Chromium Isotopes and Cr(VI) in Seawater: Implications for Ultramafic Weathering and CO₂ Variability
Prof. Edward A. Boyle's Group, MIT EAPS
06/2024 - 08/2024 & 07/2025 - 09/2025
- Conducted MC-ICP-MS measurements of Cr concentrations and δ⁵³Cr isotopic compositions in seawater profiles from the tropical Pacific Ocean.
- Performed ion exchange purification, Cr(VI) precipitation with Mg(OH)₂, and sample preparation for high-precision isotope analysis.
- Compiled published data on Cr concentrations/isotopic compositions in rivers, oceans, and weathering systems to test Kent-Muttoni and Jagoutz hypotheses on long-term climate variability.
- Supported NSF proposal development by synthesizing evidence linking ultramafic weathering patterns to atmospheric CO₂ variability and paleoclimate reconstruction.
Remote Sensing Analysis of Policy-Driven Methane Emission Reduction in Rice Cultivation Systems
Prof. Charles Taylor's Group, Harvard Kennedy School
12/2024 - Present
- Developed a web scraping pipeline to systematically collect policy approval data from Chinese provincial government websites, creating a novel county-level dataset on dryland rice seed policies (2008-2024).
- Engineered provincial phenological model in Google Earth Engine to distinguish rain-fed from irrigated rice using multi-parameter indicators (NDVI, LSWI, slope), calibrated against MIRCA-OS reference data.
- Conducted event study analysis using 39,045 county-year observations (2,432 treated, 363 controls) to evaluate causal impact of water-saving policies. Found policies achieved water conservation through technological improvements rather than production reduction (irrigated area: β=-4.09 km², p=0.92).
- Initiated TROPOMI/GOSAT atmospheric methane analysis to assess environmental outcomes of cultivation system transitions.
LiDAR-Augmented Zero-Shot Tree Crown Segmentation Using SAM 2
Prof. Lu Liang's Group, UC Berkeley CED
08/2024 - 02/2025
- Enhanced tree crown delineation by fusing SAM 2 with RGB and LiDAR data, combining bounding boxes and LiDAR point prompts with noise reduction via sampling.
- Demonstrated SAM 2's potential for efficient ITC segmentation in complex forests, with LiDAR significantly boosting precision and recall.
- Paper published in Information Geography. DOI: 10.1016/j.infgeo.2025.100025
Local Climate Zone (LCZ) and Urban Morphology Analysis of U.S. Cities
Prof. Lu Liang's Group, UC Berkeley CED
08/2024 - Present
- Developed a machine learning model for LCZ mapping using LiDAR and satellite imagery.
- Engineered a Python pipeline to process over 500,000+ Google Street View images from 20+ U.S. cities.
- Implemented deep learning to quantify urban canyon characteristics (SVF, vegetation, buildings).
- Validated SVF metrics from Street View with LiDAR data.
- Analyzed correlations between LCZ, urban morphology, and climate data (PM2.5, humidity, temp).
Nationwide Evaluation and Calibration of PurpleAir Temperature Sensors for Urban Thermal Environment Research
Prof. Lu Liang's Group, UC Berkeley CED
05/2025 - Present
- Developed machine learning calibration framework for 98 PurpleAir sensors across 31 U.S. states, integrating 797,744 hourly observations (943 days, 2022-2024) with ERA5 meteorological reanalysis through spatial-temporal matching and quality control.
- Engineered 63 features capturing sensor thermal dynamics: 32 temporal features (lagged variables, rolling statistics, cumulative radiation, thermal streak indicators) and 31 spatial/meteorological features; temporal features reduced error by 29% compared to spatial-only models.
- Implemented temperature-based stratification (cold/normal/hot regimes) achieving test-set MAE of 0.38-0.53°C and RMSE of 0.57-0.74°C, representing 32-51% improvement over unstratified national baseline; outperformed climate-zone stratification in data-sparse regions.
- Evaluated four gradient boosting frameworks (XGBoost, LightGBM, CatBoost, Random Forest) with Bayesian hyperparameter optimization; stratified XGBoost ensemble achieved R²=0.975 for nationwide deployment.
- Conducted SHAP interpretability analysis revealing sensor temperature history, solar radiation accumulation, and humidity dynamics as dominant error drivers, with distinct hierarchies across thermal regimes; findings enable hyperlocal monitoring for heat-health assessment and environmental justice applications.
- Authored research manuscript (under review) and presented poster at NSF NCAR Research Symposium: Human and Geographic Dimensions of Extreme Heat and Heat Risk.
Global Dust Emission Dynamics Under Climate Change and Land Use Management: A Multi-satellite Analysis (2003-2020)
Prof. Minghao Qiu's Group, School of Marine and Atmospheric Sciences and Program in Public Health, Stony Brook University
05/2025 - Present
- Resolved the "greening-dust paradox": naive correlation showed vegetation increase associated with MORE dust (+0.374), but causal analysis revealed strong protective effect after controlling for climate confounders (0.1 NDVI increase reduces AOD by 0.017-0.020 units, 25-30% of mean).
- Integrated 22 years of CAMS and MODIS satellite data (2003-2024) across 1,336 grid cells in northern China and Mongolia, implementing dual-method causal framework: Random Forest counterfactual analysis (R²=0.77) validated with panel regression.
- Conducted mediation analysis revealing 97% of vegetation's dust-reducing effect operates through direct physical mechanisms (canopy interception, surface stabilization) rather than indirect pathways, implying rapid policy benefits without decades of ecosystem development.
- Identified 35-fold spatial heterogeneity: barren lands show 3-6× stronger effects than grasslands/croplands; core dust regions exhibit 35× stronger effects than non-source areas, providing evidence for strategic targeting.
- Demonstrated policy implications: strategic allocation toward high-impact zones (barren lands in core dust regions) could increase aggregate dust reduction by 40-60% with the same $3-5 billion annual budget, offering cost-effective alternatives to current uniform policies.
- Validated causality through spatial placebo tests (p<0.001) and temporal stability analysis across 22 years (trend p=0.446), supporting external validity and policy persistence.
Chinese Think Tank Project: Digital Technology Industry's Role in Low Carbon Development - Key Strategies and Drivers
Prof. Xin Tian's Group, School of Environment, Beijing Normal University
08/2023 - 12/2023
- Applied Leontief input-output model and Structural Path Analysis to quantify carbon emissions across China's digital technology supply chain.
- Evaluated ICT sector inter-dependencies using backward/forward linkage analysis and influence coefficients.
- Used Structural Path Analysis (SPA) to identify how various economic activities impact carbon emissions and employed Structural Decomposition Analysis (SDA) to break down emission intensity into distinct factors.
- Co-authored 47-page policy report published by China Mobile Corporation.
Spatio-Temporal Optimization Modeling for National Photovoltaic Infrastructure Deployment
Prof. Xin Tian's Group, School of Environment, Beijing Normal University
06/2025 - Present
- Developed multi-period dynamic optimization model for 50-year PV deployment strategy across China, minimizing life-cycle costs using operations research framework.
- Integrated GIS, meteorological, and economic data to quantify cost function (CAPEX, O&M, transmission, decommissioning, ecological opportunity costs).
- Employed CatBoost for data imputation and time-series forecasting of demand and cost reduction curves.
Ecological surprises and bioremediation potential of oil pollution in coastal ecosystems: A review
Prof. Junhong Bai's Group, School of Environment, Beijing Normal University
05/2024 - 01/2025
- Conducted comprehensive literature review of bioremediation methods (phytoremediation, microbial, nanotechnology) for petroleum contamination in mangroves, salt marshes, seagrasses.
- Highlighted advanced approaches including omics technologies, engineered microbes, and real-time monitoring systems while addressing challenges such as salinity, low oxygen conditions, and nutrient limitations.
Spatial distribution, source identification, and potential risks of 14 bisphenol analogues in soil from different functional zones in Chengdu, China
Prof. Zhi Li's Group, College of Architecture & Environment, Sichuan University
02/2023 - 02/2024
- Analyzed spatial distribution of 14 bisphenol compounds across functional zones in Chengdu using 376 dust samples.
- Applied normality, correlation, and Mann-Whitney U tests for statistical analysis.
Polyethylene microplastics hamper aged biochar's potential in mitigating greenhouse gas emissions
Prof. Ke Sun's Group, School of Environment, Beijing Normal University
12/2023 - 01/2025
- Investigated interactive effects of PE microplastics and decadal biochar on CO₂/N₂O emissions through incubation and field experiments.
- Revealed microplastics reduced biochar's mitigation effectiveness by 44-82% via altered soil structure and microbial activity.
Fourteen-year field evidence reveals superior co-benefits of biochar in immobilizing heavy metals and sequestering carbon
Prof. Ke Sun's Group, School of Environment, Beijing Normal University
12/2024 - 02/2026
- Analyzed 376 soil samples from a 14-year field experiment (2007-2021) to quantify long-term effects of biochar vs. straw amendments on heavy metal dynamics in agricultural soils.
- Performed comprehensive data analyses including DGT bioavailability measurements for 6 heavy metals (Cd, Pb, Ni, Co, Cu, Mo) and BCR sequential extraction for speciation analysis.
- Applied advanced statistical modeling (PLS-SEM, VPA) on over 40 soil physicochemical and microbial parameters to deconvolve regulatory pathways, identifying that microbial properties explained 30% of metal bioavailability variance while physicochemical factors governed 12% of speciation shifts.
- Co-developed a novel Carbon-Metal Coupling Index using entropy weight method, demonstrating high-dosage biochar achieved 0.703 coupling score (vs. 0.361 for low-dosage biochar and 0.396 for straw), representing 94% improvement in synergistic benefits.
