When
Where
UPDATE -- Speakers and topics have been announced!
Nathan Strom: Targeted Surrogate Modeling for Groundwater Recharge Feasibility Assessment Using HYDRUS-1DAssessing groundwater recharge feasibility through surface ponding across varied landscapes requires simulating a wide range of surface and subsurface conditions, including soil, vegetation, and hydrologic dynamics, often using numerical models like HYDRUS-1D. However, fully exploring this high-dimensional parameter space is computationally intensive and often inefficient. This talk introduces the concept of a targeted surrogate modeling approach designed to improve predictive accuracy within regions of the parameter space that are most relevant for informing recharge feasibility, specifically, where surface ponding produces recharge that is both possible and sufficient. By using adaptive sampling to focus on transition zones in the parameter space, where the likelihood of recharge shifts from negligible to significant, the method supports faster and more accurate assessments of recharge potential across soil, plant, and ponding condition configurations that are most informative for practical decision-making and scenario evaluation. Patrick Neri: Plant ESM Complexity - How simplifying complex problems may still require complex solutionsIn earth system models (ESMs), the dynamics of plant carbon assimilation and respiration are diverse and complex. Different life-styles, such as evergreen or deciduous, are considered differently in response to the climate of the seasons, water availability, and leaf characteristics. As ESMs are usually computationally expensive to run, optimizing the various parameters that influence all these factors to better match observations is a frequent challenge in the field. One approach that is used to tackle this challenge is the creation of surrogate models. These models are trained on the output of ESMs, in order to statistically derive a prediction of said output as a function of model inputs and used parameters. These surrogates, which can range from multi-linear regressions to advanged deep learning networks, all have the goal of lowering the computational complexity of the task, speeding up the search for optimal parameters. In this brief talk, I will outline my ongoing attempt at constructing an example of a surrogate model, one which tries to separate each plant functional type (PFT) into its own surrogate, with the goal of recombining their predictions to compare and optimize against community-level measurements. I will highlight the challenges of considering only GPP, as well as others when considering GPP & LAI. Finally, I will demonstrate the success (and failure) of such a system when applied to multiple site observations at the same time. |
HASSA will be hosting the next brown bag lunch event on April 21 at 1pm.
We encourage all students to sign up to get experience presenting research or other topics in front of peers and faculty! Speaker names and topics coming soon!
Reach out to any of our officers or email us at uahassa@gmail.com to grab a presentation spot.
¤
Cheers, HASSA Officers
Tyler Maio - President
Jess Meyer - Vice President
Eden Harper - Treasurer
Ander Ortiz- Outreach Chair
Brandolyn Baeza - Social Chair
Natalie Yurek - Secretary
Jack Flanigan - Undergraduate Representative