-
Notifications
You must be signed in to change notification settings - Fork 0
Cyberseminar Case Study Modules
Hydrologists generally share an understanding of the processes that contribute to the temporal and spatial evolution of moisture fluxes (e.g. evapotranspiration, streamflow) and states (e.g. soil moisture, snow water equivalent) in a watershed. However, for any particular watershed, they do not necessarily agree on the relative importance of these processes and how this understanding should be codified into computer models. As a result, there is a profusion of hydrologic models that not only differ in their process representations and numerical solution techniques, but also in their model parameters, input requirements, and file formats. Unfortunately, this hampers the advancement of hydrologic understanding in two ways. First, models tend to differ in so many ways that it is difficult to attribute differences in model outcomes to any specific model component (e.g. Nijssen et al., 2003). Second, because of the unique data and format requirements of most models and the resulting effort expended in their implementation and analysis, hydrologists tend to gravitate towards the models they know rather than the models they need. These modules will encourage and train participants to approach their modeling activities as a formal evaluation of multiple working hypotheses on model representations of physical processes. In addition, the modules will rely on the techniques learned from the training modules in Section 5.1 to encourage hydrologic modeling best practices. The training modules will consist of pre-hackweek and hackweek activities, with the former focused on the development of CI skills and the latter focused on collaborative research using project datasets and models.
Case Study Module 1. Hydrologic Model Construction and Testing of Modeling Hypotheses (Lead developer: Bart Nijssen)
The Structure for Unifying Multiple Modeling Alternatives (SUMMA) method (Clark et al., 2015a; 2015b) has recently been advanced to address the challenges outlined above by allowing users to systematically evaluate competing process representations and different spatial configurations within a single hydrologic modeling code. Integration of SUMMA with HydroShare is part of a separate NSF-funded project (OAC-1664061), in which both Bandaragoda and Nijssen are participants, and which will develop CI in support of SUMMA configuration, application, and analysis. Part of this includes the development of a python package (pySUMMA) that will enable users to configure, instantiate, and run SUMMA models from within Jupyter notebooks, as well as the development of a SUMMA app to automate SUMMA configuration in HydroShare. We will develop a CI training module that will instruct users on the configuration, implementation and analysis of alternate SUMMA instances. Pre-hackweek training will include self-study modules that participants can use to familiarize themselves with SUMMA and its implementation in HydroShare. We will translate existing SUMMA test cases, which cover a wide range of SUMMA capabilities, into Jupyter notebook modules, which can be analyzed and modified using pySUMMA. Assessment will investigate the functionality of pre-hackweek activities, and WaterHackWaterHackWeek Week activities, for allowing participants to formulate new or expanded research questions using the project model and data.
Machine learning advancements in computer science are responsible for major contributions towards improving the optimization of hydrologic models. However, the complexity of applying optimization to high dimensional coupled physical models varies so much by application, that there is limited guidance on best practices for use. We will develop a module to describe the underlying numerical methods used in a multi-objective optimization algorithm (DREAM; Vrugt, 2016) available in the Spotypy library (Houska et al., 2015). The case study will illustrate how to calibrate a highly parameterized spatially distributed hydrologic model (DHSVM, Wigmosta et al., 1992) coupled with a Landlab (Hobley et al, 2017) hydrometeorology component that uses climate station observations (following Livneh et al, 2015) and physics-based lapse rates to match high elevation atmospheric models (WRF; Henn et al., 2017; Currier, 2016; Currier et al., 2017) in a cloud computing environment (HydroShare). Assessment of this module focuses on determining user understanding of basic optimization (machine learning) concepts and vocabulary, as well as the user’s ability to develop new or expanded optimization strategies and experimental designs for their projects.
Hydrologists often use remote sensing datasets as a source of calibration and validation for model outputs. Aligning high resolution hydrology model output with remote sensing products requires aggregation and reprojection of large datasets to common spatial and temporal scales. We will develop a module to illustrate the use of several Python tools for the intercomparison of large gridded datasets, and the sharing of access to these products using cloud resources. The case study will be drawn from an ongoing NASA-funded effort to understand hydrological changes of the High Asia region. Data from the NASA Land Information System (Kumar, 2008) will be aggregated to coarse spatial resolution grid cells for comparison to the NASA Gravity Recovery and Climate Experiment product (Luthcke et al., 2013). Python tools including Xarray (Hoyer and Hamman, 2017) and Dask will be deployed in Jupyter Notebooks to access and process the data from cloud storage. Assessment of this module will focus on participant understanding of cloud computing architecture, multi-dimensional array manipulation and the potential for enhanced collaboration through data sharing in the cloud.
The Landlab group (Tucker, CU, Boulder; Gasparini, Tulane U; Istanbulluoglu and Bandaragoda, UW) have been developing earth surface modeling components that describe a single-process and coupled models that involve multiple processes since the beginning of the Landlab project in 2014. Landlab is available on HydroShare, and about a dozen of coupled models have been developed and deployed on HydroShare as Jupyter notebooks which have been used in CSDMS annual hands on meeting clinics, and various short courses (e.g., GSA-2018) and the CUAHSI Biennial meeting in 2016. In our proposed CT activities, we will focus on several Landlab components commonly used by hydrologists, including overland flow routing by various methods, ecohydrologic vegetation dynamics, landsliding and sediment routing in river networks. Workflows will be organized mainly using these components with real-world examples from the Skagit and Puyallup river systems that drain steep erodible terrain, including peaks like Mount Rainier and Glacier Peak.