Delivery-Scale Evaluation and Modeling of Nutrient Reduction Practices
Issue
Most prior work on practice performance and nutrient loads in Iowa has been done at either plot scale or larger watershed scale. It has been challenging to properly scale up plot-level measurements to the area of practice implementation and larger watersheds in order to better assess water quality impacts across landscapes and with multiple practices. Nutrient loads and load reductions at plot scale can differ substantially from loads actually delivered to surface waters. Likewise, nutrient loads at larger watershed scales can differ substantially due to the effects of in-stream processes. Few studies have focused on delivery-scale water and nutrient monitoring, which hinders model improvement and evaluation.
However, the scale at which the load is actually delivered to surface water is the most appropriate scale for assessing impacts of management practices on agricultural non-point source loads. Few studies have focused on delivery-scale water and nutrient monitoring. It is also a critical missing link that hinders model improvement and evaluation.
Objective
This project focuses on delivery-scale evaluation and modeling of the effectiveness of in-field and edge-of-field practices including cover crops, fertilizer management, and wetlands in increasing ecosystem nitrogen and phosphorus retention, and reducing downstream nutrient loads. This research will build process-based simulation capacity by integrating a delivery-scale monitoring network, in-field management surveys, and modeling estimation.
Approach
Researchers will test, improve and validate the DLEM’s (Dynamic Land Ecosystem Model) performance in estimating delivery-scale water flow and nitrogen and phosphorus loads against long-term monitoring for four watersheds. Two of the watersheds have widespread implementation of cover crops and all four outlet to CREP (Conservation Reserve Enhancement Program) wetlands. The improved and calibrated DLEM will be used to assess the effectiveness of cover crop implementation, combined with various fertilizer management practices on delivery-scale nutrient reduction, and identify the hot spots and moments contributing to high nitrogen and phosphorus loads in these four watersheds. In addition, researchers will quantify how alternative nutrient reduction practices would affect nutrient and phosphorus loads in the future.
Project Updates
Note: Project reports published on the INRC website are often revised from researchers' original reports to increase consistency.
April 2022
FINAL REPORT
Key Findings and Outcomes
(Find the referenced figures and tables in a PDF file at https://www.cals.iastate.edu/inrc/files/project/files/inrc_final_report_project_2017-06_figures-tables_lu-crumpton-helmers_.pdf).
Model inputs development: To drive the ecosystem model and characterize weather, soil properties, topographic features, land cover and land use change, and agricultural management practices across the four catchments studied (i.e., KS and RS in Story county, and LP and WW in Floyd County), researchers developed a time-series spatially explicit database during 2000-2019 at a spatial resolution of 30m × 30m (Fig. 1). Multiple data sources were used and harmonized to develop the model input datasets. For example, data of soil properties were derived from the Soil Survey Geographic Database (SSURGO) by the Natural Resources Conservation Service (NRCS). Climate data were obtained from Iowa Environment Mesonet (IEM), which provides daily temperature, radiation and precipitation. The land use data were obtained from the Cropland Data Layer (CDL).
Different from limited management information in other areas, in this project, the project team used a field-level farmers’ survey conducted by Dr. Matt Helmer’s team. The application rate, timing and form of N fertilizer and manure from the field survey were compiled as annual fertilizer management data at the same resolution to drive the DLEM. The management survey also included tillage practices and planting of cover crops at the field level. The model input list is summarized in Table 1.
The data preparation process was programmed by the Matlab software.
Model improvements: Six major modifications were made in DLEM to improve the modeling representation of the key features/mechanisms in regulating hydrologic flows and nitrogen loads within the agricultural catchments.
(1) A new two-layer snow module with processes of snow temperature change, phase change, water movement and snow compaction was developed to better simulate snow accumulation and depletion, which improves the water-energy balance simulation.
(2) Subsurface flow was divided into quick lateral flow, slow lateral flow and groundwater flow. By routing water with different travel times, the new subsurface flow structure improved the modeled dynamics of baseflow and recession of the hydrograph.
(3) Tile drainage was added to DLEM as one of the subsurface flows where tile exists. With its own travel time and flow path, the addition of tile drainage altered the ratio of surface flow to subsurface flow and significantly enhanced model performance in estimating peak flows.
(4) Soil roughness induced by tillage was added to the model to smooth small peak flows and enhance infiltration.
(5) Prairie potholes play an important role in regulating water balance and the N cycle. The hydrologic processes of potholes were included in DLEM by re-distributing surface runoff to potholes. Results show that including potholes improved model performance in estimating water balance and N loads.
(6) Based on the field survey, planting cover crops from fall to early spring the next year was incorporated into DLEM. The simulation shows that adding cover crops has little impact on hydrology but decreases N loads. This improved model performance in capturing the N loading difference from the catchments with vs. without cover crops.
Model parameterization: The spatial and temporal heterogeneity of climate, topography, soil properties and agricultural management leads to various patterns in hydrological processes and N loading across catchments. Therefore, based on the observation data, we calibrated key parameter values that constrain crop growth, snow dynamics, hydrological processes and N leaching at each catchment. Due to the lack of field-level crop yield information, the county-level corn and soybean production survey data from Story County and Floyd County during the period 2000 to 2019 were used to calibrate the crop growth-related parameters in four catchments. Monitored snowpack depth from 2000 to 2019 was obtained from the IEM site to constrain the related parameter values and simulated snowpack depth. Part of the stream flow observation data were used to calibrate the key parameters that control flow travel time, flow rate and other hydrological related characteristics. Lastly, parameters that regulate biogeochemical processes and flow-weighted N concentrations were tuned to make the modeled nitrate-N concentration approximate observations.
Model performance: The model evaluation results indicated that the simulated daily streamflow well re-produced the observed patterns at four catchments. The simulated daily streamflow well captures the rise and fall of hydrographs (Fig. 2). However, there are some discrepancies in the modeled magnitude of peak flows and low flows by comparison with the observations. Most of the observed stream peaks are captured by the model simulation, but the latter underestimates large spikes, especially in the wet years.
By comparison, the modifications made to DLEM tremendously improved the model performance in estimating streamflow by smoothing small peaks and increasing baseflow (Fig. 3(a)). The statistical indicators for modeling daily and monthly in-flow nitrate concentration are within the satisfactory range to the very good range at four catchments. In addition, the simulated daily and monthly stream nitrate loading fall in the good and very good range -- except daily N loading in KS is under the acceptable limit (Table 2). The simulated daily nitrate concentration generally matches the observations at four catchments, well capturing the observed inter-annual and seasonal dynamics (Fig. 4). Some discrepancies are found for the low loads in winter and peak loads in early spring. For example, DLEM underestimated nitrate concentration in the winter of 2015 in RS and KS, and it overestimated that in the early spring of 2018 in RS.
Overall, the daily simulated nitrate concentration by the modified model shows a significant improvement in the magnitude and seasonal dynamics of modeled water discharge and nitrate concentrations compared to the model before improvement (Fig. 3(b)).
During the study, researchers gave seven presentations related to this project.
Follow-up work beyond this project:
Factorial Analysis: The simulated historical daily nitrate loads at four catchments from 2000 to 2019 showed large inter-annual and seasonal variations. The factorial analysis was designed to explore the contributions of different factors to this change. The factors that were considered include the changes in climate, land cover and use patterns, N fertilizer use, manure application and N deposition. A series of model simulation experiments were set up, including simulations with all factors dynamically changed and other simulations with one of the aforementioned factors unchanged from 2000 to 2019.
Strategies to reduce nitrate loads: A few agricultural in-field practices that may potentially reduce nitrate loads were tested by DLEM at the four catchments. The practices we tested include lowering fertilizer application rate, replacing N fertilizer with manure, postponing fall N application to after-planting and planting cover crops. The simulated historical N loads under these practices were compared with the simulations under current practices to quantify the effectiveness of these proposed practices in reducing N loads.
December 2019
During the second half of 2019, researchers worked on model improvement documentation, model simulation on other three catchments and further model refinement. By comparing observed discharge at the outlet of RS catchment with DLEM results, we found DLEM has an unsatisfied performance in predicting winter thawing and freezing, subsurface drainage and rainfall event peaks in the tile-drained landscape. Therefore, a series of mechanisms was introduced to improve DLEM’s representation in this region. Four major modifications were made, including a new snow module, soil structure modification, new subsurface drainage processes and grid cell-based residence time of waterbody pools. These modifications largely improved the model performance in hydrology and N transportation from land to aquatic systems. With the modified model, 12 scenarios were set up with combination of application rate, application timing and fertilizer form (NH4+-N/NO3--N) in RS catchment to fully assess the effects of best management practices on N loading reduction.
Model driving data was prepared for the other three catchments, KS in the Story County, and FX and LP in the Floyd County. RS and KS are paired catchment that received similar management practices such as high synthetic N fertilizer input combined with fall applied manure. However, RS adopted cover crop strategy to prevent N loss through leaching while KS did not grow any cover crops. Similarly, FX and LP catchments are also paired with only synthetic N fertilizer at low rate, but one has cover crops planted and the other one does not. The same approach was followed as RS to develop the input data for the other three catchments. Among four catchments, KS and LP lack N fertilizer management practices record. Since they are paired to RS and FX, respectively, we calculated the frequency of each practices in each catchment with available data and applied the probability of practices to KS and LP.
During this period, Dr. Lu received the NSF Early Career Award to support our water quality related research for next five years.
July 2019
During the first half of 2019, work was done on model-data comparison and model implementation at the catchment level. The soil water transportation process in the DLEM model was improved to capture the water table fluctuation between soil layers. By doing this, the water table close can be raised closer to the surface during heavy rainfall, and thus, more water can be drained into the tile lines after heavy rainfall events. In addition, the subsurface hydrological processes in the DLEM were revised to add subsurface lateral flow to better simulate water recession following rainfall events. The simulated baseflow in the original DLEM model can only represent total subsurface water flow. After these modifications, the model is able to represent quick, medium and slow drainage, respectively, from tile lines, lateral flow from multiple soil layers and groundwater flow from the aquifer layer.
The time-series of simulated nitrogen (N) loading was also compared with measured data at the inlet of wetland in RS catchment. The model performance in simulating daily magnitude and variations of N loading is good but some peaks were missed. This may be caused by the underestimation of rainfall intensity, which leads to less rainfall water entering the soil at sub-daily scales. Now, the team is working to improve the model estimates of N loading by allocating precipitation to sub-daily scale based on historical hourly precipitation records from weather stations in Iowa. This version of the model will be set up to run in other three catchments to test model performance, and then investigate the effect of different agricultural management, such as cover cropping, on N loading among the four catchments.
January 2019
During the second half of 2019, researchers worked on model improvement documentation, model simulation on three catchments and further model refinement. Comparing observed discharge at the outlet of RS catchment with DLEM results showed that the DLEM performed unsatisfactorily in predicting winter thawing and freezing, subsurface drainage and rainfall event peaks in the tile-drained landscape. Four major modifications were made to improve the model performance in hydrology and N transportation from land to aquatic systems: a new snow module, soil structure modification, new subsurface drainage processes and grid cell-based residence time of waterbody pools.
With the modified model, 12 scenarios were set up with combination of application rate, application timing and fertilizer form (NH4+-N/NO3--N) in the RS catchment to fully assess the effects of best management practices on N loading reduction. Model driving data was prepared for other three catchments, KS in the Story County, and FX and LP in the Floyd County. RS and KS are paired catchment that received similar management practices such as high synthetic N fertilizer input combined with fall applied manure. However, RS adopted cover crops to prevent N loss through leaching, while KS did not grow any cover crops. Similarly, FX and LP catchments are paired with only synthetic N fertilizer at a low rate, one with and one without cover crops. Researchers followed the same approach as RS to develop the input data for all catchments. Among four catchments, KS and LP are lacking N fertilizer management practices record. Since they are paired to RS and FX, respectively, the frequency of each practices in each catchment was calculated with available data and the probability of practices applied to KS and LP. Next, analysis on 12 scenarios in the RS catchment will be calculated, and the simulation of the other three catchments finished.
Outreach during this period included 1 field day, 4 presentations and 1 workshop.
The team received the National Science Foundation Early Career Award, which will support its water quality related research for the next five years.
December 2018
This quarter has been focused on model improvement and model-data comparison. A new snow module was added to the Dynamic Land Ecosystem Model (DLEM) to improve its performance in winter hydrological processes. Model representation of snow formation and melting was improved by 1) introducing snow phase change and compaction processes to track the dynamics of snow depth, density, coverage and temperature, and 2) updating liquid water flow in snow pack regarding capillary force of snow porosity. The model-estimated formation and melting of snow and ice is shown to align well with observations. Specifically, it successfully reduces discharge peak when the snow pack exists and enhances the peak when snow melting occurs. The modeled - to 3-day lag of peak time behind measurement was also fixed.
DLEM is original designed for large scale study, in which the time step of 30 minutes is reasonable. However, it is no longer fitful for small regions, such as RS catchment. So the time step was adjusted to 10 minutes, which solved the lag problem despite more computational costs. In contrast to time lag, the model-estimated magnitude of peak flow is still lower than that of the monitored data. Tile line maps and tile line processes have been incorporated into the model, but it doesn’t raise flow amount effectively. Based on a site-level modeling process check, the lower modeled peak flow has been attributed to underrepresented subsurface hydrological processes in the model. More work will be done on this in the following quarter.
The time-series of simulated nitrogen (N) loading was also compared with measured data at the inlet of the wetland in RS catchment. The model performance in simulating daily magnitude and variations of N loading is generally good but underestimates some peaks and troughs of N load before the growing season. We expect the model estimate could be improved after the subsurface processes modification.
In the next quarter, the same model will be set up to run in the paired catchment to test model performance and investigate the effect of different agricultural management practices, including cover crops, on N loading between two catchments.
March 2018
During this quarter, the team has been working on primary model setting and model testing. Based on one of the two catchments in Story County, the team has refined the procedure in the development of model input data and cross-checked the whole dataset by using multiple data sources. The team also included gridded maps of nitrogen deposition, tillage and cover crop practices to drive the process-based ecosystem model. The initial and transient model run has been completed, and the preliminary model estimates of magnitude and spatiotemporal patterns of water discharge and in-flow NO3 concentration compared with some long-term delivery-scale monitoring data. Site-level measurement data of surface/sub-surface runoff and N leaching has been used to calibrate the model.
December 2017
The focus this quarter was on input data development, and making the format compatible to a process-based land ecosystem model. The catchment in Story County, one of four catchments in the study, was chosen as the data development template. Based on the multi-source data from remotely sensed products, national survey and monitoring network, team members developed spatially explicit time-series data of climate, soil properties, land use and land cover maps, flow direction and other ancillary data. These data are used to characterize the conditions of geography, climate, soil, hydrology and land use pattern within the catchment boundary over time. The input data for the other three catchments will be built using this data development template and similar data sources.
September 2017
Planning and data sharing to facilitate regional data development is underway. Based on the research goal, a set of gridded input data is needed to drive the ecosystem model, and characterize weather, soil properties, DEM (digital elevation model), land cover and land use change, plus agricultural management practices across four catchments in Story and Floyd counties. Various data sources were complied and group meetings held to discuss the best way to produce time-series spatially explicit data during 2000 through to the present at a spatial resolution of 30m × 30m. A shared data folder is accessible to all project investigators. Spatial database, such as catchment boundary, land use history, are uploaded into the geospatial data library.