Follow this link to skip to the main content
NASA Jet Propulsion Laboratory California Institute of Technology
JPL - Home Page JPL - Earth JPL - Solar System JPL - Stars and Galaxies JPL - Science and Technology
Bring the Universe to You: JPL Email News JPL RSS Feed JPL Podcast JPL Video
MISR - Multi-angle Imaging SpectroRadiometer
  MISR Plume Height Project  
 Get Data
Access Data
MISR Plume Height Project
MISR Plume Height Project 2
MISR Data System
Product Maturity Levels
Local Mode Data
 News and Events
 Ask a Question
 About Us
 Other Resources
Data Quality Statement
MISR Plume Height Project
David Nelson, Cecelia Lawshe, David Diner, Ralph Kahn

MISR Plume Height Project
Data Quality Statement and Error Analysis

Home Page


The Multi-angle Imaging SpectroRadiometer (MISR), aboard NASA's Terra satellite, measures atmospheric and surface properties with nine cameras viewing at different angles. These multiple views allow the height and motion vectors of clouds and smoke plumes to be retrieved using stereoscopic methods.

MISR INteractive eXplorer (MINX) is used to visualize MISR data and to analyze properties of smoke plumes. MINX enables users to interactively digitize smoke plumes and automatically retrieve heights and winds, albedos, aerosol properties, and the radiative power and brightness temperatures of the fires associated with plumes.

MINX is a powerful tool capable of generating statistics on large numbers of smoke plumes which may be useful in climate modeling or in the study of plume dynamics. However, several issues have been identified as potential sources of error and/or bias in the data generated by MINX. These issues do not detract from the value of the data as long as they are taken into account when considering how best to use the data for specific applications.

The goal of this document is to identify the different sources of error and bias inherent in the data and to assess the seriousness of each source when the data are applied to climatological and plume dynamics studies.

The errors and biases have been divided into two categories. The first category addresses errors and biases introduced during the digitizing process when using the MINX tool. The second addresses systemic sources of error and bias not related to digitizing. These two categories are discussed in the following sections where each source of error and bias is discussed and then assessed as of Low, Medium, or High importance for consideration in climatology and plume dynamics studies respectively.

The range of possible errors and biases associated with user error or misuse of data are not covered here. Examples of incorrect usage might be to interpret plume median heights as being measured relative to the terrain (they are measured in meters relative to mean sea level) or to use zero-wind heights as estimates of injection heights (rather than the more appropriate wind-corrected height values). The user is encouraged to carefully read the MINX program documentation to ensure appropriate data use.

Error and bias sources in digitizing process

Issue 1 - Low quality of height retrievals for low optical depth aerosols
The stereo height retrieval technique used in MINX depends on successful identification of features across multiple cameras by performing cross-correlation between pairs of camera images. Features are identified as those elements within a scene having strong textural character. If the dominant texture visible in a scene is attributable to surface terrain, then the correlation process will match terrain features rather than smoke in the atmosphere. In this case, MINX will register zero-wind heights near ground level and wind-corrected heights will be absent.

This problem occurs wherever the optical depth of the smoke is low and features on the ground are easily visible through the smoke. It is more pronounced for cases with higher ground surface reflectance. Consequently, this problem is more noticeable in areas of grassland or thin forest that have high terrain reflectivity.

Dispersed smoke clouds that are not clearly associated with individual smoke plumes or fires are also represented in the online database. These are more likely than the dense cores of plumes to be optically thin, and they play an important role in climatological studies. Height measurements for smoke clouds are strongly biased in favor of dense smoke, and this should be taken into account if you use these data.

Climatology applications can be biased by the absence of low optical depth smoke plumes in the database. Case studies in plume dynamics should not be adversely affected by this issue, since individual plumes will generally be selected for analysis.

In MINX V2.0, there is a remedy for this problem when the optical depth of the smoke is not too low. A height retrieval option that uses blue-band data rather than red-band data has been implemented which provides significantly improved retrieval results in many cases.

Climatology applications: High
Plume dynamics case studies: Low

Issue 2 - Low quality of height retrievals for homogeneous aerosols
If a smoke plume or smoke cloud is texturally homogeneous, then the feature matching process will fail and neither zero-wind heights nor wind-corrected heights will be retrieved. This can be the case even if the smoke is optically dense. Most plumes are heterogeneous enough to enable height retrievals to succeed. Smoke clouds are generally more homogeneous that smoke plumes and sometimes pose a problem during height retrievals, especially when the smoke is far from the source.

Climatology applications: Medium
Plume dynamics case studies: Low

Issue 3 - Avoiding height retrieval contamination by water clouds
When digitizing plumes partially obscured by water clouds, the MINX operator must carefully avoid capturing data points within the clouds. If clouds are inadvertently captured as part of a plume, the height retrieval can be significantly biased, generally toward excessively large heights. The median height values in the summary files for each project may or may not reflect this bias, depending on whether an applied outlier filter has excluded the erroneous points. The raw data files for each plume will include these points. This source of error is highly operator-dependent. It should be noted that this issue is one of the most persuasive in arguing against the use of fully automated plume-detection and analysis methods.

Climatology applications: High
Plume dynamics case studies: Medium

Issue 4 - Digitizing diffuse smoke around plume cores
Smoke plumes in the database have been digitized as much as possible to capture the high-density smoke in plume cores for which MINX is best able to retrieve heights. These cores are typically the highest portions of plumes. Some plumes have margins that are gradational with respect to smoke density, and the less dense, peripheral smoke is usually lower than the cores.

The digitizing process involves a significant amount of interpretation, and digitizing operators must decide what constitutes the plume core and where to place the plume boundary. If more of the peripheral smoke is included within the plume boundary, then the median height reported in the database will usually be lower than if the plume had been digitized more conservatively. This decision-making process leads to an interpretational bias that is probably operator-dependent.

Climatology applications: Medium
Plume dynamics case studies: Low

Issue 5 - Consideration of pyrocumulus clouds associated with plumes
Pyrocumulus clouds constitute a special case of the cloud contamination issue above. Pyrocumulus clouds are frequently seen directly above the fire location and plume. Most often they occur as small equi-dimensional clouds that appear whiter than the smoke below them, but occasionally they are large and towering with a brownish cast. As a matter of policy, the online plume database has been constructed from plumes digitized by excluding pyrocumulus clouds as much as possible.

It is possible that by intentionally excluding pyrocumulus clouds, a significant contribution to smoke injection above the boundary layer is not captured in the database. This may be especially true for cases like the Alaska, 2004 data where, for 3% of plumes, pyrocumulus clouds ascended to above 5500 meters.

Climatology applications: High
Plume dynamics case studies: High

Issue 6 - Incorrect wind direction is digitized
MINX assumes that the motion of particles in a plume is horizontal and in the direction specified by the digitizing operator. If the digitized direction does not correspond to the actual direction of plume motion, then the resulting wind-corrected heights will be incorrect. Greater errors in direction result in greater errors in retrieved heights.

This problem is magnified in cases where there is turbulence within the plume. Vertical particle motion, which the retrieval algorithm assumes is negligible, and local changes in wind direction can produce a large scatter in height and wind values or can prevent MINX from finding a solution.

This is a potentially serious source of error when wind speeds are large and wind direction deviates significantly from the across-track direction (see Issue 7). Turbulence within plumes certainly occurs, and vertical movement translates into errors in determining horizontal velocity.

Climatology applications: Medium
Plume dynamics case studies: Low

Issue 7 - Plume wind direction is along-track
MINX has difficulty determining winds and wind-corrected heights when the wind direction is nearly parallel to the direction of motion of the spacecraft (the along-track direction). This is because the along-track motion of smoke particles due to wind cannot be separated from the along-track parallax caused by spacecraft motion. Heights for plumes that trend within 10-15 degrees of the along-track direction should thus be treated with caution. Even plumes whose direction is 20-30 degrees off the along-track direction may show significantly greater scatter in their results than plumes that trend across-track.

MINX automatically notes degraded digitization quality for plumes subjected to winds within 10 degrees of the along-track direction by decrementing the Data Quality parameter in both the project summary files and in the raw data files.

Climatology applications: High
Plume dynamics case studies: Medium

Other systemic sources of error and bias

Issue 1 - Spatial and temporal bias caused by spacecraft orbit and instrument field of view
MISR, with its 360 km swath width, was never intended to be a monitoring instrument capable of viewing every event worldwide. It has a ground-track repeat cycle of 16 days but obtains repeat coverage in 2 to 9 days, depending on latitude. This results in a significant spatial under-sampling of fires, especially in lower latitudes. Most short duration fire events will be missed completely by MISR.

MISR is in a sun-synchronous, descending near-polar orbit, with an equator crossing time of 10:30 AM. In most regions, fires are largest in size, more numerous in counts and grow most rapidly during the afternoon. This diurnal variation produces a potentially strong temporal bias in the plume height retrievals generated from MISR data.

These orbit and instrument-related considerations constitute perhaps the largest single contribution to biases in the plume database. Use of complementary sensors (e.g., MODIS) can provide indications of fires missed by MISR.

Although MISR provides less coverage than MODIS, it's coverage is significantly greater than space-borne lidar instruments whose swath width is very small (Kahn et al, 2008).

Climatology applications: High
Plume dynamics case studies: Medium

Issue 2 - Temporal bias in selecting periods to analyze for plume database
The online plume database is projected to contain a minimum of eight years of smoke plume results for North America. The years and months of interest are being specifically selected to favor seasons with high fire activity. Typically the months of January, February and December are not analyzed due to low fire activity. The intentional exclusion of low fire-activity months may affect the assessment of climate variability due to smoke.

Climatology applications: Medium
Plume dynamics case studies: Low

Issue 3 - Spatial bias in selecting regions to analyze for plume database
The online plume database is limited to fires in North America plus selected areas and time periods in several other parts of the world. It is not known to what extent existing data are representative of the results that could be obtained from a worldwide database.

Climatology applications: Medium
Plume dynamics case studies: Low

Issue 4 - Fire pixel frequency and power bias in selecting orbits to analyze for plume database
The MISR orbits and blocks to be inspected for each project year are selected in a preprocessing step that filters out orbits that do not satisfy certain constraints. This is done to reduce the volume of data to a manageable amount while attempting to preserve larger plumes and orbits with higher concentrations of plumes.

The constraints are:

  1. Any MISR orbit that has fewer than 7 fire pixels is rejected.
  2. Any MISR block that has fewer than 2 fire pixels is rejected.
  3. Any MODIS fire pixel that has a confidence level less than 20% is rejected.
  4. Any MODIS fire pixel that has power less than 15 MW is rejected.

This has an unknown effect on the database results. Most of the rejected plumes are probably very small. However it is likely that some important plumes are being excluded.

Climatology applications: Medium
Plume dynamics case studies: Low

Issue 5 - Climatological bias in detecting plumes
MISR and other instruments observe smoke plumes under predominantly clear-sky conditions. Fires that occur beneath clouds, which are not observed and captured in the plume database, might exhibit different rates of occurrence, levels of emissions and relationships between fire power and injection height than clear-sky fires.

Climatology applications: Medium
Plume dynamics case studies: Low

Issue 6 - Limitations of the MODIS thermal anomaly product
Although the MODIS MOD14 thermal anomaly product is generally very good at locating hot spots and associated plumes, sometimes plumes that the MINX operator can visually identify with high confidence are not identified by MODIS hot spots. The current procedure is for MINX operators to digitize obvious plumes even if no fire pixels are present. Fortunately, this problem is not common.

More common is the occurrence of a fire beneath a thick water cloud or very dense layer of smoke with the smoke plume emerging some distance away. In many of these cases, the MODIS IR sensors do not detect a thermal anomaly. Even if fire pixels are present, the operator will digitize only the tail of the plume to avoid cloud contamination which results in the exclusion of fire power information.

There is no a priori reason to expect a perfect correlation between MOD14 fire power and plume height. Several other factors may affect the relationship, including atmospheric opacity above the fire, and fire emissivity that is less than unity (e.g., "smoldering" fire). In addition, the instantaneous fire power measured by MODIS may not represent the full dynamic state of the plume, and atmospheric stability structure also plays a role in determining plume height (Kahn et al, 2007).

In all these situations, the radiant energy flux reported by the MOD14 fire power does not bear a simple relationship to plume height. Radiant energy flux is a less direct constraint on plume development models than plume height, but it contains other information that can be used to diagnose the nature of the fire itself. For example, significant correlation was found in various regions of the world between fire radiative power and a rate of smoke emission calculated with certain assumptions (Ichoku & Kaufman, 2005).

Climatology applications: High
Plume dynamics case studies: High

Issue 7 - Stereo heights vs. plume tops and injection heights
Stereo algorithms, such as the one used in MINX, return heights derived from the apparent motion, between different camera images, of texturally-rich features on or near the outer envelope of a smoke plume. These features may be at the crest of the plume, on the side, or even within the plume, if the outer layer of smoke is optically thin. All these raw height measurements are included in the calculation of the plume's median height.

For many plumes, the pixel heights in the smoke plumes are not normally distributed. Instead, there is typically one or more modes at which there are large numbers of pixels and the remaining pixels are distributed asymmetrically around them. Therefore using the median height does not necessarily result in an accurate or representative estimate of plume height.

Plume median heights can be used as a surrogate for injection heights. Although an attempt has been made to filter out outliers and to provide useful numbers for climatological purposes, the resulting values are only one estimate of what injection heights might be. The science user is encouraged to use the raw, wind-corrected plume height values to derive his/her own injection heights.

Other satellite or ground-based instruments (e.g. lidar) that use different height determination methods may derive height results that differ from these. This may make it difficult to independently verify height estimates.

Climatology applications: Medium
Plume dynamics case studies: Medium


Ichoku, C. and Y. J. Kaufman, A method to derive smoke emission rates from MODIS fire radiative energy measurements, IEEE Trans. Geosci. Rem. Sensing, 43(11), 2636-2649, 2005.

Jordan N., C. Ichoku, and R. Hoff, Estimating Smoke Emissions Over The U.S. Southern Great Plains Using MODIS Fire Radiative Power and Aerosol Observations, Atmos. Env., 42, 2007-2022, 2008.

Kahn, R. A., Y. Chen, D. L. Nelson, F.-Y. Leung, Q. Li, D. J. Diner, and J. A. Logan (2008), Wildfire smoke injection heights: Two perspectives from space, Geophys. Res. Lett., 35, L04809, doi:10.1029/2007GL032165.

Kahn, R. A., W.-H. Li, C. Moroney, D. J. Diner, J. V. Martonchik, and E. Fishbein (2007), Aerosol source plume physical characteristics from space-based multiangle imaging, J. Geophys. Res., 112, D11205, doi:10.1029/2006JD007647.