Print Version

Table of Contents

Collapse Menu | Expand Menu

Introduction

Use of Satellite Observations in NWP

Meteorological satellite observations are integral to the forecast process. Weather forecasters rely on them to develop situational awareness and conceptual models of the current state of the atmosphere, as well as to verify numerical weather prediction (NWP) analyses and forecasts.

It’s commonly understood that satellite observations are used in NWP models and are important to the quality of NWP model forecasts. However, few are familiar with the details, such as:

  • The types of observations used

    Did You Know ...   NWP models assimilate observations from polar-orbiting and geostationary satellites as well as Global Positioning Satellite (GPS) systems. Of these, polar orbiter observations are used most extensively and have the greatest impact.
  • The role that satellite observations play

    Did You Know ...   Satellites provide observations in otherwise data sparse areas, such as the oceans of the Northern Hemisphere and most of the Southern Hemisphere. Without this data, good model forecasts would not be possible at three- to seven-day lead times in both hemispheres, and sometimes even at two- to three-day lead times in the Southern Hemisphere.
  • How model limitations impact the use of satellite observations

    Did You Know ...   Limitations in NWP models and computing resources prevent 95% of all satellite observations from being assimilated.
  • The impact that satellite observations have

    Did You Know ...   Denying satellite data to NWP significantly degrades the forecast at all lead times, and has a greater impact than denying any other type of conventional observation.

These are some of the topics addressed in this lesson.

The lesson focuses on weather prediction models, although many of the concepts and processes apply to other environmental analyses and forecast models, such as those that monitor and forecast climate, air quality, oceans, water resources, and ecosystem health.

The processes depicted in the lesson come from the U.S. National Center for Environmental Prediction (NCEP) modeling center, with the Global Forecast System (GFS) model serving as the main example. These processes differ in some respects from those used in other national modeling centers, but the ideas behind them are the same.

About the Lesson

This 90-minute lesson is intended to help operational meteorologists, atmospheric science students, and other interested users understand the role and importance of satellite data in NWP analyses and forecasts.

Lesson Structure

The lesson is structured as follows.

  • Section 2 briefly describes the history of satellite observations in NWP and their impact on NWP model forecast skill.
  • Section 3 provides background information about the types of environmental satellites that provide input to NWP, the satellite observations that are assimilated, the major components of NWP models, and how they forecast atmospheric behavior.
  • Section 4 looks at how satellite observations are actually assimilated into NWP. It begins by describing how observations from new satellite instruments are vetted for use in NWP, and then examines the process of assimilating observations that have been deemed acceptable. You will see how a model’s capabilities affect its use of satellite data and how satellite data helps improve model forecasts.
  • Section 5 describes current challenges to making optimal use of satellite observations in NWP and explores advances in satellite and NWP systems that should address these challenges and improve model forecasts.
  • Section 6 summarizes the lesson.

Learning Objectives

By the end of the lesson, users should be able to:

  • Describe the impact of satellite observations on NWP model analyses and forecasts
  • Identify the primary types of environmental satellites that provide observations to data assimilation (DA) systems and NWP models
  • Describe the major components of forecast models and how they forecast atmospheric processes
  • Describe how observations from new satellite instruments are vetted for use in DA systems
  • Describe the process of assimilating satellite observations and retrievals once they have been accepted for use in DA systems
  • Provide examples of how model limitations can impact the assimilation and use of satellite observations in NWP
  • Describe expected developments in satellite, DA, and NWP systems that will address current challenges and improve NWP forecast quality

Prerequisite Knowledge

To get the most out of the lesson, users should have basic knowledge of environmental satellites and their products as well as general knowledge of NWP models and data assimilation systems. Some of this information can be found in the following MetEd lessons:

Impact of Satellite Observations on NWP

History of Satellite Observations in NWP

Vertical profiles of atmospheric temperature and moisture (soundings) were first provided by the Nimbus III research and development polar-orbiting satellite in 1969. Early on, atmospheric scientists realized that the observations could be used in NWP models. But it wasn't until the 1980s that they were actually assimilated. A number of problems had to be solved first. For example, scientists had to determine what information to assimilate, how best to assimilate it, and what its impact on the forecast was.

The graphic below shows improvements in NWP model skill from 1984 through 2012 resulting from advancements in the GFS model, the data assimilation system that feeds the initial conditions to that model, and the observational systems (mostly satellites).

Graphic shows GFS anomaly correlation (AC) scores for 500-hPa heights for the extratropical Northern and Southern Hemisphere (NH and SH, 20-60N and 20-60S), with scores from a fixed version of the Climate Forecast System (CFS) as a baseline for comparison

Plotted are GFS anomaly correlation scores (skill scores for planetary to synoptic-scale flow) for 500-hPa heights for the extratropical Northern and Southern Hemispheres, with scores from a fixed version of the Climate Forecast System (CFS) as a baseline for comparison. Since 1987, when Southern Hemisphere forecasts began, anomaly correlation scores have improved about 33% in the Northern Hemisphere and about 44% in the Southern Hemisphere. Even with the advances in models and data assimilation systems during this period, a large portion of the improvements are due to the dramatic increase in satellite observations used by data assimilation systems.

If you'd like more information on the satellite-related developments that improved NWP model forecast skill, click the link below. Some of the information is rather technical and not explained until later in the lesson, so feel free to bypass it.

More Information

NWP model forecast skill increased significantly from the mid-1980s through the mid-1990s when satellite sounding retrievals were added and improved.

The skill gap between the Southern and Northern Hemispheres decreased in the late 1990s and early 2000s when the data assimilation system started assimilating sounding retrievals derived from satellite observations, not just satellite observations.

Forecast skill improved significantly in the late 1990s and early 2000s when the first operational microwave sounders were launched on polar orbiters. They provided atmospheric profiles through cloud cover, which is particularly helpful in regions where persistent cloud cover blocks visible and infrared observations.

Impact of Satellite Instrument Type on Forecast Error

Which satellite instruments have the greatest impact on reducing forecast error? The graph below shows observational datasets from various instruments ranked by how much they reduced forecast error from September through December 2008 at the European Center for Medium-Range Forecasts (ECMWF).

We’ll examine this graphic in more detail later in the lesson, after we’ve discussed the types of satellite instruments that provide data to NWP. But for now, notice that sounding data from polar-orbiting satellites and, to a lesser extent, Global Positioning Satellites-Radio Occultation - the red bars - reduce forecast error by almost half. While these results are specific to the ECMWF model, other modeling centers have obtained similar results when adding this data.

Graph showing percent contribution of different types of observations on forecast error on the ECMWF system from Sep to Dec 2008

Next, we'll look at a case example that focuses on the impact of polar-orbiting satellite data on NWP forecast skill.

Hurricane Sandy Experiment

On 29 October 2012, Hurricane Sandy made landfall along the coastal areas of New York, New Jersey, and southern New England, causing storm surges of over ten feet. Six to seven days before the storm tore through the area, the ECMWF NWP model predicted that it would take a sudden left turn (turn towards the west), providing sufficient time to mitigate the loss of life and property.

As part of an NWP forecast post-storm assessment, the ECMWF did an Observation System Experiment (OSE) in which it excluded polar-orbiting satellite observations from the analysis that provides the starting point for the forecast. The OSE was performed to see the impact of losing these observations on the forecast.

The graphic below shows the 168-hour operational and OSE forecasts, and the verification for Hurricane Sandy at the time of landfall. The control forecast included polar orbiter observations and had the hurricane making landfall near Norfolk, VA, with strong onshore winds expected north of the storm that would likely produce a storm surge up to the New York City area. Without the satellite data set (middle panel), the forecast only called for high surf and rip currents on the northeastern U.S. coast.

Graphs showing results from 0 - 168 hr forecasts for Observing System Experiments (OSEs) withholding conventional, satellite data, and GFS operational control from 15 Aug to 30 Sep 2010

Extent of Impact From Satellite Observations

The Hurricane Sandy Observation System Experiment showed how essential polar orbiter observations were to the forecast guidance for a single event. Do all satellite observations typically have such an impact on forecast quality? If so, how far into the forecast period is this impact felt?

To answer these questions, atmospheric scientists ran two OSEs using the May 2010 version of the GFS. The GFS control forecast included all observations. One experiment omitted all conventional observations, such as radiosondes, while the second omitted all satellite observations. The results are shown below for 0-hr to 168-hour forecasts, with the Northern Hemisphere extratropics on the left and the Southern Hemisphere extratropics on the right.

Graphs showing results from 0 - 168 hr forecasts for Observing System Experiments (OSEs) withholding conventional, satellite data, and GFS operational control from 15 Aug to 30 Sep 2010

The lower panel shows that denying satellite data to NWP significantly degrades the forecast at all forecast times in both hemispheres, and has a greater impact than denying conventional observations (radiosondes, surface weather stations, ships, buoys, and aircraft). Notice, however, that the effects of withholding conventional data are hardly noticed in the Southern Hemisphere.

Question

Why does withholding conventional data have less of an effect in the Southern than Northern Hemisphere? (Choose the best answer.)

The correct answer is B.

The graphics in the tabs below show coverage for four types of observation systems. Even with aircraft data coverage, there are significantly more conventional observations (radiosondes and aircraft) in the Northern Hemisphere than Southern Hemisphere. In addition, a far greater proportion of observations in the Southern Hemisphere is from satellites, which explains their large impact on model forecast skill in that region.

Radiosondes

Graph showing temperature data from radiosondes used by ECMWF in over 124 forecast cycles from 1 Mar to 1 Apr 2012

Aircraft

Graph showing temperature from AMDAR aircraft observations used by ECMWF in over 124 forecast cycles from 1 Mar to 1 Apr 2012

LEO Soundings

Graph showing AMSU-A Sounding Data used in a single ECMWF analysis cycle ending 00 UTC 30 Mar 2012

Satellite-Derived Winds

Plot showing Atmospheric Motion Winds used in a single ECMWF analysis cycle, 12 UTC 29 Mar 2012
Please make a selection.

Satellites, NWP, and the Forecast Process

General Forecast Process

The flowchart outlines the forecast process, from observations to the human generated forecast. Satellite observations are integral to the entire process. They are included in the observations that enter the data assimilation system, which creates the best possible analysis of initial conditions for starting the NWP forecast. The downward arrow from the DA system to the model forecast refers to the use of the analysis as the starting point for the current forecast cycle. The upward arrow from the model forecast to the DA system reflects the DA system using the most recent NWP model forecast valid at analysis time (the “first guess”) as the starting point for its analysis.

Flowchart of the general forecast process, from observations to the human forecast

The rest of this section explores the satellite and NWP boxes in more detail, focusing on the types of satellite instruments and observations that are used in NWP, the major components of forecast models, and how those components forecast atmospheric processes. This sets the stage for Section 4, which describes how observations from new satellites instruments are vetted for inclusion in DA systems and how the observations deemed acceptable are actually assimilated.

GEO and LEO Imagers

Three primary types of satellites remotely sense meteorological data:

  • Geostationary earth orbiters (GEOs)
  • Low earth orbiters (LEOs), a subset of which are polar-orbiting satellites
  • Global positioning systems (GPS) in medium earth orbit (MEO)
Graphic showing the rotation of geostationary (Meteosat) and polar-orbiting (MetOp) satellites around the earth plus the orbits of GPS satellites in Medium Earth Orbit used in radio occultation

GEO and LEO satellites carry two types of instruments: imagers and sounders. Each provides observations from different parts of the earth-ocean-atmosphere system.

Visible and infrared imagers mainly observe the condition of radiating surfaces, such as soil, vegetation, water, ice, and cloud tops. A subset of infrared channels is sensitive to atmospheric water vapor and observes mesoscale to larger-scale atmospheric circulations.

The amount of energy that reaches an imager at a particular wavelength (commonly expressed as a radiance) from these objects is considered a satellite observation. This energy is typically displayed as a single channel image.

Satellite observations from multiple channels can be further processed to create products that provide information about surface and atmospheric conditions, such as snow cover, atmospheric winds, and dust.

The graphic below shows EUMETSAT’s Dust RGB product, which detect the presence of atmospheric dust. The product is made from the three infrared channels/channel differences shown on the left. The dust stands out clearly in magenta in the RGB product.

Dust RGB with its 3 input images in different color guns (IR12.0-10.8, IR10.8-8.7, IR10.8)

Sounders

Sounding instruments sense energy in portions of the spectrum where it is absorbed and re-emitted by atmospheric constituents such as carbon dioxide, water vapor, and ozone. These measurements are used to extract three-dimensional information about temperature, moisture, and other atmospheric constituents.

A new generation of LEO sounders, known as hyperspectral sounders, sense energy at very high spectral resolution, sampling over thousands of spectral bands. Hyperspectral sounders are capable of providing radiosonde-like, high-resolution profiles of the atmosphere. Partly as a result, 60% of the satellite data assimilated into NWP models comes from LEO sounders.

Among the hyperspectral sounders is the Cross-track Infrared Sounder or CrIS, which flies on the Suomi NPP and future Joint Polar Satellite System (JPSS) polar orbiters. CrIS is a high spectral resolution interferometer that senses infrared energy from the atmosphere, surface, and cloud cover. CrIS flies in tandem with the Advanced Technology Microwave Sounder or ATMS, which can sense into and through most cloud cover, giving it a three-dimensional perspective into cloud and storm systems. ATMS data started being assismilated into operational NWP models in 2012, CrIS data in 2013.

Illustration of sounding in partly cloudy environments when microwave and hyperspectral instruments are combined

As with imagers, hyperspectral sounder channels can be combined to retrieve meteorological products. Examples include total precipitable water (TPW), aerosol optical thickness, cloud top temperature and height, cloud top water phase (water versus ice), cloud optical thickness, and particle size. These are useful to forecasters in diagnosing such things as aviation and air quality hazards, and for NWP in improving the analysis of temperature and moisture. High-resolution sounding data also benefits NWP (and human) forecasts of convection and severe weather, precipitation type, and maximum and minimum temperature.

Even though GEOs, such as the U.S. GOES-East and GOES-West, continuously observe the same areas, sounders onboard GOES-8 to 15 only have 18 infrared sounder channels, resulting in GEO soundings with coarse vertical resolution. As a result, almost no GEO sounding data is used in data assimilation.

High vertical resolution hyperspectral sounders are not available on the current U.S. Geostationary Orbit Environmental Satellite (GOES) satellites, nor will they be on the next-generation GOES-R series. This degrades monitoring of the atmosphere in potentially crucial situations such as severe convection.

The graphic below compares GEO and hyperspectral LEO soundings to radiosonde profiles from the same location and time. In the LEO sounding, note the fine detail and excellent agreement of temperature, dewpoint, and convective instability to the radiosonde values.

Comparison of temperature and dewpoint profiles from RAOB, current GOES Sounder, and a simulated high spectral resolution IR sounder.

For more information on hyperspectral sounders, see the COMET lesson Advanced Satellite Sounding: The Benefits of Hyperspectral Observation - 2nd Edition. For more on the use of hyperspectral sounders in GEO vs. LEO satellites, access the COMET lesson Toward an Advanced Sounder on GOES?.

GPS-RO

Another type of sounding technology provides atmospheric profiles of temperature and moisture. It uses Global Positioning Satellite (GPS) signals that are intercepted by LEO satellites, such as the COSMIC constellation and EUMETSAT’s MetOp satellites, in a process called radio occultation (RO).

GPS radio signals bend as they move through the atmosphere. The amount of bending depends on the density of air in the signal’s path, which, in turn, depends on the temperature and moisture along the path. The process is illustrated below, with hypothetical soundings for temperature and water vapor pressure shown on the right. As of 2014, GPS satellites provide around 2,500 soundings around the world each day for assimilation in NWP, compared to about 44,000 soundings per day from the AIRS hyperspectral sounder on NASA’s Aqua polar-orbiting satellite.

Graphics showing GPS sounding through radio occultation plus a temperature and moisture sounding made from the data

For more information on GPS and the extractable meteorological information, see the UCAR COSMIC page and COMET’s COSMIC lesson.

Which Satellite Observations Are Assimilated

For purposes of NWP modeling, the earth-ocean-atmosphere system can be divided into three spheres: the lithosphere (including vegetation), hydrosphere, and atmosphere. While satellites provide massive amounts of information about all three spheres, model limitations only permit some of it to be used.

Click on each sphere on the graphic to see the observations and measurements that are assimilated into NCEP DA systems as of 2014.

While most information comes from LEO and GPS-RO sounders, GEO sounders contribute as well.

image image image image

Note that some satellite observations, such as snow cover and snow depth, are used in the analysis and allowed to evolve throughout the forecast, while climatological values derived from multi-year satellite observations are used for other parameters. The latter tend to evolve slowly over time, such as land use or vegetation, which can take years to change.

In the next section, you will see where this satellite data is plugged into the DA system, how DA systems relate to NWP models, and how NWP models generally work.

How NWP Models Forecast Atmospheric Processes

NWP models simulate the atmosphere and provide forecast guidance on the occurrence and values of weather elements such as temperature, moisture, wind, and precipitation. Weather models can forecast phenomena from tens to thousands of kilometers in size and timescales from hours to days. While this lesson focuses on NWP models, similar principles apply to climate models, which forecast at larger spatial and longer temporal scales.

The computer programs (“modules”) in an NWP model make up its “architecture.” There are three general categories of modules. One set, the data assimilation system, ingests observational data to prepare the analysis or initial conditions for the forecast, while the other two, dynamics and physical parameterizations, create the forecast. We will spend much of this lesson discussing how satellite data is used in the DA system.

Data Assimilation System

The data assimilation system provides an analysis of the current state of the atmosphere and land and ocean surfaces. To get a good NWP forecast, this analysis must be of high quality. This requires a good DA system, good data to assimilate, and, of course, a good NWP model.

At the U.S. National Centers for Environmental Prediction (NCEP), analyses for global and large-domain mesoscale models are done every six hours, starting at 00 UTC each day. For each analysis, the DA system ingests observational data over a “time window” centered on the analysis time. The data is combined with a short-range forecast or “first guess” to create a final analysis from which to start the NWP forecast.

The graphic below is an idealized schematic of how DA works. The blue lines represent forecasts for a series of cycles that evolve with time. The red line and pink shading are the true atmospheric state (“truth”) and its uncertainty. The vertical black arrows represent the DA process, where the first guess and observations are combined to create the final analysis. Notice that while forecasts always drift away from the true atmospheric state, the DA process pulls the first guess back toward the “truth” using the observations. Also notice that the new analysis is never exactly the “true” atmospheric state, but lies within the range of uncertainty in pink shading.

Evolution of a model variable and forecast error during model integration

Since the analysis relies on the short-range forecast first guess, the quality of the NWP model is important to the DA system. And because NWP forecasts are sensitive to their starting point or initial conditions, they depend on a good DA. If the model first guess were not anchored to true atmospheric conditions in the DA process each cycle, model analyses would be too far from the “truth,” which would negatively impact NWP forecasts.

Next, we’ll look at the modules that produce the forecast, dynamics and physical parameterizations.

Forecast Modules

Once the DA system completes its analysis of the initial conditions, the NWP forecast begins. The model’s dynamics and physical parameterizations each handle different aspects of atmospheric simulation.

Dynamics

A model’s dynamics forecast the evolution of atmospheric processes that can be directly “seen” at the model’s time and spatial scales. Depending on the model’s resolution, the dynamics may simulate atmospheric phenomena ranging from planetary and synoptic-scale waves to narrow precipitation bands. Some of the highest-resolution models can directly predict convective updrafts and downdrafts.

Dynamics are calculated using equations that forecast the wind and its movement of momentum, heat, moisture, and other atmospheric constituents, such as ozone and condensed cloud water.

Physical parameterizations

Some processes occur on time and/or spatial scales too small to be directly accounted for in the model. But since they significantly affect the atmosphere, their impacts must be estimated through physical parameterizations. The graphic shows the processes that require physical parameterization in most models. These include, but are not limited to, short- and longwave atmospheric radiative transfer, cloud and precipitation microphysics, land surface and planetary boundary layer (PBL) processes, and convection.

Illustration of all processes and physical model elements that are parameterized in numerical weather prediction models. Includes 20 different items, such as topography, deep convection, longwave radiation absorption and emission, microphysical processes, land surface processes and land use types, soil and vegetation processes, snow/water/ice at the earth surface, atmospheric radiation transfer, etc.

Between forecast calculation times, the NWP model adds the effects of the physical parameterizations and wind dynamics on atmospheric temperature, moisture, and momentum within a 3-D model grid box to establish the next set of forecast values.

For more information on the basics of physical parameterization, see the physical parameterization section of COMET’s NWP Model Fundamentals lesson.

Now that you are armed with the basics of DA systems and NWP models, we will explore how satellite data is vetted for use in DA and the process of assimilating the data that has been deemed acceptable.

Assimilating Satellite Observations Into DA Systems

Introduction

This section focuses on the use of satellite data in NWP, examining such topics as:

  • How observations from new satellite instruments are vetted for use in DA systems
  • How satellite observations are assimilated once they are accepted for use
  • How NWP model capabilities impact the use of satellite observations

Accepting Data From New Instruments: Flowchart

When new satellite instruments are designed and orbits determined, the needs of data assimilation and model forecasts are taken into consideration. For example, hyperspectral sounders were developed in response to the need for higher vertical resolution in the satellite soundings used in DA systems. In a similar vein, new LIDAR (Light/Laser Detection and Ranging) instruments were developed to provide more of the wind data needed to improve DA analyses.

Once an instrument is launched and begins sending observations, the data goes through a testing phase (also known as the “mission checkout phase”) that can last several months to a year or more. The process from pre-launch to acceptance as valid data for DA is outlined in the flowchart below. Click on each box for more information. On the next page, you’ll answer a series of questions about the process.

Flowchart depicting how new satellite observations are accepted into data assimilation systems

Accepting Data From New Instruments: Questions

Now that you’ve studied the flowchart, answer the following questions.

Question 1

Satellite observations continue to be monitored and assessed by satellite and DA scientists regardless of whether they are actually used in DA. Why is this done? (Choose all that apply.)

The correct answers are B and C.

Satellite observations are monitored and assessed because a model may evolve to the point where it's able to use the data. High quality observations are also critical to forecasters, who rely on them for monitoring the weather and evaluating NWP guidance, and other users. Option A is incorrect because the data does not tell us when an instrument will fail, as has been seen with the unexpected loss of satellite observations.

Please make a selection.

Question 2

What characteristics of an NWP model must be considered when assimilating any observations? (Choose all that apply.)

The correct answers are A and B.

For A, what the model can simulate largely determines the satellite data used. For example, if an NWP model only indirectly accounts for thunderstorm processes using convective parameterization, observational data reflecting convective vertical motions will not be used in the analysis. For B, if we have satellite data for a variable that the model doesn't predict, it will not be used. Option C is incorrect because the length of time between computations makes no direct difference in what data can be assimilated.

Please make a selection.

Question 3

In what ways can model architecture limit or prevent the use of real-time satellite data? (Choose all that apply.)

The correct answers are A and B.

For A, using climatology in the NWP model prevents the DA system from using real-time satellite data for the analysis of that quantity. For option B, if a model has coarse resolution in the stratosphere, it will limit the usefulness of ozone (and other) important satellite observations. Option C does not prevent assimilation of real-time satellite data; in fact, it makes it critical!

Please make a selection.

Question 4

A new instrument is launched on the latest satellite. Why might the DA system not use its observations? (Choose all that apply.)

The correct answers are A and C.

For A, high quality satellite data tested in DA systems has sometimes decreased forecast skill. This may be from interactions with other data in the model, conflict with forecast model climatology, or other reasons. For C, if an NWP model does not predict a feature, it will not be able to assimilate remotely sensed data of that feature. Option B is incorrect because data at higher resolution than the NWP model can be averaged over the model grid box (superobbing) or thinned to the resolution of the model.

Please make a selection.

Question 5

New satellite data deemed useable by a DA system is assigned a relative weight, which determines its influence over the final DA analysis. Data with larger random error has: (Choose the best answer.)

The correct answer is A.

An observation's weight in the final DA analysis is directly related to its expected random error. Data with smaller random errors is more reliable and therefore has greater weight in the analysis.

Please make a selection.

Question 6

A new instrument was launched several months ago, and the data has been tested in a DA system. The data will be incorporated into the operational DA system if it: (Choose all that apply.)

The correct answers are A and B.

Improved forecast skill is an obvious reason to incorporate data from the new instrument. Maintaining forecast skill is not as obvious but the data may have a positive impact when combined with even newer data in future testing. Data that reduces forecast skill will not be included in the current DA system. However it may be incorporated in the next DA system upgrade if there's time to adjust and retest it.

Please make a selection.

Assimilating Observations and Retrievals: Flowchart

Once satellite observations are accepted for use in a DA system, they become part of the data compared to the first guess. But how is this comparison done and how is the data used?

Satellite data is used in DA systems in two forms:

  • As satellite observations, which measure the radiance or energy scattered, transmitted, absorbed, and reemitted from the earth-atmosphere system
  • As satellite retrievals extracted from the observations, such as atmospheric motion vector winds, temperature, and humidity.

The general process for assimilating satellite data, regardless of form, is shown in the flowchart below.

Flowchart showing how satellite observations and satellite retrievals are used in data assimilation

First, each satellite retrieval or observation is checked for gross errors to remove extreme or unphysical values, such as an Earth surface temperature of 100°C. Gross errors can result from:

  • Satellite instrument problems, such as uncorrected instrument degradation or interference from cosmic radiation or commercial radio waves
  • Errors in data processing, such as assigning the wrong height for cloud motion winds
  • Cloudy fields of view for infrared channels and precipitating fields of view for microwave channels that are not accounted for and can interfere with obtaining soundings down to the surface
  • Errors in estimating the amount of radiation emitted by the surface, which results from uncertainties in emissivity (this is the radiative emission efficiency of an object when compared to an ideal emitter, also known as a blackbody, which is expressed as a value between 0 and 1)

The observations or retrievals are then compared to the model first guess, with the difference calculated. This difference is the “analysis increment,” which is given an analysis weight depending on our relative confidence in the data compared to other types of data available in the area. For example, satellite data might get less weight in regions with radiosondes and other highly accurate observations than over the oceans, where those types of observations are not available. Note that the first guess is given a heavy weight since it is from a short-range forecast of generally high quality.

Next, the analysis increment is compared or “buddy checked” with its neighbors. If the increment is very different from those around it, its weight will be reduced. It is not completely discarded in case it has uniquely captured a significant error for which the first guess requires adjustment.

Finally, the weighted increment is combined with those from other satellite and observation platforms, and used in the analysis to obtain the best model analysis possible. The forecast then begins, with the current cycle’s short-range forecast valid at the next analysis time serving as the first guess for the next forecast cycle.

Assimilating Satellite Retrievals

Recall that satellite data is processed in two forms, as observations and as retrievals. While both go through the same general assimilation process, there are some important differences.

Basically, the quantities that satellites observe, such as radiance and sea surface backscatter, are not explicitly forecast by NWP models so some processing is needed before the DA system can make the comparison. In contrast, many satellite retrieval quantities are directly forecast by NWP models and can be matched directly to the first guess fields.

We'll examine these differences in a bit more detail, starting with retrievals. We’ll use atmospheric motion vector (AMV) winds as an example.

Atmospheric motion vectors are retrieved by following cloud and water vapor features in infrared and water vapor channel imagery. The displacement over a series of images determines the direction and speed of the wind at feature level. The height is computed from observed brightness temperatures. As with other satellite retrievals, the AMV wind is compared directly with first guess data, in this case the first guess wind.

Below are AMV winds from water vapor features on the left, and the (six-hour forecast) first guess 300-hPa winds and heights from the NCEP NMM-B (Non-hydrostatic Mesoscale Model - Arakawa B-grid) on the right, both valid at the same time. The cyan-shaded AMV wind barbs are for pressures from 250 to 350 hPa.

2 plots, one showing GOES Hi-Density winds, the other the 6-hr NMM-B 300 hPa forecast for heights and winds, both for 12 UTC 12 April 2013

In the satellite retrieval on the left, the wind from the AMVs is 65kt from the northwest. That’s not far from the April climatology for this area, so this wind should pass the gross error check. In the first guess on the right, the forecast wind in the area is 80kt from the northwest, perhaps slightly more north of west than the AMV wind. Subtracting the observation from the first guess gives us an unweighted analysis increment of southeast at about 15 kt for the 300-hPa vector wind.

Depiction of how, in data analysis, the analysis increment is the difference between the observation and the first guess

This value is compared to the neighboring increments for wind from satellite and other sources (not shown). If the other increments are significantly different, the typical weight used for AMVs at this height will be reduced, with the amount of reduction depending on the size of the difference.

Assimilating Satellite Observations

Assimilating satellite observations requires some extra steps before the analysis increment can be determined. We will examine these steps, using “brightness temperature” at a particular wavelength as the satellite observation example.

First, the model first guess is converted into a simulated satellite observation (in the form of brightness temperature) using a “forward” (radiative transfer) model. Second, the difference between the observed and simulated brightness temperature is made as small as possible by adjusting other variables such as temperature and moisture. And third, the adjustments to the first guess variables from this process become the analysis increments.

The images in the tabs show the first guess, the final analysis of brightness temperature, and the satellite image for water vapor channel 3 brightness temperature for the 12 UTC 16 January 2014 NAM analysis cycle. Keep in mind that the final analysis includes adjustments for conventional data as well as satellite data. As we might expect, the first guess and analysis are generally quite close. Differences are most obvious near the U.S. East Coast, in the swirl of clouds with an upper-level cyclonic circulation in the center of the North American continent, and in the dry area in red from west of Oregon to the Gulf of Mexico and Florida. Finally, the satellite image illustrates additional details that are not included in the DA system because the NAM has coarser resolution than the satellite observations.

First Guess

Satellite-observed brightness temperature from water vapor channel 3 and the corresponding NMM-B simulated satellite brightness temperature from the first guess on the right for 1 May 2013

Analysis

GOES Ch 3 brightness temperature NAM 00H forecast for CONUS valid 00 UTC 16 Jan 2014

Water Vapor Channel 3

GOES water vapor ch 3 brightness temperature 1145 UTC 16 Jan 2014

 

The rest of the process is the same as for retrievals, with the analysis increment weighted, buddy checked, and weight adjusted if necessary before being used in the analysis with increments from all other observation sources.

In recent years, modeling centers have distributed simulated brightness temperatures as satellite look-alike products to help forecasters and other users assess short-range NWP forecasts. At NCEP, these simulated products are created by the same radiative transfer model used to produce the simulated satellite observation in the DA system.

For training on how to use synthetic imagery to forecast orographic cirrus, severe weather, low clouds and fog, and cyclogenesis, see the VISIT website. In addition, the MetEd website has training on using synthetic satellite data to forecast fog for aviation in a lesson on the use of aviation products from the Weather Research and Forecast - Environmental Modeling System (WRF-EMS) NWP model in East Africa.

Assimilation Questions

Question 1

Which statements capture the difference in how satellite observations and retrievals are assimilated? (Choose all that apply.)

The correct answers are A and D.

Satellite retrievals derived from satellite observations are compared to model data in its original form. For example, retrieved atmospheric motion vectors are compared to first guess winds. Conversely, satellite observations are compared to a model first guess that's been converted to a simulated satellite observation.

Please make a selection.

Question 2

Why can satellite retrievals, but not satellite observations, be compared directly with model first guess fields?(Choose all that apply.)

The correct answers are B and C.

For option B, NWP models that provide the first guess in DA do not directly predict the quantities observed by satellite instruments. The first guess data has to be converted into a synthetic satellite observation via a radiative transfer ("forward") model. Option C is correct because first guess fields can be compared to satellite retrievals of that field. A good example is model winds and satellite-derived atmospheric motion vectors.

Please make a selection.

Question 3

Which of the following satellite retrievals and observations is most likely to be considered a gross error if used in a DA system? (Choose the best answer.)

The correct answer is B.

It is highly unlikely that a surface temperature of 20°C could coexist with snow cover. Option A could exist in high latitudes for seawater because of its salt content. Option C could easily exist for any reasonably sized convective system. Option D is well within the climatological range of TPW for the central U.S. in March.

Please make a selection.

Question 4

A GPS-RO total precipitable water satellite retrieval is accepted after a gross error check and has a +5 mm analysis increment when compared to the short-range forecast. The nearby analysis increments range from -1 mm to +1 mm. What will happen to the GPS-RO analysis increment? (Choose the best answer.)

The correct answer is D.

The weight of an analysis increment that's significantly different from its neighbors will be reduced, not increased. This is done so it won't adversely affect the analysis but will still be included in case it reflects a needed correction missed by the other increments. Option A is incorrect because data that has passed the gross error check is not rejected even if the analysis increment is much different from its neighbors for the reason explained above. Option C is incorrect because it could negatively impact the moisture analysis if its weight were not changed.

Please make a selection.

Future Advances Using Satellite Observations in NWP

Overview

In this section, we will identify current challenges (as of 2014) to making optimal use of satellite observations in NWP due to limitations in NWP models, DA systems, and satellite systems, and look at advances in all three areas that will address them. When particular satellites are involved, we will mention them. Among these are:

  • The U.S. GOES-R series of geostationary satellites, scheduled to launch in 2016. They will bring new capabilities for lightning detection, more than triple the number of infrared channels, four times higher spatial resolution, five times faster imaging, and more accurate measurements for observing subtle features.
  • The Suomi-NPP polar orbiter, which launched in 2011, and the follow-on Joint Polar Satellite System (JPSS) satellites, the first of which is scheduled for launch in 2017. These satellites have a hyperspectral sounder (CrIS), a microwave sounder (ATMS), and a 22-channel imager (VIIRS), which includes a day-night band for high-resolution visible imaging at night.
  • CALIPSO and ADM-Aeolus satellites with Light Detection and Ranging (LIDAR) instruments. CALIPSO is a research instrument for studying the role of clouds and aerosols in regulating Earth’s weather, climate, and air quality. ADM-Aeolus will be the first satellite to directly observe wind profiles from space and will also provide information on aerosols and clouds. ADM-Aeolus is the European Space Agency’s Atmospheric Dynamics Mission, planned for launch in 2015. CALIPSO is NASA’s Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation, launched in 2006 as a research instrument. It flies in formation with NASA’s A-train constellation.
  • COSMIC-2, a GPS-RO follow-on mission with initial implementation scheduled in 2015.

We will begin by looking at the challenges related to DA and NWP, showing experimental results that address them when available.

DA System and NWP Model Challenges

Convection

CONVECTION

Challenge: If a convection-allowing model does not predict convection at a particular location in its first guess, the model will not assimilate satellite data from active convection at that location.

Expected advance: A potential solution has been found through experimental assimilation of lightning data in high resolution models. The experimental DA system builds thunderstorms in the analysis where there’s observed lightning but no storms in the model, and adjusts storm intensity based on lightning frequency.

The rationale for the assimilation is based on the observed correlation between increases in lightning frequency and convective storm intensity. Also over the next decade, new GEOs with lightning detection capabilities will provide continuous lightning coverage over much of the full Earth disk. They will be able to observe cloud-to-ground as well as in-cloud and cloud-to-cloud lightning. The GOES-R series of satellites will begin carrying the Geostationary Lightning Mapper (GLM) in 2016, and both China and EUMETSAT have plans to launch lightning imagers later this decade.

Experimental results: We see the impact in these graphics. The first two rows show the chance of convective initiation within 25 miles of a point at two- and one-hour lead times for forecasts run with and without proxies for the future GOES-R GLM lightning data. The third row shows the Storm Prediction Center’s severe weather reports during the event and radar at the two-hour forecast lead time. The experiment clearly provides better guidance than the control model run where no lighting was assimilated.

Plots showing model convective initiation probability (%) with 25 mi of a point with storm report and radar verification

Satellite/instrument data source: GOES-R will carry the GLM, which will provide lightning data to DA systems for convection-allowing NWP models. This should improve both initial and forecast convection strength and location.

Microphysics

MICROPHYSICS

Challenge: Retrieved cloud microphysical properties and precipitation hydrometeors are not assimilated in DA systems, resulting in poorer NWP forecasts of cloud and precipitation.

Expected advance: Work is being done at NCEP to assimilate microwave radiances for “all sky” conditions (clear to cloudy) by using GFS first guess profiles of cloud water and ice in calculating the simulated model radiance. Results have been mixed so far, with more work needed on:

  • Quality control
  • Thinning or averaging observations and bias-correcting the microwave radiances
  • Determining the weight and area of influence of the resulting analysis increments

When predicted rain, snow, and graupel are added to the GFS microphysics, these will be included in the calculation of first guess microwave radiance.

On a second research front, work continues on assimilating retrievals of cloud and precipitation hydrometeor data at the NOAA Joint Center for Satellite Data Assimilation. Implementation of an operational DA system that assimilates retrievals of cloud microphysics and precipitation hydrometeors, or the microwave radiances affected by these hydrometeors, appears to be at least several years away (as of 2014).

Below are examples of the satellite retrievals available in 2013 from the NOAA Office of Satellite and Product Operations (OSPO) operational MW Integrated Retrieval System (MIRS), all from NOAA-18. Included are total integrated cloud liquid water, rain water path, and ice water path. All could be assimilated into DA systems if the models providing the first guess directly forecasted these variables. Operational assimilation of these quantities would improve cloud and precipitation forecasts, particularly in the first 12 to 24 hours.

Examples of satellite retrievals from the Microwave Integrated Retrieval System (MIRS), including rain water path, graupel, and cloud liquid water, 23 Oct 2013

Satellite/instrument data source: The Microwave Integrated Retrieval System (MIRS), which uses microwave observations from multiple satellite instruments, provides profiles and integrated total quantities of moisture, cloud amount, rain, ice, snow, and graupel that are not yet assimilated into DA systems. Once testing is complete and the data begins to be assimilated, it will directly improve NWP cloud and precipitation forecasts, and indirectly improve other forecast variables affected by cloud and precipitation processes.

Natural and Anthropogenic Aerosols

NATURAL AND ANTHROPOGENIC AEROSOLS

Challenge: NCEP NWP models use an aerosol seasonal climatology, rather than retrievals of actual aerosol amount, to estimate their effects on radiation in the atmosphere and at the surface. While GEO and LEO satellites already provide aerosol retrievals, there is no predicted aerosol in the first guess with which to compare them.

Expected advance: NCEP plans to add an aerosol forecast model to the operational GFS and aerosol retrievals to the DA system in the 2015 timeframe. This will improve first guess simulated radiances, leading to better assimilation of all sounder and imager data.

Satellite/instrument data sources: CALIPSO provides information on thin clouds, aerosol profiles, and total column integrated aerosol quantities using three laser channels in experimental mode. ADM-Aeolus will provide these retrievals in real time after its anticipated launch in 2015. Once DA systems and NWP models are able to use the data operationally, we can expect improvement in the redistribution of energy in the atmosphere and improvements in near-surface and lower troposphere forecasts.

Surface Conditions: Soil moisture

SURFACE CONDITIONS: SOIL MOISTURE

Challenge: Soil moisture is retrieved from space-based microwave instruments but the products are not assimilated into DA systems. Instead, NCEP NWP models forecast soil moisture using a four-layer soil model that is either:

  • Slowly adjusted toward a seasonal climatology (GFS), or
  • Initialized with values from an independent land data assimilation system (LDAS) that is updated with analyzed precipitation and estimated evaporation values for the NMM-B (Nonhydrostatic Multi-scale Model - Arakawa B Grid)

Assimilation of real-time soil moisture retrievals would improve the analysis and forecast of near-surface and planetary boundary layer conditions, including the initiation of convection, especially during droughts and floods.

Expected advances: NESDIS (NOAA’s National Environmental Satellite, Data, and Information Service) uses surface (1- to 5-cm deep) soil moisture retrievals from LEO microwave instruments to analyze surface soil moisture in the Soil Moisture Operational Products System (SMOPS). The graphic below shows a SMOPS plot of the fraction of the top 5-cm soil layer containing water. The patchiness results from complex land cover, mountainous terrain, dense vegetation, and gaps in satellite coverage.

NOAA SMOPS blended soil moisture, daily for 19 Aug 2013

Experiments assimilating these soil moisture retrievals in the GFS were run for April 2012, when significant drought was developing in the U.S. Midwest. The resulting monthly mean soil moisture was reduced over much of the U.S., correcting a wet bias in the GFS. Forecasts of other quantities, including 500-hPa height and rainfall, were also slightly improved (not shown).

Difference in Surface Soil Moisture (% saturation), Average for April 2012 at 18 UTC

Future work on satellite soil moisture retrievals will improve the weighting and area of influence of soil moisture data and analysis increments so they can be included in the operational DA system. In addition, microwave data from the Japanese GCOM-W (Global Change Observation Mission) polar-orbiting satellite series will be added.

Satellite/instrument data sources: The NESDIS operational soil moisture analysis is made by blending data from:

  • U.S. Navy’s Windsat microwave radiometer onboard the Coriolis polar orbiter
  • EUMETSAT’s ASCAT (Advanced SCATterometer) on MetOp-A, -B, and future -C polar orbiters
  • MERIS (Microwave Imaging Radiometer using Aperture Synthesis) onboard the European Space Agency’s SMOS (Soil Moisture Ocean Salinity) mission launched in 2009

Plans are to incorporate soil moisture retrievals from the Japanese GCOM-W AMSR-2 microwave instrument during 2014 to replace data from the AMSR-E instrument that failed in 2011. Among the expected results are better near-surface temperature and moisture forecasts, improved simulation of the planetary boundary layer, and better prediction of stability parameters in convective environments.

Surface Conditions: Greenness Fraction

SURFACE CONDITIONS: GREENNESS FRACTION

Challenge: While satellites retrieve the fraction of land with green vegetation, the information is not assimilated in DA systems. NCEP NWP models use a monthly climatology from a five-year data set of greenness fraction retrievals instead.

Live green vegetation transports soil moisture from below the surface into the atmosphere as water vapor. This is an important local moisture source and needs to be accounted for in NWP models. While NCEP models account for vegetation greenness with a seasonally varying climatology based on satellite observations, real-time observations can differ considerably from that climatology. We see an example below.

Roll or hover your mouse over the image to compare the NOAA/NESDIS real-time vegetation greenness and GFS climatology images.

Both are for the same date and use approximately the same color scale. As you can see, the observed greenness fraction is greater than climatological greenness for area A and lower for area B. Both differences result from seasonal precipitation anomalies.

before after
before after

Question

Assume full sun, equal soil moisture, and otherwise identical atmospheric conditions in both locations at the outset of an NWP forecast. How will actual 2-meter temperatures be impacted by the vegetation greenness anomalies compared to the model's temperature forecasts for areas A and B? (Choose the best answer.)

The correct answer is D.

Area A has more green vegetation than prescribed in the GFS. The vegetation will use more solar energy for evapotranspiration and less for sensible heating than in the model forecast, resulting in cooler actual temperatures than in the model forecast. In area B, the opposite is true. More solar energy is available for surface heating, resulting in warmer temperatures than in the model forecast.

Please make a selection.

Expected advance: Using real-time greenness fraction is on the horizon in the NCEP NWP models. But both this change and the soil moisture change discussed previously will require testing of and adjustment to other land surface model parameters, such as soil heat and moisture conduction, vegetation evapotranspiration, and others set to work best with the climatological greenness fraction. This takes a great deal of time, staff, and computer resources.

Satellite/instrument data sources: Vegetation retrievals are presently available from the AVHRR (Advanced Very High Resolution Radiometer) imagers on several polar-orbiting satellites. Higher resolution greenness fraction is also available from MODIS (Moderate Resolution Imaging Spectroradiometer) and from the VIIRS (Visible Infrared Imaging Radiometer Suite) imager on the current Suomi-NPP and upcoming JPSS satellites. Once successfully assimilated, these retrievals will improve near-surface model forecasts of temperature and dewpoint, planetary boundary layer height and stability, and convective indices like Lifted Index and CAPE (convective available potential energy).

Forward Radiative Transfer Model

FORWARD RADIATIVE TRANSFER MODEL

Challenge: Recall that when satellite observations are assimilated into DA systems, the model fields must first be converted into simulated satellite data using a forward radiative transfer model. The forward model would provide more accurate simulated data if it incorporated more radiatively active components (trace gases and aerosols) that are already available from satellite retrievals but currently excluded from DA systems.

Expected advance: As computing resources increase, more atmospheric constituents will be included in the NWP and forward models. As a result, less bias correction will be needed to account for their absence. This will, in turn, lead to better assimilation of satellite data in DA systems and a better DA system analysis.

Satellite/instrument data sources: For aerosols, LIDAR instruments on CALIPSO and ADM-Aeolus will provide aerosol measurements that can be used in the forward model. For trace gases, data will come from the hyperspectral infrared channels on:

  • AIRS (Atmospheric Infrared Spectrometer) carried by NASA’s Aqua satellite
  • IASI (Infrared Atmospheric Sounding Interferometer) on MetOp-A, -B, and future -C polar orbiters
  • CrIS (Cross-track Infrared Sounder) on the Suomi-NPP satellite and future JPSS polar orbiter series

Analysis of Small-Scale Features

ANALYSIS OF SMALL-SCALE FEATURES

Challenge: High-resolution NWP models have difficulty both assimilating small-scale features in their analyses and correctly predicting their evolution. This results from inconsistencies between such features and the initial analysis fields, which in turn can be caused by problems properly assimilating high resolution data. For example, analyzed small-scale clouds, such as cumulus, require a consistent analysis of water vapor, upward motion, and cloud water at these same scales, and in turn the NWP model needs to be able to properly maintain these scales in its forecasts. An additional complication is that poor forecasts of small-scale features adversely affect the analysis first guess.

Expected advance: Addressing this challenge will require improvements to both DA systems and NWP models. High-resolution data from satellites (and other surface-based sources such as dual-polarization radar) will need to be properly assimilated into the analysis. With a better understanding of the small-scale features in the data, scientists will create DA analyses and NWP models that better assimilate the small-scale motions that support small-scale features, such as gravity waves and convective updrafts and downdrafts.

Satellite/instrument data sources: The following instruments will provide the small-scale observations needed to support the forecast of small-scale features:

  • Infrared imagers (for cloud tops)
  • The GOES-R Geostationary Lightning Mapper
  • High-resolution infrared and microwave sounders
  • Other non-satellite sources

Satellite Data Challenges

Effect of Clouds and Precipitation on Remote Sensing

EFFECT OF CLOUDS AND PRECIPITATION ON REMOTE SENSING

Challenge: NCEP cannot assimilate infrared soundings in cloudy regions where some of the most high-impact weather occurs. That's because first, we cannot see through clouds at IR wavelengths, and second, we cannot accurately determine cloud top height where the IR observations end in the vertical. That’s due to our inability to accurately determine cloud top height, which is where the bottom of a retrieved sounding would be.

Expected advance: Microwave data over the ocean is used to correct satellite pixels that are partially covered by clouds, in a process known as “cloud clearing.” This enables the assimilation of IR observations from pixels with cloud contamination. Ongoing experiments and better instrument data continue to improve microwave sounding radiances over the oceans, which increases the number of useful IR and MW soundings.

Satellite/instrument data sources: The joint U.S. Japanese Global Precipitation Mission (GPM), expected to launch in 2014, will carry both a dual-frequency precipitation radar and microwave imager for measuring precipitation, storm system structure, and cloud and precipitation hydrometeors. Together with the GCOM-W, MetOp, and JPSS satellites, microwave instruments will improve the characterization of clouds and their content. This will further extend satellite sounding coverage into areas affected by clouds and precipitation.

Satellite Wind Data Gaps

SATELLITE WIND DATA GAPS

Challenge: There is insufficient wind data from the current satellite suite to improve wind analyses in DA systems.

Expected advance: Atmospheric motion vectors are currently retrieved by two methods. The first method tracks cloud and water vapor features using visible and infrared imagery from LEO and GEO satellites. While useful, these motion vectors are susceptible to significant vertical placement errors. The problem can be mitigated with LIDAR, which is similar to radar but uses pulsing laser light rather than radio frequencies. Data from the joint U.S./French CALIPSO research satellite LIDAR has been used for proof of concept experiments.

Artist rendition of the Calipso satellite

By 2015, the European Space Agency will launch ADM-Aeolus and its ALADIN Doppler wind LIDAR. ADM-Aeolus is a research mission that will provide near-real-time retrievals of wind direction and speed in the lowest 20 km of the atmosphere with global coverage. Assimilation of the data should be easier because of experience with CALIPSO. Unfortunately, no follow-on missions are currently planned beyond CALIPSO and ADM-Aeolus.

The other wind retrieval method relies on the detection of microwave radar signals reflected off a roughened ocean surface. This allows for the indirect measurement of near-surface ocean winds (about 10m above the surface) under most conditions. The exceptions include moderate to heavy rainfall and actual winds over 50 knots. India’s OceanSat-2 satellite and the European MetOp satellites currently carry instruments called scatterometers that measure near-surface ocean winds across the globe at least twice daily. However, their limited swath width results in large gaps between orbits. Additional scatterometer instruments are expected to be launched over the coming decade to help provide more frequent coverage and reduce coverage gaps.

Satellite/instrument data sources: The same LIDAR instruments used to retrieve aerosol quantities on CALIPSO and ADM-Aeolus will provide high-resolution wind measurements by retrieving the direction and velocity of aerosol particle movement. For near-surface ocean winds, current OceanSat-2 and MetOp ASCAT scatterometer data will be augmented in the future with additional scatterometers that have wider viewing swaths to reduce data gaps between orbits and provide enhanced coverage.

Microwave Sounder Deficiencies

MICROWAVE SOUNDER DEFICIENCES

Challenge: The vertical resolution of microwave sounders is coarser than that of infrared and hyperspectral infrared sounders. This leads to the loss of potentially important details in temperature and moisture structure.

Expected advance: To improve the vertical resolution of microwave sounders, satellites need additional microwave channels. This will provide better sounding in and below clouds, and produce soundings that provide improved information for NWP.

Satellite/instrument data sources: The NOAA/NASA MicroMAS (Microsized Microwave Atmospheric Satellite) is a 4kg 3U CubeSat hosting a passive microwave spectrometer that could provide initial high-resolution microwave sounding data for testing in assimilation systems. The first launch is planned for 2014. MicroMAS will ultimately make up a constellation of CubeSats called the Distributed Observatory for Monitoring of Earth (DOME). DOME should achieve superior spatial, spectral, and radiometric resolution compared to current systems, and provide the needed high-resolution microwave sounding data for DA systems at a relatively low cost when compared to current conventional satellite platforms.

Lack of MW and Hyperspectral IR Sounders on GEOs

LACK OF MICROWAVE AND HYPERSPECTRAL IR SOUNDERS ON GEOs

Challenge: GEOs do not currently have microwave or hyperspectral infrared sounders to continuously monitor the atmosphere. A microwave capability in GEO orbit would offer significant benefits for observing precipitation. Hyperspectral sounders would provide critical wind and high-resolution sounding information in the lower atmosphere for forecasting convection.

Expected advance: Unfortunately, no sounders will be on the next generation GOES-R, a significant gap for forecasters and DA systems. Data from the GOES-R Advanced Baseline Imager (ABI) will replace current infrared sounder data to produce legacy products only. These products will have higher horizontal and temporal resolution, and expanded geographical coverage when compared to the current GOES sounders. This is expected to partially compensate for the reduction in sounding channels and coarse vertical resolution.

A new generation of GEO hyperspectral infrared sounders will come online during the next decade as EUMETSAT and the China Meteorological Administration move forward with plans for their next generation GEO satellites.

For more information on GOES-R ABI, see the COMET lesson “GOES-R ABI: Next Generation Satellite Imaging.” For more information on forecast problems that could be addressed with hyperspectral sounders on GOES satellites, see the COMET lesson “Toward an Advanced Sounder on GOES?”.

Loss of Satellite Coverage

LOSS OF SATELLITE COVERAGE

Challenge: The potential loss of satellite coverage, particularly from LEOs, could degrade DA analyses and NWP forecasts.

Expected advance: Instrument failure is always a risk with satellite platforms, and the risk increases as an instrument ages. To mitigate the possibility of data gaps, we usually have additional satellites in orbit, with instruments that can replace failing ones. Plus, overlap between current satellites and the launch of new series is typically planned to minimize the risk of complete data loss.

While GOES-R satellites will begin replacing the current operational GOES series in 2016, NOAA's next generation operational LEOs, beginning with JPSS-1, will not start launching until 2017. This creates the potential for a data gap as the current LEO satellites exceed their life expectancy around that time. There are currently no plans to accelerate the deployment of NOAA's next generation operational LEOs.

Summary

Satellite Data, DA Systems, and NWP Model Forecasts

Data assimilation systems

Data assimilation systems produce the starting point or analysis for an NWP forecast. To do this, they take a short-range NWP forecast or first guess valid at the analysis time, and use observations within a time window centered on that time to adjust the first guess to the best analysis possible. This becomes the starting place for the current forecast cycle.

Forecast modules

The forecast part of the NWP model consists of two modules. One deals with dynamical equations, the other with physical parameterizations. Dynamical equations forecast processes that are large enough to be analyzed and forecast at the model's resolution, such as shortwave troughs. Physical parameterizations estimate the effect of processes that are smaller than the model can forecast directly, such as convection and atmospheric transmission of long- and shortwave radiation.

Observations from new instruments

The needs of DA systems and NWP help drive new instrument design. Observations from new instruments are quality tested before and after launch. They are then tested in the DA system, and NWP forecasts are run from these test analyses. If the quality of the forecasts improves or is maintained, the new data are accepted for operational use. Otherwise, they will not be used.

All satellite data are continuously monitored, even after they are accepted into operations. Any degradation in data quality is corrected if possible. Otherwise, the data are removed from the DA observational stream. Other satellites with the same or similar instruments may be considered to replace the lost data.

Satellite observations and retrievals

Satellite data are assimilated operationally in two forms, as observations and as retrievals. When retrieval products are used, they are compared directly to the matching first guess data to determine an analysis increment. When observations are used, the first guess data is converted into simulated satellite observations and adjusted to bring it and the actual observations as close to each other as possible. These adjustments are then used as analysis increments. From this point forward, observations and retrievals are handled in the same manner: weighted, buddy checked, and weight adjusted if they're significantly different from neighboring increments. In the last step, the increments from all observational platforms are combined to make the final analysis.

Satellite data are essential for making good forecasts in high-impact situations, and for improving overall model forecast skill. This is especially true in and downstream from data-sparse areas.

Future challenges

To improve satellite data and NWP and DA systems in the future, challenges need to be met in three areas. NWP models and their DA systems need to make better use of existing satellite data, particularly in the areas of clouds and precipitation, explicitly predicted convection, natural and human-made aerosols, and real-time surface conditions such as soil moisture and vegetation greenness. Preparatory work needs to continue so the satellite observations from GOES-R and the next generation of NOAA LEO satellites will be used effectively.

In the area of satellite remote sensing, we need to increase the resolution of microwave sounders, improve the usefulness of sounder data in areas of clouds and precipitation, increase the amount of satellite wind data, and add hyperspectral infrared instruments to geostationary satellites. In addition, we need to address the potential loss of U.S. LEO data if there's a gap between the decommissioning of any current satellites and the launch of next generation satellites.

Main Points

These are the main points to take away regarding how satellite data impact NWP analyses and forecasts.

  • Satellite data is vital for DA systems to produce good analyses for NWP forecasts.
  • Satellite data is tested extensively before being used in operational DA systems.
  • Satellite data is monitored continuously to assure good quality, regardless of whether it’s used in operational DA systems or not.
  • Before satellite data is useable in DA, it has to match data in the NWP model, which depends on the model’s dynamics and physics.
  • Satellite data is vital to the quality of NWP forecasts for individual events and over the long term.
  • Satellite data provides important guidance to forecasters on using NWP in the forecast process.

You have completed the lesson! Please consider taking the quiz and the lesson survey.