1. Introduction
Marshlands are one of the most vital ecosystems on the planet as they provide a number of ecosystem services to humans and serve as essential habitats for a diverse range of species. Marshes are, however, at the frontlines of sea level rise (SLR). According to the Intergovernmental Panel on Climate Change (IPCC), sea levels could rise anywhere from 1 to 3 m in the coming centuries [
1]. The sheer speed of this change could be disastrous to the pertinent coastal wetlands due to the effect of “coastal squeeze”, a phenomenon in which wetlands are unable to migrate inland both due to the speed of SLR and the presence of coastal human-made infrastructure that blocks the migration path of wetlands [
2]. It is thus necessary to assess the vulnerability of wetlands for communities so that policymakers can produce informed policies that prevent marshland shrinkage and even disappearance, considering the ecosystem services that marshes provide to plant and animal species [
2].
The study of the potential loss of marshland has intrigued researchers from different domains of study, including biology, economics, ecology, anthropology, and geography. A number of domains can benefit from being able to efficiently analyze and quantify the amount of marshland that could be lost in a specific study area, which requires the use of SLR modeling.
SLR modeling began in the 1980s as awareness of Antarctic ice sheet instability grew. Early estimations were simple analyses of sea level change versus global temperature, dubbed “bathtub models”, which eventually grew into estimations using models and larger amounts of contextual information in the late 1980s [
3]. Considering the difficult task of estimating an exact rate of SLR over the coming decades, studies examining SLR consequences often examine the effects of multiple rates of SLR to determine trends and differences between scenarios (see [
4]). To support such analyses, a series of models have been developed to evaluate SLR impacts, such as DIVA, SimCLIM, BTELSS, or SLAMM [
5]. To quantify potential land cover change as seas rise, a spatial representation based on raster data is often used to provide measurements of loss due to the easily calculated grid of cells for explicit representation of spatial processes [
6].
The Sea Level Affecting Marshes Model (SLAMM), developed in the 1980s [
7], can quantify the land cover change in response to SLR by analyzing each raster cell within a study area against its neighboring cells in tandem with elevation and slope data. The model is recognized for its ability to simulate inundation, erosion, soil saturation, and barrier island overwash. SLAMM can run simulations with input data that are freely available, making it useful for researchers, particularly with budget constraints. The model creates GIS-compatible outputs, rendering it well-suited for researchers with GIS experience or backgrounds. SLAMM has been used extensively in scientific research and known bugs are solved as the program is continuously updated with new releases [
8].
Examples of SLAMM being used for the study of SLR impact include Linhoss’ study [
4] on inundation and marsh migration in Northeastern Florida which found significant land cover change for upstream rivers driven by the migration of wetland [
4]. Another example is Akumu et al.’s study in 2011 [
9] on the potential impacts of SLR in a rapidly growing region of New South Wales, suggesting approximately 25% freshwater marshes in the region could be lost by the end of this century [
9]. The model has been met with criticism due to the lack of empirical outputs [
10]; however, incorporating SLAMM into workflows can increase quantitative output data for verifiable land cover change results. For more examples of SLAMM applications in wetland studies, please refer to [
11,
12,
13]. Furthermore, SLAMM can be used to gauge marshland loss for time periods in the past by using historical data. Results can be compared with observed sea level and marshland loss trends, a practice called hindcasting, in which researchers conduct retrospective analyses (see [
14,
15,
16]).
However, SLR modeling, represented by SLAMM, often includes analyzing different study areas at multiple spatiotemporal scales and with various SLR rates for scenario analysis. As a result, large amounts of time must be spent on preprocessing, analyzing, and post-processing geospatial data associated with SLR modeling that often needs to be repeated a number of times. These shortcomings often discourage those without geospatial data analytics backgrounds from studying the impact of SLR on marshland, which calls for a solution for SLR modeling.
Cyberinfrastructure technologies, more specifically, high-performing computing (HPC), massive data handling capabilities, and virtual organization [
17], have the power to automate and integrate data analytics and modeling modalities related to SLR studies. Cyberinfrastructure technologies encompass computing systems, advanced data management and visualization techniques, and collaborative software tools for innovative and transformative research [
17]. A cyberinfrastructure technology common in both academia and industries is the scientific workflows [
18,
19], or automation technologies that can integrate and execute various models or tasks without (or with minimal) human intervention. Scientific workflows can be easily repeated to support reproducible scientific studies, allowing for, for example, verification and validation of results by other professionals collaboratively [
20], and can be implemented in a way that new data and parameters can be plugged in as needed [
21]. The implementation of scientific workflows can range from the use of basic scripts that call upon data and models to produce outputs in an automatic manner, to complex programming with GUI that chains data and models together [
22].
Scientific workflows are created for the purpose of supporting and automating tasks so that they can be used or re-used by people throughout the research community [
20,
23]. Scientific workflows have been increasingly repurposed and reproduced so that scientists with less familiarity with scientific workflows can use them in their own research. There have been ongoing improvements to scientific workflow validation, minimizing potential imprecision of results. Parallel computing has been incorporated into scientific workflows which allows for results to be produced much faster than sequential computing (also known as serial computation), and cloud computing has allowed researchers to schedule scientific workflows to be executed on remote virtual machines [
18]. Scientific workflows are often conceptually represented as directed acyclic graphs (DAG), which represent relationships between tasks (as nodes) and are used for various business and scientific purposes. Nodes in a scientific workflow graph can represent any computational task from data processing to complex analytics or models [
24]. Edges in the graph show how data move throughout a workflow and are processed. These improvements to scientific workflows over the past two decades have made them a useful tool for research in geospatial data sciences considering the large amounts of data that are needed for reproducible processing and analysis [
22].
As an example of scientific workflows implemented in wetland studies, Tang et al. in 2017 [
25] developed a cyber-enabled spatial decision support system (SDSS) framework to locate basecamps for the purpose of inventorying mangrove forests in the Zambezi River delta of Mozambique. Tang et al. utilized cloud computing and high-performance computing within scientific workflows to work with large amounts of data. Further, Tang et al. [
26] aimed to estimate the biomass of mangroves throughout the world due to previous studies mainly focusing on mangrove biomass at local scales. Mangroves are an important habitat in relation to climate change studies, as they sequester large amounts of carbon from the atmosphere [
26]. The study examined the biomass of mangroves throughout the world, which requires the handling of massive datasets that needed efficient and automated processing.
Further application of scientific workflows in marshland studies was reported by Felton et al. [
27], in which an automatic approach was used to detect wetlands present in existing data for large study areas. Their automatic approach allows for cost-effective and efficient wetland mapping. Felton et al. were able to create a workflow that was less likely to falsely predict wetlands than the National Wetlands Inventory dataset. Wu et al. [
28] explored the development of automated systems to map the flooding dynamics of wetlands using light detection and ranging (LiDAR) and Google Earth Engine. Wu et al. [
28] found that their workflow was able to delineate wetland flooding status as well as illustrate the hydrological dynamics of the wetlands within the study area. Wu et al. stressed that their workflow can be replicated and scaled up or down to study various wetland areas. These examples show scientific workflows being implemented in wetland studies to improve data processing and analytics to help understand environmental trends. However, scientific workflows have yet to be studied to improve the efficiency of the SLR modeling.
Scientific workflows have become more prevalent in interdisciplinary areas of research, as big data and data science converge into interdisciplinary study domains [
26,
29]. Specifically, in the realm of geospatial and environmental research, the amount of data available to researchers has grown extensively. This growth allows for more expansive data-driven research, but also leads to computationally demanding and labor-intensive geospatial data processing or analytics. These efficiency shortcomings apply to the study of marshland and SLR modeling. By combining scientific workflows with sea level prediction models, data (input and output) can be analyzed efficiently and repeatedly, allowing for the dissemination of SLR predictions, and potentially supporting policymakers to enact climate change mitigation policies or adaptive strategies. Thus, in this study, we will investigate how scientific workflows can automate data analytics and modeling techniques specifically for the purpose of estimating potential change (mostly loss) of marshland and associated spatial characteristics in response to SLR.
3. Methodology
In this section, we present a spatially explicit framework of using scientific workflows for automated SLR modeling. Then we focus our discussion on four scientific workflow modules that constitute the entire framework. Last, we described our implementation of this scientific workflow framework.
3.1. Framework
The overall spatially explicit framework, named GSWAM-SLR (GIS-based Scientific Workflows for Automated Modeling of Sea Level Rise; see
Figure 3) consists of four major modules: data preprocessing, SLR modeling, data post-processing for landscape pattern analysis, and parallel computing. The purpose of implementing the overall framework includes minimizing human interaction, preventing potential human error, along with decreasing the overall time to analyze the data. In the data preprocessing module, the collected data inputs are processed within GIS environments and transformed into the required spatial datasets for the SLAMM model. For the module of SLR modeling, model runs for scenario analysis are configured and executed. We implement a landscape pattern analysis module during post-processing to further analyze the GIS-based SLR modeling results for their spatial characteristics. Considering the computational intensity of SLR modeling and landscape pattern analysis, we used a parallel computing module to accelerate these two computationally demanding steps.
3.2. Scientific Workflow for Data Preprocessing
SLAMM requires a minimum of three spatial datasets: elevation, land cover, and slope. The finalized datasets used for SLAMM are derived from a DEM dataset, the National Land Cover Dataset (NLCD), and the National Wetlands Inventory (NWI). Prior to preprocessing all of the data, the NWI data requires to be reclassified to corresponding categories that SLAMM uses (refers to
Table 2) to simulate the land cover change of the study area. For example, regularly flooded marsh is category 8 for SLAMM, and any regularly flooded marsh from the NWI dataset required this category added as an attribute.
The pre-processing scientific workflow (see
Figure 4) begins by re-projecting the DEM dataset to Albers Conical Equal Area (WGS1984) to match the land cover dataset. The DEM is then resampled from a spatial resolution of 3 m to 10 m. This reprojected and resampled DEM is used in the calculation of slope. Both DEM and slope rasters are exported to ASCII format for use in SLAMM. The NWI data are converted to raster data format after reprojection to Albers Conical Equal Area (again, to match the land cover dataset). The NLCD dataset is clipped and snapped (through an Extract by Mask tool in ArcGIS Pro) using the resampled DEM. This step is necessary to reduce the computing time of the following steps by limiting the national land cover dataset to only cover the study area. This step is followed by reclassification to match the required SLAMM input land cover categories. The NWI and NLCD datasets are combined into a new raster dataset of the land cover. In the case of two overlapping pixels, the NWI Wetlands dataset is chosen as it contains more detailed SLAMM categories. The land cover mosaic raster is then clipped and snapped to the resampled DEM for consistency. Finally, this combined land cover raster is converted to ASCII format. The preprocessing model for this study was implemented inside ArcGIS Model Builder (in ArcGIS Pro 2.7.2) but can be accomplished inside any GIS software.
3.3. Scientific Workflow for Sea Level Rise Modeling
SLR model used in this study is SLAMM. The SLAMM model can create spatially explicit outputs of various SLR scenarios, making it useful for comparing how different rates of SLR could impact a study area. Conceptually, the SLAMM model is based on a raster-based spatial representation to examine each cell within a study area against its neighboring cells for processes associated with SLR. For example, if a marshland cell is surrounded by ocean and developed land, then the marshland in that cell is unable to migrate and will be lost to SLR [
8].
SLAMM predicts land cover change using its decision trees built into the model to examine geometric and qualitative relationships between land cover classes. The decision trees examine each cell individually and the neighboring cells, considering the cell’s proximity to different bodies of water, elevation, and its land classification, to determine if the cell will be converted and what the new land type will be. SLAMM takes into account six possible processes that affect wetland dynamics in response to SLR: inundation, erosion, overwash, saturation, accretion, and salinity. The SLAMM uses spatial processes to decide how marshland could potentially be converted. For the conversion of wetlands, the possible movement or loss of a wetland cell is a function of the slope of the cell and the minimum possible elevation of wetlands as seas rise.
To determine the potential land cover change in response to SLR, the SLAMM model uses varying scales of neighborhood analysis. Technically, when the potential of accretion of sediment is evaluated as seas rise, a cell’s distance to a river or tidal channel will play a factor in whether or not a wetland cell could experience sedimentation. On the other hand, when estimating the potential of overwash, cells directly adjacent to water have the maximum potential for erosion [
8]. For example, an inland fresh marsh will be converted to a transitional salt marsh when it falls under its lower elevation boundary. However, if it is adjacent or near to water bodies and is within a tropical region, it will convert to mangrove. Moreover, as the sea level rises, both inland-fresh marshes and tidal-fresh marshes can be converted into transitional salt marshes or irregularly flooded marshes. Subsequently, with further sea level increases, these marsh types will transit into regularly flooded marshes. Finally, they become a tidal flat area. In short, the conversion of a cell does not only depend on the type of the cell but is also related to its neighbors with respect to SLR.
The automation of SLAMM involves two main steps: model configuration and model execution. This two-step process is necessary due to the requirement of creating a parameter set that governs how the simulations are conducted. The parameter set is stored within a text file and contains parameters that define the configuration of the simulation. By automating SLAMM (in Python scripts here), researchers can efficiently explore various SLR scenarios with minimal human interactions. This streamlined approach allows for a more systematic analysis of the potential impacts of SLR on the study area, enabling researchers to assess the effects of different SLR scenarios on marshes.
While running SLAMM by hand one or two times is not extremely time-consuming, running it for many SLR scenarios manually can be prone to error due to, for example, its complicated user interface (see
Figure 5). First, the user must call the program, manually input each data source, set the parameters for each scenario, and run through the model output maps, which can be time consuming for each scenario. Therefore, automating the entire model configuration process can considerably cut down the time. Model configuration can be run either with scripts (see
Figure 6) or by hand to generate a parameter file, and that parameter file is used in model execution to run SLAMM many times without the requirement of clicking through each scenario by hand.
The second part of the scientific workflow, model execution, is the main component of the entire framework of this study. This scientific workflow (see
Figure 7) creates a batch file containing scripts that are used to run SLAMM automatically. The scientific workflow alters the desired parameter file to reflect various SLR scenarios. The SLAMM model is then run with specified parameters. This allows SLAMM to be run in the background as many times as needed. For example, if the scientific workflow is set to run SLAMM for 0 m of SLR by 2100, and increase the sea level by 0.05 m up to 3 m, then a total of 61 scenarios will be examined. The scientific workflow runs SLAMM, adds 0.05 m to the SLR within the parameter set, and runs SLAMM again, repeating this process until the SLR reaches 3.0 m. A user could easily adapt this scientific workflow for their study area. The model execution component of the framework could be deployed in a parallel computing environment. For example, rather than running all scenarios one by one, we could break the set of model runs into subsets each handled by a computing node—parallel computing (to be discussed in detail in
Section 3.5). Overall, scientific workflows of model configuration and execution are fundamental in the automation of SLR modeling.
3.4. Scientific Workflow for Landscape Pattern Analysis as Data Post-Processing
To reveal how the habitat of marshes spatially respond to SLR in the study area, it is essential to analyze how spatial patterns of different marsh types change across SLR scenarios. Therefore, we conducted landscape pattern analysis that uses landscape metrics to quantify characteristics of spatial patterns at different levels, including patch, class (type), and landscape. Landscape metrics allow us to evaluate spatial patterns in terms of landscape composition and configuration [
37], and have been applied (though not extensively) into sea level studies. For example, Torio and Chmura [
38] conducted landscape pattern analysis in the SLR study in Maine, USA. A set of landscape metrics were introduced to characterize landscape configuration and fragmentation. Wu et al. [
39] compared the thresholds of sea-level rise rate with respect to wetland area and landscape metrics such as mean patch size and mesh size. Wu et al. stressed that while the use of landscape metrics may lead to different thresholds, it allows for considering landscape characteristics and ecosystem dynamics into SLR studies [
39].
In this study, we chose to use three types of landscape metrics: area-, shape-, and aggregation-related so that we can evaluate spatial characteristics of land cover patterns from different aspects. Since landscape metrics falling into the same type can be highly correlated, we selected several representative metrics as follows. Area-related metrics quantify the amount of land cover at patch or class level. Shape metrics assess the compactness of a land cover patch. Aggregation metrics reveal the degree of landscape fragmentation. We selected the following landscapes metrics that describe the corresponding aspects of spatial patterns at different levels (patch, class, and landscape).
where A
i is the area of the ith patch of a land cover class (type). It is a simple statistic to summarize the area for each class of interest. We use it to evaluate changes of areas in different scenarios.
where E
i is the perimeter of the ith patch of a class; min(E
i) is the minimum perimeter for a hypothetical patch with the same area as the ith patch. The mean shape index examines the complexity of patches for a class. This index is characterized by a range of values starting from 1, which indicates the round shape. It has no upper limit where the higher values are, the more complex the shapes become.
where D
ij is the distance from patch j to the nearest neighbor belonging to the same class i. mENN describes the level of isolation among the patches. The higher the value is, the more isolated the pattern becomes.
where E is the total perimeter of a class; min(E) is the minimum perimeter for a hypothetical patch with the same total area of the corresponding class. LSI shares a similar formula with mSHAPE. However, when the total perimeter of a class is used, it depicts the degree of landscape fragmentation for the patterns of patches within a class. As the LSI value increases, it indicates a decrease in compactness, implying that the patches of the class become less compact and more fragmented.
where p
i is the percentage of the ith class within the landscape. Shannon diversity index measures the diversity of the landscape. A higher value of this index implies a higher diversity within the landscape. In other words, the landscape becomes less likely to be dominated by one class as the value increases.
3.5. Scientific Workflow for Parallel Computing
Conducting a scenario simulation of SLR with a fine interval, along with landscape pattern analysis for each scenario, presents a significant computational challenge due to the number of simulation scenarios and fine spatial resolution of GIS data. It is often beyond the computing capability of a single computer to conduct an expected number of scenarios along with the landscape pattern analysis within a reasonable amount of time. To address the computational challenge posed by the analysis of all SLR scenarios, we use parallel computing that breaks a large analysis task into a set of computationally manageable subtasks. By deploying these subtasks over HPC clusters, we can execute scenario analysis in parallel (see
Figure 3). This parallelization enhanced the efficiency of the analysis, as multiple scenarios could be processed simultaneously on various computing nodes on an HPC cluster.
In this study, the SLR model and landscape pattern analysis were implemented in different operating systems: Windows and Linux. Considering that the SLAMM software (version 6.7 beta) for SLR modeling is only available within a Windows environment, we used a Windows-based HPC solution for SLR modeling. As the SLAMM model runs for various scenarios that are independent of each other, this makes the scenario analysis of SLR become an embarrassingly parallel computing problem (no or little communication exists among computing nodes; see [
40]). We implemented the parallel SLR modeling using Windows batch scripts. Further, landscape pattern analysis is capable of being deployed on Linux-based HPC environments often with a large number of computing nodes available. Parallel landscape pattern analysis is then conducted using Linux Shell scripts: landscape pattern analysis for each SLR simulated land cover data is wrapped as an independent computing task that is ported to a single CPU on a Linux HPC cluster. Specifically, the SLR model with the model input and scenario configuration for a scenario is fed to each computing node on a Windows-based computing cluster—i.e., the running of all scenarios is parallelized. Then, the output of each scenario simulation (simulated land cover pattern) serves as the input for the subsequent landscape pattern analysis that is conducted on a Linux-based computer cluster. The final output of the scientific workflow for high-performance computing delivers both simulated land cover at a given scenario of SLR and the landscape pattern metrics for corresponding land cover classes of interest.
3.6. Implementation
To execute the entire scientific workflow for this study, a suite of software platforms was utilized. For data preprocessing, ArcGIS ModelBuilder and Notebooks were used for spatial data processing inside ArcGIS Pro 2.7.2. For model configuration and model execution, a web-based Jupyter Notebook file was used to run Windows batch scripts for automating SLAMM execution inside Google Chrome (version 91.0.4472.124). The SLAMM version used was 6.7 Beta (build number 6.7.0242). Landscape pattern analysis was written in C/C++ programming language.
5. Discussion
In this study, the Central South Carolina coast where the City of Charleston is located was used as our study region to exemplify how scientific workflows can be of help on SLR studies. Our experimental results indicate a significant loss of various marsh types within the SLR range of 0 to 1 m. This suggests that these marshes are highly vulnerable to even relatively small increases in sea level in the study area. Furthermore, the results indicate that from 1 to 3 m of SLR, there is no noticeable increase in the area of most marsh types. This suggests that these marshes are not able to find sufficient suitable habitats to compensate for the losses experienced within this range. In contrast, transitional marshes show a gradual increase in response to the rising sea levels. This indicates that transitional marshes are more resilient to SLR compared to others in our study area.
By analyzing landscape pattern changes in terms of shape, landscape fragmentation, and landscape diversity, we can gain a better understanding of the spatial dynamics of marshes in response to SLR. The use of landscape metrics can be of great help in quantifying spatial characteristics and their change with SLR. Further, thresholds of SLR with respect to various metrics (nonspatial or spatial) can be different for the same study region. This is because spatial patterns of land cover types are modified as sea level rises, which leads to nonlinear responses of alternative land cover types that interact with each other.
The threshold of SLR could potentially be different for various study areas, as coasts throughout the world will experience SLR differently. The overall land cover pattern of a study area affects how the land will change, which will in turn drive the development of the land cover pattern—nonlinearity and feedback loops exist in coastal ecosystems driven by SLR. Even for the same study region, the threshold of SLR may be different, depending on study purposes (e.g., minimizing the loss of marshland or protecting of specific marshlands for landscape conservation in need of consideration of spatial characteristics [
37]). In other words, the use of multiple landscape metrics (including area—patch area) is recommended while analyzing landscape response to SLR.
6. Conclusions
SLR is a substantial threat to the sustainability and resilience of our coasts. The IPCC predicts that seas could rise anywhere from 1 to 3 m by the end of the century. The rate of SLR can potentially increase the negative consequences of climate change, creating an opening for SLR modeling automation that can quickly and accurately predict the potential land cover change of coastlines. Making SLR modeling automated can also encourage discourse on how SLR will affect domains such as economics, politics, anthropology, and more. The computationally intensive task of environmental modeling, represented by SLR modeling in this study, can be automated and accelerated by incorporating scientific workflows in this study. Our study suggests the following findings:
Scientific workflows are the cyberinfrastructure technology beneficial to SLR modeling, which are analysis programs that can be replicated by researchers. Scientific workflows are prevalent throughout data science and are becoming increasingly popular in interdisciplinary studies. In geospatial studies, scientific workflows are used in many ways, including wetland and SLR analysis;
Creating a scientific workflow to automate SLR models provides solid support for analyzing, for example, how different rates of SLR affect shorelines and landscape dynamics;
The use of scientific workflows for SLR model automation lowers the difficulty of SLR modeling which enables researchers to conduct SLR studies through the lens of their respective domains;
This study creates an easily repeatable scientific workflow that automates the preprocessing of input data, data analysis, and the post-processing of data to produce repeatable results for SLR modeling. With GIS inputs and automation support, a researcher can easily and efficiently investigate potential SLR effects on land cover types (e.g., marshland in this study) for a desired study area.
The entire workflow framework in this study integrates a set of individual scientific workflows. For data preprocessing, the workflow processes geospatial data and generates input data for SLAMM. For spatial modeling, the scientific workflow of model configuration specifies the set of parameters and input data of the SLR model for different scenarios. The model execution workflow allows for running SLAMM repeatedly for various SLR scenarios. For data post-processing, the model extracts land cover data from the SLAMM outputs and conducts landscape pattern analysis to quantify the spatial characteristics of landscape patterns in response to SLR. The parallel computing module of the framework provides support for accelerating geospatial data analytics and SLR modeling, which are computationally demanding. Computing time for SLR-related data analytics and modeling can be substantially reduced as illustrated in this study (from days down to hours and from hours down to minutes as high-performance computing clusters are used). Only through the combination of scientific workflows and parallel computing capabilities, SLR modeling and associated geospatial data analytics can be conducted efficaciously.
This scientific workflow framework has substantial managerial implications for coastal sustainability and resilience. The framework and its software package could serve as a useful spatial decision support tool for coastal management and conservation that considers the impact of SLR. With support from this framework, systematic scenario analysis can be conducted to provide a quantitative assessment of land cover types in coastal regions over time. The spatial results of SLR from the framework can be used to guide climate-related policies or decisions related to, for example, urban or transportation planning, landscape conservation, and stormwater and wastewater management for coastal regions. Stakeholders or professionals can use this framework to evaluate quantitatively the potential ecological or environmental impact of SLR on coastal ecosystems, which will be of great help for the recommendation of coastal management practices and strategies for implementing sustainable and resilient coastal ecosystems. This is a particularly important topic of concern as climate change has been receiving considerable attention from coastal communities, the government, and the public.
While our study provides solid support for automating repetitive scenario analysis for SLR studies, limitations exist for our scientific workflow framework. The first limitation is that our framework is based on the loose coupling of multiple geospatial data analytics and model capabilities. Users need computing and GIS background to better utilize the scientific workflow framework for the automation of SLR modeling. The second limitation is that SLR modeling (SLAMM here) and associated geospatial data analytics (including landscape pattern analysis) were deployed on HPC clusters over different operating systems (Windows and Linux). This may introduce a certain level of difficulty for data handling if users are not familiar with cross-platform computing.
Thus, several directions exist for the future work of our study. First, we plan to deploy our scientific workflows to a web-based platform. Turning this workflow into an easily executed online application could allow professionals or the public outside of research areas to quickly model SLR for an area of interest. For example, someone working in a local municipality could quickly identify what areas of their town might be severely affected by various rates of SLR via a web-based solution. Even those working in environmental disaster relief could benefit from a web application based on the SLAMM model to plan how to mitigate SLR effects for high-risk areas. The second direction of future work is to convert the SLAMM model, which is open-source now, to a Linux-based operating system environment. This will allow us or users to leverage more HPC power from Linux-based computing clusters, which are more common than Windows-based counterparts. The third option that could take this research a step further is to apply our approach to larger study regions and work with larger amounts of data, which would be beneficial for understanding overall SLR patterns for an area such as the east coast of the U.S. or another large coastline.