings
- Removing the sequential trend causes the largest degradation (MSE 0.87 to 1.46 for temperature), worse than removing proximity or flow trends.
Experimental setup
Datasets: ECMWF ERA5 reanalysis (temperature, relative humidity)
Benchmarks: ERA5 temperature (East Asia, 2020 test) ERA5 relative humidity (East Asia, 2020 test)
Metrics: MSE MAE
Baselines: ConvGRU ConvLSTM TrajGRU PredRNN MIM PhyDNet SA-ConvLSTM SimVP TAU
Builds on: ConvGRU SKNet PredRNN ConvLSTM
Closest papers by meaning
- On the Integration of Spatial-Temporal Knowledge: A Lightweight Approach to Atmospheric Time Series Forecasting (NeurIPS 2025 · similarity 0.85)
- SNN-PDE: Learning Dynamic PDEs from Data with Simplicial Neural Networks (AAAI 2024 · similarity 0.84)
- Expand and Compress: Exploring Tuning Principles for Continual Spatio-Temporal Graph Forecasting (ICLR 2025 · similarity 0.84)
- Explanation of Dynamic Physical Field Predictions using WassersteinGrad: Application to Autoregressive Weather Forecasting (NeurIPS 2026 · similarity 0.84)
- How Different from the Past? Spatio-Temporal Time Series Forecasting with Self-Supervised Deviation Learning (NeurIPS 2025 · similarity 0.83)
- Spatiotemporal-aware Trend-Seasonality Decomposition Network for Traffic Flow Forecasting (AAAI 2025 · similarity 0.83)
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.