/EcoSLIM_CONUS

Under development

Primary LanguageFortran

EcoSLIM_CONUS

  • Long-term effective: Welcome collaborations from experts good at parallelization of Lagrangian method and its load balancing! Please contact me via cy15@princeton.edu
  • Reviewers please refer to input_scripts

A lagrangian particle tracking code

image
A particle tracking code simulates water ages and source-water mixing, working seamlessly with the integrated hydrologic model ParFlow-CLM (Maxwell et al., Ecohydrology, 2019).

Backward particle tracking based on ParFlow CONUS 2.0 model

image

10 representative publications using ParFlow and/or EcoSLIM

  • Yang C., Condon L., Maxwell R., 2023. Unraveling groundwater-stream connections at the continental scale. Nature Water, in revision
  • Yang C., Tijerina-Kreuzer D., Tran H., Condon L., Maxwell R., 2023. A high-resolution, 3D groundwater-surface water simulation of the contiguous US: Advances in the integrated ParFlow CONUS 2.0 modeling platform. Journal of Hydrology, https://doi.org/10.1016/j.jhydrol.2023.130294
  • Yang C., Maxwell R., McDonnell J., Yang X., Tijerina-Kreuzer D., 2023. The role of topography in controlling evapotranspiration age. Journal of Geophysical Research: Atmospheres, https://doi.org/10.1029/2023JD039228
  • Yang C., Ponder C., Wang B., Tran H., Zhang J., Swilley J., Condon L., Maxwell R., 2023. Accelerating the Lagrangian particle tracking in hydrology to continental-scale. Journal of Advances in Modeling Earth Systems, https://doi.org/10.1029/2022MS003507
  • Yang C., Maxwell R., Valent R., 2022. Accurate load balancing accelerates Lagrangian simulation of water ages on distributed, multi-GPU platforms. Computers & Geosciences, https://doi.org/10.1016/j.cageo.2022.105189
  • Yang C., Zhang Y.-K., Liang X., Olschanowsky C., Yang X., Maxwell R., 2021. Accelerating the Lagrangian particle tracking of residence time distributions and source water mixing towards large scales. Computers & Geosciences, https://doi.org/10.1016/j.cageo.2021.104760
  • Yang C., Li H.-Y., Fang Y., Cui C., Wang T., Zheng C., Leung L. R., Maxwell R., Zhang Y.-K., Yang X., 2020. Effects of Groundwater Pumping on Ground Surface Temperature: A Regional Modeling Study in the North China Plain. Journal of Geophysical Research: Atmospheres, https://doi.org/10.1029/2019jd031764
  • Tran H.*, Yang C.*, Condon L., Maxwell R., 2023. The Budyko shape parameter as a descriptive index for streamflow loss. Frontiers in Water, https://www.frontiersin.org/journals/water/articles/10.3389/frwa.2023.1258367/full
  • Tijerina-Kreuzer D., Swilley J., Tran H., Zhang J., West B., Yang C., Condon L., Maxwell R., 2023. Continental scale hydrostratigraphy: basin-scale testing of alternative data-driven approaches. Groundwater, https://ngwa.onlinelibrary.wiley.com/doi/abs/10.1111/gwat.13357
  • Swilley J., Tijerina-Kreuzer D., Tran H., Zhang J., Yang C., Condon L., Maxwell R., 2023. Continental scale hydrostratigraphy: comparing datasets to analytical solutions. Groundwater, https://ngwa.onlinelibrary.wiley.com/doi/full/10.1111/gwat.13354

Leveraging the latest parallel architecture, to accelerate the understanding of water cycle in the changing world!

  • A parallel version EcoSLIM based on domain decomposition using the latest multi-GPU with CUDA-aware MPI technique.
  • Halo cells are used around each subdomain to store particles out of boundary and then transfer them to neighbors.
  • This development aims to handle the particle tracking at the continental US scale with long timescale.
  • It can be applied to real cases now. Irregular boundaries are supported. HDF5 is supported.
  • Optimization continues. Technical support will be provided in about October 2022 after we release the first stable version.
  • Users are welcome to download and use per interests at current time. Please refer to README.md in src folder for details.
  • Enjoy!

Tests across three spatial scales

image

Test on CONUS2.0 without LB (particle distribution)

image

Subdomain demonstration

image

Parallel performance

Currently, optimization continues, so only the particle loop kernel got tested on the Della-GPU cluster at Princeton University. Each GPU node is equipped with 2 NVIDIA A100 GPUs and 2 2.60-GHz AMD EPYC 7H12 sockets. Each socket has 64 cores without hyperthreading.
image

  • Tests used ER_Shrub and LW_Shrub cases based on the Hillslope model in Maxwell et al. (Ecohydrology, 2019).
  • Speedup is calculated by comparing the wall-clock time used by 128-thread to that used by 2-A100.
  • Results show speedup of ~10-fold for ER_Shrub and ~12.5-fold for LW_Shrub on one node.
  • The more the particles, the higher the speedup. Particle numbers for ER_Shrub and LW_Shrub are 5.6- and 17.4-million, respectively.
  • Results show good parallel scalability across two nodes, ~10-fold to ~20-fold for ER_Shrub and ~12.5-fold to ~25-fold for LW_Shrub.
  • LB is Load Balancing. Sn represents LB schemes. Speedup by S3 using 4-GPU is smaller due to the uneven split in y direction (5 grid-cells).
  • Latest parallel performance tests based on the whole code show as good performance as this single kernel tests. This part will be updated soon.

Presentations

Activities

Acknowledgments

  • Thanks so much to the following software engineers for their guidance in the code development:
    NVIDIA, Carl Ponder; Princeton University/NVIDIA, Bei Wang
  • Thanks so much to the CONUS2.0 team for offering the CONUS2.0 ParFlow model for code tests
  • Thanks so much to Prof. Reed Maxwell at Princeton University and Prof. Laura Condon at the University of Arizona for their support in the application of computational resources: We won the NACR Accelerated Scientific Discovery program 2021 fall. We will run particle tracking based on the CONUS2.0 ParFlow model on the coming NCAR supercomputer Derecho using 100 NVIDIA A100 GPUs