While atmospheric reanalysis datasets are widely used in climate science, many technical issues hinder comparing them to each other and to observations. The reanalysis fields are stored in diverse file architectures, data formats, and resolutions. Their metadata, such as variable name and units, can also differ. Individual users have to download the fields, convert them to a common format, store them locally, change variable names, regrid if needed, and convert units. Even if a dataset can be read via the Open-Source Project for a Network Data Access Protocol (commonly known as OPeNDAP) or a similar protocol, most of this work is still needed. All of these tasks take time, effort, and money. Our group at the Cooperative Institute for Research in the Environmental Sciences at the University of Colorado and affiliated colleagues at the NOAA's Earth System Research Laboratory Physical Sciences Division have expertise both in making reanalysis datasets available and in creating web-based climate analysis tools that have been widely used throughout the meteorological community. To overcome some of the obstacles in reanalysis intercomparison, we have created a set of web-based Reanalysis Intercomparison Tools (WRIT) at www.esrl.noaa.gov/psd/data/writ/. WRIT allows users to easily plot and compare reanalysis datasets, and to test hypotheses. For standard pressure-level and surface variables there are tools to plot trajectories, monthly mean maps and vertical cross sections, and monthly mean time series. Some observational datasets are also included. Users can refine date, statistics, and plotting options. WRIT also facilitates the mission of the Reanalyses.org website as a convenient toolkit for studying the reanalysis datasets.
Reanalysis (or retrospective analysis) is a scientific technique whereby observations taken over a long time period are combined objectively with a model forecast to form time series of fields representing the state of the system, “analyses.” Reanalysis fields for the atmosphere often include pressure, temperature, wind, and humidity throughout the troposphere and lower stratosphere. Some now extend even higher. Reanalysis fields for the ocean often include pressure, temperature, currents, and salinity throughout the depth of the ocean. The numerical model used in this data assimilation procedure is kept constant over the entire time period, which differs from operational analysis in numerical weather prediction, where the model is frequently changed to improve forecasts. Using the same model greatly reduces trends arising from model differences, allowing reanalysis datasets to be potentially useful for studying climate processes over an extended period. Additionally, data assimilation fills in gaps in output variables—especially in data-sparse regions—and constrains variables to be more physically consistent. There are 12 major atmospheric reanalyses currently available (see www.reanalyses.org/ for an up-to-date listing), with more becoming available soon. These reanalysis datasets have been used in an enormous number of studies, with more than 19,900 peer-reviewed publications relating to reanalysis available as of November 2013.
Despite their widespread use throughout geosciences, reanalysis products have known issues. Even though the model and data-assimilation system are not changing, observational data density, type, and quality can vary in time. These observational variations can introduce spurious variations into the reanalysis fields. Interaction between model biases and the time-varying observing system can also introduce spurious variations.
In addition to these differences, there are differences in model characteristics. For example, each reanalysis uses a different model with different model physics and different resolutions, as well as different assimilation algorithms and input observations. Reanalysis systems can be optimized for different goals; for example, the National Centers for Environmental Prediction (NCEP)'s North American Regional Reanalysis is optimized for North America, while NASA's Modern Era Reanalysis for Research and Applications (MERRA) emphasizes the Earth's water budget.
The differences highlight the need for improved understanding of the limitations and strengths of the reanalysis datasets and how they can be used for weather and climate studies. For this and other purposes, it would be useful to easily compare these datasets. A host of issues make this difficult. The reanalysis data are available from different organizations, are stored in different formats with different file-naming schemes, and have different resolutions, file attributes, variable names, and units. Generally speaking, users have to download the data, convert it to a standard format, store it locally, change variable names, regrid if needed, convert units, and write code general enough such that their applications—which might include NCL, GrADS, IDL, or others—can be used to read each dataset and compare the desired variables.
Even if the dataset can be read via a remote protocol such as Open-source Project for a Network Data Access Protocol (OPeNDAP), most of this work is still needed. All of these tasks take time, effort, and money. Our group at the Cooperative Institute for Research in Environmental Sciences (CIRES) at the University of Colorado and affiliated colleagues at the NOAA's Earth System Research Laboratory Physical Sciences Division (NOAA/ESRL/PSD) have spent several years developing expertise both in making reanalysis datasets available and in creating web-based climate analysis tools that have been widely used throughout the meteorological and oceanographic community. To overcome some of the obstacles in reanalysis intercomparison, we have created a set of Web-based Reanalysis Intercomparison Tools (WRIT), which allow users to quickly plot and compare reanalysis datasets. Users may still want to download and analyze the data, but by using the tools, they can efficiently test hypotheses, generate results, and create plots to narrow down scientific questions for further research. As it is easy to use, it also provides a valuable teaching resource where students can find out for themselves, for example, how the Hadley cell circulation varies throughout the annual cycle. WRIT also facilitates the mission of the www.reanalyses.org/ website, which promotes reanalysis research and intercomparison, by providing a convenient toolkit for studies involving these datasets.
The reanalysis datasets we have made available so far via WRIT are the NCEP/National Center for Atmospheric Research (NCAR) Reanalysis I (R1), NCEP/Department of Energy (NCEP/DOE) Reanalysis 2 (R2), European Center for Medium-Range Weather Forecasts (ECMWF) Interim Reanalysis (ERA-Interim), MERRA, NCEP Climate Forecast System Reanalysis (CFSR), and the NOAA-CIRES Twentieth Century Reanalysis V2 (20CR). We are adding the Japan Meteorological Agency 55-year reanalysis (JRA-55). Basic attributes of each dataset are given in Table 1. Several pressure-level variables are currently available via WRIT: geopotential height, zonal and meridional wind, omega (vertical pressure velocity), air temperature, and relative and specific humidity. These are distributed by the data providers on different levels starting at 1,000 mb and can go up as high as 1 mb. Because of these different levels, we use only a common set of levels that are available in the datasets being compared when calculating cross-section differences. Single-level variables include 2-m air temperature, 10-m zonal and meridional wind, sea level pressure, sea surface temperature, precipitable water, evaporation rate, and precipitation rate.
To compare reanalyses with observational datasets, WRIT includes the University of Delaware V3.01 land air temperature and precipitation dataset, NOAA's Global Land Surface Temperature Anomalies (GHCN_CAMS) air temperature, Global Precipitation Climatology Project V2.2 satellite and station merged precipitation fields (GPCP), NOAA's Precipitation Reconstruction over Land (PREC/L), University of East Anglia Climatic Research Unit air temperature and precipitation time series fields (CRU_TS3.10), the Wave and Anemometer-based Sea Surface Wind dataset (WASWind), the Hadley Sea Level Pressure (HadSLP2) dataset, and the Hadley Centre Sea Ice and Sea Surface Temperature dataset (HadISST1.1). (Note that HadISST is used as boundary conditions for 20CR). Basic attributes for each dataset are given in Table 2.
Information on the reanalyses and how to download them are available from www.reanalyses.org/. Additionally, user plots, suggestions, and comments can be posted there. Currently, WRIT includes three web tools. These are available at www.esrl.noaa.gov/psd/data/writ/ and include maps, time series, and trajectories. A summary of their features is given in Table 3.
WEB PAGE DEVELOPMENT.
Our goal with each of the WRIT webpage tools was to create an easy-to-use single-page interface that allows users to quickly test hypotheses or plot basic climate products. As such, we designed each page to focus on a single product. To maximize use of our resources, we drew upon commonly used web and climate processing software. The WRIT code performs the steps needed to complete the requested calculations. These steps include handling the different grids and grid resolutions, and accounting for attributes, units, and levels. When necessary, map or cross-section differences are interpolated to the same horizontal grid, converting to the higher resolution of the compared fields. The output includes the visualization image, postscript files, and NetCDF and KML files containing the visualized data. (KML files are used by Google Earth and Google Maps). It also includes links to background information and references for the datasets chosen.
Originally, the project was envisioned to use OPeNDAP to read files from remote sites. While we have many datasets stored courtesy of NOAA/ESRL/PSD, we cannot store all possible datasets of interest to the climate community due to space considerations. We hoped that by accessing datasets remotely, WRIT could analyze more datasets and variables. This was only partially successful. While we could easily read data stored at PSD via OPeNDAP, datasets from other providers often took several minutes or more to read. Metadata provided in some OPeNDAP files was also an issue. In some cases, the metadata were lacking, hard to use, or in error. We hope that other data suppliers can improve these issues so that web tools such as ours can do more than plot simple maps from data accessed remotely. Given these issues, all data available to WRIT are stored locally courtesy of NOAA/ESRL/PSD.
WEB TOOL FEATURES.
WRIT Maps Plotting Tool.
The WRIT Maps interface, shown in Fig. 1a, allows users to select the datasets, level(s), type of plot (map, latitude/height, or longitude/height), the statistic (monthly mean, anomaly, or long-term mean), and plotting options. Users can also select a season and one or more years to average in a composite. These years do not have to be sequential or in any particular order, as illustrated in Fig. 1. If two datasets are chosen, a map or cross section is calculated for each dataset and then those results are differenced. Results are returned as images, NetCDF files, and KML. As an example, one can create a composite of 2-m air temperature anomalies averaged over January to March from 20CR using only selected years corresponding to warm anomalies in the central equatorial Pacific (i.e., El Niño Boreal winters). Figure 1b shows this composite. Figure 2 shows the longitude-height cross section of zonal wind anomalies averaged over five Januaries corresponding to recent La Niña events. The cross section is latitudinally averaged from 4°S to 4°N over the tropical Pacific for CFSR (Fig. 2a), the MERRA reanalysis (Fig. 2b), and their difference (Fig. 2c). A strengthened Walker circulation is seen in the composite of both reanalyses, but with the CFSR anomalies showing differences from MERRA as illustrated in Fig. 2c. The CFSR has stronger westerly anomalies in many regions of the tropospheric cross section.
WRIT Time Series Plotting Tool.
The WRIT time-series interface (Fig. 3a) is similar to that of WRIT Maps. The available reanalysis datasets are the same, but there are fewer observational datasets (more will be added). The user can choose one or two datasets to plot. For each dataset, users can choose a variable, a level, monthly or seasonal averaging, year range, and statistic to plot: monthly mean (original data), anomaly (difference from user-specified climatology), or pregenerated climatology (long-term mean). When comparing datasets, users can choose the output type for the two time series (same or different y-axis), or the user can display the difference of the two time series, a scatterplot, or cross-correlation plot of the data. For the selection of single datasets, users can plot a time series, its autocorrelation, or its probability density function as a histogram. Statistics are returned corresponding to the period requested for each plot. These include mean, standard deviation, and slope of linear trend (Fig. 3c) over the period requested. When two datasets are selected, users also are given the correlation and root-mean-square difference of the two time series. As the spatial resolution of each variable and dataset may differ, the exact coordinates used in any area averaging are also returned. Figure 3b shows the time series of globally and monthly averaged 300-hPa air temperature anomalies from the 20CR and R1 datasets. The comparison illustrates the apparent discontinuity and the subsequent beneficial effect that the satellite data introduced in the late 1970s had on R1 but not the surface-pressure based 20CR.
WRIT Trajectories Tool.
The Trajectory plotting tool uses three different reanalysis datasets for which we have 6-hourly data readily available: R1, R2, and 20CR (with ERA-Interim being added). Users can select start and end times, an initial pressure level or set of pressure levels, and a starting and ending location. Users then plot a trajectory or set of trajectories based on the three-dimensional winds from the selected reanalysis dataset. The trajectory can be forward or backward in time for up to 7 days. It uses a time step of 1 h by interpolating the 6-hourly data. It then plots the positions every 6 h. For single pressure-level trajectories, the vertical position is indicated using color, while for multiple levels, each pressure level is plotted as a different color. The trajectory analysis code was obtained from the University of Australia in Melbourne courtesy of Ian Simmonds. It is based on a 3D advective model that is solved using a 4th-order Runge-Kutta scheme. The results are plotted on a map and are also available for viewing via a Google Earth web plugin. Shown in Fig. 4a is a screenshot of the trajectory options used to plot Fig. 4b. This illustrates the Colorado flood event of September 2013 using R1 (back trajectories ending 40°N, 255°E; 12 September 2013). The trajectories show low-level southeasterly airflow directly into the Colorado Front Range (from a very moist southeast United States; not shown). Users can download the NetCDF or KML files containing the trajectory points for offline use.
We appreciate the IT support, programming, and web interface discussions with the NOAA/ESRL/PSD IT group and with D. Murray and D. Allured from PSD and the University of Colorado/CIRES. Assistance from the NCAR NCL support desk is also appreciated. For help with data preparation, S. Lubker of NOAA/ESRL/PSD is gratefully acknowledged. We would like to acknowledge the help of the data support staff at the institutions supplying the reanalysis and observation data: ECMWF, NOAA/NCEP, NASA, and particularly D. Ostrenga at NASA for assistance with metadata and access. Three anonymous reviewers and editor Bryan Etherton made valuable suggestions on an earlier version of the manuscript. Support for the Twentieth Century Reanalysis Project dataset is provided by the U.S. Department of Energy, Office of Science Innovative and Novel Computational Impact on Theory and Experiment (DOE INCITE) program, and Office of Biological and Environmental Research (BER), and by the NOAA Climate Program Office. G.P. Compo received support from the Office of Science (BER), U.S. Department of Energy, and the NOAA Climate Program Office.