Quantitative mapping of the soil rubification process on sand dunes ...

Geoderma 131 (2006) 1 – 21 www.elsevier.com/locate/geoderma

Quantitative mapping of the soil rubification process on sand dunes using an airborne hyperspectral sensor E. Ben-Dora,T, N. Levina, A. Singerb, A. Karnielic, O. Braund, G.J. Kidrona a

Department of Geography and Human Environment, Tel-Aviv University, Ramat Aviv, P.O.B. 39040, 69978, Israel b Seagram Center for Soil and Water Sciences, Hebrew University of Jerusalem, Rehovot, 76100, Israel c Bloustein Center for Desert Research, University of Ben-Gurion, Negev, Israel d Bar-Kal Systems Engineering, Ltd., Netanya, Israel Received 15 July 2004; received in revised form 26 January 2005; accepted 25 February 2005 Available online 2 June 2005

Abstract Soil rubification is defined as a pedogenesis stage in which iron is released from primary minerals to form free iron oxides that coat quartz particles in soils with a thin reddish film. This study used an airborne hyperspectral sensor to spatially map soil rubification on the surface of coastal sand dunes in Israel. The area selected for this study is located south of the city of Ashdod, Israel, and is one of the last remaining coastal sand dune areas in the country. The study area occupied about 60 km2, and was covered by two flight lines of the CASI 48-channel sensor from an altitude of 2650 m, providing a pixel size of about 3 m. A traditional way to estimate rubification is to chemically measure the iron oxide concentration in the sand. The Dithionite Citrate Bicarbonate (DCB) method, which is a laboratory bwetQ-based procedure, was used to precisely measure the free iron oxide status of selected locations along the area. The soil reflectance properties of these samples were measured in the laboratory across the Visible and Near InfrarRed (VIS–NIR) region and was used with the DCB-Fe data to evaluate a spectral-based model for assessing the rubification extent solely from spectroscopy. After the CASI data were atmospherically, BRDF and geometrically corrected, they were run against the spectral model on a pixel-by-pixel basis, generating a rubification map from a far distance. Spectral validation of independent samples and field measurements of dune movement over 17 months showed that the map is reliable and significantly correlated with known stabilization processes throughout the area (the overall accuracy estimated as 78%). It was concluded that soil spectroscopy, either from field or air, enables the detection of small changes in the Fe absorption feature across the VIS region that provides information regarding the iron oxide minerals and content. This supports the utilization of a sensitive airborne hyperspectral sensor to rapidly and quantitatively evaluate spatial information concerning important pedogenic processes. D 2005 Elsevier B.V. All rights reserved. Keywords: Remote sensing; Reflectance spectroscopy; Rubification; Sand dune; Free iron oxide; Soil mapping

T Corresponding author. Tel.: +972 3 6407049; fax: +972 3 6406243. E-mail address: [email protected] (E. Ben-Dor). 0016-7061/$ - see front matter D 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.geoderma.2005.02.011

2

E. Ben-Dor et al. / Geoderma 131 (2006) 1–21

1. Introduction Remote sensing of soils plays a major role in both soil survey and soil mapping applications. Air photos and satellite imagery are the major database from which final products, such as soil maps, are produced. The combination of remote sensing and traditional soil survey methods provides a spatial overview of large areas and effectively sheds light on regional temporal processes. Air photos and satellite images are based on the spectral reflectance information of soils, which relates to the soil chemical and physical characteristics. Unfortunately, both air photos and satellite imagery suffer from low spectral resolution and, thus, provide only limited spectral information about the targets explored. The high spectral resolution capability of remote sensing from airborne sensors means, significantly broadens the utility of this tool for further mapping of the soil surface from a more precise chemical and physical point of view (Ben-Dor, 2002). During the past few years it has been shown that soil spectra across the Visible–Near InfraRed–Short Wave InfraRed (VIS–NIR–SWIR) spectra region are characterized by significant chromophores (e.g. OH, Fe3+, CO3 and COOH) enabling quantitative analysis of soil properties (Ben-Dor, 2002). Accordingly, remote sensing of soils by high spectral resolution sensors is receiving more and more attention to rapidly and quantitatively map soils from far distances (Ben-Dor et al., 1999, 2002, 2004; Malley et al., 2004; Baptista and Netto, 2001). Imaging Spectrometry (IS) or hyperspectral technology is an advanced tool that provides high spectral resolution data (near-laboratory-quality reflectance and emittance data) for each single picture element (pixel) from a far distance (Goetz et al., 1985). This information allows the identification of objects based on the spectral absorption features of the chromophores and has been found to be very useful in many terrestrial and aquatic applications (Clark and Roush, 1984; Goetz et al., 1985). One of the most important phenomena in the soil formation process over a sand dune environment is the rubification process. Rubification is defined as a pedogenesis stage in which iron is released from primary minerals to form free iron oxides that coat quartz particles in soils with a thin reddish film (Buol et al., 1973). The free iron oxides coat the quartz

particles and provide a reddish chroma to the matrix as well as stability (Ben-Dor and Singer, 1987). There is abundant evidence that many dune sands become reddened with time. That process is promoted by warm temperatures, oxidizing conditions and periodic presence of moisture (Norris, 1969). Williams and Yaalon (1977) have demonstrated reddening in sand dunes under laboratory weathering conditions. Their research showed that organic matter is not necessary to initiate the process. On the coastline of Israel sand has been accumulated to form the coastal dune strip. The sand in this strip is brought from the Nile delta via a counter clockwise long-shore current along the eastern shore of the Mediterranean Sea (Nir, 1989), as presented in Fig. 1. Along the beaches of Israel, the color of the sand changes progressively from the Gaza Strip in the south to Athlit, ca. 130 km, to the north, ranging from 7YR7/4 (reddish light yellow) to 10YR7/10 (yellowish gray). According to Emery and Neev (1960), the change in color reflects a reduction in iron oxides within the sand transported by the longshore currents. As the sand is transported inland by the south westerly winds of the winter storms (Tsoar, 1990), the sand grains gain stronger red colors via the soil rubification process. During the rubification process, sand reddening occurs as a result of an interaction between electromagnetic radiation and free Fe oxides. Fe in the free iron oxides is spectrally active across the VIS–NIR region via the electron transition (of 6A1YT1g between 750–950 nm and 6A1YT2g between 550–650 nm) and is responsible for the Fe absorption of radiation that gives the soil its red color. Based on the free iron oxides that redden the sand, and using spectral color indices or linear mixing models, Madeira et al. (1997) and White et al. (1997, 2001) showed that it is possible to account for iron oxide status using the Landsat Thematic Mapper data (only six bands in the VIS– NIR–SWIR region) over lateritic soils in Brazil, and sand dunes at the Namib, and at the Northern Rub’ Al Khali (United Arab Emirates), respectively. Field spectroscopy was used by Bullard and White (2002) to quantify iron oxide coatings on dune sand in the Simpson-Strzelecki Desert, Australia. However, the iron content detected by all of the studies above was relatively higher than what may be found in the initial stages of the rubification process as presented


3

Fig. 1. The counter clockwise long-shore current transporting the sand from the Nile river towards Israel, and yearly estimates of the amount of sand transported at various locations (following Nir, 1989).

by the younger (ca. 1000 years) coastal dunes of Israel (Tsoar, 1990). Thus, the question whether such an initial process can be monitored from afar, remains open. Accordingly, it is strongly assumed that high spectral resolution data, such as provided by the high signal to noise field instruments and IS technology will enable such a detection. We therefore proposed to use a detailed reflectance spectroscopy across the VIS–NIR region to check whether small spectral changes based on the occurrence of free iron oxides can be detected along the coastal sand dunes of Israel. We also hypothesized that if such a relationship occurs, a sensitive IS sensor could be able to precisely detect such a phenomenon from a far distance and thus be able to spatially map the rubification. Therefore the purpose of this study was then, to apply a detailed spectral examination of the free iron oxide content of sand dunes along the

coastline of Israel from laboratory field and air perspective to answer the above questions.

2. Material and methods 2.1. The selected area The selected area lies south of the city of Ashdod near the coast of the Mediterranean Sea (locally termed as the Ashdod dunes). It borders the cities of Ashdod and of Ashkelon in the north and south, respectively, the Mediterranean Sea in the west, and a main north–south highway in the east (located at the interface between the dunes and agricultural fields). The total surface area is ca. 60 km2 with an average width of ca. 5 km. Fig. 2 provides a color composite of ETM+ image that covers the entire dune area

4



between the cities of Ashdod and Ashkelon and shows its position along the state of Israel. The occurrence of vegetation (reddish pixels) along the dune strip is relatively scarce, but visible in several pockets. As observed from the ETM+ image, there is no significant (reddish) color sequence at a west–east or south– north cross section of the bare dunes, except in the interface between the beach and the dunes and in several discrete pockets. These pockets are paleosols which were exposed to the sensors’ eye as a result of ongoing civil engineering activity to expand the city of Ashdod southbound, or due to the quarrying of sand and aelionites (bkurkarQ). Other reddish pockets are inter-dune vegetated and stabilized areas where iron oxide coating has been more intensive. As the total dune coverage along the coast of Israel has significantly diminished over the past 60 years from 462 km2 (Tsoar, 1990), to only 185 km2 at the end of the 1990s (Frumkin-Ahiron et al., 2003), the area selected by us represents one of the last bare sand dune areas that is still active (Fig. 2a). Although active, the study area is undergoing a process of stabilization by vegetation. The stabilization process is attributed to the cessation of grazing (Levin and Ben-Dor, 2004) and traditional agricultural practices (Tsoar and Zohar, 1985) that took place since the establishment of the state of Israel in 1948 (Tsoar and Blumberg, 2002). The direction of both sand movement and the prominent stormy winds is toward the northeast. Based on this pattern, we distributed our sampling sites along cross sections that will be related to the processes of dune development and sand movement (see Fig. 2b). 2.2. Airborne data acquisition 2.2.1. The selected sensor The CASI sensor was selected for this study. It is an airborne push broom sensor, which is sensitive to the VIS–NIR region and can be mounted on light aircraft. The sensor can be adjusted to carry up to 256

5

spectral bands across the VIS–NIR region based on the acquisition modes selected (spectral or spatial) (Anger et al., 1994). In October 1999 the CASI sensor was mounted onboard a Piper Aztec two-engine aircraft and acquired data from an altitude of 2650 m over the area of Ashdod, Israel (Fig. 2b). Two flight lines were acquired from south to north, each having a spatial swath of 1.2 km and ground resolution of about 3 m. The spectral configuration of the sensor was set to consist of 48 spectral channels equally divided across the VIS–NIR spectral region (0.4–1.0 Am) of the sensor’s detector sensitivity. 2.2.2. Ground sampling and measurements Parallel to the flight coverage, ground truth data were collected along the entire flight line coverage (10 km length). These data included atmospheric conditions (optical depth, humidity and water vapor content) and analytical data of the soil sample collections. Because the research area was composed of only a subset of the entire flight paths (4 km length), more soil samples that are mostly allocated within the research area were also collected the following year and during the same season for further study (all together 324 samples). All these soils were sampled from the upper surface of the dune (0–2 cm). For each location, a mix of several samples taken from an area of 1 1 m was used for further application. The soil samples were then brought to the laboratory, air dried, and analyzed for reflectance. The reflectance was measured by an ASD spectrometer (ASD, 2001) that covers the VIS–NIR–SWIR spectral region (0.4–2.5 um) using a closed chamber with constant and self Tungsten illumination. For each sample, twenty spectral measurements were taken and averaged to present the spectral characteristic of the sample. Whereas Fig. 2 shows the map of the area along with the flight coverage and the position of the soil sampling location, Table 1 provides a subdivision of the 324 soils into sub-groups that were used for the

Fig. 2. The study area on the ETM+ (Landsat 7) image (encoded RGB as 4,3,2) from August 7th, 1999, showing a wide range overview of the central dune area in Israel (a) and the enlargement of the study area (b). Also provided on the enlargement view (b) is the distribution of the sites from which sand samples were collected. Sites used for calibration and validation of the model are colored in yellow, orange and red, as described in the map legend. White rectangulars are at the positions of the erosion pins used to measure the sand stability. For all the samples for which percentage of Fe-DCB was extracted a red label indicating their serial number is given in the map (see Table 2 for more information). The grey polygons present the extent of the area covered by the two CASI images. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

6


Table 1 Schematic description of the soil samples used

Altogether 324 soil samples were collected across the dunes of Ashdod.

194 of them taken from active dune areas.

29 calibration samples between spectral indices and % Fe DCB 13 validation samples 27 calibration samples between CASI & ASD spectral indices

42 samples for which % Fe DCB was measured

40 samples located inside the CASI passes in areas with 0% vegetation cover

analysis. As seen from Fig. 2b, parts of the soil samples were allocated outside the CASI coverage whereas others were positioned nearby or inside vegetated inter-dune areas. From a careful analysis it was postulated that 194 of the soil samples were located over active dunes while the remaining 110 samples were located over other parts of the dune areas (e.g. vegetated inter-dune areas, exposed paleosols). Thus, the 194 samples that were located at the dunes were selected for further analysis. From these 194 samples, 42 soils were selected for the Dithionite Citrate Bicarbonate (DCB) method (see later). These samples were divided into two subgroups: for calibration (29 samples) and validation (13 samples), all located in un-vegetated areas of the CASI flight lines. These 42 samples (termed as Fe-DCB) were collected based on a complete pedogenetic sequence represented in the study area, ranging from pure beach sand dune to Haploxeralf (USDA) (or Hamra soil in a local terminology). In both groups, analysis for free iron oxide content using the Dithionite Citrate Bicarbonate (DCB) extraction method (Mehra and Jackson, 1960) was performed along with spectral measurements using the ASD spectral device. The 42 soil samples were also defined for their color indices using the Munsell color chart that was done under similar illumination conditions (see Table 2 for details). The accuracy of the color indices derived from the CASI image with respect to the ASD measurements was determined based on 40 samples located from areas with 0% vegetation cover in the CASI image; these were divided into two subgroups: calibration (27 samples) and validation (13 samples

that were the same samples that were used for validation of the Fe-DCB content, as described above). In order to further examine the in situ dune stabilization process, 315 erosion pins were installed along four parallel west–east transects (covering a total distance of 12.1 km). The pins served for erosion and deposition measurements (Arens et al., 2004). Measurements were conducted during a period of 17 months from December 2002 to May 2004 in 2–6 week intervals. The exact location of the erosion pins was determined using a differential GPS. Approximately 56% of the pins were allocated within the CASI flight line whereas the rest well characterized the entire dune area of Ashdod. These locations are shown overlaying the Landsat ETM+ image in Fig. 2b. 2.2.3. Imaging Spectrometry (IS) data manipulation The raw CASI data were converted into radiance values, using calibration measurements acquired in the laboratory prior to the flight. The conversion of the radiance into reflectance values was done according to the Empirical Line (EL) calibration method (Roberts et al., 1985), using five selected targets that represented the entire scene albedo variations and situated on a flat terrain. This conversion removes the atmospheric attenuation, such as aerosol scattering effects and gas absorption, but still maintains topographic effects in the data. Using a detailed Digital Terrain Model (DTM) of the area (at 0.5 m vertical resolution), the radiance was normalized to a nadir view, following Smith et al. (1980). Nevertheless, although the above steps are mandatory for sufficient data analysis, another source of critical errors that may be encountered in the data is that of the Bidirectional Reflectance Distribution Function effect (BRDF) (Pellikka et al., 2000). This effect is basically a product of different illumination and viewing angles that changes the spectral signature of a given material. To perform an empirical correction of this effect, the following procedures were applied: (1) Both CASI flight lines were georectified to the Israel Transverse Mercator (ITM) coordinate system (Mugnier Clifford, 2000) using 259 ground control points (GCPs) collected from a digital orthophoto for the western flight line, and 249 GCPs for the eastern flight line. A local affine


7

Table 2 Characteristics of the 42 sand samples for which percentage of Fe-DCB was extracted Sample Distance Munsell number from sea color (m)

Percentage BI: brightness CI: soil HI: hue RI: redness SI: spectral ABS1 590- ABS2 716of Fe-DCB index color index index index index 556-489 nm 534-477 nm measured

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 101 102 103 104 105 106 107 108 1031 2020 2021 2035 2074 2086 2095 2106 3054 3057 3067 3073

0.0370 0.0399 0.0366 0.0398 0.0407 0.0323 0.0355 0.0938 0.0261 0.0260 0.0309 0.0287 0.0311 0.0316 0.0267 0.0355 0.0261 0.0300 0.0246 0.0218 0.0240 0.0465 0.0733 0.1948 0.1893 0.3838 0.3423 0.3159 0.3926 0.0688 0.108 0.0496 0.0635 0.0473 0.0274 0.0348 0.0334 0.0394 0.0229 0.0381 0.0309 0.0472

3301 3209 2921 3142 2674 2609 2433 954 1150 1266 725 290 112 1209 668 460 535 123 200 60 7 2346 1228 1190 1346 521 98 218 821 2211 2816 513 549 991 2231 2634 3016 3399 865 745 523 370

10YR 6/3.5 10YR 7/5.5 10YR 7/5 10YR 7/5.5 10YR 7/4 10YR 7/4.5 10YR 7/4 9YR 5/6 10YR 7/4 10YR 7.5/4 10YR 7/4 10YR 8/3.5 10YR 8/2.5 10YR 7/4 10YR 7/4 10YR 7/4 10YR 7/4 10YR 8/2.5 10YR 7/4 10YR 8/2 10YR 8/2 10YR 6/6 10YR 5/4 10YR 4.5/4 7.5YR 4/5 10YR 4/4 10YR 4/4 7.5YR 4/6 7.5YR 4/6 10YR 4/4 10YR 4/4 10YR 4.5/3.5 10YR 4.5/3.5 10YR 6/3 10YR 5.5/4 10YR 6/3 10YR 6/4 10YR 6/4 10YR 6/4 10YR 6/3 10YR 7/4 10YR 5/3

0.252 0.338 0.356 0.347 0.354 0.348 0.366 0.237 0.381 0.397 0.388 0.395 0.431 0.357 0.367 0.368 0.358 0.413 0.336 0.425 0.450 0.349 0.230 0.205 0.186 0.172 0.184 0.185 0.144 0.241 0.184 0.185 0.204 0.244 0.275 0.308 0.366 0.336 0.342 0.299 0.394 0.249

0.150 0.212 0.198 0.218 0.199 0.201 0.193 0.279 0.184 0.181 0.184 0.163 0.130 0.180 0.185 0.189 0.192 0.129 0.194 0.121 0.120 0.229 0.256 0.273 0.266 0.264 0.278 0.293 0.284 0.237 0.272 0.213 0.216 0.192 0.216 0.156 0.184 0.183 0.161 0.164 0.167 0.172

3.07 3.69 3.58 3.74 3.58 3.56 3.51 4.37 3.42 3.46 3.51 3.19 2.99 3.48 3.47 3.49 3.50 2.91 3.49 2.83 2.81 3.71 4.37 4.41 4.84 4.84 4.93 5.01 5.19 3.94 4.56 4.18 4.03 3.82 3.80 3.04 3.51 3.37 3.26 3.23 3.25 3.67

47.6 42.8 34.4 42.6 35.1 37.1 31.4 153.0 26.9 24.0 25.7 21.5 13.8 29.1 29.2 30.1 32.4 15.1 37.6 13.4 11.8 47.5 127.6 191.9 200.5 230.7 229.0 259.7 385.3 103.5 228.5 133.6 115.6 66.3 65.6 33.7 28.8 34.7 27.7 37.4 22.2 54.5

0.344 0.439 0.416 0.447 0.417 0.425 0.410 0.532 0.396 0.385 0.389 0.367 0.301 0.379 0.395 0.402 0.407 0.307 0.413 0.293 0.290 0.478 0.482 0.517 0.470 0.465 0.488 0.514 0.486 0.476 0.502 0.401 0.419 0.379 0.437 0.363 0.388 0.399 0.353 0.364 0.371 0.347

0.994 0.970 0.977 0.968 0.978 0.979 0.982 0.935 0.988 0.986 0.984 0.993 0.999 0.986 0.985 0.981 0.980 1.002 0.980 1.005 1.001 0.964 0.951 0.937 0.926 0.938 0.930 0.911 0.918 0.959 0.951 0.984 0.976 0.978 0.980 1.000 0.981 0.983 0.997 0.997 0.990 0.994

1.047 1.002 1.012 0.998 1.011 1.014 1.017 0.939 1.025 1.022 1.019 1.037 1.042 1.020 1.021 1.018 1.016 1.048 1.019 1.052 1.050 1.000 0.948 0.939 0.924 0.921 0.911 0.900 0.894 0.980 0.934 0.982 0.983 0.996 0.998 1.040 1.015 1.024 1.032 1.035 1.035 1.013

All spectral indices were measured using the ASD.

transformation (a linear combination of translation, rotation and scaling adapted locally using the triangulation technique available in ENVI 3.5 package, 2001; Research Systems, Inc., 2001) was then applied to account for the non-systematic geometric effects caused by the dunes’ topography (5–20 m

high) and the low flight altitude (about 2300 m above surface) from which the aerial photos were taken (Buiten and van Putten, 1997). (2) The values of the spectral indices found for those sand samples which are located inside the CASI flight lines in areas with 0% vegetation cover, were compared to

8


those measured using the ASD; (3) The differences in the values of the two sets of spectral indices were used to calculate linear empirical correction equations from the nadir viewpoint to all other viewing angles on each line of the image. These parameters were then used to correct the rectified spectral indices maps to the illumination residual effects on a line-by-line basis. An empirical offset factor was then applied to the east flight line to adjust it to the west flight line using similar ground targets. Validation of this correction was performed by comparing the similarities between the imaging spectra and the laboratory spectra using samples that were not incorporated in the EL process.

Image processing steps Digital Terrain Model

CASI image DN values

CASI image Radiance values

Simulated shading Cosine (incidence angle) values

CASI image Reflectance values

Following radiometric atmospheric rectification, a field examination was carried out to verify that the occurrences of objects such as sand, vegetation, water and soils actually existed on the ground. Only after the previous correction stages, the sub-scene from each flight line (20 km length over 400 m swath) covering the dune area only (7 km length over 400 m swath) was further used for this study. To summarize the above technical stages, we provide a flow chart diagram (Fig. 3) that describes in detail the procedures performed to obtain a nadir view point reflectance data of the dune, clean from atmospheric attenuation and other physical artifacts.

Field & lab work Placement of erosion pins

Measurement of erosion and deposition of sand

Measurement of the reflectance spectra

Collection of soil and sand samples

Measurement of % Fe content (42 samples)

CASI image Reflectance values normalized to a flat surface

Calculation of various color & vegetation spectral indices

Images of spectral indices Without BRDF effects

Calculation of various color spectral indices

Map of % Fe content across the dunes area

Fig. 3. A flow chart that summarizes the analytical and technical stages of this study that were performed to obtain the final map of Fe-DCB percentage across the dune area.


9

2.3. Synthetic sand dune

3.2. Synthetic sand dune spectra

In order to study the spectral behavior of the iron oxides within the study area, a mixture of synthetic iron oxide and quartz was analyzed. For that purpose, we selected hematite and goethite, which represent the color agent minerals, and quartz, which represents the major mineral component in the soil matrix. The spectra of these minerals were taken from the Grove et al. (1993) spectral library. In order to enhance the absorption features in each mineral spectrum, we applied a spectral continuum removal (CR) algorithm (Clark et al., 1987) in which the monotonous spectra signals are characterized by a unity reflectance value and absorption features are presented relative to a continuum of interpolated reflectance values that connect between two absorption edges. In addition, in order to evaluate our capability to distinguish between those three minerals, we spectrally mixed the three components (hematite, goethite and quartz) to yield a new synthetic bduneQ spectrum.

The spectral characteristics of goethite, hematite and quartz are presented in Fig. 4a. As can be seen, hematite and goethite in their pure forms hold significant spectral signatures, which make them easily distinguishable. The quartz in this discussion is represented by rather featureless components across the study’s spectral region (VIS–NIR). That was, however, not the case following the application of the CR technique (that removes the base line effect). As can be seen in Fig. 4b, the absorption feature in the CR spectra of the iron minerals is more enhanced and indicates more clearly the occurrences of iron oxides. For example, after applying the CR algorithm, the hematite shows an absorption feature at 535 nm. Due to the electronic transition of 6A1YT2g, goethite shows an absorption feature at 485 nm. The slight shift of the Fe absorption feature by 50 nm enables the recognition of both hematite and goethite occurrences in a given matrix. Also noticeable is the absorption peak at 900 nm, which can serve as a similar indicator. However, as it is located outside the area of favorable sensitivity for the CASI sensor, the discussion over this wavelength is beyond the scope of the present research. Fig. 5 shows the result of a linear mixture of 90– 60% quartz with complementary fractions of hematite and goethite, both in its original (a) and continuumcorrected CR (b) spectral domains. It can clearly be seen that the slope of the original spectra changes from low to high as the iron mineral fraction increases (in the original spectra), while the peak depth of Fe increases as well (most clearly seen in the CR spectra). Also visible is the position of the peak, which shifts relative to the pure absorption features based on the relative concentration of the two minerals. Since they present equal amounts, the absorption position was located between the original absorption of the pure mineral absorption (485 and 535 nm yielded 510 nm). These spectral trends strongly suggest that the spectral reflectance properties of iron oxides with similar components would be capable of providing information regarding the amount and species of iron oxides. One should note however that in the natural soil environment, the formation of free iron oxides might not be a separate phase, but instead be part of a coating over the soil

3. Results and discussion 3.1. General From Table 2 it is postulated that very low Fe-DCB values were measured along the dune areas in question (from 0.02% to 0.05%) with no significant visible-on-spot color change, except for the area in immediate adjacency to the coastline. Also, no significant discrimination was obtained between the sand samples using the Munsell parameters. The low Fe-DCB values of the sand samples agrees with the previous observation that showed only a weak red color sequence in either the satellite or ground view domains. This strongly suggests that methodologies other than visual observation to account for the FeDCB changes should be adopted. In this sense, and based on previous success by Bullard and White (2002) to assess free iron oxides from field-based reflectance spectroscopy, we previously suggested that the potential of the spectral reflectance indices be systematically examined as a tool for large area mapping purposes and, thus, performed various steps in which we studied the pure and natural spectral information of iron oxides in the selected dunes.

10


Fig. 4. The reflectance spectra of quartz, goethite and hematite as taken from Grove et al. (1993) before (a) and after (b) applying the continuum removal (CR) logarithm (see text for more details).

particles. Free iron oxides also act as a cementing agent between other minerals such as clay particles (Ben-Dor and Singer, 1987). As a result, the matrix in question can no longer be represented by a simple linear mixture model, except to provide a trend against which the spectral information can be checked. 3.3. Laboratory (Field) dune spectra Based on the previous finding that a mixture of iron oxides and quartz can provide significant spectral

information, we further examined the real bsoilQ environment, using the laboratory information obtained by both spectral and chemical means. For this stage, the original laboratory spectra of the field samples (in the 42 samples subgroup presented in Table 1) consisting of 2151 spectral channels (between 350 and 2500 nm) and their re-sampled spectra (consisting of 48 spectral CASI channels) were investigated. In Fig. 6, two representative samples are shown both in the original ASD and CASI spectral configurations. As can be seen, the


11

Fig. 5. The reflectance spectra of the linear mixture of quartz, goethite and hematite presented in Fig. 5 before (a) and after (b) applying the continuum removal (CR) algorithm.

spectral channel reduction (from 2151 to 48) has not significantly affected the spectral quality in terms of the spectrum’s slopes or of the Fe absorption features. This is due to the fact that, across the VIS–NIR region, the soil spectra are rather monotonous and the 48-channel configuration is able to keep the spectral information in the scene. Further, while checking spectral degradation, Ben-Dor and Banin (1994) concluded that several soil properties are better

predicted from reflectance information if the original information (consisting of 3113 channels) has been reduced into 15–63 channels. Thus, the 48 spectral channels of the CASI do not seem to be an obstacle for the examination of the potential to quantitatively estimate the DCB-Fe content. This was done first in the reflectance spectra of four representative soil samples taken in the laboratory (in the 48 CASI channels) with low and high Fe-DCB content.

12


Fig. 6. The reflectance spectra of two representative samples (8 and 21) before and after re-sampling the laboratory spectra into the CASI spectral configuration.

Fig. 7 shows the above spectra before and after applying the CR algorithm. From this figure it is postulated that, alongside the absorption intensity of the Fe at around 499 nm, the CR enables the exact position of the Fe absorption feature of the real soil samples. Based on Fig. 4 and on the discussion about the synthetic mixture mineral composition, the observed wavelength in Fig. 7, 499 nm, is attributed to a goethite–hematite mixture with approximately 70% goethite and 30% hematite in the selected samples, and probably in the entire population. As can be seen from Fig. 7, a spectral sequence exists between the selected samples. In spite of the visual observation that the spectra show a good representation of iron oxides, further quantitative study is called for to estimate the DCB-Fe content solely from the spectral data. Based on the above, two types of spectral parameters were calculated: (A) absorption depth (using the CR spectra), termed ABS 1 (based on the wavelengths of 489, 556 and 590 nm), and ABS 2 (based on the wavelengths of 477, 534 and 716 nm). (B) Color indices using three bands representing the red (R, 693 nm), green (G, 556nm)

and blue (B, 477 nm) bands following Mathieu et al., 1998: pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Brightness Index; BI ¼ ðB2 þ G2 þ R2 Þ=3 ð1Þ Colouration Index; CI ¼ ð R GÞ=ð R þ GÞ

ð2Þ

Hue Index; HI ¼ ð 2TR G BÞ=ðG BÞ Redness Index; RI ¼ R2 = BTG3

ð3Þ ð4Þ

Saturation Index; SI ¼ ð R BÞ=ð R þ BÞ

ð5Þ

In order to select the best representative index that can predict the Fe-DCB content solely from reflectance, a correlation analysis was performed using the laboratory Fe-DCB chemical data and the laboratory spectral data. Following this application, the following sequence of R 2 values on a log-normal scale were fo un d: RI ( 89. 2%) NA BS 2 ( 87 .8 % ) NH I (83.9%) N ABS1 (83.4%) N CI (71.0%) N BI (69.0%) N SI (52.3%), suggesting RI to be the best parameter to spectrally describe the Fe-DCB content. In Fig. 8 the RI (determined from the ASD) is plotted against the Fe-DCB (determined in the laboratory). One should


13

Fig. 7. The reflectance spectral of four representative sand samples from the study area as obtained in the laboratory before and after applying the continuum removal (CR) algorithm (a—laboratory and b—image spectra). Also provided is the DCB-Fe content of each sample as obtained by the DCB method in the laboratory.

note the redness saturation in the sensitivity of the various spectral indices that has occurred. A similar saturation was reported by Torrent et al. (1983) in Brazil based on Munsell color charts with respect to the hematite content of lateritic soils. 3.4. Image dune spectra To further apply the best spectral index found in the previous stage (RI) to the image, it was first necessary to examine how the RI from the ASD data

is correlated with the CASI data. For that purpose, we extracted from the image the spectral data of each of the sand sample locations using the differential GPS data, from the geometrically, radiometrically and atmospherically rectified image. Thus a spectrum of the selected locations in the CASI domain was obtained, for which only those samples that were found over areas with 0% vegetation cover (according to the Soil Adjusted Vegetation Index, SAVI; Huete et al., 1992) were taken. Based on a total of 40 samples, the obtained sequences of R 2 values between the

14


Fig. 8. A scatter plot presenting the correlation between percentage of Fe-DCB as measured in the laboratory and the RI (Redness Index), using 42 sand samples.

spectral indices (as measured by the ASD) and the CASI images on a linear scale were: CI (91.1%) N SI (88.4%) NRI (88.1%) NHI (87.8%) NABS1 (79.1%) N ABS2 (63.2%) N BI (49.4%). From this finding it is postulated that the ASD and CASI data are significantly correlated with each other. They are more correlated for the RI and less for the ABS2 parameters (see Fig. 9 for the RI index case). The good correlations between the two data sets suggest that the atmospheric correction of the CASI data is reliable for the RI parameter and factors such as BRDF or topographic errors were effectively removed from the image. Based on the above findings, we further selected the RI index for describing the spatial intensity of the rubification process. For that purpose, we first examined the feasibility of this index to track the rubification process over a west–east transaction, starting at the coastline and ending at the inner land position, and then applied it to the image data. To assess our hypothesis that a progressing reddening of the dunes may take place from west to east, spectral parameters of 194 samples, located on the active dunes, were plotted against their distance from the coastline. The obtained R 2 sequences on log– log scale were: RI (77.6%) N SI (72.9%) N CI (71.1%) N HI (61.8%) N ABS1 (57.3%) N ABS2

(55.0%) N BI (53.1%). Again, the RI performed the best correlation amongst other spectral parameters with the distance from the coastline. This can be clearly seen when data taken from both the ASD laboratory (194 samples) data and the CASI data (40 samples) were plotted against the distance from the coastline (Fig. 10). It can be noticed that a significant and rapid change occurs within the RI index (and hence with the sand redness color) in the first 600 m close to the coastline that turns into a moderate to almost small linear slope farther away from the coastline (R 2 = 78% for the ASD data, and R 2 = 88% for the CASI data). Assuming that higher RI values may also indicate areas where the dunes are more stabilized, we plotted the sum of absolute changes in the sand’s height on all those erosion pins located on the lee slopes of dunes along our central transect (n = 32) against the distance from the coastline. As can be seen in Fig. 11, sand erosion and deposition decrease exponentially as a function of the distance from the coastline (R 2 = 78%, p b 0.01). This is probably due to lower wind speed as the roughness of the surface increases inland (Troen and Peterson, 1989). Indeed, when plotting these two variables one against the other, i.e. the RI parameter and the sum of absolute changes in the sand


15

Fig. 9. A scatter plot presenting the correlation between the values of RI as calculated from the ASD spectra and the BRDF corrected CASI image (n = 40).

accumulation height, a mirror-like picture representing a negative logarithm correlation was found (R 2 = 51%, p b 0.01; see Fig. 11). This pattern clearly reveals that the most rapid spatial changes, both in the sand color and in the sand movement, occur closer to

the coastline, the place that supplies the sand grains into the dunes system, and from which the south westerly winds blow. In order to check whether the prediction of the FeDCB content is possible directly from the CASI

Fig. 10. Rubification of the dune sand as a function of the distance from the coastline, as measured by the RI (Redness Index) using 194 sand samples measured by the ASD, and 40 of the same samples as measured from the CASI image (after the BRDF correction).

16


Fig. 11. A scatter-plot presenting the correlation between the sum of absolute changes in the sand’s height with respect to: (1) the distance from the coastline, as measured between December 2002 and May 2004, for 32 erosion pins located on the lee slopes along the central transect of sand dunes in the Ashdod dunes; (2) the Redness Index (RI).

image, we applied the RI index to the CASI image on a pixel by pixel basis. Using the 29 samples of the calibration group we obtained the following equation: Fe DCB ¼ 0:0244e0:0089RI R2 ¼ 87:6% ð6Þ

In order to calculate the overall accuracy we further manipulated the validation results to give the Relative Spectral Prediction Error (RSPE) as follows:

This equation was then applied to every pixel in the image to reveal a gray scale image, in which the intensity level of each pixel correlates with the DCBFe content. Fig. 12 shows the two geo-rectified flight lines after applying the above equation and after the gray scale image was encoded with a color table that presents the iron oxide (Fe-DCB) as reddish in color. In contrast to the manipulated Fe-DCB image, areas in the gray scale image that are not characterized as dune or soil (pixel with than less 5% vegetation as calculated based on SAVI parameter (Huete et al., 1992) were not encoded with the selected color table, because their calculated values lie outside the FeDCB calibration scale. A validation examination to check the calibration reliability was performed using the remaining 13 samples. Fig. 13 thus provides a scatter-plot presenting the real Fe-DCB values against the predicted values of the above 13 validation samples. The obtained R is relatively good (0.73) suggesting that the examined methodology is reasonable for IS data.

where FeMAX and FeMIN are the maximum and minimum values of the Fe-DCB population and RMSE (Root Means Squared Error) is calculated using the following equation: 0:5 RMSE ¼ AðFeðpÞ FeðtÞÞ2 =n 1 ð8Þ

RSPE ¼ RMSE=ðFeMAX FeMINÞ

ð7Þ

where Fe(p) is the Fe-DCB content predicted from the RI index and Fe(t) is the true Fe-DCB content as obtained from the wet chemistry analysis. Accordingly, the calculated RMSE found for the iron content was 0.0040, representing an error value of RSPE = 22% relative to the 13 validation samples (covering sand dunes only). The overall accuracy is therefore 78%, if only the dune areas are taken into consideration. This error suggests that the spectral information from remote sensing is a favorable tool to determine the free iron oxides (Fe-DCB) on a spatial domain using the RI parameter. The Fe-DCB image shows a match between the field-laboratory cross section using the 194 field samples and their laboratory RI ASD


17

Fig. 12. Map of the predicted values of percentage of Fe-DCB (*100), based on the RI parameter, as calculated from the CASI image. Black areas: vegetation.

values. A reddish trend from low to high Fe-DCB (or RI) content going from west to east (from the coastline to the dunes inland, respectively) is thus observed. We can thus summarize and say that the reddish color on the processed image (Fe-DCB content) fairly well reflects the stabilization process of the dune

under investigation. In other words, the more stable the sand, the more effective are the soil weathering processes that produce free iron oxide minerals. It should be pointed out that another mechanism may affect the reddish color in the same way the weathering process does. This is the aeolian abrasion process (Bullard et al, 2004) in which iron-rich

Fig. 13. A scatterplot representing the modeled values of percentage of Fe (based on the RI parameter) vs. those measured in the laboratory, for the 13 validation samples.

18


coatings of the sand particles are removed by the particle-to-particle friction driven by the wind energy. As both mechanisms (chemical weathering and physical abrasion), change the reddish color, it is only obvious that the sensitive reflectance information identifies such color changes long before it is visible to the naked eye. As a result, it can be concluded that reflectance spectroscopy can be used as a tool for tracking early rubification processes in active dune areas. As it stands, the spectral information is sensitive enough to detect relatively low FeDCB content from both field and airborne domains. Today, when field spectrometers are widely used and airborne imaging spectrometers are commercially available, the findings provide additional and important insight into the pedology–geomorphology processes of sand dunes. 3.5. Possible errors that might affect similar mapping The 22% error estimated for the Fe-DCB image is an average value that does not necessarily represent local errors (e.g. errors from mis-positioning of the sample in the image and on the ground). Local and regional factors that might affect the results could be the existence of chromophores other than Fe which are active across the VIS spectrum as well. Accordingly, organic matter, vegetation and soil moisture may be those components that affect the Fe-DCB results. Organic matter may change the spectral slope of the VIS–NIR region (Ben-Dor et al., 1997) and accordingly the RI index. However, this can happen only when the organic matter content is reaching a significant threshold (above 2%; Baumgardner et al., 1985). If the organic matter is fresh, then signatures of chlorophyll can also be presented at around 680 nm (Ben-Dor et al., 1997). The organic matter content (fresh and mature) over a free sand dune area (far from vegetation spots) is fairly low in the selected area (b 1%) and thus, not spectrally significant. However close to vegetation pockets, caution must be taken in using the proposed spectral method. Along these areas both organic matter and chlorophyll content may be masking out the spectral response of the Fe. Another factor that may affect the Fe-DCB analysis is soil moisture. In this regard the overall soil reflectance is known to decrease across the entire

spectral region while the moisture increases and thus diminishes the Fe absorption feature accordingly (Baumgardner et al., 1985). High moisture content could bias the results to a lower DCB-Fe content both by using the color indices and/or the ABS parameters. During the flight time, the sand dune was dry (as it took place before the beginning of the rainy season) and, therefore, any potential effect from the above moisture error seems to be negligible, if any. Biogenic crust may also affect the spectral DCB Fe estimation especially at around 680 nm (Karnieli and Tsoar, 1994). This may hence increase the RI parameters, biasing the DCB-Fe results toward higher values. Around the Ashdod dunes, the biogenic crust is only weakly developed, especially if the sand is dry, and, consequently, error from this source seems to be minimal as well. Other possible errors may arise from the fact that atmospheric attenuation has not been completely removed. In this regard, if aerosols are not properly removed, the RI parameter may decrease and the corresponding spectral DCB-Fe content can decrease accordingly. Around the ABS-1 peak (499 nm) a weak absorption peak of Ozone is generally presented (Ben-Dor et al., 1999) and hence may decrease the RI values as well. This is true, of course only if the Ozone is not distributed constantly along the area, which is not the case here. Water vapor absorption is active around 650 and 740 nm, which, again, can reduce the RI values and thus must be completely removed prior to any spectral analysis. We assume that this type of atmospheric attenuation is minimal too as all validation checks showed that the atmospheric removal procedure was favorably done. If BRDF effects have not been removed they may increase or decrease the overall reflectance and hamper quantitative analysis. This effect may cause tone variation from image edges to the center. In this study we have corrected this effect by an empirical method that showed relatively good results (using a set of 27 calibration samples against 13 validation soil samples). Nevertheless, in any future analysis similar to this work, significant attention must be paid to this effect. Other sources of errors are parameters related to the sensor performances such as: signal-to-noise ratio, spectral resolution, band width and position, and geometric rectification process. It can be concluded that in spite of the above-mentioned restrictions, it is


still possible to spatially map the occurrence of free iron oxides using a hyperspectroscopy approach and observe interesting spatial processes related to the dune stabilization as governed by free iron oxides.

4. Summary and conclusions Reflectance spectroscopy offers new insights into the pedogenesis process, long before it is visible to the naked eye. The imaging spectroscopy technique, which provides near-laboratory-quality reflectance information, also has the capability to obtain nonvisible information and thus to produce a spatial overview of the pedogenesis processes in large scales. The current study showed that it is possible to detect even a narrow and low Fe-DCB concentration range by both spectral methodologies (point and image) and quantitatively estimate iron oxide minerals fractions (hematite and goethite) in the soil. Two parameters were found to be correlated with the Fe-DCB content: the color index RI and the scaled reflectance (ABS2). The RI parameter was further selected to predict the Fe-DCB content as it represents higher correlation with other parameters and holds better validation power. Applying the predictive RI equation on a pixel by pixel basis revealed a spatial Fe-DCB image with an estimated accuracy of ~ 80%. Although some spectral degradation occurred while going from the laboratory to the spectral image domain, the spatial overview of the Fe-DCB images revealed a reasonable view of the Fe-DCB distribution, taking into account the sand movement potential, vegetation occurrence and human activities (e.g. quarrying of sand and the exposure of paleosols in construction sites) that were verified on the ground. Ground validation has not only consisted of spectral measurements of 194 samples, but also of a careful geomorphology study lasting 17 months to track the sand erosion/deposition. A correlation between the sand accumulated near the erosion pins and the spectroscopy confirm the paper’s hypothesis. We may therefore conclude that high spectral information from either field or airborne domains can provide valuable information regarding premature soil formation stages that are represented by the rubification process. This information can be used as a tool to assess sand dune

19

stabilization along areas with similar characteristics. Caution must be taken, however, when organic matter content is relatively high, when the sand is wet or when biogenic crust is well developed. Also, air and space images must be carefully treated to properly remove the atmospheric attenuation and BRDF effects along with the usage of high quality IS data (high signal-to-noise ratios) and well geo rectified images prior to sample location in the model.

Acknowledgments The authors wish to thank Mrs. Ziva Hochman from the Soil and Water Department, Faculty of Agriculture, Hebrew University of Jerusalem for carrying out the Fe-DCB analysis. We also thank Ravid Pik and Yoav Eshel from the Israeli Green Patrol (Ha Sayeret ha Yeruka) for the use of the differential GPS.

References Anger, C., Mah, S., Baney, S., 1994. Technological enhancements to the Compact Airborne Spectrographic Imager (CASI). Proceedings of the First International Airborne Remote Sensing Conference and Exhibition, Strasbourg, France, 12–15 September 1994 II, 205 – 213. Arens, S.M., Slings, Q., de Vries, C.N., 2004. Mobility of a remobilized parabolic dune in Kennemerland, The Netherlands. Geomorphology 59, 175 – 188. ASD, 2001. Analytical Spectral Devices website: http://www.asdi. com/. Accessed on November 13, 2003. Baptista, G.M., Netto, J.S.M., 2001. RCGb index: a tool for mapping the degree of weathering in the tropical soils in Brazil. In: Green, O.G. (Ed.), Proceedings of the Tenth JPL Airborne Earth Science Workshop. JPL Publication. 02-141-50. Baumgardner, M.F., Silva, L.F., Biehl, L.L., Stoner, E.R., 1985. Reflectance properties of soils. Advances in Agronomy 38, 1 – 44. Ben-Dor, E., 2002. Quantitative remote sensing of soil properties. Advances in Agronomy 75, 173 – 243. Ben-Dor, E., Banin, A., 1994. Visible and near infrared (0.4–1.1 Am) analysis of arid and semi arid soils. Remote Sensing of Environment 48, 261-27. Ben-Dor, E., Singer, A., 1987. Optical density of vertisol clays suspensions in relation to sediment volume and dithionite– citrate–bicarbonate extractable iron. Clays and Clay Minerals 35, 311 – 317. Ben-Dor, E., Inbar, Y., Chen, Y., 1997. The reflectance spectra of organic matter in the visible near infrared and short wave infrared region (400–2,500 nm) during a control decomposition process. Remote Sensing of Environment 61, 1 – 15.

20


Ben-Dor, E., Irons, J.A., Epema, A., 1999. Soil spectroscopy. In: Rencz, A. (Ed.), Manual of Remote Sensing, Third edition. J. Wiley & Sons, Inc., New York, pp. 111 – 189. Ben-Dor, E., Patkin, K., Banin, A., Karnieli, A., 2002. Mapping of several soil properties using DAIS-7915 hyperspectral scanner data. A case study over clayey soils in Israel. International Journal of Remote Sensing 33, 1043 – 1062. Ben-Dor, E., Goldshalager, N., Braun, O., Kindel, B., Goetz, A.F.H., Bonfil, D., Agassi, M., Margalit, N., Binayminy, Y., Karnieli, A., 2004. Monitoring of infiltration rate in semiarid soils using airborne hyperspectral technology. International Journal of Remote Sensing 25, 1 – 18. Buiten, H.J., van Putten, B., 1997. Quality assessment of remote sensing image registration—analysis and testing of control point residuals. ISPRS Journal of Photogrammetry and Remote Sensing 52, 57 – 73. Bullard, J.E., White, K., 2002. Quantifying iron oxide coatings on dune sands using spectrometric measurements: an example from the Simpson-Strzelecki Desert, Australia. Journal of Geophysical Research 107 (B6), 2125 – 2138. Bullard, J.E., McTainsh, G.H., Pudmenzky, C., 2004. Aeolian abrasion and modes of fine particle production from natural red dune sands: an experimental study. Sedimentology 51, 1103 – 1125. Buol, S.W., Hole, F.D., McCracken, R.J., 1973. Soil Genesis and Classification. The Iowa State University Press, Ames, p. 360. Clark, R.N., Roush, T.L., 1984. Reflectance spectroscopy: quantitative analysis techniques for remote sensing applications. Journal of Geophysical Research 89 (B7), 6. Clark, R.N., King, T.V.V., Gorelick, N.S., 1987. Automatic continuum analysis of reflectance spectra. Proceedings of the Third AIS Workshop, 2–4 June 1987, JPL Publication, vol. 8730. JPL, Pasadena C, pp. 138 – 142. Emery, K.O., Neev, D., 1960. Mediterranean beaches of Israel. Ministry of Development, Geological Survey, Bulletin, vol. 26. Geological Survey of Israel, Jerusalem, Israel, pp. 1 – 22. Frumkin-Ahiron, T., Frumkin, R., Roudich, R., Melloul, A., Levin, N., Papay, N., 2003. Conservation of the Coastal Sand Dunes – a Policy Report, The Surveys Unit – Open Landscape Institute—The Society for the Protection of Nature in Israel, Ministry of Environment, The Nature and Parks Authority, The Jewish National Fund, The Water Commission and the Jerusalem Institute for Israel Studies. 126 p., (in Hebrew). Goetz, A.F., Vane, H.G., Solomon, J., Rock, B.N., 1985. Imaging spectrometry for earth remote sensing. Science 228, 1147 – 1153. Grove, C.I., Hook, S.J., Paylor, E.D., 1993. Laboratory Reflectance Spectra of 160 Minerals, 0.4 to 2.5 Micrometers. JPL Publication, vol. 92–2. Jet Propulsion Laboratory, Pasadena. Huete, A.R., Hua, G., Qi, J., Chehbouni, A., van Leeuwen, W.J.D., 1992. Normalization of multidirectional red and NIR reflectance with the SAVI. Remote Sensing of Environment 41, 143 – 154. Karnieli, A., Tsoar, H., 1994. Spectral reflectance of biogenic crust developed on desert dune sand along the Israel–Egypt border. International Journal of Remote Sensing 16, 369 – 374.

Levin, N., Ben-Dor, E., 2004. Monitoring sand dune stabilization along the coastal dunes of Ashdod-Nizanim, Israel, 1945–1999. Journal of Arid Environments 58, 335 – 355. Madeira, J., Bedidi, A., Cervelle, B., Pouget, M., Flay, N., 1997. Visible spectrometric indices of hematite (Hm) and goethite (Gt) content in lateritic soils: the application of a thematic mapper (TM) image for soil-mapping in Brasilia, Brazil. International Journal of Remote Sensing 18 (13), 2835 – 2852. Malley, D.F., Martin, P.D., Ben-Dor, E., 2004. Application in analysis of soils in near infrared spectroscopy in agriculture. In: Roberts, A.C., Workman, J.J., Reeves, J.B. (Eds.), Agronomy Monograph no. 44, pp. 729 – 784. Mathieu, R., Pouget, M., Cervelle, B., Escadafal, R., 1998. Relationships between satellite-based radiometric indices simulated using laboratory reflectance data and typic soil colour of an arid environment. Remote Sensing of Environment 66, 17 – 28. Mehra, O.P., Jackson, M.L., 1960. Iron oxide removal from soils and clays by dithionite–citrate system buffered with sodium bicarbonate. Clays and Clay Minerals 7, 317 – 327. Mugnier Clifford, J., 2000. Grids and datums: the state of Israel. Photogrammetric Engineering and Remote Sensing 66 (8), 915 – 917. Nir, Y., 1989. The seashores of the Mediterranean Sea along Israel and north Sinai: Sedimentology aspects: Ministry of Energy and Infrastructure. The Geological Survey of Israel, Jerusalem, GSI/39/Report 88. Norris, R.M., 1969. Dune reddening and time. Journal of Sedimentary Petrology 39 (1), 7 – 11. Pellikka, P., King, D.J., Leblanc, S.G., 2000. Quantification and reduction of bi-directional effects in aerial CIR imagery of deciduous forest using two reference land surfaces types. Remote Sensing Reviews 19, 259 – 291. Research Systems, 2001. ENVI User’s Guide, Envi Version 3.5. Roberts, D.A., Yanaguchi, Y., Lyon, R.J.P., 1985. Calibration of airborne imaging spectrometer data to the percent reflectance using field spectral measurements. Proceedings of the Nineteenth International Symposium on Remote Sensing of Environment, Ann Arbor, Michigen 21–25 October 1985, 463 – 472. Smith, J.A., Lin, T.L., Ranson, K.J., 1980. The Lambertian assumption and Landsat data. Photogrammetric Engineering and Remote Sensing 46 (9), 1183 – 1189. Torrent, J., Schwertmann, U., Fechter, H., Alferez, F., 1983. Quantitative relationships between soil colour and hematite content. Soil Science 136 (6), 354 – 358. Troen, I., Peterson, E.L., 1989. European Wind Atlas, Commission of the European Community. Riso National Laboratory, Roskilde, Denmark. 656 pp. Tsoar, H., 1990. Trends in the development of sand dunes along the southeastern Mediterranean coast. In: Bakker, Th.W., Jungerius, P.D., Klijn, J.A. (Eds.), Dunes of the European Coasts, Catena Supplement, vol. 18, pp. 51 – 60. Tsoar, H., Blumberg, D.G., 2002. Formation of parabolic dunes from barchan and transverse dunes along Israel’s Mediterranean coast. Earth Surface Processes and Landforms 27, 1147 – 1161. Tsoar, H., Zohar, Y., 1985. Desert dune sand and its potential for modern agricultural development. In: Gradus, Y. (Ed.), Desert Development. D. Reidel Pub. Co., pp. 184 – 200.

E. Ben-Dor et al. / Geoderma 131 (2006) 1–21 White, K., Walden, J., Drake, N., Eckardt, F., Settle, J., 1997. Mapping the iron oxide content of dune sands, Namib Sand Sea, Namibia, using Landsat Thematic Mapper data. Remote Sensing of Environment 62, 30 – 39. White, K., Goudie, A., Parker, A., Al-Farraj, A., 2001. Mapping the geochemistry of the Northern Rub’ Al Khali using multispectral

21

remote sensing techniques. Earth Surface Processes and Landforms 26, 735 – 748. Williams, C., Yaalon, D.H., 1977. An experimental investigation of reddening in dune sand. Geoderma 17, 181 – 191.