LS-NTP: Unifying long- and short-range spatial correlations for near-surface temperature prediction

Area of Science:

• Environmental Science
• Computer Science
• Data Science

Background:

• Near-surface temperature prediction (NTP) is crucial for preventing temperature crises.
• Existing methods struggle to simultaneously model long- and short-range spatial correlations, limiting prediction accuracy.

Purpose of the Study:

• To develop a novel approach for NTP that explicitly captures both long- and short-range spatial correlations.
• To introduce a new convolution operator, Long- and Short-range Convolution (LS-Conv), for enhanced spatial-temporal forecasting.

Main Methods:

• A novel Long- and Short-range Convolution (LS-Conv) operator is proposed, integrating Node-based Spatial Attention (NSA), Long-range Adaptive Graph Constructor (LAGC), and Long- and Short-range Integrator (LSI).
• LS-Conv unifies a Long-range aware Graph Convolution Network (LR-GCN) for long-range correlations and a Short-range aware Convolution Neural Network (SR-CNN) for short-range correlations.
• A new model, Long- and Short-range for NPT (LS-NTP), is developed based on the LS-Conv operator.

Main Results:

• The proposed LS-NTP model demonstrates superior performance compared to state-of-the-art techniques on two real-world datasets.
• The LS-Conv operator effectively captures both long- and short-range spatial dependencies crucial for accurate NTP.

Conclusions:

• The novel LS-NTP model, utilizing the LS-Conv operator, significantly advances the accuracy of near-surface temperature prediction.
• This research addresses a critical gap in spatial-temporal forecasting by simultaneously modeling diverse spatial correlations.