Mussel farm product quality remains a challenging problem for operational marine science. In an operational scenario, the model chain, orchestrated in a work ow fashion, produces a huge amount of predicted spatially-referenced (big) data. These work ow components have been integrated into the Framework to Advance Climate, Economic, and Impact Investigations with Information Technology (FACEIT), a work ow engine and data science portal based on Galaxy and Globus technologies. We describe how FACE-IT work ows can be used to couple many simulation/prediction models, leveraging high-performance and cloud computing resources to enable fast full system modeling in order to produce operational predictions about the impact of pollutants spilled out from both natural and anthropic sources in mussels farming high density areas.
Projections of future food quality require data from
climate models to forecast future conditions, coupling weather
models, wind-driven sea waves models and ocean
circulation/river advection models leveraging, transport and
diffusion of pollutants on projections about future
infrastructures, as new shery and mussel farm installments, land-use
and land cover [
A system supporting this kind of decisions would require: the management of di use and point pollution sources; the ability to scale in terms of domain size; the computational performance and e ectiveness needed to provide results for decision support.
We describe here a real world operational and on-demand
application for mussel farm food quality prediction and
assessment [
The rest of this paper is as follows. x2 introduces the general FACE-IT infrastructure and how it has been developed in the context of agricultural modeling and food quality and extended in order to support the described application; x3 details the application work ow and how di erent models have been implemented to t the proposed work ow infrastructure; x4 discusses computational and environmental issues as carried out from data analysis; and nally x5 presents conclusions and proposes future work. 2.
We have previously developed the Framework to Advance Climate, Economic, and Impact Investigations with Information Technology (FACE-IT) for crop and climate impact assessments.
This integrated data processing and simulation framework
enables data ingest from geospatial archives; data
regridding, aggregation, and other processing prior to simulation;
large-scale climate impact simulations with agricultural and
other models, leveraging high-performance and cloud
computing; and post-processing to produce aggregated yields
and ensemble variables needed for statistics, for model
intercomparison, and to connect biophysical models to global
and regional economic models. FACE-IT leverages the
capabilities of the Globus Galaxies platform [
The FACE-IT infrastructure is extended in order to support applications related to weather, sea wave conditions, ocean circulation and pollutant transport and di usion.
FACE-IT Galaxy work ows provide a robust and e
ective integration of earth science features in Globus Galaxies
in order to prepare a stable environment for virtually
endless data work ow-based applications. Mainly leveraging on
the datatype enforcement using the NetCDF Schema, the
latest FACE-IT Galaxy improvements are focused on static
and dynamical maps for data visualization, dedicated tool
parameters and new data sources [
Improving the evaluation of pollution e ects in aquatic ecosystems is important for economic pro t and to improve environmental sustainability. To achieve this target we designed and the implemented WaComM, a three dimensional decision support model enabling the simulation and prediction of pollutant spills, transport and dispersion in both inshore and o shore environments.
WaComM is driven by a complex chain of outputs from observational data, weather and oceanic models. The computational process is shown in Figure 1. First, data for the region is used to initialize the Weather Research and Forecast (WRF) model. WRF is used to compute wind conditions that are one of the forcing of sea surface current forecasted by Regional Oceanic Modeling System (ROMS). The nal result of WRF-ROMS coupled models is a hourly modeling simulation of the 3D hydrodynamic ow that we used as input data for WaComM to follow the pollutants Lagrangian transport.
The model chain has been integrated into FACE-IT Galaxy to be a qualitative and (semi) quantitative tool for human health risk assessment that could be helpful to achieve a better management of o shore human activities and a making decision support tool to select the best marine areas where these activities could be deployed (Figure 3). 3.1
The Weather Research and Forecasting (WRF) [
WRF experiment. 4. UnGrib wraps the WRF software component with the same name. UnGrib decode the GFS data in an intermediate le used by other tools. 5. MetGrid shares the name with its regular WRF counterpart which is the wrap. It interpolates data produced by the UnGrib tool on the domains produced by GeoGrid. 6. Real prepares the data produced by MetGrid for the real case simulation. It is responsible of the creation of the boundary and initial conditions for the WRF tool. 7. WRF wraps the core model. In the current implementation it relays on a only OpenMP enabled WRF compiled binary or on a hierarchical parallelism enabled on MPI distributed memory, OpenMP shared memory and GPGPUs for some model components. 8. WRF Aggregator aggregates WRF outputs of a speci ed domain in a single NetCDF le remapped on a equal spaced regular latitude longitude grid. 9. WRF Plot: it is used for simply, fast and reliable
WRF output plotting.
Integration of WRF into the FACE-IT Galaxy framework eventually an interpolation of MyOCEAN data [
We have developed a tutorial to disseminate the use of WRF/FACE- 5. WRF to ROMS: While the initial and boundary IT. conditions provided are provided by the Copernicus project with the \OCEAN to ROMS" tool, the wind 3.2 Regional Ocean Model System friction data is gathered directly from WRF outputs. The wind eld and other variables are regridded and eventually interpolated on the actual ROMS domains.
Nevertheless previous experiences with the Princeton Ocean
Model (POM) parallelization [
We developed the following tools: 1. Make ROMS from WRF Experiment reads a
WRF experiment description dataset and produces a ROMS experiment description dataset. The ROMS experiment is created to be compliant with the WRF domains, initial and boundary conditions and the simulation starting time and duration. 2. Get MyOCEAN: While we are able to download initial and boundary conditions for the WRF model from NCEP/GFS, there is no analogous worldwide service for ocean modeling. Because our mussel farms application is focused on the central Tyrrhenian sea east sector, we download data from the services provided by the Copernicus European Project for marine environment monitoring. 3. Sea Grid is a toolbox backend that we developed to emulate the behavior of WRF's GeoGrid module.
SeaGrid produces the digital bathymetry model of the computation domain de ned by the ROMS experiment descriptor dataset. 4. MyOCEAN to ROMS acts in the same shape of
WRF's MetGRID module performing a regridding and 3.3
WaComM is an evolution of the Lagrangian Assessment
for Marine Pollution 3D (LAMP3D) model [
We developed the following tools: 1. Make WaComM experiment from ROMS reads a ROMS experiment description dataset and produces a WaComM experiment description dataset. The WaComM experiment is created to be compliant with the ROMS domains, initial and boundary conditions and the simulation starting time and duration. 2. WaComM wraps the WaComM executable. At the present it support restarts and shared memory parallel implementation based on OpenMP. This tool produces a WaComM output as a NetCDF-based datatype and a comma-separated-values le with the particles status. This le could be used as input in order to implement restarting. 3. WaComM Aggregator aggregates WaComM outputs of a speci ed domain in a single NetCDF le remapped on a equal spaced regular latitude longitude grid. 4. WaComM Plot is used for simply, fast, and reliable
WaComM output plotting.
The FACE-IT application on mussel farms modeling for food quality assessment and human health protection have practical implications from both computational and environmental point of view. 4.1
The FACE-IT application described in this paper is computationally intensive, involving loosely coupled parallelization at the work ow level and tight coupled parallelization at the tool level. The need for an operational system drove us to design and implement two di erent deployment scenarios.
The rst deployment scenario, FACE-IT Amazon Web Services, is that normally used by FACE-IT. It involves EC2 machines onto which tasks are scheduled by HTCondor integrated by an ad-hoc daemon monitoring the Condor queue. This daemon analyzes the ClassAd requirements for each submitted job and manage the instantiation of the needed computation resources.
The second deployment scenario is a Dedicated HPC Cluster. FACE-IT Galaxy can be deployed on dedicated HPC clusters using the DRMAA interface to Torque if the system local scheduler match the interface requirements. If this is not the case of the current deployment, we developed a custom job runner interacting with the local scheduler with a custom implementation of the queue management. We used this approach successfully providing HPC resources to a distributed / shared memory and GPU accelerated version of WRF. This approach also supports shared memory parallel versions of ROMS and WaComM. 4.2
Weather conditions in uence the transport of pollutants near mussel farms. Our case study focuses on mussel farms in the Punta Terone (between Capo Miseno and Punta del Poggio) and Centocamerelle (Gulf of Pozzuoli, Campania Region) areas. These mussel farms are classi ed as \type A" in accordance with current Italian legislation. The concentration of Escherichia coli must be less than 230 MPN (most probable number) and the concentration of Salmonella must be zero.
We ran simulations during the historical period from December 7th{21st, 2015, chosen to correspond with available eld measurements. As shown in Figure 4.2, the forecasted average areal distribution of tracers falls within a region bounded by Lonmin 14.08, Lonmax 14.1, Latmin 40.76 and Latmax 40.81. Chemical analysis on the mussels (Mytilus galloprovincialis) showed that on December 9th the concentration of E. coli was greater than the legal limits, while on December 21st it was lower than the legal limits).
Comparison between numerical forecasts and chemical analysis show a remarkable similarity in trends, although more observations will be needed before the method can be fully assessed. These early results do con rm that the system holds promise as a decision support tool for many applications that depend on sea quality, and require forecasts in support of local observations and measurements (Figure 4). 5.
We described our use of the Globus Galaxy-based FACEIT technology in a project that extends FACE-IT's initial focus on climate, agricultural and socio-economic interactions to a tactical pollutant prediction application.
We will continue to maintain and improve the FACE-IT
core infrastructure [
We thank the Globus Galaxies, Globus, and Galaxy teams for their outstanding work on those systems and for their assistance with this project.
This work is supported in part by NSF cyberSEES program award ACI-1331782; NSF Decision Making Under Uncertainty program award 0951576; DOE contract DE-AC0206CH11357; project IZSM ME04/12 RC/C78C 120017001, \Mapping Escherichia Coli and Salmonella pollution in mussel farm areas and model prediction comparisons"; and European Union Grant Agreement number 644312-RAPID { H2020-ICT-2014/H2020-ICT-2014-1 \Heterogeneous Secure Multi-level Remote Acceleration Service for Low-Power Integrated Systems and Devices," using GVirtuS open source software components. EC2 resources were generously provided by Amazon.