Research

Research

My research focuses on improving the representation of land-atmosphere interactions in land surface models, with emphasis on water, carbon, and energy cycle processes. I use the Community Land Model (CLM5) as a framework to conduct perturbed parameter ensemble experiments, drawing on NEON flux tower observations to quantify model uncertainty and identify key process sensitivities. This work seeks to improve how models represent vegetation-climate feedbacks and to advance predictions of water and carbon cycle dynamics under changing climate and land use.

A central component of my research is testing how parameter choices influence model performance across scales. By combining ensemble modeling with site-level observations, I evaluate how well models capture processes such as evapotranspiration, soil moisture dynamics, and energy balance closure. The ultimate goal is to refine process representation in land surface models so they better align with observations and can more reliably inform climate projections.

In addition to process-based modeling, I explore applications of machine learning to environmental systems. This includes using data-driven methods to analyze and predict hydrological and climate variability, such as soil moisture and salinity dynamics. These approaches complement mechanistic models by offering alternative perspectives on system behavior and by highlighting nonlinearities or drivers that may be underrepresented in traditional models.

My broader motivation is to bridge observations, models, and data-driven approaches to improve our understanding of land–climate interactions and to provide insights that support sustainable water and ecosystem management in a changing environment.

Methods

• Perturbed Parameter Ensembles (PPEs): I conduct large ensembles of CLM5 simulations, systematically varying parameters to assess their influence on water, carbon, and energy fluxes. This framework helps identify the most sensitive processes and quantify model uncertainty.

• Observational Integration (NEON Flux Towers): I use eddy-covariance data from NEON sites to evaluate model behavior, test process representation, and assess energy and water balance closure at site and regional scales.

• Uncertainty Quantification: I employ ensemble statistics, sensitivity analysis, and emulators to assess the robustness of model predictions and to inform parameter optimization strategies.

• Machine Learning Applications: I apply methods such as recurrent and residual neural networks to environmental time series, testing their ability to capture hydrological and climate-driven variability. These data-driven approaches complement process models by providing alternative perspectives on system dynamics.

Overall, my research is driven by a desire to deepen our understanding of the Earth’s climate system and to develop effective strategies for mitigating and adapting to the impacts of changing environment.