User:Jhurley/sandbox

Downscaled High Resolution Datasets for Climate Change Projections

Global climate models (GCMs) have generated projections of temperature, precipitation and other important climate change parameters with spatial resolutions of 100 to 300 km. However, higher spatial resolution information is required to assess threats to individual installations or regions. A variety of “downscaling” approaches have been used to produce high spatial resolution output (datasets) from the global climate models at scales that are useful for evaluating potential threats to critical infrastructure at regional and local scales. These datasets enable development of information about projections produced from various climate models, about downscaling to achieve desired locational specificity, and about selecting the appropriate dataset(s) to use for performing specific assessments. This article describes how these datasets can be accessed and used to evaluate potential climate change impacts.

Related Article(s):

• Climate Change Primer

Contributor(s): Dr. Rao Kotamarthi

Key Resource(s):

• Use of Climate Information for Decision-Making and Impacts Research: State of our Understanding

• Applying Climate Change Information to Hydrologic and Coastal Design of Transportation Infrastructure, Design Practices

• Statistical Downscaling and Bias Correction for Climate Research

• Downscaling Techniques for High-Resolution Climate Projections: From Global Change to Local Impacts

Downscaling of Global Climate Models

Some communities and businesses have begun to improve their resilience to climate change by building adaptation plans based on national scale climate datatsets (National Adaptation Plans), regional datasets (New York State Flood Risk Management Guidance), and datasets generated at local spatial resolutions. Resilience to the changing climate has also been identified by the US Department of Defense (DoD) as a necessary part of the installation planning and basing process (DoD Report on Effects of a Changing Climate). More than 79 installations were identified as facing potential threats from climate change. The threats faced due to changing climate include recurrent flooding, droughts, desertification, wildfires and thawing permafrost.

Assessing the threats climate change poses at regional and local scales requires data with higher spatial resolution than is currently available from global climate models. Global-scale climate models typically have spatial resolutions of 100 to 300 km, and output from these models needs to be spatially and/or temporally disaggregated in order to be useful in performing assessments at smaller scales. The process of producing higher spatial-temporal resolution climate model output from coarser global climate model outputs is referred to as “downscaling” and results in climate change projections (datasets) at scales that are useful for evaluating potential threats to regional and local communities and businesses. These datasets provide information on temperature, precipitation and a variety of other climate variables for current and future climate conditions under various greenhouse gas (GHG) emission scenarios. There are a variety of web-based tools available for accessing these datasets to evaluate potential climate change impacts at regional and local scales.

Figure 1. Typical processes and spatial scales of Regional scale Climate Models. The models may calculate circulation in the atmosphere, cloud processes, precipitation, and land-atmospheric and ocean-atmospheric processes on a limited portion of the Earth, with boundary conditions provided by a Global Climate Model.

Methods for Downscaling

Table 1. Physical and chemical properties of TCP

Property	Value
Chemical Abstracts Service (CAS) Number	96-18-4
Physical Description (at room temperature)	Colorless to straw-colored liquid
Molecular weight (g/mol)	147.43
Water solubility at 25°C (mg/L)	1,750 (slightly soluble)
Melting point (°C)	-14.7
Boiling point (°C)	156.8
Vapor pressure at 25°C (mm Hg)	3.10 to 3.69
Density at 20°C (g/cm3)	1.3889
Octanol-water partition coefficient (logKow)	1.98 to 2.27 (temperature dependent)
Organic carbon-water partition coefficient (logKoc)	1.70 to 1.99 (temperature dependent)
Henry’s Law constant at 25°C (atm-m3/mol)	3.17x10-4 to 3.43x10-4

Table 2. Advantages and limitations of TCP treatment technologies

Technology	Advantages	Limitations
ZVZ	Can degrade TCP at relatively high and low concentrations Faster reaction rates than ZVI Material is commercially available	Higher cost than ZVI Difficult to distribute in subsurface in situ applications
Groundwater Extraction and Treatment	Can cost-effectively capture and treat larger, more dilute groundwater plumes than in situ technologies Well understood and widely applied technology	Requires construction, operation and maintenance of aboveground treatment infrastructure Typical technologies (e.g. GAC) may be expensive due to treatment inefficiencies
ZVI	Can degrade TCP at relatively high and low concentrations Lower cost than ZVZ Material is commercially available	Lower reactivity than ZVZ, therefore may require higher ZVI volumes or thicker PRBs Difficult to distribute in subsurface in situ applications
ISCO	Can degrade TCP at relatively high and low concentrations Strategies to distribute amendments in situ are well established Material is commercially available	Most effective oxidants (e.g., base-activated or heat-activated persulfate) are complex to implement Secondary water quality impacts (e.g., high pH, sulfate, hexavalent chromium) may limit ability to implement
In Situ Bioremediation	Can degrade TCP at moderate to high concentrations Strategies to distribute amendments in situ are well established Materials are commercially available and inexpensive	Slower reaction rates than ZVZ or ISCO

There are two main approaches to downscaling. One method, commonly referred to as “statistical downscaling”, uses the empirical-statistical relationships between large-scale weather phenomena and historical local weather data. In this method, these statistical relationships are applied to output generated by global climate models. A second method uses physics-based numerical models (regional-scale climate models or RCMs) of weather and climate that operate over a limited region of the earth (e.g., North America) and at spatial resolutions that are typically 3 to 10 times finer than the global-scale climate models. This method is known as “dynamical downscaling”. These regional-scale climate models are similar to the global models with respect to their reliance on the principles of physics, but because they operate over only part of the earth, they require information about what is coming in from the rest of the earth as well as what is going out of the limited region of the model. This is generally obtained from a global model. The primary differences between statistical and dynamical downscaling methods are summarized in Table 1.

It is important to realize that there is no “best” downscaling method or dataset, and that the best method/dataset for a given problem depends on that problem’s specific needs. Several data products based on downscaling higher level spatial data are available (USGS, MACA, NARCCAP, CORDEX-NA). The appropriate method and dataset to use depends on the intended application. The method selected should be able to credibly resolve spatial and temporal scales relevant for the application. For example, to develop a risk analysis of frequent flooding, the data product chosen should include precipitation at greater than a diurnal frequency and over multi-decadal timescales. This kind of product is most likely to be available using the dynamical downscaling method. SERDP reviewed the various advantages and disadvantages of using each type of downscaling method and downscaling dataset, and developed a recommended process that is publicly available. In general, the following recommendations should be considered in order to pick the right downscaled dataset for a given analysis:

• When a problem depends on using a large number of climate models and emission scenarios to perform preliminary assessments and to understand the uncertainty range of projections, then using a statistical downscaled dataset is recommended.
• When the assessment needs a more extensive parameter list or is analyzing a region with few long-term observational data, dynamically downscaled climate change projections are recommended.

Uncertainty in Projections

Table 2. Downscaling model characteristics and output

Model or Dataset Name	Model Method	Output Variables	Output Format	Spatial Resolution	Time Resolution
Statistical Downscaled Datasets
WorldClim	Delta	T(min, max, avg), Pr	NetCDF	grid: 30 arc sec to 10 arc min	month
Bias Corrected / Spatial Disaggregation (BCSD)	Empirical Quantile Mapping	Runoff, Streamflow	NetCDF	grid: 7.5 arc min	day
Asynchronous Regional Regression Model (ARRM v.1)	Parameterized Quantile Mapping	T(min, max), Pr	NetCDF	stations plus grid: 7.5 arc min	day
Statistical Downscaling Model (SDSM)	Weather Generator	T(min, max), Pr	PC Code	stations	day
Multivariate Adaptive Constructed Analogs (MACA)	Constructed Analogues	10 Variables	NetCDF	grid: 2.5 arc min	day
Localized Constructed Analogs (LOCA)	Constructed Analogues	T(min, max), Pr	NetCDF	grid: 3.75 arc min	day
NASA Earth Exchange Downscaled Climate Projections (NEX-DCP30)	Bias Correction / Spatial Disaggregation	T(min, max), Pr	NetCDF	grid: 30 arc sec	month
Dynamical Downscaled Datasets
North American Regional Climate Change Assessment Program (NARCCAP)	Multiple Models	49 Variables	NetCDF	grid: 30 arc min	3 hours
Coordinated Regional Climate Downscaling Experiment (CORDEX)	Multiple Models	66 Variables	NetCDF	grid: 30 arc min	3 hours
Strategic Environmental Research and Development Program (SERDP)	Weather Research and Forecasting (WRF v3.3)	80+ Variables	NetCDF	grid: 6.5 arc min	3 hours

A primary cause of uncertainty in climate change projections, especially beyond 30 years into the future, is the uncertainty in the greenhouse gas (GHG) emission scenarios used to make climate model projections. The best method of accounting for this type of uncertainty is to apply a climate change model to multiple GHG emission scenarios (see also: Wikipedia: Representative Concentration Pathway).

The uncertainties in climate projections over shorter timescales, less than 30 years out, are dominated by something known as “internal variability” in the models. Different approaches are used to address the uncertainty from internal variability. A third type of uncertainty in climate modeling, known as scientific uncertainty, comes from our inability to numerically solve every aspect of the complex earth system. We expect this scientific uncertainty to decrease as we understand more of the earth system and improve its representation in our numerical models. As discussed in Climate Change Primer, numerical experiments based on global climate models are designed to address these uncertainties in various ways. Downscaling methods evaluate this uncertainty by using several independent regional climate models to generate future projections, with the expectation that each of these models will capture some aspects of the physics better than the others, and that by using several different models, we can estimate the range of this uncertainty. Thus, the commonly accepted methods for accounting for uncertainty in climate model projections are either using projections from one model for several emission scenarios, or applying multiple models to project a single scenario.

A comparison of the currently available methods and their characteristics is provided in Table 2 (adapted from Kotamarthi et al., 2016). The table lists the various methodologies and models used for producing downscaled data, and the climate variables that these methods produce. These datasets are mostly available for download from the data servers and websites listed in the table and in a few cases by contacting the respective source organizations.

The most popular and widely used format for atmospheric and climate science is known as NetCDF, which stands for Network Common Data Form. NetCDF is a self-describing data format that saves data in a binary format. The format is self-describing in that a metadata listing is part of every file that describes all the data attributes, such as dimensions, units and data size and in principal should not need additional information to extract the required data for analysis with the right software. However, specially built software for reading and extracting data from these binary files is necessary for making visualizations and further analysis. Software packages for reading and writing NetCDF datasets and for generating visualizations from these datasets are widely available and obtained free of cost (NetCDF-tools). Popular geospatial analysis tools such as ARC-GIS, statistical packages such as ‘R’ and programming languages such as Fortran, C++, and Python have built in libraries that can be used to directly read NetCDF files for visualization and analysis.

See Also

Climate Change Impacts to Department of Defense Installations, SERDP Project RC-2242