Before the Streams: Modeling the Effectiveness of In-Field and Edge-of-Field Practices in Reducing Nitrogen Loads
Issue
A wide variety of in-field and edge-of-field practices have been developed and implemented to reduce nitrogen (N) loads in Iowa and other Corn Belt states. Despite progress in both measurement and modeling, one of the bottlenecks in obtaining reliable modeling estimates is improper scaling up from field-level measurements to the watershed scale.
Objective
This project focuses on scaling up monitoring data at delivery scale and riparian buffers from a single catchment to the entire Bear Creek Watershed. The research team will assess and predict the relative contributions of in-field and edge-of- field practices to reduce NO3 loads. This will reduce scaling uncertainty in water quality models, and promote accurate modeling assessments and predictions of nutrient reduction practices across agricultural landscapes.
Approach
Some team members were involved in the earlier development of a coupled hydrological-biogeochemical model, Dynamic Land Ecosystem Model (DLEM). In this model, river network and in-stream processes have been incorporated into a terrestrial modeling scheme, and the design of grid-to-grid connection allows researchers to track lateral water flow and nutrient movement from land to aquatic systems. In this project, model performance will be tested against the long-term monitoring of water flow and NO3 loads in the tile outlets, diverted into the buffers and NO3 concentration changes within the buffers. The in-situ measurements in crop fields, at delivery scales, in tiles, and riparian buffers will constitute long-term benchmark data along hydrological flow routing for the team to calibrate and validate DLEM’s estimates of N dynamics before entering streams. The improved model will distinguish and quantify the impacts of individual and combined in-field and edge-of-field practices on N load reduction.
Project Updates
Note: Project reports published on the INRC website are often revised from researchers' original reports to increase consistency.
March 2023
FINAL REPORT
Key Research Questions
1) How have N loads at delivery scale varied among catchment?
2) Whether and how can we represent the key processes regulating dynamics water and nutrient export from tile-drained agriculture dominant catchment in process-based modeling?
3) Can we quantify the effectiveness of in-field and edge-of-field N reduction practices through modeling approaches?
Research Findings
We developed a catchment-level hydro-ecological model and tested it at four catchments that drain into first-order streams with drainage areas ranging from 2.1 to 4.6 km2. Two out of four catchments are in the Des Moines Lobe region of Iowa (DML-IA) with dense prairie potholes, and as such each catchment features multiple, drained depressional wetlands. In contrast, the other two catchments are situated outside of the DML-IA, and do not include drained depressional wetlands. Land use was similar in the four catchments with more than 80% of the area used for corn-soybean (CS) and continuous corn (CC) cropping systems. These four catchments are well representative of tile-drained catchments with and without prairie potholes, and featured with a variability of climate, catchment characteristics, and agricultural management, which makes them especially suitable to develop and evaluate model predictions regarding water discharge and nitrate export. • Using historical weather conditions, field-level practice information, and other input drivers to force the model, we confronted the modeled results with daily observations of water discharge and NO3-N export from 2015 to 2019. Model simulations were conducted at a 30-m by 30-m resolution, in which the time-series gridded input driver data were developed to drive the model at the same resolution over the period 2000 to 2019. The catchment-level simulations show that the improved model well reproduced the daily and monthly dynamics of water discharge and NO3-N concentration and loading measured from 2015 to 2019 in all four catchments (Cao et al., in review). The model estimation shows that subsurface flow (tile flow + lateral flow) dominates the discharge (70%-75%) and NO3-N loading (77%-82%) over the years. However, the contributions of tile drainage and lateral flow vary remarkably among catchments due to different tile-drained area percentages and the presence of farmed potholes. Furthermore, we found that agricultural management (e.g. tillage and fertilizer management) and catchment characteristics (e.g. soil properties, farmed potholes, and tile drainage) play an unignorable role in predicting the spatial distributions of NO3-N leaching and loading. The simulated results reveal that the model improvements in representing water retention capacity (snow processes, soil roughness, and farmed potholes) and tile drainage integratively improved model performance in estimating discharge and NO3-N export amount at daily timestep, while improvement of agricultural management mainly impacts the NO3-N export prediction. • Our simulations show that grid cells that have high-level N input rates and tile drainage demonstrate the highest NO3-N leaching rates. Interestingly, farmed potholes with tile installed underneath could intercept NO3-N in surface runoffs that route from the neighboring grid cells, forming hotspots of NO3-N leaching in the areas with potholes. Meanwhile, by connecting and delivering local NO3-N leached from soils, the DLEM-catchment is capable to track the NO¬3¬-N loading dynamics along the flow pathway, which reflects the net changes between NO3-N leaching, transport and removal through the flow routing network over a given time period. For example, in-flow NO3-N loads are slowly accumulated in central RS (< 1.5 Mg N yr-1, 1Mg = 106 g) due to general low NO3-N leaching rates (< 5 g N m2 yr-1), whereas NO3-N loads exceeding 1.5 Mg N yr-1 from relatively smaller contributing areas are extensively found in KS. This implies that the prioritized catchments might be sought to cleaning water and reducing N footprint from agricultural landscape. • Using the improved DLEM-catchment model, we conducted four scenario simulations to evaluate how alternative tile drainage installation plans could affect water discharge and N loading at the delivery scale. The four scenarios include: 1) Original setting as we know in one of the tested Iowan catchments (original), 2) Increase50%: Increase the tile space from 20 m to 30 m, which decreases tile drainage intensity, 3) Decrease50%: Decrease the tile space from 20 m to 10 m, which increases tile drainage intensity, and 4) No Tile: All grids are set to non-tiled. . The model-estimated differences in water and N movement among scenarios are quantified. • We are also developing a water quality prediction tool at an event-scale based on a dynamic time window defined by consecutive days of rain. Using delivery-scale water quality monitoring data from 4 Iowa catchments (drainage area ranging from 2.1 – 4.6 km2), we trained and tested stacked random forest models. The event-scale stacked random forest model is a robust machine learning approach that can be used for regression and classification in nonlinear relationships. Daily data was split sequentially for training and testing (3 years, 2 years). In addition to total event precipitation, some of the variables identified as important factors include event duration, amount of extreme precipitation days, the length of the dry period before the event, and the season. Initial results across all catchments show good performance for simulating peak and total discharge (r_peak= 0.85, NSE_peak= 0.71, r_total= 0.84, NSE_total= 0.71). Our ensemble model also predicts the nitrate concentration change following rainfall events as a categorical variable (increase, decrease, no-change) with 54% accuracy (Calderon et al., 2023).
Project Activities
- 10 presentations
- 1 workshop
Publications / Journal Articles
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. https://doi.org/10.1029/2021EF002141 2. Lu, C, J. Zhang, H. Tian, W. Crumpton, M. Helmers, W. Cai, C. Hopkinson, S. Lohrenz. 2020. Increased extreme precipitation challenges nitrogen load reduction to the Gulf of Mexico. Communications Earth & Environment, https://doi.org/10.1038/s43247-020-00020-7
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.
Leveraged Dollars
$2,535,000
Other Accomplishments
Thesis/Dissertations:
Peiyu Cao, “Tracking agricultural nitrogen fertilizer use and nitrogen loss pathways across scales through data synthesis and modeling approach.” Iowa State University, 2022, PhD dissertation
Aishwarya Sarkar, "Transfer learning approaches for knowledge discovery in grid-based geo-spatiotemporal data." Iowa State University, 2022, MS thesis
December 2022
Based upon DLEM 2.0, we developed a new DLEM version (named DLEM-catchment) to track water and nutrient dynamics in typical tile-drained delivery-scale agricultural catchments in the Midwestern US. The DLEM-catchment has better representations of management practices and unique features to mimic flow-specific water and N transport through farmed potholes and tile lines. We applied the DLEM-catchment in four data-rich tile-drained catchments in Iowa at a 30-m by 30-m resolution. Using historical weather conditions, field-level practice information, and other input drivers to force the model, we confronted the modeled results with daily observations of water discharge and NO3-N export from 2015 to 2019. We further calibrated the key parameters related to plant growth using the county average corn and soybean yield from 2000 to 2019, which is essential for simulating water balances (e.g. transpiration) and N balances (e.g. N uptake and plant residual). In this work, the snowpack depth obtained from nearby weather stations was used to calibrate the parameters that determine the snowfall/rainfall ratio. The model was first run to get the baseline of C, N, and water pools that represents the status of 2000, by repeatedly using the 20-year (1980-1999) mean climate and land use of 2000. The model was then run to simulate the daily discharge and in-flow NO3-N concentration using driving data from 2000 to 2019. The model estimation shows that subsurface flow (tile flow + lateral flow) dominates the discharge (70%-75%) and NO3-N loading (77%-82%) over the years. However, the contributions of tile drainage and lateral flow vary remarkably among catchments due to different tile-drained area percentages and the presence of farmed potholes. Furthermore, we found that agricultural management (e.g. tillage and fertilizer management) and catchment characteristics (e.g. soil properties, farmed potholes, and tile drainage) play an unignorable role in predicting the spatial distributions of NO3-N leaching and loading. The simulated results reveal that the model improvements in representing water retention capacity (snow processes, soil roughness, and farmed potholes) and tile drainage integratively improved model performance in estimating discharge and NO3-N export amount at daily timestep, while improvement of agricultural management mainly impacts the NO3-N export prediction.
Publication / Journal Articles:
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.
Proposals Submitted:
2022, Co-PI on a proposal to INRC, "Spatially delineated carbon credit potential and implied nutrient reduction co-benefit: An assessment with integrated ecological and economic modeling framework", budget requested: $97,690 2022,
Co-PI on a proposal to USDA, "Women in Agriculture Adopting Climate-Smart Practices", budget requested: $4,932,809 2022,
PI on a proposal to NSF, "Collaborative Research: A Physics-Informed Flood Early Warning System for Agricultural Watersheds with Explainable Deep Learning and Process-Based Modeling", budget requested: $499,986 2022,
Co-PI to a proposal to USDA, "Quantifying and mitigating indirect greenhouse gas emissions from tile-drained agricultural catchments", budget requested: $650,000 .
Other Accomplishments:
Thesis/Dissertations:
Peiyu Cao, “Tracking agricultural nitrogen fertilizer use and nitrogen loss pathways across scales through data synthesis and modeling approach.” Iowa State University, 2022, PhD dissertation
Aishwarya Sarkar, "Transfer learning approaches for knowledge discovery in grid-based geo-spatiotemporal data." Iowa State University, 2022, MS thesis
June 2022
Our long-term monitoring data across catchments suggests that tile drainage and farmed potholes play an important role in regulating the hydrological processes and N loading in this region. In this reporting period, we havd made substantial changes in modeling structure and codes to separate water flows and track water routed through tile lines and spillover from potholes. Current results show that subsurface flows, mainly as tile flow, comprise up to ~80% of annual discharge, which are close to the subsurface contribution in the tile-drained watershed from other publications. Given the limited length of water quality monitoring data, we calibrated the key parameter values by using annual range of observations, and further evaluated the model performance in simulating discharge, NO3-N concentration and loading against daily and monthly observations from 2015 to 2019 in four catchments. The predicted daily discharge over the study period is satisfactory for all the four catchments according to NSE metric, whereas PBIAS and KGE values indicate good model simulations. All three criteria indices for monthly discharge indicate the good to very good model performance at four catchments. The simulated daily discharge reasonably captures the rise and fall of the observed hydrographs. However, some discrepancies in stream peaks and low flows are apparent. For example, the model underestimates peak flows for some storm events in the KS and WW catchments, especially for simulations of wet years.
Publications / Journal Articles:
Lu, C., Yu, Z., Zhang, J., Cao, P., Tian, H., & Nevison, C. (2022). Century-long changes and drivers of soil nitrous oxide (N2O) emissions across the contiguous United States. Global Change Biology, 28, 2505– 2524. https://doi.org/10.1111/gcb.16061
Zhang, J., Lu, C., Tian, H., Pan, S., Miao, R., Do we trade water quality for crop production?: Tradeoffs between crop production and riverine nitrogen loading in the Mississippi River Basin, USA. Nature Food, under review.
Proposals submitted:
2022 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 Activities:
Lu, C., J. Zhang, P. Cao, Y. Bo. Dynamics of soil N2O emission intensity in the US croplands. AGU Fall Meeting, Chicago, IL, USA. Dec. 12-16, 2022. https://agu.confex.com/agu/fm22/meetingapp.cgi/Paper/1195308 2. Cao, P., B. Yi, Y.
Zhang, C. Lu. Half Century Crop-Specific Manure Management History in the Contiguous United States: Manure Nitrogen and Phosphorus Application Rate, Timing, and Method. AGU Fall Meeting, Chicago, IL, USA. Dec. 12-16, 2022. https://agu.confex.com/agu/fm22/meetingapp.cgi/Paper/1067414
August 2021
To strengthen the modeling representation of hydrological processes and N movement in this region, we modified the model structure by updating the snow module, reconstructing subsurface flows, and including tile drainage and farmed potholes. The modified DLEM has been comprehensively calibrated and validated in 4 catchments, with one in Bear Creek Watershed during this reporting period by comparing model simulations against various observations. The simulated snow depth in two catchments shows that the new snow module well captures the snow accumulation, ablation, and duration, indicating a more accurate estimation of soil temperature, moisture, and hydrological processes, such as snow-melt runoff in winter. The data-model comparison also shows that DLEM has a good performance in estimating stream discharge and NO3--N concentration.
Project Activities:
- 3 presentations
- 1 workshop
Publications / Journal Articles:
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. https://doi.org/10.1029/2021EF002141
Submitted - A Sarkar, J Zhang, C Lu, A Jannesari. HydroDeep--A Knowledge Guided Deep Neural Network for Geo-Spatiotemporal Data Analysis. arXiv preprint arXiv:2010.04328
Proposals submitted:
2021 PI on proposal to NSF, “Combined Deep Learning and Process-Based Modeling Framework to Predict Floods in Agricultural Watersheds” Under review. Total requested budget: $521,052 2021 ISU Postdoctoral Seed Grant, Predicting short-term and long-term land-to-aquatic nitrogen loading variations in agriculture-dominated watersheds in Iowa (PI: jien Zhang, advisor: Chaoqun Lu) $1,800
Leveraged Dollars:
$522,852
Other Accomplishments:
2021 Panelist in the Iowa Nutrient Research Center session at the Iowa Water Conference (virtual).
July 2021
During July to December 2021, researchers strengthened the modeling representation of hydrological processes and N movement in this region. Modifications included updating the snow module, reconstructing subsurface flows and including tile drainage and farmed potholes. The modified DLEM has been comprehensively calibrated and validated in four catchments, with one in Bear Creek Watershed during this reporting period by comparing model simulations against various observations. The simulated snow depth in two catchments shows that the new snow module well captures the snow accumulation, ablation, and duration, indicating a more accurate estimation of soil temperature, moisture and hydrological processes, such as snow-melt runoff in winter. The data-model comparison also shows that DLEM has a good performance in estimating stream discharge and NO3--N concentration.
The analyses imply that tile drainage and farmed potholes play an important role in regulating the hydrological processes and N loading in this region. In this reporting period, substantial changes were made in modeling structure and codes to separate water flows and track water routed through tile lines and spillover from potholes. Current results show that subsurface flows, mainly as tile flow, comprise up to ~80% of annual discharge, which are close to the subsurface contribution in the tile-drained watershed reported in other publications. With the modeled tile flow approximating the observed patterns, follow-up work will focus on calibrating and validating DLEM’s performance in simulating the effects of saturated riparian buffer on tile flow N removal. Specifically, two catchments within Bear Creek Watershed will be tested. The team will: 1) extract model input driving data from the Bear Creek Watershed dataset developed for the two catchments; 2) quantify tile drainage contribution based on observations; 3) compare performance of DLEM with vs. without saturated riparian buffer for N leaching, plant N uptake and denitrification against observations.
Other outreach/accomplishments included three presentations and one workshop.
December 2020
To incorporate the effect of the saturated riparian buffer practice in DLEM and apply it in the Bear Creek Watershed, researchers modified the DLEM to incorporate the tile drainage process in the model based on water table level and the parameters of tile lines (e.g. depth, width, space between lines). The incorporated tile drainage calculates the volume of water entering the pipes and the flow rate. A pool in each grid cell was created to store water and nutrient drained by tile lines and transport the mass through pipes to the stream along the flow direction. This technique allows water with high N concentration in the tile lines to spread over the riparian buffer. Instead of flowing the water and leached N in tile lines to surface drainage in each grid cell, tile lines are connected between upstream and downstream grid cells, allowing water and dissolved N to drain to the river channel through the continuous tile lines. At the outlet of tile lines, if there is a saturated riparian buffer, the water is spread out to irrigate the vegetation on the buffer.
A saturated riparian buffer map has been generated based on satellite images.
Outreach during this period included 1 field day, 4 presentations and 1 workshop.
The team received the NSF Early Career Award to support its water quality related research for the next five years.
July 2020
To gain the DLEM’s representation in prairie potholes regions, the research team introduced the impact of potholes in DLEM by obtaining the area and depth maps of potholes and resampling them into 30 meter resolution as DLEM’s input. DLEM's structure was modified by delivering surface flow into potholes and trapping the water in them. Instead of combining surface and subsurface flows in every grid cell, these two components flow separately to the downstream grid cells based on flow direction. When the potholes exist in the downstream grid cell, the water of surface flow is distributed to and stored in the pothole temporarily. The potholes also collect water from rainfalls on the grid cell. Water stored in the pothole can be lost as evaporation, infiltration and overland flow when the water depth exceeds the maximum depth.
Nitrogen biogeochemical processes were also included, related to pothole distributions, to improve the estimation of nitrogen leaching. The performance of DLEM simulation before and after modification shows that the modified DLEM improved the peak estimation. For example, small peaks during non-rainy days, such as in fall and spring, are eliminated due to the pothole absorption, which matches well with observations. By using the original version of the DLEM model, annual N yield, N loading and N input was quantified for the Bear Creek Watershed from 1979 to 2015. The modeled results represent the baseline scenario of the N budget during the study period. The total N inputs, including N fertilizer, N fixation and atmospheric N deposition, increased from 926 tons year-1 in 1979 to 1100 tons year-1 in 2015, with an average of 1041 tons year-1. Compared to the N input, both the N leaching and N loading from the watershed showed a larger inter-annual variation, due to the year-to-year corn-soybean rotation. Annual N loading ranged from 232 to 932 tons year-1 from 1979 to 2015, with an average of 632 tons year-1, accounting for 61% of the annual N input.
These modeling results together suggest that the total N input and N loss from the Bear Creek Watershed increased over time and the inter-annual variations in N loading were larger than the N input. Future work will include updating DLEM to fully incorporate surface runoff, lateral flow, tile drainage and baseflow to estimate the contributions of these water flows to the N loading given different N management practices.
July 2019
During January-June 2019, geospatial data at a 30-m resolution was developed to characterize key environmental changes and drive ecosystem modeling in the Bear Creek watershed. These datasets include gridded input data of climate, soil properties, DEM (digital elevation model), land cover and land-use change (LULC), river network, spatial nitrogen (N) deposition and agricultural management practices across the entire study region. The daily spatial weather data from 1979 to 2018, including temperature, precipitation and shortwave solar radiation, were interpolated from six weather stations in the Iowa Environmental Mesonet. The Inverse Distance Weighting (IDW) method was used to interpolate station-derived weather data to gridded maps. 1979 was used as the starting year because of the radiation data availability. The LULC data were extracted from the 30-m resolution Cropland Data Layer (CDL) dataset from 2000 to 2018 (pre-2000 LULC were assumed the same as in 2000). The river network was constructed based on a recent 30-m DEM dataset. The 30-m gridded N deposition input data were resampled from the NADP database with a ~4 km resolution. Tile-drained soil for Bear Creek was created based on the SURRGO soil dataset. Because there is no available surveyed N fertilizer data of corn (the major N consumer) in Bear Creek watershed, six corn N fertilizer input scenarios were developed from the surveyed N fertilizer information in the RS catchment. First, the maximum, mean, and minimum values of the N fertilizer input in the RS catchment was calculated for 2016, assuming the 2016 N fertilizer input in the Bear Creek watershed has a similar amount as in the RS catchment. Using the county-level NuGIS N database, the average N fertilizer amount of the Story and Hamilton counties was then calculated (from 1987 to 2013, and N fertilizer from 2014-2016 was assumed the same as in 2013). The trend of the averaged NuGIS N fertilizer use rates from 1987-2016 was used to back-calculate three time series by using the maximum, mean and minimum values of 2016 N fertilizer amount in the RS catchment. Since there were two N fertilizer application timings in the RS catchment, the three N fertilizer time series were partitioned based on these two timings, before-planting vs. after-planting, to create the six N fertilizer scenarios. The ratio between ammonia and nitrite N in Bear Creek was kept the same as the ratio in the RS catchment. Using these input data, initial model runs were completed (to get initial status of ecosystem carbon, nutrient and water storage) in the Bear Creek Watershed. The first transient model simulation has been carried out to validate model performance against the field observations.
During the reporting period, one project-related workshop was presented and one paper was accepted:
Lu, C., J. Zhang, P. Cao, J. Hatfield. Are we getting better in using nitrogen? — Variations in nitrogen use efficiency of two cereal crops across the United States. Earth’s Future DOI: 10.1029/2019EF001155
December 2018
During this quarter, available data sources were collected to develop geospatial data that can drive ecosystem modeling in the Bear Creek watershed. Based on the research goal, a set of gridded input data is needed to characterize weather, soil properties, digital elevation model (DEM), land cover and land use change (LULC), and agricultural management practices. Data sources were compiled and group meetings held to discuss ways to produce time-series spatially explicit data during 2000 to present. Land cover and cropping system data from 2000 to 2015 will be developed from maps from the USDA National Agricultural Statistics Service (NASS) cropland data layer. Data to characterize agricultural management practices is based on the field- and county-surveys, literature and farmer-interviews. Daily weather data will be interpolated from the US-Historic Climatology Network (HCND) database. Topographic information will be derived from the Shuttle Radar Topography Mission (SRTM, Farr et al., 2007). County-level Soil Survey Geographic Database (SSURGO) data from USDA Geospatial Data Gateway will be resampled. We will finish data preparation and pre-processing by the end of February 2019.
September 2018
During the first quarter, a team met to kick off the project. Team members who conducted field monitoring at tile outlet and riparian buffers have shared the published data and relevant records in the Bear Creek Watershed. The data sets include monitored discharge and nitrate (NO3) concentration at each outlet, water table depth (from groundwater wells), and denitrification and nitrous oxide (N2O) emissions at the riparian zone. The data have been collected and compiled into excel files that will support HUC 12-level model test and calibration. In addition to these measurement data, digital maps of tile lines and unified riparian buffer zones developed by Co-PI Isenhart’s team were also shared. The new data will be used to characterize distribution of saturated and traditional riparian buffers, riparian vegetation cover and tile lines across the Bear Creek Watershed.