What AI Weather Models Can (and Can’t) Tell Us TodayModel Notes
Oct. 2026How data-driven weather models work — what they do well, where they blur, and what forecasters should trust.
I am currently pursuing a professional master’s degree at the Department of Atmospheric and Oceanic Sciences (AOS), Fudan University, advised by Prof. Wen Zhou. My undergraduate research explored AI for Science, with a focus on hydrometeorological forecasting and drought prediction.
My recent interests lie in weather and climate predictability, especially East Asian monsoons, heatwave-to-heavy-rainfall transitions, and combining machine learning with physical-process information. Current projects include drought forecasting, quasi-biennial oscillation (QBO) experiments, and Antarctic station forecasts.
Professional Master’s in Meteorology · Department of Atmospheric and Oceanic Sciences · Advisor: Prof. Wen Zhou
B.Eng. in Spatial Information and Digital Technology · GPA 3.59/4.0, ranked 2/91 (Top 5%)
A ResNet-Transformer runoff forecasting model fusing meteorological, DEM, and upstream-runoff data; NSE 0.992, ≈13.8% above the LSTM baseline. First-author paper in Water Resources Management.
An AI platform unifying meteorological visualization, forecasting, and analytics: ECharts/Cesium frontend, SpringBoot + MySQL/Redis backend, and a CNN-LSTM temperature module.
Undergraduate comprehensive internship at a STAR Market-listed maker of high-precision intelligent manufacturing equipment (precision dispensing systems for electronics assembly and IC packaging). Contributed to developing and maintaining PC-side (host) control software for dispensing equipment (C#/Qt) — device status monitoring, process parameter configuration, and production data acquisition modules; wrote Python scripts for production data processing and statistical reporting; assisted engineers with software integration and on-site commissioning; gained hands-on familiarity with industrial automation workflows and manufacturing informatics.
2025
Water Resources Management, 39 (2025), 6073–6092 · JCR Q1, IF = 5.7 (Springer)
A multi-source data fusion ResNet-Transformer model for runoff forecasting in the Jinsha River Basin; average NSE 0.992 across forecast horizons (≈13.8% above the LSTM baseline), with SHAP-based interpretability analysis.
DOI: 10.1007/s11269-025-04241-3How data-driven weather models work — what they do well, where they blur, and what forecasters should trust.
Why physical constraints matter more than raw accuracy — and how to tell genuine skill from climatological mimicry.
Heatwave-to-heavy-rainfall transitions: circulation, air–sea memory, and their predictability.
Since joining the group in 2026, I have been building a physics-guided deep-learning hybrid model for daily-scale drought forecasting in the Jinsha River Basin, extending the multi-source fusion approach of my runoff study to drought. Code and data on GitHub.
I am exploring the QBO as a predictor of extended-range variability, tracing how QBO phases connect to circulation anomalies that influence East Asian climate. Code on GitHub.
I am applying AI weather models to forecasting for Antarctic stations: verification, site-specific correction, and products supporting fieldwork.
2D/3D meteorological geospatial visualization toolkit for Leaflet and Cesium (v0.9.0).
Polar forecasting system for Antarctic stations.
Cryptocurrency market monitoring system.
For inquiries, please contact me via email.
Code & data for the Jinsha River Basin daily drought forecasting project: a physics-guided hybrid model (PyTorch).
PythonOrigin/OriginPro plotting skill for meteorological and hydrological research: 40 scientific palettes, 12 native Origin examples, and a 30-paper evidence base with journal figure recipes and validation tools.
Python
ArcMap scientific cartography skill: a 30-paper design study, 10 native map examples with editable MXD/LYR files, reproducible synthetic data and validation.
Python
Windows GUI toolkit for downloading and organizing climate data: CMIP6, NEX-GDDP-CMIP6, CORDEX, NASA POWER, NOAA CPC and NCEP/NCAR reanalysis, with batch tasks, pause/resume and checksum verification.
PythonPyTorch CNN-LSTM temperature time-series forecasting experiments behind the Intelligent Meteorological Map’s real-time predictions.
PythonInteractive Plotly dashboards for NOAA weather-station extreme-temperature data.
Python