NASA Global Satellite and Model Data Products and Services for Tropical Cyclone Research

The lack of observations over vast tropical oceans is a major challenge for tropical cyclone research. Satellite observations and model reanalysis data play an important role in filling these gaps. Established in the mid-1980s, the Goddard Earth Sciences Data and Information Services Center (GES DISC), as one of the 12 NASA data centers, archives and distributes data from several Earth science disciplines such as precipitation, atmospheric dynamics, atmospheric composition, and hydrology, including well-known NASA satellite missions (e.g., TRMM, GPM) and model assimilation projects (MERRA-2). Acquiring datasets suitable for tropical cyclone research in a large data archive is a challenge for many, especially for those who are not familiar with satellite or model data. Over the years, the GES DISC has developed user-friendly data services. For example, Giovanni is an online visualization and analysis tool, allowing users to visualize and analyze over 2000 satellite- and model-based variables with a Web browser, without downloading data and software. In this chapter, we will describe data and services at the GES DISC with emphasis on tropical cyclone research. We will also present two case studies and discuss future plans. We present two case studies to show how our datasets can be used in tropical cyclone research. In the first case, three different precipitation datasets were compared during the landfall of Hurricane Katrina. Results show that large differences exist among the three datasets. The MERRA-2 modeled precipitation has a large wet bias compared to the other two datasets. The areas of heavy precipitation for both modeled and observation-corrected are different from those of 3B42. The results suggest that these differences need to be considered when using the model precipitation products. The second case is an example of exploring different satellite- and model-based variables from existing data services and tools such as Worldview, Panoply, and Giovanni.


Introduction
Tropical cyclones form over vast tropical oceans where in situ observations are sparse and discontinuous. The lack of observational data over these areas historically has been a major obstacle for tropical cyclone research and other weather and climate-related studies. Understanding the complex atmospheric and oceanic processes, and their interactions at multiple scales (e.g., convective, synoptic) over the life cycle of tropical cyclones, requires multiscale, multi-platform observational networks. It has been a great challenge to design, deploy, and maintain such networks without interruption, particularly in the harsh environments imposed by these extreme phenomena. Since the satellite era began, data collected from satellites, along with model reanalysis data, have played an important role in providing continuous global observations, filling in data gaps, and enabling research on weather and climate on different scales.
The concept of using satellites to observe Earth's weather and climate was developed as early as 1946 [1]. NASA launched the first successful, weather satellite, TIROS-1 (Television InfraRed Observational Satellite) on April 1, 1960 [1,2]. In 1964, the Nimbus project was initiated and a total of 7 experimental meteorological satellites were launched over a 14-year time period (1964)(1965)(1966)(1967)(1968)(1969)(1970)(1971)(1972)(1973)(1974)(1975)(1976)(1977)(1978) [1,2]. Since then, weather and climate research have come to rely heavily on long-term, consistent satellite observations from multiple operational space-borne platforms to continuously observe the Earth's atmospheric and surface conditions. The Earth-observing satellite era began in earnest after NASA transferred the technology to the National Oceanic and Atmospheric Administration (NOAA) in the 1970s. This was followed by several operational weather satellites series, including the Polar-orbiting Operational Environmental Satellites (POES) and the Geostationary Operational Environmental Satellites (GOES), to provide continuous global weather observations. Meanwhile, the Defense Meteorological Satellite Program (DMSP), also launched in the 1970s, provided additional observations of global weather events. These series evolved into the constellation of weather satellites operating today, using research and operational satellites from domestic and international organizations to provide the frequent, global observations necessary for improved understanding and forecasting of Earth's complex weather systems.
Atmospheric 3-D winds, air and sea surface temperatures, pressure, precipitation, water vapor, aerosols, etc. are among the fundamental variables for tropical cyclone research and applications. As aforementioned, few in situ observations are available over vast and remote tropical oceans. Direct measurements of these variables are difficult both from surface and space. Over the years, satellite-based algorithms have been developed and improved to derive these essential variables from few key measurements such as radiances observed onboard satellites, and their datasets are archived and distributed to support tropical cyclone research.
Established in the mid-1980s (Table 1), the Goddard Earth Sciences Data and Information Services Center (GES DISC), as one of the 12 NASA Distributed Active Archive Centers (DAACs), archives and distributes satellite and model data for a range of Earth science disciplines [3], such as precipitation, atmospheric dynamics, atmospheric composition, and hydrology, derived from well-known NASA Earth's satellite missions (e.g., the Tropical Rainfall Measuring Mission (TRMM), the Global Precipitation Measurement (GPM)) as well as model assimilation projects (MERRA-2, NLDAS). These data have been widely used in tropical cyclone research.
To facilitate data access, the GES DISC has developed user-friendly data services for researchers around the world (Figure 1). For example, the Geospatial  Interactive Online Visualization ANd aNalysis Infrastructure (Giovanni) [4][5][6][7] is a powerful online visualization and analysis tool, allowing users to visualize and analyze over 2000 satellite-and model-based variables with a Web browser, without downloading data or software. Locating datasets suitable for tropical cyclone research in a large data archive is a challenge for many users, especially those who are not familiar with satellite data. For example, a search for "precipitation" in the GES DISC Web site [3] can return over 400 results. To facilitate data access, the GES DISC has recently developed a "data list" (also known as "variable set") concept that groups relevant variables from different datasets together to serve specific research needs. A prototype data list targeting hurricane study has been implemented.
The chapter is organized as follows: first, we give a brief overview of NASA satellite and model data at GES DISC, followed by introducing datasets for tropical cyclone studies, data services, case studies, and summary with future directions. The GES DISC archives and distributes data from a range of satellite observations, models, ground measurements, and field campaigns in multiple Earth science disciplines including global precipitation, atmospheric dynamics, hydrology, and atmospheric composition with a total volume of 2.3+ Petabytes consisting of 100+ million data files covering 3000+ public and restricted collections. Over 1200 data collections are being curated at GES DISC.  including the past, current, and future satellite missions. Table 3 lists basic data services and user support. More details about data services are described in Section 4.
The GES DISC is a certified trusted repository of Earth science data. Increasingly, funding organizations and publishers require data to be published to certified data repositories adhering to FAIR principles-(Findability, Accessibility, Interoperability, and Reusability). The GES DISC is a regular member of the International Council for Science (ICSU) World Data System (WDS). Established to archive and distribute data from the 1957-1958 International Geophysical Year, WDS spans a range of scientific disciplines data at 52 centers in 12 countries who adhere to the established principles. The GES DISC is also registered as a scientific data repository through re3data.org and meets the repository criteria including DOI assignments, dataset landing pages, dataset documentation, redundant archive (backups), data integrity checks, and user services. This registry is used by highimpact journals such as Nature Scientific Data [8].

Datasets for tropical cyclone research
There are many research areas associated with tropical cyclones such as cyclone genesis, intensification, track forecasting, rainfall amounts, etc. While GES DISC archives many variables required to conduct such research, other variables may be located at other NASA DAACs (such as sea surface state and temperature). It can be challenging enough to find relevant variables for a specific research area from a large collection of satellite observations from a single data archive-locating relevant datasets from multiple data centers is even more challenging. Due to the  page limit, in this section, we can only present a brief overview of several key data collections at GES DISC for tropical cyclone research.

Brightness temperature collection
Brightness temperature, derived from radiance measured from satellite instruments, is a fundamental physical variable in satellite meteorology. Infrared satellite images have been available since the Nimbus era to support weather analysis and forecast. For example, infrared images are used in tropical cyclone monitoring and forecast operations at the NOAA National Hurricane Center and the U.S. Navy Joint Typhoon Warning Center. Animations made from infrared images are frequently used in daily local TV weather news, online weather news, and scientific presentations. The GES DISC archives brightness temperatures from infrared instruments from the Nimbus era up to more recent and current passive microwave satellite instruments from domestic and international research and operational satellites.
Datasets from the Nimbus data rescue project [9] consists of digitized black-andwhite film images ( Figure 2) and radiance data obtained by the Nimbus satellites during the 1960s, 70s, and 80s [10]. Related instruments onboard the Nimbus satellites  Table 4. Negatives of photo facsimile 70-mm file strips were scanned and saved as JPEG 2000 digital files. There are 20 datasets from the Nimbus satellites (1-7) beginning on Aug. 28,1964 and ending on May 9, 1985. Tropical cyclone research requires frequent and continuous observations of the Earth's atmosphere to analyze and understand event development and processes. Operational geostationary satellites make such observations from cloud tops possible, although it is still challenging to continuously observe changes inside a weather system. The first geostationary satellite in operation is GOES 1, which was launched on October 16, 1975 [1]. As more operational geostationary satellites were added by different international agencies, infrared data from these satellites can be stitched and provide a near-global (60° N-S) coverage of the Earth's atmosphere [11,12].
With support from the NASA Global Precipitation Climatology Project (GPCP) and by the Tropical Rainfall Measuring Mission (TRMM), the NOAA National Weather Service (NWS) Climate Prediction Center (CPC) has developed a globally merged (60° N-S) pixel-resolution IR brightness temperature dataset (equivalent blackbody temperatures), merged from all available domestic and international geostationary satellites [11,12]. This half-hourly and 4 km x 4 km resolution dataset is also called the merged IR and is available at the GES DISC from 2000 onward [12]. Figure 3 is a sample of the dataset, showing two tropical cyclones (Cilida and Kenanga) on December 20, 2018. In addition to tropical cyclone and other research, the merged IR dataset has been an important input for a number of algorithms that derive near-global IR-based precipitation estimates in several well-known satellitebased global precipitation products [13,14] such as the Integrated Multi-satEllite Retrievals for GPM or IMERG [13].
Currently in operation, the GPM Microwave Imager (GMI) [15] is used as the reference standard to generate Level-1C common calibrated brightness temperature products from the GPM constellation consisting of both domestic and international satellites, based on the algorithms developed by the GPM intercalibration (X-CAL) working group [16]. These Level-1C products are transformed from their equivalent Level-1B radiance data. There are many applications of these passive microwave brightness products. For example, the GPM profiling algorithm (GPROF) uses these Level-1C products to generate hydrometeor profiles and surface precipitation estimates (Figure 4) used as input data in IMERG.

TRMM and GPM precipitation dataset collection
Launched in November 1997, the TRMM satellite (40° N-S), a joint mission between NASA and JAXA (the Japan Aerospace Exploration Agency), carried  several precipitation-related instruments, including the first spaced-borne Ku-band precipitation radar (PR), a passive TRMM microwave imager (TMI), a visible and infrared scanner (VIRS), and a lightning imaging sensor (LIS) [17]. TRMM ended in April 2015. Over a 17-year period, TRMM provided observations that are used to produce groundbreaking 3-D images of rain and storms over vast and remote tropical oceans and continents. TRMM provides opportunities for researchers to understand characteristics of atmospheric systems through instantaneous measurements in different wavelengths from the onboard instruments.
TRMM data available at GES DISC [18] are listed in Table 5. They can be categorized in different processing levels, ranging from Level 1 to 3 [19]. Level-1 TRMM datasets consist of reconstructed and unprocessed instrument data at full-resolution data at Level-1, 1A, 1B, and 1C [19] from the three TRMM  instruments. Level-2 TRMM datasets are derived geophysical variables at the same resolution and location as Level-1 source data such as the GPROF hydrometeor profiles and surface precipitation estimates. Level-3 TRMM datasets are Level-2 variables that are mapped on uniform space-time grid scales, ranging from 3 hourly to monthly. For example, the TRMM Multi-satellite Precipitation Analysis (TMPA) datasets [20][21][22] are Level-3 products, including both nearreal-time and research grade products. The TMPA datasets have been widely used in research and applications around the world, especially in gauge sparse regions. TRMM products processed with GPM algorithms are also available [3]. Their data format and naming convections are consistent with those of GPM [3]. TRMM LIS data are archived at the Global Hydrology Resource Center (GHRC) [23]. Studies to investigate the relationship between lightning and precipitation have been reported (e.g., [24]).
Built on the success of TRMM, GPM [15] is another joint mission between NASA and JAXA to continue key measurements after the TRMM era. The main concept of GPM is to form an international constellation of research and operational satellites and use GPM as a core satellite that carries advanced radar and passive microwave radiometer instruments to measure precipitation from space as well as serve as a reference standard to unify precipitation measurements from other domestic and international satellites in the constellation [15].
A list of TRMM datasets at GES DISC. TRMM products processed with GPM algorithms are also available [3]. Their data format and naming convections are consistent with those of GPM. More information is available in each dataset landing page.
In addition to the passive GPM microwave imager or GMI, a dual-frequency precipitation radar (DPR) has been added in GPM (Figure 5a). A new frequency (Ka-band) in the DPR is capable to detect light rain from space, which is one of challenges in satellite precipitation retrieval algorithms. The GMI carries four additional high frequency channels for measuring falling snow, compared to the TMI. The GMI's spatial resolution is improved significantly with a 1.2 m diameter antenna [15].  GPM data products at GES DISC [25] are categorized also in three processing levels ( Table 6). Like the TRMM era, a new multi-satellite, multi-retrieval product suite (IMERG) has been developed, with significant improvements in both spatial (0.1 x 0.1 deg.) and temporal (half hourly) resolutions over TMPA. There are three dataset categories in IMERG, Early, Late, and Final. The IMERG-Early provides near-real-time (latency: ~4 hours) global precipitation estimates, which are suitable for various research and applications such as flood watching. As more data are available, IMERG-Late (latency: ~12 hours) provides better estimates on precipitation than the Early. The IMERG-Final (latency: ~3.5 months) is a research-grade dataset that is bias corrected with ground gauge data from the Global Precipitation Climatology Centre (GPCC).

MERRA-2 dataset collection
Datasets from the Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2) are developed by the NASA Global Modeling and Assimilation Office (GMAO) to place NASA's Earth Observing System (EOS) satellite observations in a climate context and to improve the representation of the atmospheric branch in the hydrological cycle from previous reanalysis or MERRA [26]. MERRA-2, available from 1980 onward, also includes the first long-term global aerosol reanalysis through assimilating satellite-based observations and representing their interactions with other physical processes in the Earth's climate system [26]. There are 95 product groups, and the file format is in NetCDF-4. Key meteorological variables in MERRA-2 for tropical cyclone studies such as wind, air temperature, and geopotential height are available at GES DISC. The spatial resolution is about 0.5 deg. x 0.625 deg. in the latitudinal and longitudinal directions, respectively, with 42 pressure levels and 72 model levels. MERRA-2 temporal resolutions range from hourly, 3 hourly, daily to monthly. Initial evaluation of MERRA-2 has been done and is available on the GMAO Web site [27].

Tools
In research, data evaluation is often the first step to examine and understand a new physical dataset. Due to the complexity of satellite-based datasets, it is not an easy task to conduct such a task, especially for those without some prior knowledge about the dataset. Over the years, many tools have been developed by different organizations to facilitate such tasks. In this chapter, we describe two popular tools, GES DISC Giovanni and NASA GISS Panoply.

GES DISC Giovanni
Giovanni [4][5][6][7]28] provides a simple and easy way to analyze and visualize more than 2000 satellite-and model-based physical parameters archived at GES DISC without downloading data and software. Variables of many well-known satellite missions and projects mentioned earlier such as TRMM, GPM, MERRA2, etc. are included in Giovanni.
Giovanni was first developed at the beginning of the TRMM era when TRMM TMPA datasets were available to the public [6]. Precipitation is a very popular variable and is used in many disciplines such as hydrology and agriculture. At that time, standard archived TMPA files were written in HDF4, a format that was not well known outside the remote sensing community. As a result, many TRMM users had difficulties to handle such format, which was a major barrier for TMPA data access and utilization. Recognizing this problem, scientists and software engineers at GES DISC worked closely with the TMPA product provider and developed a Web-based tool, the TRMM Online Visualization and Analysis System (TOVAS) [6]. With a Web browser, one can obtain average and accumulated rainfall maps as well as time series plots and Hovmöller diagrams in their areas of interest. Users can download results in several commonly used formats such as ASCII, which can be directly imported into Microsoft Excel for further processing. Later, several MODIS atmospheric products (e.g., aerosols, atmospheric water vapor) were added to TOVAS.
To meet an increasing demand for adding more analytical functions and variables, GES DISC developed Giovanni, allowing functions and variables to be added through a Web-based interface [4,5]. In current version of Giovanni [7], more advanced information technologies have been implemented in the development, such as having all variables accessible in one Web interface, facet searching, sorting, provenance, etc. As of this writing, over 2000 variables from different Earth science disciplines are available and searchable in Giovanni, including datasets curated through other DAACs. More than 1300 referral research papers have been published by users around the world, with help from Giovanni. In short, Giovanni provides an easy way to evaluate and explore Earth science data at GES DISC.
With over 2000 variables in Giovanni, it is necessary to provide flexible search capabilities. Frist, users can type in key words such as IMERG and see variables only related to the key words. Often results from a search can contain many variables. For example, a search for precipitation returns 143 variables. To locate those of interest, one can sort the results based on source, spatial and temporal resolutions, begin or end dates. Facets have been developed to help narrow down search results, including disciplines, measurements, platform/instrument, spatial resolutions, temporal resolutions, wavelengths, etc.
Giovanni provides an interface for selecting date range. Users can either pick a date from the Web interface or type in their own date information. Likewise, users can draw their region of interest or type in the longitude and latitude coordinates in the interface, or they can select predefined shape files from a list including countries, land/sea masks, U.S. states, and major hydrological or watershed basins around the world.
There are 22 built-in analytical functions that are grouped into 5 categories based on their analysis types such as maps, time series, and comparisons. Once all the selections are done in the Web interface, a click on the "Plot Data" button will direct the user to the visualization result page where the user can find different options to fine tune their maps or plots. The output page provides a browse history in which users can return back to the input Web interface, download the results in graphic or NetCDF formats, or visit the lineage to see the provenance information or download data in each process. Figure 6 is an example of using Giovanni to generate the accumulated rainfall from Hurricane Harvey during August 24-31, 2017.
Giovanni training materials have been developed over the years. The Giovanni user guide [29] is available through the help button along with release notes, browser compatibility, and known issues. Users can also visit YouTube for Giovanni-related How-to videos [30]. The NASA Applied Remote Sensing Training (ARSET) project also provides materials for Giovanni online training [31], used in live webinars that are free of charge for users around the world. If users have questions or suggestions about Giovanni, they can submit them thorough the feedback button in the landing page and a staff member at GES DISC will provide assistance. Acknowledgment policy is also available at the bottom of the output page.

NASA GISS Panoply
Although users can access over 2000 variables through Giovanni, all these variables are in Level-3. While adding data variables from other levels in Giovanni are being considered, users can use Panoply [32], developed by NASA Goddard GISS, to view Level-2 and Level-3 data. Panoply is another powerful tool for viewing NASA data. Panoply can be installed in several platforms and operating systems (e.g., macOS, Windows), requiring Java 8 or later version installed in their systems. Most datasets archived at GES DISC can be viewed by Panoply. Once Panoply is installed, there are several ways to import data to Panoply. If NetCDF-3 or NetCDF-4 is the default setting for opening files in the system, Panoply will automatically open the file when you download it from GES DISC. Popular Level-2 and Level-3 datasets are often available in OPeNDAP and GrADS Data Server (GDS). Users can directly use their dataset links in Panoply to visualize the variables. Although analytical functions in Panoply are quite limited, it is so far an easy way to visualize datasets that are currently not available in Giovanni (Figures 3, 4 and 5a).

Data services
Data services are essential for data archive centers. The totally redesigned GES DISC Web site [3] (Figure 1) makes datasets, documents, and help information easy to find. Due to a large volume of datasets and information archived at GES DISC, a search capability is necessary to facilitate dataset discovery and exploration. From the search box (Figure 1), users can search datasets and information in the following categories: data collections, data documentation, alerts, FAQs, glossary, How-to's, image gallery, news, and tools. The category for data collections is the default since many users come to GES DISC for datasets. Users can also browse data by category, including subject, measurement, source, processing level, project, temporal resolution, and spatial resolution. As of this writing, the GES DISC archives more than ~2.3 PB data with over 117 million files. Over 2.4 billion files have been distributed with data volume over 23 PB.
After a user types in a key word in the search box (Figure 1), the search results are listed. Faceting and sorting are available for identifying datasets of interest, similar to those in Giovanni. For example, a search for IMERG returns 15 datasets and users can use Version to sort different versions and find out the datasets of the latest version. "Get Data" or "Subset/Get Data" right below a dataset name provides a direct link to the data download interface. A click on a dataset name leads to the dataset landing page with more information on dataset summary, data citation, documentation, and more data access methods including links to online archive, Giovanni, Web services such as OPeNDAP, GDS, and THREDDS (the Thematic Real-time Environmental Distributed Data Services) available for some popular datasets.
Most NASA datasets are global coverage. Many study areas are either local or regional; therefore, spatial subsetting is important to reduce download volumes, permitting the user more time to do research. Spatial subsetting is available for a large number of datasets at GES DISC. Users can use a computer mouse to drag an area of interest or type in the geolocation coordinates in the subsetter interface. Some subsetters can do more than spatial subsetting such as parameter subsetting, i.e., selecting wanted variables from a list, which can also help reduce data to be downloaded. For MERRA-2 data, the subsetter can also regrid the original MERRA-2 data into different grid structures with a list of interpolation methods (e.g., bilinear, bicubic). Furthermore, the MERRA-2 subsetter can subset data at pressure levels. NetCDF-4 is available for many datasets at GES DISC, which can be handled by many software packages or tools such as Panoply, ArcGIS.
Level-2 data subsetting is available for popular and high-volume datasets such as the Level-2 GPM dual-frequency product from DPR (2ADPR) that provides general characteristics of precipitation, correction for attenuation, profiles of precipitation water content, rain rate, as well as particle size distributions of rain and snow. The dual-frequency observations from DPR provide better estimates of rainfall and snowfall rates than the single-frequency TRMM PR [33] with additional information for particle size and melting layer height from the Ka band. Variables in 2ADPR are available in all three scan modes of DPR: a) normal scans (NS), b) matched scans (MS) (Figure 5a), c) and high sensitivity (HS), and their swath sizes range from 120 km to 245 km. Each file contains over 400 variables with size close to 300 MB. The subsetting service, developed at GES DISC, provides both variable and spatial subsetting capabilities, which help reducing the file size by several orders of magnitude, depending on selections. Three spatial subsetting capabilities (Figure 5b) are currently available: a rectangular latitude/longitude box, a circle, and a point. Users can pick one of them in the interactive subsetting Web interface and create an area or point of interest. These spatial and parameter subsetting capabilities facilitate ground validation and evaluation activities. For example, users can pull DPR data over a time period for a location where a rain gauge is located for validation or evaluation.
For decades, the GES DISC has archived and distributed a large amount of Earth science data, information, and services to diverse communities including the tropical cyclone community. From searching, discovering to assessing such "Big Data," i.e., heterogeneous and immense scientific data (particularly, satellite or model products) in order to timely and properly examine and assess those natural devastating weather events with an imminent goal for better understanding their natures and reducing the resultant disaster risk, it has, nonetheless, become a daunting task for science researchers and application users (and decision makers).
Aiming to substantially assist our users in their online effectively (i.e., quickly and properly) acquiring the data they want and/or need a "one-stop shop" with a minimum effort from our large data collection for their investigating and assessing the targeted disastrous weather such as hurricanes, the GES DISC has recently developed a value-added and knowledge-based data service prototype by preparing/presenting the "List" of relevant data and the pertinent resources accordingly. Such a data service framework, termed as "Datalist" (currently containing "Hurricane Datalist" only), which basically consists of suites of annotated Web addresses (URLs) that point to the proper and relevant data and resources. Figure 7 (concept based on [34,35]) shows a basic workflow of how a user can online, via accessing Hurricane Datalist, acquire and assess the respective datasets (down to the variable level such as wind and air temperature, etc.) or services (such as Subsetting and Giovanni) they want or need relevant to their targeted hurricane event, e.g., Hurricane Sandy (October 22-29, 2012) over the US continent and coast area at one stop. Basically, through visiting the Hurricane Datalist page (Figure 7), users can readily choose and apply those handy "Subsetting" options of (1) refining data temporal range; (2) selecting spatial domain; (3) choosing targeted variables; and (4) acquiring and downloading the data they want and/or need. Moreover, a useful "window shopping" service is offered to users, allowing them to utilize Giovanni  to plot and view their interested variables pre-"data downloading" that help them make proper decisions. Figure 8 shows two maps produced with Giovanni for the two sampled data variables, i.e., the MERRA-2 wind speeds (Figure 8a) and the AIRS air temperature (Figure 8b).
Users must register the NASA Earthdata login system before downloading data from any NASA data center (including GES DISC). A help feature [3] is available to assist users when they have questions about data and services, which can be very helpful for those who are not familiar with NASA satellite datasets and may not know where to begin. Using this feature, GES DISC staff and supporting scientists can guide users regarding questions related to datasets, tools, documentation, and services. FAQs and How-To recipes are also available and searchable.

Case 1: Evaluation of MERRA-2 precipitation during hurricane Katrina landfall
Evaluation (e.g., comparison) of a dataset prior to download is important to understand (for example) any biases or systematic differences in datasets, which is quite common for remote sensing and model datasets. Over oceans, few in situ observations are available, especially for precipitation, making it very difficult to assess their biases. MERRA-2 datasets provide over 39 years of continuous analysis ranging from hourly to monthly, as mentioned earlier, and can be used to study events and environmental changes (e.g., trends) in tropical oceans and other regions as well.
There are two types of precipitation parameters in MERRA-2: a) precipitation from the atmospheric model (variable PRECTOT in the MERRA-2 dataset collection) and b) observation-corrected precipitation (variable PRECTOTCORR) [36,37]. Observational data are introduced in the latter parameter due to considerable errors that propagate into land surface hydrological fields and beyond [38].
Bosilovich et al. [37] have conducted a general evaluation of MERRA precipitation estimates, including precipitation climatology, interannual variability, diurnal cycle, Madden-Julian Oscillation (MJO) events, global water cycle, and U.S. summertime variability. Although the preliminary evaluation provides a basic understanding of the MERRA-2 precipitation products, evaluation for extreme weather events is still needed to better understand MERRA-2 precipitation behavior and characteristics.  shows 3 daily precipitation maps from the MERRA-2 modeled, observation-corrected, and bias-corrected TMPA 3B42 precipitation products on Aug. 29,2005, when the deadly Hurricane Katrina, as currently ranked as the 3rd most intense landfalling hurricane in the U.S. history, made a landfall near New Orleans, Louisiana. Katrina claimed at least 1245 lives, making it the deadliest U.S. hurricane since the Hurricane Okeechobee in 1928. Apart from the obvious difference in the dataset spatial resolutions (~0.5° in MERRA vs. 0.25° in 3B42), it is seen that large differences exist among three precipitation products (Figure 9). Figure 9a shows that the modeled precipitation has the largest systematic differences against TMPA 3B42 (Figure 9c) in terms of intensity and structure. Significant differences still exist even after observational data were applied for bias correction (Figure 9b and c). More importantly, the centers of heavy rainfall are mismatched in all datasets. In TMPA, there are two centers and all are over the ocean (Figure 9c). By contrast, the center from the modeled is located on land (Figure 9a). The center from the observationcorrected is between the modeled and TMPA (Figure 9b). Also, it seems that both MERRA-2 precipitation products do not do well in the light rain regions in Figure 9. The overestimation issue in this case seems to be consistent with the evaluation conducted by Bosilovich et al. [37]; however, other important issues (e.g., the heavy rain centers in wrong places) do exist.

Case 2: Acquiring Earth science data measurements all at once to study hurricane events
Collecting and evaluating data are important activities for tropical cyclone research. A one-stop shop for acquiring these activities can save time and is very desirable. However, such system is still being developed and there are many obstacles to overcome. This case study is to demonstrate, with existing tools, one can conduct such activities.
With Giovanni, users can explore over 2000 satellite-and model-based variables. For example, Hurricane Maria is a deadly Category-5 hurricane and caused a heavy damage on local economy in island countries in the Caribbean. Before using Giovanni, one can obtain the true color satellite image from NASA Worldview (Figure 10a) and the IR image from the merged IR dataset at the GES DISC and Panoply (Figure 10b). The latter provides uninterrupted IR data at 4-km spatial resolution available every 30 minutes, which is very helpful for tracking the hurricane evolution. There are many variables in Giovanni from TRMM, GPM, MERRA-2, etc. Once the beginning and ending times as well as the geolocation are decided, one can input such information in Giovanni. The next step is to select variables of interest. In this case, four MERRA-2 variables are selected: (a) cloud top temperature (Figure 10c); (b) surface wind speed (Figure 10d); (c) total column ozone (Figure 10e); and (d) surface specific humidity (Figure 10f). In Figure 10b and c, large differences in cloud top temperatures exist between the merged IR dataset and MERRA-2. For example, the MERRA-2 cloud top temperatures appear to be cooler than those of the merged IR north of the coast of Venezuela. High wind speeds are found near the hurricane center (Figure 10d) where low total column ozone (Figure 10e) and high surface specific humidity (Figure 10f) are located.

Summary and future plans
In this chapter, we present an overview of basic datasets and services at GES DISC for tropical cyclone research. The collection at GES DISC includes datasets from major NASA satellite missions (e.g., TRMM, GPM) and projects (e.g., MERRA-2, GPCP) with emphasis on precipitation, hydrology, atmospheric composition, atmospheric dynamics, etc. The GES DISC provides user-friendly data services to facilitate data evaluation and download, including a) Giovanni, an online visualization and analysis tool for access over 2000 variables without downloading data and software; b) data subsetting services that allow spatial and variable subsetting of Level-2 and Level-3 datasets; c) different data access methods (online archive, OPeNDAP, GDS, THREDDS) and data formats for a wide variety of users with various technical expertise in handling complex remote sensing datasets; d) information about documentation and data citation; and e) user services to answer data or service-related questions.
We present two case studies to show how our datasets can be used in tropical cyclone research. In the first case, three different precipitation datasets were compared during the landfall of Hurricane Katrina. Results show that large differences exist among the three datasets. The MERRA-2 modeled precipitation has a large wet bias compared to the other two datasets. The areas of heavy precipitation for both modeled and observation-corrected are different from those of 3B42. The results suggest that these differences need to be considered when using the model precipitation products. The second case is an example of exploring different satellite-and model-based variables from existing data services and tools such as Worldview, Panoply, and Giovanni.
Two main areas are focused in future plans: datasets and services. First, as mentioned earlier, NASA Earth data are archived at 12 discipline-oriented data centers across the United States. Datasets at other NASA data centers are also important for tropical cyclone research. For example, the Physical Oceanography DAAC at JPL archives key measurements for tropical cyclone research such as satellite-based ocean surface wind, sea surface temperature, etc. NASA airborne and field campaign datasets for hurricane research, archived at the Global Hydrology Resource Center, play an important role in product validation and case studies. One challenge is to facilitate data discovery and access across the DAACs because relevant datasets are located in different or multiple centers and each center has its unique Web interface for ordering data as well as tools for customized analysis and visualization, which may create problems to some users. It would be more user friendly and efficient to have a one-stop Web interface with data services for acquiring datasets from different data centers. Prototypes have been developed specifically for tropical cyclone research, but they have not been fully integrated into operation and only limited datasets are available; therefore, the datasets can be incomplete. Currently, only NASA Earthdata allows searching datasets from the 12 NASA data centers with very limited data services available. As well, tools can be consolidated and further developed to facilitate access to datasets at different data centers. More on this is to be elaborated in the services.
Analysis-ready data can save time and expedite research and discovery because data are pre-processed at a data center based on research needs. These data are friendly to user's written analysis software or publicly available software packages or tools. For example, time series data subsetting can be a challenging issue for some communities (e.g., hydrology). NASA data are file based, one-time step per file with complex data structures containing few or more variables, which is not optimal for time series data access [39]. To make data analysis ready, data may need to be re-organized for efficient access, such as is demonstrated by the data rods concept [39]. Increasingly, machine learning (ML) and artificial intelligence (AI) algorithms are being used in many areas including tropical cyclone research. Training data play an important role in both ML and AI development and applications. Making training data from collections at DAACs analysis ready (e.g., providing event-based data subsets) can save time for downloading data and processing.
The NASA's Earth Observing System Data and Information System (EOSDIS) cloud evolution [40] is a project to deploy NASA Earth science data and services into a commercial cloud environment to improve data accessibility and serviceability across all NASA DAACs. Prototypes are currently in development to demonstrate cross-DAAC data discovery, access, and servicing. Some datasets are moving natively and operationally into cloud environments as of this writing, such as those from the upcoming Surface Water and Ocean Topography (SWOT). Selected GES DISC datasets are scheduled to be deployed into a cloud environment in 2020, including the IMERG and MERRA-2 collections. Placing the EOSDIS archive collectively in the cloud will, for the first time, place NASA Earth Observation (EO) data "close to compute" and improve management and accessibility of these data while also expediting science discovery for data users. This will also enable largescale data analytics for data users, especially allow more efficient use data from multiple DAACs.
As mentioned earlier, NASA Earth science data collected from satellites, model assimilation, airborne missions, and field campaigns are large, complex, and evolving. It can be a daunting task to obtain data from different centers with different interfaces or tools. Data services increasingly play an important role to facilitate data discovery, access, and exploration. Data services at GES DISC are evolving as well. For example, our currently predefined Datalist by no means can contain variables in all possible research topics in hurricane research. User-defined Datalist would allow users to define their own Datalist is the next thing to be developed. The user-defined Datalist not only can save time for dataset search but also can be sharable to other collaborators or scientists.
Since tropical cyclone research can be categorized based on different spatial and temporal scales. For mesoscale or synoptic scale, event case studies are heavily conducted in tropical cyclone research, requiring that an event information can be easily retrieved from a database (e.g., Hurricane best track data or HURDAT) in order to locate relevant datasets for subsetting, analysis, and visualization with that information (spatial and temporal constrains). Furthermore, datasets can also be searched and located based on other information such as track, intensity, or criteria set by users. At present, neither Giovanni nor the GES DISC Web interface has such capabilities. Adding such event databases can also create data subsets for other research activities such as wild fires, volcanic eruptions, heat waves, snow storms, and nor'easters.
Integration, analysis, and visualization for NASA Level-2 and airborne products are challenging because there are a lot of data-related issues such as formats, structures, terminology, etc. Furthermore, integration of Level-2 and Level-3 products is also needed since model and multi-satellite products (MERRA-2, IMERG) are gridded Level-3 products. The first challenge is to be able to locate (then collocate) available datasets (e.g., swath) for an event and subset the data for the area of interest. The latter work has been done for some GPM products at GES DISC, as mentioned earlier. Customized development is necessary to deal with different data structures from different satellite missions as well as airborne and field campaigns, requiring a close collaboration among data centers. At present, there is no Level-2 dataset in Giovanni. Adding Level-2 datasets and airborne data in Giovanni is important to significantly expand the capabilities of Giovanni in tropical cyclone research because observations are very limited over vast tropical oceans, and all these available observations are important for a wide variety of research activities, regardless. Participation and feedback from users or stakeholders always play a key role to ensure development results to be user friendly and useful.
For climate scale, data services need to provide essential information and datasets including climatological datasets (e.g., sea surface temperature, ocean surface wind speed), anomalies, trends, etc. to help researchers to understand changes and trends in environmental conditions over tropical oceans where tropical cyclones are born and developing. Long-term datasets are important, such as MERRA-2 datasets provide over 39 years of global assimilation analysis (1980-present), which is suitable for generating climatological datasets. Giovanni provides on-the-fly generation of climatology and time series plots for several key datasets, such as MERRA-2, TMPA, and IMERG, for tropical cyclone research.