Location: Watershed Physical Processes Research
2024 Annual Report
Objectives
1. Develop physically based multidimensional numerical models and technology for watershed erosion and sediment transport processes, water quality issues due to agro-pollution, and surface-groundwater interactions in agriculture landscapes.
1.1. Expand CCHE1D model capabilities.
1.2. Simulate rain-induced watershed soil erosion over tilled and non-tilled agricultural land and sediment transport.
1.3. Develop a coupled surface-subsurface water model.
1.4. Improve CCHE water quality and pollutant transport models.
2. Enhance the Decision Support System for the web-based Agricultural Integrated Management System (AIMS) by integrating watershed and channel network technology with geospatial and remote sensing data for effective watershed runoff, sediment, and water quality management.
2.1. Improve AIMS and AnnAGNPS integration.
2.2. Integrate CCHE1D channel network delineation model and multiple-scaled channel networks and flow model into AIMS.
2.3. Develop assessments of the AIMS integrated technology with case studies.
2.4. Explore cloud computing capability for AIMS.
Approach
In order to better assess and predict the direct impacts of water availability and soil erosion from agricultural fields, gullies and channels, existing technology must be developed and upgraded to assess and improve predictions for more realistic conditions. This includes technology that is needed for decision making in identifying, targeting and implementing conservation practices that protect the landscape. The first objective of the project plan focuses on the development and improvement of numerical models associated with watershed erosion and sediment transport processes, water quality issues due to agro-pollution, and surface-groundwater interactions in agriculture landscapes. This objective relates to the ARS National Program Action 211, Component 1 (Effective Water Management in Agriculture) and Component 2 (Erosion, Sedimentation, and Water Quality Protection). The second objective focuses on the development and improvement of a watershed management tools to assess runoff, sediment, and water quality conservation practice impacts. This objective is related to NP 211 Component 1 and Component 4 (Watershed Management to Improve Agroecosystem Services).
Progress Report
The program code of CCHE1D_WQ model has been modified under the PGI Fortran system which is consistent with the updated CCHE1D flow/sediment model. The updated CCHE1D_WQ model has been linked with the recent version of AnnAGNPS watershed model to simulate flow, sediment, and water quality in channel networks. This linked modeling system has been validated using field measurements from the Goodwin Creek Experimental Watershed that were provided by NSL researchers to NCCHE researchers. A conference paper was presented at the 2024 ASCE/EWRI World Environmental & Water Resources Congress.
The CCHE2D-CUDA model for GPU (Graphic Process Unit) is being developed, based on which the computing efficiency of the CCHE2D soil erosion model will be enhanced. In CCHE2D-CUDA, the parallelized ADI (Alternate Direction Implicit) solver has been refined and the parallelized SIP (Strong Implicit Procedure) solver is under development. A technical manual describing this CCHE2D-CUDA model has been completed. A journal paper has been drafted and a conference abstract has been presented at the 2024 ASCE/EWRI World Environmental & Water Resources Congress.
The CCHE2D-Hybrid for sediment transport model has been developed. This model is based on a hybrid unstructured mesh system consisting of triangle and quadrilateral cells. It adopts a conservative accurate edge interpolation method and a multi-point momentum interpolation correction method. This single-phase non-equilibrium sediment transport model aims for non-uniform non-cohesive sediments in unsteady turbulent flows. A journal paper describing this model and its validation cases has been published at Journal of Applied Sciences.
A three-dimensional meshless numerical model (CCHE3D-GW-Meshless) based on the moving least squares method was developed, validated and successfully applied to a field case of pumping groundwater near a meandering river in Shellmound, MS. The package of the model was uploaded online (https://www.ncche.olemiss.edu/cche3d-gw-meshless/), and a manuscript detailing the methodology and the verification of the model as well as the results for the field pumping case was published in the Environmental Modelling & Software. With this meshless model, the time-consuming mesh generation for complex topography, such as the meandering river, can be circumvented, which in turn facilitates the setup of numerical simulations for field cases in the Mississippi River Valley Alluvial Aquifer, benefiting the research studies as well as decision makings. Furthermore, it is much easier and more flexible to add and/or move discretization points without the cumbersome remeshing, making it well suited for a dynamic decision support system.
The newly developed CCHE3D-GW-RBF was then applied to a Bayesian experimental design of the monitoring wells for estimating streambed conductivity of the Tallahatchie River (the river bend) at Shellmound, Mississippi. This is a follow-up work of our paper published in Journal of Hydrology, which studied the aquifer-stream interaction, and it is also an application of our meshless numerical model to field studies. Streambed conductivity is a crucial parameter controlling groundwater-surface water interaction while a direct measurement is difficult in the field. Thus, this parameter is often obtained through inverse calculation, and the locations of monitoring wells are important for the estimation result. As a result, CCHE3D-GW-RBF was used to run hundreds of potential designs so that the optimal observation positions can be found. It was easy to set up these simulations with the meshless model because of its flexibility in moving the monitoring wells without remeshing. The results were presented at the 2023 American Geophysical Union Fall Meeting (AGU), and a journal paper is in preparation.
CCHE3D-GW-RBF was also applied to develop a tool that can automatically estimate agricultural irrigation schemes. Field measurements have revealed that agricultural practices, especially irrigation, substantially affect the local groundwater level, which in turn implies a significant impact on the regional groundwater flow. As a result, it is imperative to have a comprehensive understanding of the irrigation scheme so that the numerical model can fully capture the details, particularly for a long-term simulation. However, data for agricultural irrigation are usually scarce, so a tool that can infer the pumping rate as well as irrigation periods from the measured groundwater drawdown was developed. The efficiency and robustness of this tool was first verified with a hypothetical case. It was then applied to the experimental site of the groundwater pumping, transfer and injection project located at Shellmound, Mississippi. These results were presented at the 2023 American Geophysical Union Fall Meeting (AGU), and a journal paper is in preparation.
A new water quality module has been developed to simulate the particulate and dissolved nutrients. The phosphate (PO4), organic phosphorus (OP), and organic nitrogen (ON) were each partitioned into particulate and dissolved forms that are simulated separately. The processes of adsorption/desorption, bed release, and exchange due to sediment deposition/resuspension were considered. The concentrations of particulate/dissolved nutrients as well as total nutrients can be simulated by this module. This module is being integrated into CCHE_2D/3D_WQ model for nutrient simulation.
The CCHE_MIX_WQ model and AnnAGNPS watershed model have been linked to study the response of lake water quality to the upland agricultural practices. The AnnAGNPS model was applied to simulate runoff, loads of sediment and nutrients from upland watersheds under alternative agricultural practices. The outputs of AnnAGNPS were set as inlet boundary conditions for CCHE_MIX_WQ to simulate water quality in the receiving waterbody. The CCHE water quality and pollutant transport models have been tested and validated using field measurements provided by NSL researchers to NCCHE researchers. These field data included flow, sediment and nutrients in Beasley Lake and upland watersheds. This model has been tested in Beasley Lake watershed to study the lake water quality responses to alternative tillage practices and crop planting in the watershed. A journal paper describing this model and its test case has been published in Journal of Environmental Modelling and Software.
Agricultural Integrated Management System (AIMS) is a Web-based decision support tool, developed to evaluate the impacts of agricultural and channel conservation management practices within any watershed in the United States. AIMS currently offers a user-friendly Web-GIS framework which allows users to create accounts, create projects and scenarios under projects, as well as visualize various geospatial data layers. AIMS will soon enable convenient interaction with geospatial data layers on a seamless map server, automated input data preparation for AnnAGNPS model and visualizing watershed simulation results on any device with internet access. Running the AnnAGNPS model requires various datasets including topographic, soil, land use and land cover, climate, management data. The assimilation of the management data to AIMS has been completed. AIMS (beta) is now live and can be accessed at: www.aims.ncche.olemiss.edu, which provides TOPAGNPS generated topographic data, soil data and some additional geospatial layers including USGS stream network, HUC boundaries and NLDAS grid for the entire United States. Backend tools for running AnnAGNPS solely from AIMS database have been developed and tested for several case studies. More recently, efforts have been focusing on preparing large scale AnnAGNPS simulations without routing, using a set of predefined input variables. The purpose of these simulations is to generate an AnnAGNPS output database for AIMS which can be used to visualize baseline loads for any watershed in the United States. A set of preliminary simulations were performed to test the speed and memory requirements using watersheds of varying sizes.
Since TOPAGNPS output for the entire United States has been stored in the main AIMS data server, users can now view the AnnAGNPS cells and stream network for any watershed of HUC12 size or larger using the AIMS web interface. The geometry can be downloaded in the form of both .csv and geoJSON data types over the AIMS interface. With a single user-identified outlet point on the map page, AIMS automatically assembles the watershed for to the user-provided outlet.
Integration of CCHE1D Model is also ongoing. Data input/output connections between TOPAGNPS and CCHE1D has been completed. The intermediate Python modules can convert TOPAGNPS generated stream network into CCHE1D geometry input with additional user provided data.
The NASA North American Land Data Assimilation System (NLDAS) climate data have been successfully tested against land-based station data for the Goodwin Creek Experimental Watershed, located in North Mississippi. A manuscript out of this pilot study has been submitted to the Journal of Environmental Modelling and Software. A follow-up paper for Goodwin Creek Watershed was carried out using evapotranspiration (ET) forcing data derived from remote sensing products has been drafted. The progress has been presented at the 2024, European Geophysics Union (EGU) General Assembly in Vienna, Austria.
Accomplishments
1. An efficiency-enhanced CCHE2D sediment transport model. The CCHE2D sediment transport model includes the flow module, the suspended-load transport module, the bed-load transport module and the soil erosion module. When simulating multi-sized sediments or long-term events or high-resolution domains, computing efficiency remains a challenge. To alleviate this problem, the parallel computing technology based on CUDA framework on GPU (Graphics Processing Units) was applied by ARS researchers at Oxford, Mississippi, to CCHE2D sediment transport model. Two implicit numerical solvers, namely, the ADI (alternating direction implicit) and the SIP (strong implicit procedure), have been parallelized. The CUDA-parallelized CCHE2D sediment transport model can achieve 3.73-4.98 times speedup in flow simulations, 2.17-3.65 times speedup in sediment transport simulations than its serial version. This model can be applied for soil erosion control and management in agricultural fields.
2. Development and manuscript of CCHE3D-GW-RBF. The unsaturated zone is the shallowest part of an aquifer near the ground surface where the aquifer is not fully saturated with water. Movement of water in the unsaturated zone strongly affects the flow of groundwater that may be caused by a well that pumps water out of or into the aquifer. Traditionally, numerical groundwater models, under these conditions focus on only pumping in the fully saturated zone, merely simulate variably saturated groundwater flow without pumping in the aquifer, or require a complex computational mesh, which is an array of points representing geographic locations at regular intervals where the model is applied. To fill this gap, a three-dimensional meshless model, CCHE3D-GW-RBF, was developed by ARS researchers at Oxford, Mississippi, for variably saturated groundwater flow with the consideration of pumping and injection. The localized radial basis function (RBF) method was employed in this model. This new model complements our previous meshless model (CCHE3D-GW-Meshless) that was based on the moving least squares (MLS) method. Compared to MLS, RBF is more flexible and easier for programming and possesses the nodal interpolation property that allows more accurate representation of real-world conditions. After verification and validation, CCHE3D-GW-RBF was successfully applied to the Mississippi River Valley alluvial aquifer at an experimental groundwater pumping and injection operation at Shellmound, Mississippi, USA, on April 14-19, 2021. CCHE3D-GW-RBF provides a new decision-support tool that allows scientists to more efficiently model groundwater resources management problems.
3. Application of CCHE3D-GW-RBF to a Bayesian experimental design. Streambed conductivity is a critical parameter controlling the intensity of stream-aquifer interactions, which in turn affects the riverine ecosystem. Streambed conductivity can be estimated through the inverse calculation of a pumping test near a river, and the locations of the monitoring wells are crucial for the accuracy of the estimation result. It is therefore essential to conduct an optimal design of the monitoring wells in which numerical groundwater models are often used. Traditional numerical groundwater models are usually mesh-based, but mesh generation for complex real-world topographies, such as a meandering river, can be troublesome. Furthermore, for the mesh-based numerical model, the monitoring wells can only be located at the grid nodes. Consequently, a remeshing of the study area by ARS researchers at Oxford, Mississippi, is generally needed for different designs of monitoring wells, which can be time-consuming for a complicated stream-aquifer system. Therefore, the newly developed meshless numerical model, CCHE3D-GW-RBF, was implemented into this Bayesian experimental design of the monitoring wells for estimating the streambed conductivity. Relative entropy is used to quantify the information gained from the drilling of the new observation well so that the optimal location can be identified. The study area was chosen near a river bend of a meandering river located at Shellmound, Mississippi, USA. It was found that setting the monitoring wells at the opposite side of the pumping well as well as near the river can help improve the estimation of streambed conductivity. With this Bayesian experimental design, scientists can thoroughly and efficiently evaluate every potential proposal before implementing that to the field, which is cost effective.
4. Application of CCHE3D-GW-RBF to develop a tool for automatic estimation of agricultural irrigation schemes. Unsustainable use of groundwater for irrigation in the Mississippi River Valley alluvial aquifer has resulted in long-term groundwater depletion, posing a challenge for the regional agricultural productivity and watershed ecosystems. To mitigate this issue, it is necessary to implement sustainable water resources management schemes, which demand a better understanding of complex surface-groundwater flow interactions in agricultural watersheds. Irrigation schemes are a crucial input parameter for both watershed and groundwater models. However, irrigation data with high spatial and temporal resolutions are usually not available. As a result, simplifications of the irrigation scheme are often made, such as using a regional averaged irrigation strategy for the whole watershed, which in turn affects the accuracy of the simulation results. On the other hand, using measured drawdowns of the groundwater head allows irrigation schemes to be inferred through the inverse calculation. Therefore, ARS researchers at Oxford, Mississippi, developed a new tool to automatically estimate the irrigation pumping schedule and the pumping rate by calibrating the newly developed CCHE3D-GW-RBF with the measured groundwater head data. Owing to the flexibility of moving the discretization points in the meshless model, simulations for this calibration can be done easily without time-consuming remeshing. The efficiency and robustness of this tool were revealed by a hypothetical case. The tool was then applied to a field case at Shellmound, Mississippi, USA. With this tool, scientists can fill in the missing irrigation data, which are crucial not only for the long-term simulation of groundwater flow but also for watershed model, such as the Annualized Agricultural Non-Point Source (AnnAGNPS) watershed pollution model.
5. Development of a model for the transport of conservative contaminant for anisotropic dispersion in heterogeneous porous media. Hyporheic zone is located adjacent to river, in which the interaction between the inflowing surface water and the local groundwater is strong. Owing to the infiltrated surface water, advection is generally strong in this area. Traditional numerical models based on the Eulerian method usually cannot precisely capture the advection process due to the numerical diffusion, which makes it not well suited for simulating the transport and exchange processes accompanied by the hyporheic flow. On the other hand, the smoothed particle hydrodynamic (SPH) method, which is a Lagrangian method, can precisely capture the advection without artificial diffusion. However, the accuracy of the previous SPH models for the transport in the groundwater flow is not satisfactory. Furthermore, SPH models have been found to result in an unphysical process, i.e., transport from low concentration to high concentration, in anisotropic dispersion, resulting in areas with negative concentration. These two issues were resolved by ARS researchers at Oxford, Mississippi, with the development of model in which the transport equation for the SPH model was reformulated to prevent the unphysical process, and the reproducing kernel particle method was adopted to improve the accuracy. With this model, scientists can accurately simulate the transport of water temperature adjacent to river (the hyporheic zone) so that the stream-aquifer exchange can be better understood.
6. The CCHE_MIX_WQ model and AnnAGNPS watershed model have been linked to study the response of lake water quality to upland agricultural practices. ARS researchers at Oxford, Mississippi applied the AnnAGNPS model to simulate runoff and loads of sediment and nutrients from upland watershed. The simulated results were used as boundary conditions for CCHE_MIX_WQ to simulate water quality constituents in surface waterbodies. Two models were calibrated and validated using measured data in Beasley Lake Watershed. This integrated modeling system was applied to analyze the effects of alternative cropping and tillage practices on the water quality in Beasley Lake. This integrated modeling system is an effective tool to analyze the influence of upland watershed practices on the lake’s water quality.
7. A new computational module has been developed to estimate wave-induced erosion of embankments. The model is designed to accept either wind parameters or wave parameters. Wind data include the mean wind speed at a 10 m elevation and the effective fetch length, which is the offshore distance over the water surface in the direction of the wind. Wave data include the spectral description of the significant wave height (four times the standard deviation of the wave signal) and peak wave period (the wave period corresponding to the spectral peak). When wind information is provided as input, the model calculates the significant wave height and peak wave period based on JONSWAP spectrum. The wind-wave prediction algorithm using JONSWAP spectrum is tested using an existing field dataset. The model also considers wave height variation along the shoreline due to processes such as refraction, shoaling, breaking and runup, and calculates the wave related shear stress along the embankment face. This module has been successfully tested by ARS researchers at Oxford, Mississippi, for a set of hypothetical bank erosion situations. A field monitoring site for measuring wind, waves and retreat was constructed in an irrigation reservoir near Shelby, MS. The reservoir embankments were recently repaired with 1:3 inner slopes and refilling was in progress during the writing of this report.
8. All of the necessary datasets for running AnnAGNPS into AIMS have been generated. Some of these datasets have already been stored in the AIMS data server, while others currently steamed from various resources (i.e. NLDAS-2 climate). Topographic data has been generated by ARS researchers at Oxford, Mississippi, for the entire United States by running TopAGNPS, a topographic parameterization program for AnnAGNPS. TOPAGNPS runs were completed for 4,799 watersheds (THUCs) using DEMs of 1 arc-second (~30 m) resolution. Soil data is obtained from NRCS Soil Data Access service and processed to produce aggregated data for AnnAGNPS. Historical climate data is derived from the North American Land Data Assimilation System Phase 2 (NLDAS-2) obtained from Hydrology Data Rods. Where NLDAS-2 is unavailable or incomplete, the climate data is supplemented using Daily Surface Weather and Climatological Summaries (DAYMET). The assimilation and aggregation of the management data for the entire United States is completed. Having the datasets prepared, it is now possible to produce ‘ready-to-run’ AnnAGNPS simulations using only data from AIMS. Automated methods to prepare required input files for geometry, soil, climate, and management for AnnAGNPS runs have been developed which enables simulations for any watershed in the United States using only AIMS datasets and with minimum user intervention. AnnAGNPS simulations were carried out on the Iowa River basin to assess ability of the AIMS cluster to handle simulations of watersheds with various sizes find out the maximum size of watershed that can be processed. Reach-by-reach simulations on the AIMS computational cluster significantly reduced the computational time and memory requirements, by enabling the parallelization of the problem.
9. A python library for CCHE1D and AnnAGNPS integration into AIMS has been developed. This library is an object-oriented framework used by ARS researchers at Oxford, Mississippi, to describe a watershed into a set of reaches and cells/catchments. The library uses reach and cell data sections, and the reach raster produced by the topographic analysis tool for AnnAGNPS, TopAGNPS, as input, and generates a set of input files for CCHE1D that links the geometry and source terms/boundary conditions between the two models. The CCHE1D delineation model/tool based on the watershed-merging algorithm has been integrated into CCHE1D-GUI. A Fortran version of this tool is available as well for further integration into AIMS.
10. New Agricultural Integrated Management System (AIMS) Computer System Updates have been implemented. The AIMS system was migrated by ARS researchers at Oxford, Mississippi, to a new machine (AIMS). The old machine (Menderes) is updated as the testing environment with debugging mode. Databases were also migrated and from Menderes to the AIMS machine and optimized. Operating system and software updates on the new AIMS machine were completed, and performance of both systems was tested. A monitoring system for Menderes and Live system was established. A separate system is setup for tile servers that can be used for AIMS and other web tools of NCCHE. The tile server currently houses USGS-HUC boundaries, NLDAS grid and USDA stream gauges. AnnAGNPS cells and network tile servers are generated tested for the Iowa River basin. The new tile servers will replace the visualization and behavior of the existing cell and network layers on the map page on AIMS. The new update will provide much faster response and eliminate the security issues. Cell and reach tile servers were generated for THUC-1055 and reach level categorization based on Strahler number is completed.
11. New AIMS-NLDAS Case studies have been conducted by ARS researchers in Oxford, Mississippi. AnnAGNPS simulations for The Upper Pearl River Watershed (UPRW) were carried out to demonstrate AIMS’ capabilities and test linkage between AnnAGNPS and CCHE1D. Due to its size, only selected small sub-watersheds within the UPRW were used for validation. Input data for AnnAGNPS were prepared from AIMS, and the results were compared with observations from the USGS station. Preliminary results showed that precipitation was accurately captured, while daily and monthly total streamflow compared reasonably good. Ongoing validation tasks include incorporating MODIS and NOAH ETs data and simulating more sub-watersheds within the UPRW where USGS observed data are available. To link AnnAGNPS and CCHE1D, the entire UPRW was delineated using AIMS-TOPAGNPS. Input data for AnnAGNPS was prepared from AIMS. The channel network for CCHE1D was generated from the TOPAGNPS reach network, with channels above the sixth Strahler Order converted into the computational channel network for CCHE1D. Ongoing work involves running AnnAGNPS simulations and providing input for CCHE1D as boundary conditions at specified nodes.
12. A test case for Goodwin Creek Watershed was carried out by ARS researchers at Oxford, Mississippi, using evapotranspiration (ET) forcing data derived from remote sensing products to AnnAGNPS model. ET sources including NLDAS-2 land surface models NOAH and MOSAIC, as well as ECOSTRESS and MODIS products were compared quantitatively, and their suitability was determined for use with AnnAGNPS in the Goodwin Creek Watershed. A sensitivity analysis of the model to ET datasets was performed. A selection of model outputs were compared against observed data acquired from 1982 to 1991. The results showed that providing actual ET directly to the model led to an improvement in the estimation of runoff and sediments in several scenarios. AnnAGNPS simulation for Pelahatchie Bay watershed were performed to test the model’s sensitivity to local ET values.
13. Successful expansion of the NCCHE AIMS Cluster. The AIMS cluster has been expanded by ARS researchers at Oxford, Mississippi, and is now comprised of 12 compute nodes in a 42U rack system located at the Mississippi Center for Supercomputing Research (MCSR) on the UM campus, which is a secure facility that provides a UPS system that features backup generators for high availability. This cluster now has a total of 192 high-performance CPU cores and 768GB of memory. Each compute node of this cluster consists of 2TB of NVMe local storage, as well as 50TB of shared network-attached storage (NAS). The cluster uses the SLURM workload manager software for resource management and job scheduling. For storage, the cluster also features a 50 terabyte Synology RS1221RP+ rack-mount RAID network-attached storage (NAS) unit with 2 terabytes of solid-state read/write cache. This storage is available to all cluster nodes via NFS mount. The cluster makes use of several different node types, including both Intel and AMD processors. While the master node and the first three compute nodes (the default queue) consist of Intel Xeon Silver processors, newer compute nodes have been added to the cluster which make use of AMD’s 3D V-Cache technology. These processors feature a larger (192MB) L3 cache that is shared among the physical cores, as well as a burst frequency of 5.7 GHz. They have been shown to be particularly good at handling TopAGNPS and AnnAGNPS workloads and, accordingly, they have been configured to be part of the SLURM “highperf” queue. Due to the performance and cost-effectiveness of these newer systems, additional compute nodes of this type are planned for the future as well.
Review Publications
Zhang, Y., Al-Hamdan, M., Bingner, R.L., Chao, X., Langendoen, E.J., Vieira, D.A. 2023. Generation of 1D channel networks for overland flow simulations on 2D complex domains. Journal of Hydrology. 628:1-15. https://doi.org/10.1016/j.jhydrol.2023.130560.
Fang, J., Al-Hamdan, M.Z., O'Reilly, A.M., Ozeren, Y., Rigby, J.R. 2024. A three-dimensional numerical model for variably saturated groundwater flow using meshless weak-strong form method. Environmental Modelling & Software. 175:1-22. https://doi.org/10.1016/j.envsoft.2024.105982.
Fang, J., Al-Hamdan, M.Z., O'Reilly, A.M., Ozeren, Y., Rigby, J.R., Jia, Y. 2023. A novel floodwave response model for time-varying streambed conductivity using space-time collocation Trefftz method. Journal of Hydrology. 625(A). Article 129996. https://doi.org/10.1016/j.jhydrol.2023.129996.
Zhang, Y., Al-Hamdan, M., Bingner, R.L., Chao, X., Langendoen, E.J., O'Reilly, A.M., Vieira, D.A. 2024. Application of a 1D model for overland flow simulations on 2D complex domains. Advances in Water Resources. 188. Article 104711. https://doi.org/10.1016/j.advwatres.2024.104711.