Baitian Liu

Focusing on interdisciplinary research across computer science and meteorology,
exploring innovative AI4Science applications for weather forecasting and efficient decision-making

sonderlau@foxmail.com Max1ndex
Baitian Liu

Education

Nanjing University · School of Atmospheric Sciences · NJUHML Lab Ph.D. Student | Atmospheric Science
Advisor: Prof. Huiling Yuan 2026.06 -- Present
Hangzhou Dianzi University · School of Computer Science M.Eng. Student | Software Engineering
GPA 3.24/4.00 Advisor: Prof. Haiping Zhang 2023.09 -- 2026.06
Hangzhou Dianzi University Information Engineering School · School of Computer Science B.Eng. | Software Engineering
GPA 4.2/5.0 2019.09 -- 2023.06

Experience

Zhejiang Branch, Chinese Academy of Meteorological Sciences Algorithm Development and Research Intern
Hangzhou, Zhejiang 2024.09 -- Present
  • Scope：Based on data resources and expert guidance from the Chinese Academy of Meteorological Sciences, designed and developed deep learning models for precipitation nowcasting, covering data preprocessing, model training, and integration into operational scenarios
  • Outcomes：Developed two precipitation nowcasting models for operational pain points and started integrating them into practical meteorological workflows.
    1. Fast multi-lead precipitation nowcasting model (0--2 h): avoids error accumulation from autoregressive forecasting. With adversarial training and multi-lead information embedding, it outperforms traditional numerical weather prediction while providing higher spatiotemporal resolution (1 km, 6 min).
    2. Extreme precipitation forecasting model (0--1 h): shifts modeling from the numerical domain to the frequency domain and introduces a new architecture and optimization strategy for multi-scale precipitation features. It performs strongly on extreme events, especially preserving details at longer lead times.

Publications & Patents

Selected Projects

NJU-Earth Global Short-to-Medium Range Forecast Foundation Model Core Code Contribution & Model Optimization
NJUHML Lab, Nanjing University 2026.06 -- Present
  • Overview：Participated in the NJU-Earth global deterministic forecast foundation model project at NJUHML Lab, responsible for model training and tuning, online visualization, and daily inference pipeline, supporting AIM-FDP operations
  • Contribution：Built the NJU-Earth training pipeline: CRA data processing, dataset construction, model pre-training, and fine-tuning.
    Established automated workflows covering data processing, model training, inference, and product publishing; built a visualization website displaying daily forecasts, typhoon cases, and verification metrics
  • Outcomes：Participating in the CMA AIM-FDP (AI Model Forecast Demonstration Program, medium-range track). Future plans include extending NJU-Earth to global high-resolution forecasting, regional downscaling, heavy precipitation prediction, and typhoon track/intensity assessment
AI-based Meteorological Forecasting Models Core Algorithm Development and Research
Partner: Zhejiang Meteorological Service 2024.09 -- 2026.02
  • Overview：Collaborated with Zhejiang Meteorological Service to develop deep learning-based meteorological forecasting models for operational forecasting needs, aiming to improve the accuracy and efficiency of precipitation nowcasting
  • Contribution：Responsible for radar data preprocessing, quality control, and completeness checks; designed and optimized deep learning models; developed forecast visualization tools and completed operational scenario evaluations
  • Outcomes：The model outperformed traditional numerical weather prediction during the 2025 Meiyu season evaluation, reaching 1 km and 6 min spatiotemporal resolution with much lower inference time than numerical models, addressing the real-time limitations of short-term forecasting. Research on diffusion-based probabilistic forecasting is ongoing and is expected to further improve heavy-tailed precipitation prediction
Computer Vision-based Bayberry Object Detection Core Algorithm Development and Research
Partners: Zhejiang Meteorological Service, Xianju Meteorological Bureau 2024.12 -- 2025.10
  • Overview：Collaborated with Zhejiang Meteorological Service and Xianju Meteorological Bureau to develop visual algorithms for bayberry maturity detection and deploy them in bayberry orchards
  • Contribution：Responsible for algorithm research, feasibility exploration, feature analysis, and post-processing of the collected dataset
  • Progress：Built an initial YOLO-based object detection model that meets expected recognition performance in open orchard scenarios across multiple seasons. Maturity prediction and large-scale statistical analysis are under development and may support bayberry sales in the Xianju region
Glass Plane Velocity Measurement with Binocular Telecentric Lenses Algorithm Development and Device Integration
Partners: Shanghai CEC Electronic System Engineering Co., Ltd.; Rainbow (Hefei) LCD Glass Co., Ltd. 2023.06 -- 2024.11
  • Overview：Developed a high-precision velocity measurement algorithm for 0.3 mm glass sheets in high-temperature and high-humidity industrial environments, controlling error within 0.6 mm/s
  • Contribution：Led the design and implementation of the velocity measurement algorithm, completed telecentric camera and temperature sensor debugging, designed the UI, and deployed the algorithm on an industrial mobile computing platform
  • Outcomes：Solved recognition challenges caused by glass transparency. Through precise edge localization and clock synchronization, the system achieved the target measurement accuracy. It ran stably at 80°C and 75% relative humidity, and the insulated enclosure plus self-checking functions significantly reduced defect rates in industrial production
Fast and Accurate Measurement of Australian Redclaw Crayfish via Skeleton Recognition Algorithm Development and Device Integration
2023.10 -- 2024.7
  • Overview：Implemented fast and accurate daily growth measurement for crayfish to support and accelerate breeding of new varieties
  • Contribution：Led algorithm design and implementation, including data processing, model training, and supporting device design
  • Outcomes：Used skeleton keypoint recognition together with telecentric lenses to accurately locate and measure crayfish claw length. Combined with a staged detection workflow, the system enabled fast and stable crayfish data acquisition and analysis
YOLO-based Rural Miscellaneous Pile Recognition Algorithm Development Lead
Partner: Yunzhihaihui Information Technology Co., Ltd. 2023.06 -- 2024.12
  • Overview：Developed a YOLOv5-based detection system for rural road trash bins, garbage, and miscellaneous piles, supporting closed-lid and overflow status detection for automated supervision of rural streets
  • Contribution：Responsible for dataset collection and augmentation, YOLOv5 model tuning to improve miscellaneous pile detection accuracy, and development of dedicated modules for trash bin status detection
  • Outcomes：The system achieved over 95% detection rate and 98% accuracy for trash bin closed-lid and overflow detection. It was successfully applied in the digital platform for beautiful countryside governance, significantly improving rural environmental supervision efficiency

About

During my graduate studies, I have focused on AI and software engineering, leading and contributing to multiple interdisciplinary projects from development to deployment. I am experienced in deep learning algorithm design, data processing, and system integration. As a primary contributor to projects in meteorological forecasting, industrial velocity measurement, and environmental monitoring, I helped deliver systems that were successfully applied in real-world scenarios. I have strong independent problem-solving ability and solid teamwork skills.

For my doctoral research, I plan to continue working on weather prediction and AI4Science, applying deep learning to real-world forecasting and scientific problems to produce verifiable, deployable outcomes.