Multimodal Time‑Series Dataset for Cold Stress Prediction in a Cold‑Region Open Field

This dataset and code support the paper "Early Prediction of Cold‑Region Open Field Cold Stress Based on Multimodal Time‑Series Data and LSTM Network". Data were collected from an open field in Heilongjiang, China (45.8°N) across two growing seasons (2024–2025). The dataset contains 319 daily records with five modalities: meteorology, soil, pest image counts, lure counts, and spore counts. The prediction task uses the past 7 days to forecast cold stress 3 days ahead. Cold stress is defined as daily minimum temperature < 0°C or daily mean temperature < 5°C. This upload includes: the raw dataset (final_dataset.csv), all source code (Python scripts), pre‑trained LSTM model weights (lstm_model_complete.pth), and a requirements.txt file to reproduce the environment. Code is licensed under MIT. Data is licensed under CC BY‑NC 4.0.