Natural-Color Image Mosaics of Afghanistan: Digital Databases and Maps

The 50 images in this published set are natural color renditions (Landsat bands 3, 2, 1) of the calibrated 6-band Landsat 7 ETM+ mosaics.  We applied the hybrid MODIS-Munsell natural-color transform to the Landsat 28.5-m image data that were produced as 1º x 2º map-tile mosaics by Davis (2006).

The purpose of the dataset is to form a functional and consistent digital natural color image base to support the USGS and USAID natural resource assessments for the country of Afghanistan.

Reference: Davis, P. A., 2006, Calibrated Landsat ETM+ nonthermal-band image mosaics of Afghanistan.  U.S. Geological Survey Open File Report 2006-1345, 18 p.

Complete
Data update frequency: As needed

Who created the data: Davis, P.A., Hare, T.M.

File name: Q2960_28m_GEOG_NC.tif
Type of data: remote-sensing image
Data processing environment: Microsoft Windows XP Version 5.1 (Build 2600) Service Pack 2; ESRI ArcCatalog 9.1.0.722

Contents last updated: 20070215 at time 16111600

Process description: The calibration of the Landsat ETM+ maps produced by Davis (2006) are relative reflectance and need to be "grounded" with ground reflectance data, but the difficulties in performing field work in Afghanistan precluded ground reflectance surveys. For natural color calibration, which only involves the blue, green, and red wavelength bands of Landsat, we could use ground photographs, Munsell color readings of ground surfaces, or another image base that accurately depicts the surface color. Ground photography is the least reliable of these three approaches. On four separate field trips to the north-central part of Afghanistan, we acquired ground photographs and Munsell color readings for a variety of ground surfaces, but the field areas did not include the wide range of surface colors found in Afghanistan, especially the desert areas. Therefore, we had to supplement our ground color-sample data with a less direct source of color information.
Process software and version: ENVI 4.x and ArcMap 9.x
Source used: Davis, P. A., 2006, Calibrated Landsat ETM+ nonthermal-band image mosaics of Afghanistan. U.S. Geological Survey Open File Report 2006-1345, 18 p.
Source used: Kalluri, S., Grant, D., Tucker, C., Policelli, F., Dykstra, J., and Bearden, P., 2000, NASA creates global archive of ortho-rectified Landsat data. EOS, Transactions, American Geophysical Union, v. 81, no. 50, p. 609, 617-618.
Source produced: Escadafal, R., Girard, M.-C., and Courault, D., 1989, Munsell soil color and soil reflectance in the visible spectral bands of Landsat MSS and TM data. Remote Sensing of Environment, v. 27, p. 37-46.
Source produced: Goddard, E. N., Trask, P. D., De Ford, R. K., Rove, O. N., Singewald, J. T., and Overbeck, R. M., 1980, GSA Rock Color Chart, Munsell Corporation, 115 color chips,16 p.
Source produced: Munsell Color, 2000, Munsell Soil Color Chart, Baltimore, Maryland, 322 color chips, 22 p.
Process date: 2005-2007

Process description: The MODIS (Moderate-Resolution Imaging Spectroradiometer) satellite collects 36 spectral images at relatively low resolution (250 m to 1 km per picture element), but the instrument's spectral bands allow accurate atmospheric corrections. The MODIS science team has provided ground-calibration sites that allow the sensor data to be calibrated to ground reflectance. The MODIS science team has prepared "true color" image mosaics for many countries in the world, including Afghanistan, and these are posted on www.truecolorearth.com website by WeatherStreet.com. We obtained the "true color" image mosaic for Afghanistan from this website and registered the image to our Landsat image mosaic so we could determine the correspondence of the blue, green, and red band digital values in the Landsat mosaic with their respective color values in the MODIS "true color" mosaic. The relation between Landsat band value and MODIS color value is best modeled with a third-order polynomial, which we have also observed in "true color" calibrations for IKONOS satellite data. Previous studies have found a linear relation between surface reflectance values of Landsat visible bands and Munsell red, green, and blue values for various surfaces (Escadafal et al., 1989), but the surface reflectance values used in that study did not exceed 40 percent (a digital value of 102 in our Landsat dataset), which is near the change from the linear trend at lower reflectance to a third-order polynomial at higher reflectance. In addition, closer examination of the data collected by Escadafal et al. (1989) shows that their reflectance-Munsell "linear" relation becomes less distinct with increasing reflectance and may actually be better modeled by a second- or third-order polynomial.

Process description: Our field studies were conducted after our Landsat-to-MODIS calibration was performed. During our field work, we recorded Munsell colors (Munsell Color, 1975; Goddard et al., 1995) for more than 100 rock and soil surfaces. The field work was not intended for this purpose, nor could field work be conducted in many parts of Afghanistan, so our Munsell color samples do not cover the wide range of colors that exist within Afghanistan. We extracted the blue, green, and red digital values from the MODIS-calibrated Landsat data for the sites that we sampled in the field. We converted the sample Munsell values to blue, green, and red values using a Munsell conversion program (http://colorpro.com/info/software/munsell.htm). The relations between the MODIS-calibrated Landsat blue, green, and red image values and those values derived from the Munsell samples. The red and green values correspond quite well, but the blue values in the MODIS-calibrated Landsat data are lower than that of the ground-surface Munsell values. We performed a linear least-squares regression analysis on the set of blue values and applied the resulting equation to the Landsat blue-band data and displayed it with the MODIS-calibrated Landsat green- and red-band data. The resulting color-composite image had too much red in many areas of Afghanistan. Therefore, we applied a transformation that is halfway between the MODIS calibration and the Munsell calibration, which produced much better results. This hybrid calibration replaced the subtle green tint in the MODIS-calibrated images of the alluvium and deserts with more realistic shades of gray, yellow, and red and its colors also more closely correspond to the colors of the rock outcrops that we sampled and photographed during our field studies in the Khamard district of the Bamian Province.

Process description: We applied the hybrid MODIS-Munsell natural-color transform to the Landsat 28.5-m and 14.25-m image data that were produced as 1º x 2º map-tile mosaics in UTM projection by Davis (2006). The resulting natural-color map tiles were then reprojected to the WGS84 geographic reference system using a second-order polynomial transformation and over 5,000 grid control points for the 28.5-m data and over 20,000 grid control points for the 14.25-m data in order to retain the 50-m positional accuracy of each dataset. Each map tile overlaps adjacent tiles by at least 100 picture elements. The digital map files begin with the map tile name, which starts with the letter "Q" (for quadrangle) and four digits that designate the lower-left corner latitude and longitude coordinates of the 1º x 2º map tile (as shown in Figure 5). The map tile index is followed by the map's resolution and coordinate system (UTM for Universal Transverse Mercator or GEOG for WGS84 geographic coordinates) and the designation "NC" for natural color. The following shows the two filename examples for map tile Q3460 from this Data Series publication. Q3460_14m_UTM_NC.tif Q3460_28m_GEOG_NC.tif All of the 28.5-m and 14.25-m map tiles for Afghanistan (in both UTM projection and WGS84 geographic reference system) are stored as lossless-compressed, embedded geotiff files (with associated world files). The files can be downloaded from http://gisdata.usgs.net/website/afghan. Only the 14.25-m UTM and 28.5-m WGS84 geographic Geotiff data sets are available on this Data Series publication to decrease the number of DVDs needed.

Media: CD-ROM
Scale denominator: 28.5-m spatial resolution
Contribution: USGS and NASA LandSat Program

The map tiles retained the original 28.5-m spatial resolution and were constructed from orthorectified Landsat scenes that have a positional accuracy of 50 meters.