Geomorphometry of Lut Mega-Yardangs

Introduction
Yardangs due to intensive wind erosion are exclusive landforms on the earth's desert and possibly occur on Mars and Venus. The recent advances in the remote sensing technique and easily available of high resolution satellite data e.g. QuickBird provide useful information of remote area. Hence the mega-yardangs with tens meter high and hundred meters long are easily identifiable on satellite images and their global distribution and properties can be mapped.
The Lut desert (Dasht-e Lut) in the south east of Iran is described as the “thermal pole of the Earth” (Mildrexler et al., 2006). With an area of about 80,000 square km it is regarded to the hottest and the driest desert in the world (Alavi Panah et al., 2007; Gabriel, 1938; Mildrexler et al., 2006). Yardangs are streamlined forms up to 150 km long and 75 m in height resulting from a number of formative processes, including wind abrasion, deflation, fluvial incision, desiccation cracks, slumping, weathering and mass movement (Goudie, 2007; McCauley et al., 1977; Ward and Greeley, 1984). A limited number of morphometric investigations have been done on yardangs. Goudie (2007) identified mega-yardangs in hyper-arid environments with total rainfall less than 50 mm including central Asia, the Lut desert in Iran, northern Saudi Arabia, Bahrain, the Libyan Desert in Egypt, the central Sahara, the Namib desert, the high Andes and Peruvian desert. According to Goudie, these features develop in a wide range of rock types e.g. sandstones, ignimbrites, limestones and basement rocks by a relatively unimodal wind direction.

Materials and Methods
In this study 3 arc second DEM of version 3.0 SRTM data (~ 90 m) with geographic projection acquired was used. That was re-projected to UTM grid with WGS84 Datum.We used 90 m DEM produced from version 3 SRTM 3 arc second data and the SOM algorithm for identification of yardangs in the western part of Lut desert. Self Organizing Map (SOM) is an unsupervised and nonparametric artificial neural network algorithm that clusters high dimensional input vectors into low dimensional (usually two dimensional) output map which preserve topology of the input data. In geomorphic studies of landscapes, the first and second order derivatives of DEM are the basic components for morphometric analysis (Evans, 1972). The second order derivatives of DEM are affected by geomorphological processes (Evans, 1972; Wood, 1996b). To calculate the morphometric features, a local window is passed over the SRTM DEM and the change in gradient of a central point in relation to its neighbors is derived by a bivariate quadratic function. Wood (1996a) defined a set of criteria to classify DEMs into morphometric classes. Yardang identification and analysis in the study area was performed using the parameters proposed by Wood (1996a). A local window of 5×5 is passed over the DEM and slope, cross- sectional curvature, maximum and minimum curvatures are derived by fitting a bivariate quadratic approximation surface. The derived morphometric parameters were used as an input to SOM. Frequency histograms and average quantization error of the results are compared. Two dimensional plots of mean values of morphometric parameters (feature space), oblique views of map units draped over the DEM, Landsat ETM+ data and high resolution QuickBird images were used to study the mega yardangs in other places than Iran.

Results and Discussion
Learning of the SOM was performed with four morphometric parameters as the inputs and a two-dimensional output of 10 neurons. The map with initial radius of 3, final neighborhood radius of 0.01 and 1000 iterations shows the best performance for yardang identification. The output map units from SOM are just numbers and need to be analyzed and interpreted. Studying the spatial relationship between different map units along with their morphometric parameters using feature space analysis allowed us to interpret and label them corresponding to morphometric features e.g. yardangs. Major morphometric features (corridor, planar and yardang) are identified in two-dimensional feature space plots of mean values of maximum curvature (x-axis) and minimum one (y-axis). The analysis of the results and corresponding satellite images shows the effectiveness of the method to identify the overall pattern of the morphometric features in the Lut desert. It is clear that the pattern of yardangs and their corridors running from NNW to SSE direction is parallel to the prevailing wind known as “wind of 120 days or Bad-i-sad-o-bist rooz-e Systan.

Conclusion
The results show that digital terrain analysis methods applied on SRTM in the proposed way in this study could extract morphotectonic features from SRTM along Dehshir fault and they contributed to the tectonic interpretation of the study area. According to the evidences extracted from SRTM along Dehshir fault, for example: fault traces, deflected and beheaded drainages, pattern of network drainages, erosion surfaces of uplifted and back erosion of drainages because of the location (situated in quaternary landforms), they are neotectonic evidences for activity of Dehshir fault during quaternary. Our results reveal that the morphometric analysis and feature space analysis of the first and second order derivatives of DEM such as slope, cross-sectional curvature, maximum curvature and minimum curvature led to the description of SOM outputs as yardangs, corridors and planar. Slope allows distinguishing among morphometric features in sub levels. The optimal self organizing map with suitable learning parameters should be selected for feature identification. The lowest average quantization error of 0.1040 was achieved with an initial radius of 3, a final neighborhood radius of 0.01 and 1000 iterations. Result of this method revealed that from the total 6481 km2 coverage of the study area, about 2035 km2 (34%) are classified as yardang while corridors, in total cover 2732 km2 (43%).