Web Supplement
Geomorphometry from SRTM: Comparison to NED
Peter L. Guth
Photogrammetric Engineering & Remote Sensing, Mar 2006, 72(3): p.269-277.
Click figure for larger version.
Figure 3. Map showing the
distribution of Elevation-Relief ratio (ELEV_RELF) over the
 |
 |
 |
 |
Click figures for larger version.
Figure 6. Correlation coefficients of the best fit line for the 4 SRTM data sets compared to NED. (A) shows 1 NED and all data in the United States, (B) shows 1 NED and only areas with average slope greater than 5%, (C) shows 2NED and all data in the United States, and (D) shows 2 NED and areas with average slope greater than 5%. Decimation/thinning of 1 NED created the 2 NED. Note than restricting the analysis to slopes above 5%, or comparing SRTM to 2 NED, improves the correlations. These graphs show all 33 parameters listed in Appendix A below; the version in the paper only shows the 12 parameters deemed most important. Note that some of these parameters on the bottom of each figure never show a correlation coefficient above 0.2.
 |
 |
 |
|
Click figures for larger version.
Figure 7. Slopes of the best fit line for the 4 SRTM data sets compared to NED. (A) shows 1 NED and all data in the United States, (B) shows 1 NED and only areas with average slope greater than 5%, (C) shows 2NED and all data in the United States, and (D) shows 2 NED and areas with average slope greater than 5%. Decimation/thinning of 1 NED created the 2 NED. With the exception of the PLANC_STD parameter for the 2 NED, the slopes are everywhere less than 1, meaning that SRTM always computes smaller values of the statistics than NED. These graphs show all 33 parameters listed in Appendix A below; the version in the paper only shows the 12 parameters deemed most important.
The 12 parameters in bold are those discussed in the printed version of the paper. The others are included in Figures 6 and 7 here.
ELEV_AVG, ELEV_STD, ELEV_SKW, ELEV_KRT: the first four moments of the elevation distribution, computed with the formulas in Press et al. (1986). ELEV_STD correlates strongly with slope.
SLOPE1_AVG, SLOPE1_STD, SLOPE1_SKW, SLOPE1_KRT: the first four moments of the slope distribution in percent (100*rise/run), computed with the formulas in Press et al. (1986). Slopes were computed with an eight neighbors unweighted algorithm (Evans, 1998; Florinsky, 1998; Sharpnack and Akin, 1969). The algorithm has little effect on the results; Guth (1995) showed extremely high correlations between all available slope algorithms.
SLOPE4_AVG, SLOPE4_STD, SLOPE4_SKW, SLOPE4_KRT: the slope measures computed in degrees. Slopes were computed with an eight neighbors unweighted algorithm (Evans, 1998; Florinsky, 1998; Sharpnack and Akin, 1969). The algorithm has little effect on the results; Guth (1995) showed extremely high correlations between all available slope algorithms. Because of the non-linear relationship between slope in degrees and percent, the average slope computed in degrees does not equal the average slope in percent converted to degrees with the arc tangent function, but the two units produce very similar results.
PLANC_AVG, PLANC _STD, PLANC _SKW, PLANC _KRT: the first four moments of the plan curvature distribution, computed with the formulas in Press et al. (1986). Curvature computed with the equations in Wood (1996) based on earlier suggestions from Evans.
PROFC_AVG, PROFC _STD, PROFC _SKW, PROFC _KRT: the first four moments of the profile curvature distribution, computed with the formulas in Press et al. (1986). Curvature computed with the equations in Wood (1996).
S1S2, S2S3, FABRIC_DIR: Computed after Guth (2003, following Chapman, 1952 and Woodcock, 1977) using logs of the eigenvectors of the normal vector distribution. S1S2 measures flatness (a logarithmic inverse of slope), S2S3 measures terrain organization, and FABRIC_DIR gives the dominant direction of ridges and valley. Because FABRIC_DIR measures circular angles, its statistics have anomalies.
SHAPE, STRENGTH: Fisher et al. (1987, pp.48, 159) defined these ratios of the logs of the eigenvectors, defined somewhat differently that those used by Woodcock (1977) and Guth (2003).
ELEV_RELF: the elevation relief ratio ( [AveZ-MinZ] / [MaxZ MinZ]) is computed for a region (Pike and Wilson, 1971; Etzelmuller, 2000) and is equivalent to the coefficient of dissection (Klinkenberg and Goodchild, 1992, after Strahler, 1952).
MAX_SLOPE: the largest slope (percent) in the sampling region. While this is largely of value for detecting blunders during DEM creation, it also has geomorphic significance.
GAMMA_NS, GAMMA_EW, GAMMA_NESW, GAMMA_NWSE: Nugget variance, Co, from the variogram (Curran, 1988). This is a measure of the elevation difference from each point to its nearest neighbor in four directions; smaller values reflect smooth terrain, and high values rougher terrain.
ROUGH_FAC: Measure correlating strongly with slope (Mark, 1975; Etzelmuller, 2000).
RELIEF: difference between the highest and lowest elevations within the sampling region (Drummond and Dennis, 1968).
Additional references for Appendix A.
Curran, P.J., 1988. The semivariogram in remote sensing: an introduction: Remote Sensing of Environment, 24:493-507.
Drummond, R.R., and Dennis, H.W., 1968. Qualifying relief terms: Professional Geographer,20(5):326-332.
Fisher, N.L, Lewis, T., and Embleton, B.J.J., 1987. Statistical analysis of spherical data: Cambridge University Press, 330 p.
Mark, D.M., 1975. Geomorphometric parameters: A review and evaluation: Geografiska Annaler, 57A(3-4):165-177.