Talks and presentations

From Predictions to Patterns with AI: A Differentiable SPARROW Framework for Improved Water Quality Prediction and Attribution

December 17, 2025

Conference Presentation, AGU2025dPL, New Orleans, LA

Nutrient loss is a major contributor to downstream water quality degradation, including the Gulf hypoxia. Effective nutrient management in upstream watersheds requires models that can identify both the magnitude and spatial distribution of pollutant sources and controlling factors. Traditional modeling efforts often assume spatially homogeneous parameters, which may ignore important regional variability. In this study, we developed an artificial intelligence (AI) framework that integrates SPARROW (SPAtially Referenced Regressions On Watershed attributes) with differentiable parameter learning and deep learning techniques. The differentiable Parameter-Learning SPARROW model (dPL-SPARROW) was used to assess whether key source parameters vary spatially and what landscape features drive this variation. The model was trained to estimate NO23 loads across 104 monitoring sites in the Upper Mississippi River Basin from 2001 to 2020.

Spatiotemporal Changes and Driving Factors of Riverine Nutrient Export at HUC12 Scale in the Mississippi/Atchafalaya River Basin

December 15, 2025

Conference Presentation, AGU2025MARB, New Orleans, LA

Effective management of nutrient pollution requires a detailed understanding of how nutrient export evolves across spatial and temporal scales, particularly in response to hydrological variability and anthropogenic activity. In this study, we developed a high-resolution SPARROW (SPAtially Referenced Regressions On Watershed attributes) model for the Mississippi/Atchafalaya River Basin (MARB) to quantify annual total nitrogen (TN) and total phosphorus (TP) exports at the HUC12 scale from 1980 to 2020. Model calibration was conducted using nutrient loads estimated with WRTDS-K at monitoring stations throughout the basin, enabling source attribution analysis. To examine long-term trends, we compared SPARROW-estimated nutrient loads and yields across three periods: 1980–1996, 2001–2010, and 2011–2020. A bootstrap-based scenario analysis was employed to isolate anthropogenic and hydrological influences on nutrient yield changes. Results show widespread increases in local incremental TN and TP yields across all HUC2 sub-basins in the MARB. Within the MARB, TN contributions from agricultural sources rose from 40.4% to 43.8%, while urban contributions increased from 20.9% to 26.2%. TN yields increased by 0.45 and 1.67 kg-N/ha/yr for 2001-2020 and 2011–2020 compared to the 1980–1996 baseline, with hydrologic changes explaining over 50% of the increase. The Conservation Reserve Program may have contributed to reducing TN land-to-water delivery. For TP, within the MARB, urban sources rose from 30.4% to 34.9%, while agricultural contributions fluctuated between 30% to 35% of the TP load. TP yields increased by 0.12 and 0.27 kg-P/ha/yr for 2001–2010 and 2011–2020, with hydrology accounting for over half of the increase. Among anthropogenic drivers, urban-related sources, including wastewater treatment plant discharge and urban nonpoint sources, were the dominant contributors to TP export.

How Do Anthropogenic Activity and Hydrological Variability Control the Spatiotemporal Patterns of Nitrogen and Phosphorus Export in the Mississippi/Atchafalaya River Basin?

May 22, 2025

Conference Presentation, SFS2025, San Juan, PR

Excessive nutrient exports from the Mississippi/Atchafalaya River Basin (MARB) significantly degrade water quality and lead to the hypoxia zone in the Gulf of Mexico. Quantifying the sources, fate, and transport of nutrients from headwaters to large rivers is critical to diagnose the trend of exported nutrient load and guide efficient and effective conservation planning for nutrient loss reduction. This study investigates the changes in riverine nutrient loads and yields at the HUC12 scale in the MARB from 2001 to 2020. The WRTDS (Weighted Regressions on Time, Discharge and Season) and the SPARROW (SPAtially Referenced Regressions On Watershed attributes) models are integrated to quantify the spatiotemporal patterns and driving factors of riverine nutrient export from the MARB. The integration of the WRTDS model and the SPARROW model allows us to identify and quantify contributions from various sources and examine driving factors for both spatial and temporal variations in nutrient dynamics throughout the MARB. This high-resolution analysis at the HUC12 scale identifies nutrient export hotspots and differentiates areas where nutrient export increases are attributed primarily to anthropogenic activities or hydrological variability. By providing detailed assessments of nutrient source contributions and transport mechanisms, this study offers essential insights for water quality management within the MARB. The identification of local hotspot areas with high and increasing nutrient yields, along with their driving factors at the HUC12 scale, provides crucial local context for targeted conservation strategies to reduce riverine nutrient load in the MARB.

Spatio-temporal Changes and Driving Factors of Riverine Nitrogen Export in an Agriculture-dominated Watershed: the Illinois River Basin

December 13, 2024

Conference Presentation, AGU2024, Washington, DC

Riverine nutrient load from the Mississippi/Atchafalaya River Basin (MARB) plays a crucial role in the development of the Gulf of Mexico’s hypoxic zone. Agricultural land in the MARB predominately contributes to nitrogen (N) and phosphorus (P) loading in the river system. Quantifying the sources, fate, and transport of nutrients from headwaters to large river basins is critical to diagnose the trend of exported nutrient load and guide conservation planning for nutrient loss reduction. Here taking the Illinois River Basin as an example, this study aims to quantify the changes in nitrate and nitrite (NO3+NO2) loads and yields at HUC12 scale from 2001-2020. We analyzed riverine loads and yields of nitrate and nitrite at 40 USGS gauge stations and simulated high-resolution nitrate and nitrite export throughout the Illinois River Basin using the SPARROW (SPAtially Referenced Regressions On Watershed attributes) model for two time periods (2001-2005, and 2016-2020) to quantify the spatiotemporal pattern and driving factors of changes in nitrogen export from the Illinois River Basin. We found that the five-year averaged total loads of nitrate and nitrite in the Illinois River Basin increased by 28% from 2001 to 2020, along with a 47% increase of discharge flow. Our analysis also revealed a complex spatial pattern of incremental yield of nitrate and nitrite: decreasing yields in the upper Illinois River contrasted with increasing yields in the lower Illinois and Kankakee Rivers. The simulation from the SPARROW model allowed us to identify and quantify contributions from various sources and examine driving factors for the observed spatial and temporal variations in nitrate and nitrite dynamics within the basin. By identifying areas of increasing and decreasing yields, as well as the factors driving these changes, our study provides crucial information for targeted conservation strategies to reduce nutrient load in the Illinois River Basin.

Forecasting Alpha Return from 8-K reports via deep learning

March 25, 2022

Conference Presentation, SDSC, Waco, Texas

In this study, we propose a novel machine learning approach with dimension reduction stacking to forecast the stock returns of the SP 500 companies by min-ing their SEC 8K reports via different NLP techniques. The proposed method achieves better forecasting performance than the peer methods and can be used as a concrete technology in security return analysis in practice.