Asteroid spectrum classification using Bus-DeMeo taxonomy

• Stephen M. Slivan
• MIT Dept. of Earth, Atmospheric, and Planetary Sciences
• 77 Massachusetts Avenue, Rm. 54-410
• Cambridge, MA 02139, USA

Implementation of the Bus-DeMeo taxonomy (DeMeo et al., 2009), an asteroid taxonomy with 25 classes based on principal components analysis of combined visible and near-IR spectral data spanning wavelengths from 0.45 to 2.45 microns.

Classify a spectrum

User information

(2013 August 21) What's New in Version 4

The flowchart has been modified from the previous version.

VISIR classification: Improvements have been made for the O-type, B-type, and C, Ch, Xk-type outcomes. A check for the Xn class, a class of neutral spectra with a Nysa-like absorption feature at 0.9 μm has been added in multiple cases. These modifications are highlighted in green on the updated flow chart in Table 1. (PDF file; requires Acrobat Reader.)

Near-IR only classification: Version 4 of the Web tool is unchanged from Version 3.

(2009 June 19) What's New in Version 3

The flowchart has been modified from the previous version. Text has been added to describe the features that distinguish between classes when the Web tool output gives multiple possible classes and visual inspection is needed to determine a non-ambiguous taxonomic assignment to the spectrum.

VISIR classification: Version 3 of the Web tool is fully consistent with the published manuscript DeMeo et al., 2009, including all final changes made in proof.

Near-IR only classification: Minor modifications have been made in test parameters for D-type and for ambiguous outcomes (S-complex) since the publication of DeMeo et al., 2009. These modifications are highlighted in red on the updated flow chart in Table 2. (PDF file; requires Acrobat Reader.)

Spectrum file input format

A spectrum for classification must span either one of the following two wavelength ranges:

• The full wavelength classification algorithm requires a spectrum with combined visual and near-IR data (“VISIR”) that spans wavelengths from 0.45 to 2.45 microns.
• A spectrum with the near-IR wavelengths from 0.85 to 2.45 microns only (“IR”) can be submitted for a best estimate of where it falls within the full wavelength classification.

Spectrum data are input from a text file uploaded by the user, in which each line of spectrum data contains a (wavelength, reflectance) value pair, with whitespace separating the two value fields. Wavelengths and reflectances must be positive numerical values; data lines with negative or non-numerical values in either of the first two fields are ignored. An optional third field with a positive reflectance uncertainty value may be included; otherwise, additional fields following the reflectance are ignored. Blank lines, and any other lines which don't begin with a valid (wavelength, reflectance) value pair, are ignored.

Wavelength units can be microns, nanometers, or Ångströms. Units information is specified by the user separately on the input page.

If classifying a spectrum causes the error message Unable to sample spectrum; need wavelengths from --- to --- microns, inclusive,

• Check your data file to confirm that data points completely span the indicated wavelength range.
• In some cases, you may need to make sure that the last line of the file ends with a newline character so that the last datum is read.
• If the available data don't reach all the way to the boundary wavelengths needed, then a possible workaround is to use short extrapolations to estimate spectrum reflectance values at the boundary wavelength(s), assign them large error bars, and add them to the input file. To gauge confidence in the resulting taxonomic classification, check whether it's insensitive to any plausible variation of the extrapolated data value(s). When reporting such a result, also report the extent to which any extrapolation was necessary in order to use this classification tool.

Spectrum file input format if previously smoothed and sampled

If the user submits a previously-smoothed spectrum that has also already been sampled and normalized for classification, then the user may choose to classify that spectrum without any additional smoothing steps. In this case the (previously smoothed) spectrum must already be sampled across the expected wavelength range as follows:

• A VISIR input spectrum must already be normalized to unit reflectance at 0.55 micron (equivalently, 550 nm or 5500 Å). The classification algorithm requires that the spectrum consist of reflectances for the 41 wavelengths from 0.45 micron to 2.45 microns at 0.05 micron intervals, except that the normalized point (0.55,1.00) may be omitted, leaving 40 (wavelength, reflectance) pairs.
• A near-IR only (“IR”) input spectrum must already be normalized to unit reflectance at 1.20 microns. The classification algorithm requires that the spectrum consist of reflectances for the 33 wavelengths from 0.85 micron to 2.45 microns at 0.05 micron intervals, except that the normalized point (1.20,1.00) may be omitted, leaving 32 (wavelength, reflectance) pairs.

Here is a sample smoothed, sampled spectrum file with wavelengths in microns:

Smoothed and sampled input data for (349) Dembowska.
Wavelengths are microns.
These data have NOT been corrected for slope.

      0.45     0.815
      0.50     0.904
      0.55     1.000
      0.60     1.072
      0.65     1.149
      0.70     1.214
      0.75     1.212
      0.80     1.102
      0.85     0.978
      0.90     0.907
      0.95     0.900
      1.00     0.954
      1.05     1.028
      1.10     1.109
      1.15     1.185
      1.20     1.239
      1.25     1.287
      1.30     1.346
      1.35     1.408
      1.40     1.455
      1.45     1.496
      1.50     1.515
      1.55     1.506
      1.60     1.471
      1.65     1.430
      1.70     1.383
      1.75     1.341
      1.80     1.316
      1.85     1.306
      1.90     1.305
      1.95     1.310
      2.00     1.326
      2.05     1.363
      2.10     1.412
      2.15     1.463
      2.20     1.506
      2.25     1.537
      2.30     1.565
      2.35     1.593
      2.40     1.621
      2.45     1.645
   

The normalized reflectance (at 0.55 micron for this VISIR spectrum) could be omitted, in which case the first few data lines would be

Example of omitting unity reflectance at 0.55 micron
      0.45     0.815
      0.50     0.904
      0.60     1.072
      0.65     1.149
      0.70     1.214
            .
            .
            .
   

Finally, if the file wavelengths were nanometers instead of microns then the above file fragment would be

Example of wavelengths in nanometers
     450 0.815
     500 0.904
     600 1.072
     650 1.149
     700 1.214
            .
            .
            .
   

Spectrum smoothing parameter

A cubic spline model (Reinsch, 1967) is used to smooth an uploaded spectrum prior to classification, unless the user specifies that the spectrum has already been smoothed and sampled.

The “smoothing parameter” is a non-negative floating-point value which controls the amount of smoothing. Setting the smoothing parameter to zero will turn off the smoothing; in other words, the output will be equal to the input. Setting this parameter to a positive value controls smoothing by “limiting the sum of squares of differences between calculated and actual reflectance values divided by the uncertainty in reflectance.” An appropriate value for the smoothing parameter therefore depends on the number of data points and their uncertainties, and must be determined empirically by the user for a given spectrum.

Calculation of spectrum slope

Here is the algorithm used for calculating and removing spectrum slope from a smoothed, sampled VISIR spectrum:

First, both slope and intercept of a straight line are fitted to the spectrum data points. In general this best-fit line will not pass through the VISIR normalization reference point (0.55,1.00). The line is translated in the y-direction to pass through (0.55,1.00), and the spectrum is divided by the translated line.

For IR-only classification the steps are the same but use a normalization reference wavelength of 1.20 instead of 0.55.

Overall slope removed from input spectrum

A spectrum from which the overall slope has already been removed can be classified by also entering the slope value. The application will restore the slope to the data and then apply the classification algorithm in the usual way.

To correct the error condition Unable to parse slope "---" as floating-point, either enter a valid floating-point slope value in the text box provided, or select “No” in answer to the question “Reflectances already divided by slope?”.

Here is a sample input file, obtained by removing the best-fit slope from the example spectrum above. The removed slope value of 0.3164 will need to be entered in the input page form by the user.

/* Smoothed and sampled input data for (349) Dembowska.
   Wavelengths in microns.
   Slope 0.3164 has been removed. */

0.45  0.8416
0.50  0.9185
0.55  1.0000
0.60  1.0553
0.65  1.1138
0.70  1.1590
0.75  1.1399
0.80  1.0212
0.85  0.8932
0.90  0.8166
0.95  0.7989
1.00  0.8351
1.05  0.8876
1.10  0.9446
1.15  0.9959
1.20  1.0277
1.25  1.0537
1.30  1.0879
1.35  1.1236
1.40  1.1467
1.45  1.1644
1.50  1.1649
1.55  1.1441
1.60  1.1042
1.65  1.0608
1.70  1.0141
1.75  0.9720
1.80  0.9431
1.85  0.9254
1.90  0.9144
1.95  0.9079
2.00  0.9090
2.05  0.9243
2.10  0.9474
2.15  0.9713
2.20  0.9895
2.25  0.9995
2.30  1.0073
2.35  1.0150
2.40  1.0225
2.45  1.0274
   

E-mail smoothed spectrum and results

The user may choose to e-mail the smoothed spectrum and classification results to a specified e-mail address. If this option is chosen then the following information is sent in a plain text message:

1. name of the uploaded file of spectrum data
2. value of smoothing parameter used for smoothing step prior to classification (0 for no smoothing)
3. determined taxonomic class
4. values of overall slope and the five principal components PC1' through PC5'
5. number of points in model spectrum, and wavelength units used
6. sampled spectrum as (wavelength,reflectance) value pairs, normalized to unit reflectance at 0.55 micron (for VISIR) or 1.20 microns (for IR)

The first line is a one-line summary with the file name, the slope, the principal component values, and the taxonomic class, each separated by spaces. The rest of the message is in a two-column format as shown below. In this example the “SMOOTHING 0.00” line confirms that no additional smoothing was performed on the spectrum before classification.

   SUMMARY    spectrumfile.txt   0.3164  -0.1999   0.4292   0.1205  -0.0623  -0.0098  R-type
   INPUTFILE  spectrumfile.txt
   SMOOTHING  0.00
   BDCLASS    R-type
   SLOPE       0.3164
   PC1'       -0.1999
   PC2'        0.4292
   PC3'        0.1205
   PC4'       -0.0623
   PC5'       -0.0098
   NUMLAMBDA  41
   LAMBDAUNIT micron
   BEGIN      SPECTRUM
       0.45   0.815
       0.50   0.904
       0.55   1.000
       0.60   1.072
       0.65   1.149
       0.70   1.214
            .
            .
            .
       2.40   1.621
       2.45   1.645
   END        SPECTRUM
   

Classification result: visual inspection needed: either L-type or Xe-type

A non-ambiguous taxonomic assignment should be possible by following the steps described here. An additional check is needed to distinguish L-type from Xe-type: Is there evidence of an absorption feature at 0.49 μm; that is, a concave-up curvature shortward of 0.55 μm? This feature was identified in the taxonomy defined by Bus (1999, pp. 122-124), who notes that

... recognizing its presence usually requires the fitting of a smoothing function, such as a spline, to the data. ... In many cases, only by examining the fitted (smoothed) spectrum can these features be properly identified.

If the feature is present then the classification is X-complex, Xe-type; otherwise, the classification is L-type.

Classification result: visual inspection needed: either K-type or Xe-type

A non-ambiguous taxonomic assignment should be possible by following the steps described here. An additional check is needed to distinguish K-type from Xe-type: Is there evidence of an absorption feature at 0.49 μm; that is, a concave-up curvature shortward of 0.55 μm? This feature was identified in the taxonomy defined by Bus (1999, pp. 122-124), who notes that

... recognizing its presence usually requires the fitting of a smoothing function, such as a spline, to the data. ... In many cases, only by examining the fitted (smoothed) spectrum can these features be properly identified.

If the feature is present then the classification is X-complex, Xe-type; otherwise, the classification is K-type.

Classification result: visual inspection needed: either X-type or Xk-type or Xe-type or C-type or Xn-type

A non-ambiguous taxonomic assignment should be possible by following the steps described here. Final classification of this object is based on the presence of small-scale spectral features that are best recognized by visual inspection. There are four features whose presence or absence distinguish among X-, Xk-, Xe-, C-, and Xn-types. These features are called the “0.49 μm feature”, the “0.8 to 1.0 μm feature”, the “0.9 μm feature”, and the “1.0 to 1.3 μm feature”. Each is described below:

“0.49 μm feature”: This is an absorption feature shortward of 0.55 μm. It exhibits a concave-up curvature, where the band center is located at about 0.49 μm. (Bus 1999, pp. 122-124) (Bus and Binzel 2002, Fig. 13 & Table II)

“0.8 to 1.0 μm feature”: This is a shallow absorption feature that appears as a concavity over the range 0.8 to 1.0 μm, in what is otherwise a generally linear spectrum.

“0.9 μm feature”: This is a weak to moderate narrow feature centered at 0.9 μm, similar to the spectrum of (44) Nysa.

“1.0 to 1.3 μm feature”: This is a broad and shallow absorption feature centered between 1.0 and 1.3 μm.

Perform these steps in order, until a classification is made.

1. If the “0.49 μm feature” is present then the classification is X-complex, Xe-type.
2. If the “0.8 to 1.0 μm feature” is present then the classification is X-complex, Xk-type.
3. If the “0.9 μm feature” is present then the classification is X-complex, Xn-type.
4. If the “1.0 to 1.3 μm feature” is present then the classification is C-Complex, C-type.
5. Otherwise, the classification is X-complex, X-type.

Classification result: visual inspection needed: either Cgh-type or Cg-type or Xk-type or Xn-type

A non-ambiguous taxonomic assignment should be possible by following the steps described here. Final classification of this object is based on the presence of small-scale spectral features that are best recognized by visual inspection. There are three features whose presence or absence distinguish among Cgh-, Cg-, Xk-, and Xn-types. These features are called the “0.7 μm feature”, the “0.8 to 1.0 μm feature”, and the “0.9 μm feature”. Each is described below:

“0.7 μm feature”: This is a moderately shallow absorption feature that appears as a concavity over the range of 0.6 to 0.8 μm, centered near 0.7 to 0.8 μm. (Bus 1999, pp. 115 & 119) (Bus and Binzel 2002, Fig. 10 & Table II)

“0.8 to 1.0 μm feature”: This is a shallow absorption feature that appears as a concavity over the range 0.8 to 1.0 μm, in what is otherwise a generally linear spectrum.

“0.9 μm feature”: This is a weak to moderate narrow feature centered at 0.9 μm, similar to the spectrum of (44) Nysa.

Perform these steps in order, until a classification is made.

1. If the “0.7 μm feature” is present then the classification is C-complex, Cgh-type.
2. If the “0.8 to 1.0 μm feature” is present then the classification is X-complex, Xk-type.
3. If the “0.9 μm feature” is present then the classification is X-complex, Xn-type.
4. Otherwise, the classification is C-complex, Cg-type.

Classification result: visual inspection needed: either Xk-type or Xc-type or Xe-type or C-type or Ch-type or Xn-type

A non-ambiguous taxonomic assignment should be possible by following the steps described here. Final classification of this object is based on the presence of small-scale spectral features that are best recognized by visual inspection. There are five features whose presence or absence distinguish among Xk-, Xc-, Xe, C-, Cgh-, and Xn-types. These features are called the “0.49 μm feature”, the “0.8 to 1.0 μm feature”, the “0.9 μm feature”, the “1.0 to 1.3 μm feature”, and the “0.7 μm feature”. Each is described below:

“0.49 μm feature”: This is an absorption feature shortward of 0.55 μm. It exhibits a concave-up curvature, where the band center is located at about 0.49 μm. (Bus 1999, pp. 122-124) (Bus and Binzel 2002, Fig. 13 & Table II)

“0.8 to 1.0 μm feature”: This is a shallow absorption feature that appears as a concavity over the range 0.8 to 1.0 μm, in what is otherwise a generally linear spectrum.

“0.9 μm feature”: This is a weak to moderate narrow feature centered at 0.9 μm, similar to the spectrum of (44) Nysa.

“1.0 to 1.3 μm feature”: This is a broad and shallow absorption feature centered between 1.0 and 1.3 μm.

“0.7 μm feature”: This is a moderately shallow absorption feature that appears as a concavity over the range of 0.6 to 0.8 μm, centered near 0.7 to 0.8 μm. (Bus 1999, pp. 115 & 119) (Bus and Binzel 2002, Fig. 10 & Table II)

Perform these steps in order, until a classification is made.

1. If the “0.49 μm feature” is present then the classification is X-complex, Xe-type.
2. If the “0.8 to 1.0 μm feature” is present then the classification is X-complex, Xk-type.
3. If the “0.9 μm feature” is present then the classification is X-complex, Xn-type.
4. If the “1.0 to 1.3 μm feature” is present then the classification is C-Complex, C-type.
5. If the “0.7 μm feature” is present then the classification is C-complex, Ch-type.
6. Otherwise, the classification is X-complex, Xc-type.

Classification result: visual inspection needed: either D-type or A-type

A non-ambiguous taxonomic assignment should be possible by following the steps described here. Final classification of this object is based on the presence of a spectral feature that is best recognized by visual inspection. The distinguishing feature is called the “1 μm feature” and is described as:

“1 μm feature”: This is an extremely wide, extremely deep absorption band centered around 1 μm.

If the “1 μm feature” is present then the classification is A-type. If the “1 μm feature” is not present then the classification is D-type.

Classification result: visual inspection needed: either Ch-type or Xk-type or Xn-type

A non-ambiguous taxonomic assignment should be possible by following the steps described here. Final classification of this object is based on the presence of small-scale spectral features that are best recognized by visual inspection. There are three features whose presence or absence distinguish among Ch-, Xk, and Xn-types. These features are called the “0.7 μm feature”, the “0.8 to 1.0 μm feature”, and the “0.9 μm feature”. Each is described below:

“0.7 μm feature”: This is a moderately shallow absorption feature that appears as a concavity over the range of 0.6 to 0.8 μm, centered near 0.7 to 0.8 μm. (Bus 1999, pp. 115 & 119) (Bus and Binzel 2002, Fig. 10 & Table II)

“0.8 to 1.0 μm feature”: This is a shallow absorption feature that appears as a concavity over the range 0.8 to 1.0 μm, in what is otherwise a generally linear spectrum.

“0.9 μm feature”: This is a weak to moderate narrow feature centered at 0.9 μm, similar to the spectrum of (44) Nysa.

Perform these steps in order, until a classification is made.

1. If the “0.7 μm feature” is present then the classification is C-complex, Ch-type.
2. If the “0.8 to 1.0 μm feature” is present then the classification is X-complex, Xk-type.
3. If the “0.9 μm feature” is present then the classification is X-complex, Xn-type.

Classification result: visual inspection needed: either Cgh-type or Xk-type or Xn-type

A non-ambiguous taxonomic assignment should be possible by following the steps described here. Final classification of this object is based on the presence of small-scale spectral features that are best recognized by visual inspection. There are three features whose presence or absence distinguish among Cgh-, Xk, and Xn-types. These features are called the “0.7 μm feature”, the “0.8 to 1.0 μm feature”, and the “0.9 μm feature”. Each is described below:

“0.7 μm feature”: This is a moderately shallow absorption feature that appears as a concavity over the range of 0.6 to 0.8 μm, centered near 0.7 to 0.8 μm. (Bus 1999, pp. 115 & 119) (Bus and Binzel 2002, Fig. 10 & Table II)

“0.8 to 1.0 μm feature”: This is a shallow absorption feature that appears as a concavity over the range 0.8 to 1.0 μm, in what is otherwise a generally linear spectrum.

“0.9 μm feature”: This is a weak to moderate narrow feature centered at 0.9 μm, similar to the spectrum of (44) Nysa.

Perform these steps in order, until a classification is made.

1. If the “0.7 μm feature” is present then the classification is C-complex, Cgh-type.
2. If the “0.8 to 1.0 μm feature” is present then the classification is X-complex, Xk-type.
3. If the “0.9 μm feature” is present then the classification is X-complex, Xn-type.

Classification result: visual inspection needed: either C-type or Ch-type or Xk-type or Xn-type

A non-ambiguous taxonomic assignment should be possible by following the steps described here. Final classification of this object is based on the presence of small-scale spectral features that are best recognized by visual inspection. There are four features whose presence or absence distinguish among C-, Ch-, Xk-, and Xn-types. These features are called the “0.7 μm feature”, the “0.8 to 1.0 μm feature”, the “0.9 μm feature”, and the “1.0 to 1.3 μm feature”. Each is described below:

“0.7 μm feature”: This is a moderately shallow absorption feature that appears as a concavity over the range of 0.6 to 0.8 μm, centered near 0.7 to 0.8 μm. (Bus 1999, pp. 115 & 119) (Bus and Binzel 2002, Fig. 10 & Table II)

“0.8 to 1.0 μm feature”: This is a shallow absorption feature that appears as a concavity over the range 0.8 to 1.0 μm, in what is otherwise a generally linear spectrum.

“0.9 μm feature”: This is a weak to moderate narrow feature centered at 0.9 μm, similar to the spectrum of (44) Nysa.

“1.0 to 1.3 μm feature”: This is a broad and shallow absorption feature centered between 1.0 and 1.3 μm.

Perform these steps in order, until a classification is made.

1. If the “0.7 μm feature” is present then the classification is C-complex, Ch-type.
2. If the “0.8 to 1.0 μm feature” is present then the classification is X-complex, Xk-type.
3. If the “0.9 μm feature” is present then the classification is X-complex, Xn-type.
4. If the “1.0 to 1.3 μm feature” is present then the classification is C-Complex, C-type.
5. Otherwise, the classification is C-complex, C-type.

Classification result: visual inspection needed: either O-type or Q-type

A non-ambiguous taxonomic assignment should be possible based on the best match found by comparing the distinct characteristics of the O- and Q-types:

O-type: Very rounded and deep, “bowl” shape absorption feature at 1 μm as well as a significant absorption feature at 2 μm such as seen for (3628) Božněmcová or (7472) Kumakiri.

Q-type: Distinct 1-μm absorption feature with evidence of another feature near 1.3 μm; a 2-μm feature exists with varying depths between objects.

Classification result: non-unique outcomes for IR-only input data

The Bus-DeMeo taxonomy classes are formally defined by data spanning the entire wavelength range 0.45 to 2.45 microns. For many classes, IR-only data from 0.85 to 2.45 microns do not yield a unique outcome in Principal Component Analysis (PCA) and the object cannot formally be classified. The caption and the corresponding plots with residuals for each possible type may allow the user to make an intelligent selection from the ambiguous outcome of PCA. The possible types are ranked in order of their prevalence within the data set defining this taxonomy, but this order may not be representative of the user's sample.

For any taxonomic type assigned and reported using this described inspection method, it is recommended that the type name be written with a colon (:) to indicate that there is a remaining ambiguity in the classification. For example, the resulting type may be written as “X:”, “C:”, “Xk:”, “L:”, etc.

Acknowledgment for Publications

Users of this Web tool are encouraged to reference the relevant publications pertaining to the Bus-DeMeo Taxonomy. Users are also kindly asked to add the following text to the Acknowledgment section of any refereed publication:

Taxonomic type results presented in this work were determined, in whole or in part, using a Bus-DeMeo Taxonomy Classification Web tool by Stephen M. Slivan, developed at MIT with the support of National Science Foundation Grant 0506716 and NASA Grant NAG5-12355.

Privacy information

Spectrum data submitted to the site for classification are not archived and do not appear in the server's access log. Nevertheless, users who are concerned about maintaining maximum privacy of data that they are considering submitting for classification, need to be aware that temporary disk files containing submitted spectrum data are created on the server for use during the classification process. These disk cache files are deleted when the user clicks the “Empty disk cache >” button after a classification. However, if the user uses their browser's “Back” button after a classification instead, or if a session is interrupted by a browser crash, files with spectrum data might remain intact on the server disk for as long as 24 hours before they are deleted in a daily sweep. Similarly, if a server error occurs during classification then the attempt that is automatically made to delete the disk cache files might not succeed depending on the error, again possibly leaving files intact until the next daily sweep.

Because of the nature of the IP and HTTP protocols, when you access a Web page on our server the IP address of the client (remote host) from which the request appears to have been sent is captured automatically by the server's access log. We do not link IP addresses to personally identifiable information.

Credits

• This application makes use of the CGIC Library, an ANSI C-language library for the creation of CGI-based World Wide Web applications.

  CGIC License Terms
  
  Basic License
  
  CGIC, copyright 1996, 1997, 1998, 1999, 2000, 2001, 2002 by Thomas
  Boutell and Boutell.Com, Inc.. Permission is granted to use CGIC in
  any application, commercial or noncommercial, at no cost. HOWEVER,
  this copyright paragraph must appear on a "credits" page accessible in
  the public online and offline documentation of the program.  Modified
  versions of the CGIC library should not be distributed without the
  attachment of a clear statement regarding the author of the
  modifications, and this notice may in no case be removed.
  Modifications may also be submitted to the author for inclusion in the
  main CGIC distribution.
  

• This application makes use of the GD Library, an open source code library for the dynamic creation of images by programmers.

Classify a spectrum

Page footer:

• 
• Last modified 2019 Dec 14
• Copyright (c) 2017 Massachusetts Institute of Technology. All rights reserved.