Lingling Zhang

Mitacs Globalink Research Internship, University of Sherbrooke, Quebec, Canada

Supervised by Professor Marie-Amélie Boucher

• Integrated LSTM into hydrological state variable assimilation framework to enhance runoff prediction accuracy
• Developed the RavenModel class to manage Raven executable operations, manipulate state variables, and adjust initial conditions, enabling seamless coupling between physical hydrological models and deep learning models
• Constructed training datasets from meteorological forcings and Raven open-loop outputs, trained LSTM to capture dynamic meteorological-hydrological relationships, and applied assimilation corrections using observed runoff data
• Reimplemented TensorFlow codebase in PyTorch, improving robustness and scalability while ensuring cross-platform compatibility across Windows, Linux, and macOS
• Extended the assimilation framework to support multiple hydrological models (HMETS, GR4J, HBV-EC), expanded spatial applicability across different climate conditions, and provided modular, platform-independent code for future research

It was an unforgettable journey working as a Mitacs Global Research intern in Professor Marie-Amélie's lab this summer. Her creativity and passion for hydrology inspired me greatly. During my internship, I also worked closely with Francis Lapointe, who is pursuing his doctorate under her supervision. He is a cheerful person who taught me and helped me tremendously throughout my internship.

First Author, Key Laboratory of GIS, Ministry of Education, ECNU

Supervised by Professor Hongkai Gao and Professor Xi Chen

• Collected and integrated multi-source data including USGS runoff, SMAP soil moisture, NLDAS meteorological forcing, and MODIS vegetation indices using Google Earth Engine and Python
• Built an Entity-Aware LSTM (EA-LSTM) model with static variable input layer to incorporate meteorological time series and geographic features, with a feature fusion layer to integrate temporal and static features
• Designed comparative experiments to analyze the impact of soil moisture on discharge prediction using Nash-Sutcliffe Efficiency (NSE), iteratively adjusted model architecture and parameters
• Optimized input variables by incorporating soil moisture and vegetation index based on feature importance analysis, improving NSE from 0.408 to 0.640
• Completed a 5,000-word academic paper submitted to Geophysical Research Letters

National College Students' Innovative Entrepreneurial Training Program, Core Member

• Developed and implemented an unequal-weight ensemble optimization algorithm in Python for multiple high-resolution global climate models in HighResMIP
• Created a ResNet-18-LSTM deep learning framework for spatiotemporal data analysis and prediction of extreme heat events
• Designed network architecture incorporating feature extraction, temporal processing, and spatial reconstruction to handle 8 time-step features
• Trained model for 100 epochs using AdamW optimizer and SmoothL1Loss, achieving 0.0025 loss and 0.0706 RMSE on validation set
• Completed predictions of compound extreme heat and heatwave events in China from 2015 to 2050, revealing increasing trends

College Students' Innovative Entrepreneurial Training Program, Core Member

• Collected high-resolution remote sensing images of the Yangtze River Delta and preprocessed data using Python OpenCV
• Built a YOLOv7 rotated object detection framework, annotated four types of stadiums, and trained model to automatically extract precise locations and rotated bounding boxes
• Applied path planning combined with Gaussian two-step floating catchment area method using 15-minute cycling radius to quantify accessibility
• Identified accessibility-deficient areas and proposed targeted facility optimization plans

Course Project

• Developed Python-based housing price prediction system integrating 90,000+ property listings and POI geospatial data
• Designed web crawlers, SQLite databases, handled outliers, engineered spatial and temporal features, and optimized deep learning models
• Implemented multi-dimensional feature engineering and enhanced Deep & Cross Network model
• Achieved accurate housing price prediction with RMSE of 0.0117, improved data processing efficiency by 40%
• Repository: GeoDeepHouse

Course Project (Awarded as Excellent Course Assignment)

• Integrated data from 402 medical institutions in Shanghai and OSM road network to optimize spatial distribution of medical resources
• Applied spatial pattern analysis and Enhanced Two-Step Floating Catchment Area (E2SFCA) method for accessibility assessment
• Designed three-layer GAT model with multi-task learning framework, achieving ROC-AUC of 0.82 on test set
• Proposed location plan for 30 new facilities, improving accessibility by 11.7-17.4% and reducing spatial Gini coefficient by 0.013
• Repository: MedicalLocationOptimization