Building cross-scale predictability of land-to-aquatic nitrogen loads in agriculture-dominated watersheds

Aug 2021


Modeling assessment is becoming increasingly important to developing quantitative insights into the processes governing nutrient exports from land to water bodies, and to making predictions about future conditions. However, the efficacy and attribution capacity (i.e., “get right answer for right reasons”) of the existing water quality models is uncertain owing to limited model testing against long-term observations of nutrient concentration, flow and loading within a hydrologic hierarchy system.


Researchers propose to develop a data-model integration framework to identify and quantify the issues leading to model incapability in accurately estimating N delivery across scales.


In this project, researchers will test and improve a process-based hydro-ecological model to simulate in-stream N transport and decay processes and benchmark the model’s performance using long-term water quality monitoring data at the delivery scale, along ditches and stream channels downstream of the outlets of subsurface drainage networks, and at the scale of a United State Geological Survey Hydrologic Unit Code 12 (HUC 12) watershed (~5000 ha). This work will be conducted in the HUC 12 Walnut Creek watershed located south of Ames, Iowa. The Walnut Creek watershed has been subject to extensive monitoring of nutrient loads and flows over decades at various scales, at multiple locations along the mainstem of Walnut Creek and some of its tributaries. In this project, researchers will employ a portion of the historical flow and nutrient concentration time-series monitoring data to improve and calibrate the model and validate it with the remainder. Then, the improved model will be applied in the entire watershed to quantify the relative contributions of in-field and edge-of-field management practices, and in-stream processes on nitrate-N reductions observed at the watershed outlet under current climate conditions. Finally, researchers will identify “hot-spot” areas within the watershed that are prone to losing excessive quantities of N through hydrological flushing.

Project Updates

Note: Project reports published on the INRC website are often revised from researchers' original reports to increase consistency.

July 2023

Due to limited monitoring data in the past two years, we applied for a one-year, no cost extension for this project. With the data accumulation in 2023, we start compiling the existing data to inform the model testing and improvement. On the modeling side, we continue applying the improved model to investigate the impacts of in-field practices on reducing N loading.

Across the four tested catchments (RS, KS, LP, WW), We conducted two simulation experiments. The first one reflects our best-estimates of nitrate loading at the catchment level, considering historical factors like daily weather, CO2 concentration, N deposition, land use changes and agricultural practices (crop rotation, fertilizer use, cover crop planting, tillage and tile-drainage), based on farmer surveys. According to field surveys, five out of 19 fields in RS and five out of nine fields in LP had cover crops, while KS and WW had none. This information was used to drive the first model simulation. The second experiment assumes ideal conditions with cover crops planted after row crops in each field every year. The model calculates the impact of cover crops on nitrate loading by comparing these two experiments. We find that the impact of cover crops on reducing nitrate loads varied across catchments and years. Cover crop effects were estimated to be 5.3% (-2.6% - 11%, range from first to third quartile of annual estimates) in KS, 6.4% (4.7% - 6.8%) in LP, 18% (13% - 21%) in RS, and 12% (6.1% - 17%) in WW.

The actual effectiveness of cover crops in reducing nitrate loading is likely higher in RS and LP since our estimates only account for the additional potential we can gain from existing cover crop planting to full adoption in the catchment. Generally, greater reductions in nitrate loads were observed in wet years compared to dry years, while changes in corn planting area had minimal impact on nitrate load reduction."

Publications submitted:

Cao, P., C. Lu, W. Crumpton, M. Helmers, D. Green, G. Stenback. Improving model capability in simulating spatiotemporal heterogeneity and hydrological transport of nitrate export in tile-drained catchments. In revision in Water Research Zhang, J., C. Lu, H. Feng, R. Miao, D. Hennessy, S. Pan, H. Tian. Do we trade water quality for crop production?: Unraveling the relationships between crop production and riverine nitrogen loading. In revision in Environmental Research Letters.


December 2022

By using the catchment-level water quality monitoring data, we developed and tested a tool that predicts event-scale water and nutrient movement, enhancing our understanding of how rainfall affects runoff and the connection between nitrate levels and water flow (c-Q relationships). We identified key factors (such as precipitation intensify, dry period length), and turning points of chemo-static to chemo-dynamic shift, which will guide the enhancement of process-based models. The event-scale stacked random forest model is a robust machine learning method that works well for understanding complex relationships in data, both for predicting and categorizing outcomes. We divided daily monitoring data into two periods for training and testing (3 years and 2 years). Alongside total event rainfall, important factors included event length, extreme rainfall occurrences, time without rain before an event, and the season. In initial results from all catchments, the model performed strongly in simulating peak and total discharge (r_peak= 0.85, NSE_peak= 0.71, r_total= 0.84, NSE_total= 0.71). Our combined model also forecasted nitrate concentration changes after rain events as categories (increase, decrease, no-change) with an accuracy of 54%. We will test this approach with more data from additional catchments and from monitoring stations at the main river stems in the Walnut Creek Watershed.


Cao, P., C. Lu, W. Crumpton, M. Helmers, D. Green, G. Stenback. Improving model capability in simulating spatiotemporal heterogeneity and hydrological transport of nitrate export in tile-drained catchments. In review in Water Research Zhang, J., C. Lu, H. Feng, R. Miao, D. Hennessy, S. Pan, H. Tian. Do we trade water quality for crop production?: Unraveling the relationships between crop production and riverine nitrogen loading. In review in Environmental Research Letters.

Proposals submitted:

Co-PI to INRC, “Spatially delineated carbon credit potential and implied nutrient reduction co-benefit: An assessment with integrated ecological and economic modeling framework” $97,690 budgeted PI to NSF, "A Physics-Informed Flood Early Warning System for Agricultural Watersheds with Explainable Deep Learning and Process-Based Modeling” , $700,000 budgeted Co-PI to USDA “Quantifying And Mitigating Indirect Greenhouse Gas Emissions From Tile-Drained Agricultural Catchments” $650,000 budgeted Co-PI to NSF, “REU Site: Summer Research Experiences in Agricultural Nutrient Management” $210,000 budgeted Co-PI to USDA “Women in Agriculture Adopting Climate-Smart Practices” $4.9M budgeted

Total requested - $ 6,560,000.

Other accomplishments:

Peiyu Cao, “Tracking agricultural nitrogen fertilizer use and nitrogen loss pathways across scales through data synthesis and modeling approach.” Iowa State University, PhD dissertation. 2022.

July 2022

We finalized the hydro-ecological model at the catchment level (DLEM-catchment) and assessed its performance across four catchments that contribute to first-order streams, encompassing drainage areas ranging from 2.1 to 4.6 km². Land use patterns were comparable across all four catchments, with over 80% of the land area utilized for corn-soybean (CS) and continuous corn (CC) cropping systems. These four catchments offer a representative depiction of tiled-drained areas both with and without prairie potholes. The DLEM-catchment is designed to better mirror real-world management practices and unique characteristics, allowing it to replicate how water and nitrogen (N) move in farmed potholes and tile lines. We used historical weather data, information about field practices, and other inputs to make the model work. We then compared the model's results with actual daily measurements of water flow and nitrate-N (NO3-N) movement from 2015 to 2019. We ran the model at a 30-meter by 30-meter resolution, using detailed input data covering the years 2000 to 2019. The simulations for the catchments demonstrated that our improved model accurately recreated the day-to-day and month-to-month patterns of water flow and NO3-N levels observed between 2015 and 2019 in all four catchments (Cao, 2022, PhD dissertation).


J. Zhang, Lu, C, W. Crumpton, C. Jones, H. Tian, G. Villarini, K. Schilling, D. Green. 2022. Nitrogen loading to the Gulf of Mexico will increase due to intensifying extreme precipitation during the 21st Century. Earth’s Future.

Proposals submitted:

PI on proposal to ISU PIRI, “Revolutionizing Agricultural Data Management and Utilization through a Blockchain-enabled “Information to Decision to Action” Platform” Total requested budget: $750,000.

Other accomplishments:

As the primary organizer and convener, the PI Lu has led a session at the 2021 AGU fall meeting, entitled “Data and Modeling Advances Toward Sustainable Food-Energy-Water Development on a Manageable Scale.” The research results from this project were presented at the 2021 AGU fall meeting: 2021 Peiyu Cao, Chaoqun Lu, MJ Helmers, and WG Crumpton. Modeling nutrient reduction strategies to reduce nitrate loading in tile-drained agricultural catchments. 2021 AGU Fall Meeting, New Orleans, Louisiana (poster presentation).

December 2021

Due to dry weather, we don’t get sufficient monitoring data across the stations in the Walnut Creek Watershed this year. The primary efforts have been focused on historical data compilation and some data-driven analysis at the catchment level. We developed a deep learning (DL) model that incorporates outputs from a process-based model (DLEM-Dynamic Land Ecosystem Model) for the prediction of river discharge. Our initial findings indicate that this hybrid model outperforms the standalone DL model. This supports that process-based models can offer valuable guidance to DL models based on domain-specific principles, thereby enhancing prediction accuracy. The outcomes further affirm that the hybrid methodology, encompassing domain-specific knowledge integration during pre-processing and training stages, enhances the DL model's learning (with NSE from 0.4 to 0.6), as demonstrated by Sarkar et al. in 2021. We plan to test this approach using the catchment-level monitoring data before receiving more cross-scale water quality data. Moreover, we set up a series of experiments using DLEM to examine the effects of climate and human activities on crop production and N loading in the RS and KS catchment (Story County, Iowa). The factors that were considered are climate, N deposition, N fertilizer, land use and land change, and manure N. The contribution of each factor was obtained by subtracting the corresponding experiment from All drivers. The net change of all factors was calculated as the difference between All drivers and Base line. We find that Climate was the dominant driver controlling N loading in RS over 20 years, while human activities, including N fertilizer, manure N, and land use change, played the same important role as climate in determining variations in N loading in KS. In some years such as 2014 and 2017, human activities even turned into the dominant drivers. Our simulations indicate that adopting better management practices has a higher potential in reducing N loading in KS (Cao et al., 2021).

Project Activities:

- 1 presentation


A. Sarkar, J. Zhang, C. Lu, A. Jannesari. 2021. Transfer Learning Approaches for Knowledge Discovery in Grid-based Geo-Spatiotemporal Data. arXiv preprint arXiv:2110.00841.

Proposals submitted:

Co-PI to NSF, REU Site: Summer Research Experiences in Agricultural Nutrient Management, ~$210, 000.