Using Imaging Spectroscopy for the Quantitative Determination of Soil Iron Content in Partially Vegetated Areas

We determine the soil iron content in a semiarid region of Spain using airborne imaging, field and laboratory based spectroscopy. By assessing the influence of vegetation fractional and fully covered areas are separated for specific analysis each. The derived distribution of soil iron content is spatially discontinuous so spatially discontinuous quality indicators are assigned depending on the retrieval accuracy of iron content. The test site is located at the slopes of the El Hacho Mountain, near Álora in Southern Spain, showing a large variety in iron content. Orchards with olive trees are the dominating land use of this site. There are significant gaps between vegetation covered and bare soil patches, which are assessed using a spatial resolution that allows mapping of three categories, bare soil, partly vegetated and fully vegetated areas. Since the accuracy of the iron prediction depends on the gap fraction or fractional vegetation cover (fCover) of the orchard, a spectral unmixing based approach is used to discriminate several classes of vegetation cover. The iron content of areas with a low fractional vegetation cover can be determined accurately using a quantitative statistical based model. If the vegetation fraction is significant, accurate determination of the iron content is not possible, and consequently these areas need to be interpolated or excluded from the spatial iron map. For the areas with an intermediate vegetation cover, correction of the mixed spectrum is performed, making use of the vegetation fractions determined in the first step and a vegetation spectrum derived from olive trees. This results in a residual soil spectrum, which is used to determine the soil iron content with intermediate quality. The final result is a spatial distribution of iron content with prediction uncertainty. The user of the geographical database can determine which points may be used during further analysis, depending on the required accuracy. INTRODUCTION The slopes of the El Hacho mountain complex near Álora – Southern Spain show a large variety in soil iron content. However the role of iron in plant physiology is not fully understood iron is important for agriculture. It is involved in chlorophyll synthesis and is a component of other plant tissues [1]. Both excessive and deficient iron content of soils, and the form in which it occurs, can make agricultural crops vulnerable to both qualitative and quantitative damages. Furthermore the iron concentration, and directly connected to this the soil colour, is an indicator for the age of the soil, which has a large influence on the fertility. Determining the spatial distribution of different types of iron with traditional fieldwork and laboratory analysis is time-consuming and expensive [2]. Remote Sensing has proven to be a useful tool for determining the presence of iron in large areas and various research fields [3]; [4]; [5]. Large areas can be mapped and soils are left undisturbed during sampling. The influence of iron on the electromagnetic spectrum has been widely investigated and it is shown that quantification of the amount of iron with spectral measurements is possible e.g. [6]; [7]; [8]. However, research is often limited to practically bare soil areas. Vegetation has a large influ© EARSeL and Warsaw University, Warsaw 2005. Proceedings of 4th EARSeL Workshop on Imaging Spectroscopy. New quality in environmental studies. Zagajewski B., Sobczak M., Wrzesień M., (eds) ence on the soils reflectance spectrum, and results in highly inaccurate iron quantity estimations [9]. El Hacho Mountain consists of Tertiary Marls and sands, deposited on a continental slope, which are indicated with the generic term flysch deposits. On top thick cemented deposits of sand and gravel (conglomerate) occur [10]. Conglomerate blocks are now scattered over the flysch slope, where Cambisols are formed. Behind the blocks unweathered flysch is exposed. Iron content is varying with the distance down slope from these blocks, due to variation in erosion, iron deposition and leaching. The presence of gullies further increases the spatial heterogeneity. Hematite is the major iron mineral in the area, while just behind the conglomerate blocks some goethite may be present. METHODOLOGY AND MATERIALS Soil Samples In summer 2003, 35 bare soil plots are sampled and measured for iron content. The samples are positioned in two down slope transects and cover the variation over the slopes. Fractional cover of soil, rocks and vegetation is determined and a mixed sample of the topsoil (0-2 cm) collected. In the major study area for all plots, four field spectra and samples for determination of iron content and measuring laboratory spectra are taken. The plots are divided in a training set (19 plots) and a reference set (16 plots), to allow cross-validation of the results. The total iron concentration of all 35 soil samples is determined using a dithionite extraction [11]. The concentration of iron in the extraction fluids is determined with an ICP-AES. Field and laboratory spectra Field and laboratory spectra are acquired with an ASD Fieldspec Pro FR, covering the 350 – 2500 nm wavelength region. For the field spectra four measurements per plot are taken of the surface. The field of view was 25° and measurements are taken from nadir at a distance of about 30 cm from the surface. The soil samples used for the laboratory analysis are air-dried and sieved at 2 mm. The incidence angle of the lamp is set to 30° of nadir and a 3° fore optic is used at a 30 cm distance from the target. The sample is rotated 90° in between the four measurements, which are done per sample. For further analysis these reflectance field and laboratory spectra are limited to and rescaled to the bandwidth of ROSIS. Image data Airborne hyperspectral data are collected in June 2001, during the DAISEX campaign [12]. ROSIS is build for the detection of fine spectral structures especially in coastal and inland waters [13], but it is also used for land applications[14]. The processing of ROSIS images to reflectance values is done by DLR, using the ATCOR model[15]. Due to the coarse resolution of the used DEM for ortho rectification, the image geometry within the test area had high distortions. The use of a handheld GPS for determination of the location appeared to be inaccurate too. Therefore the positioning of the plots on the ROSIS image is done visually. In this way correct spectral information was achieved for the sampling points. This has no influence on the spectral information and further analysis. Iron content Laboratory and field measurements are used to establish the regression between total iron content and reflectance in the visible and near-infrared wavelength range. The properties of the absorption dip in the visible part of the electromagnetic spectrum are used for the quantification of the iron content. Continuum removal [16] is used to normalize the reflectance, so the total absorption, described with the area of the absorption dip around 550 nm (referred to as: Area D550) and the Standard Deviation of the continuum removed values in the same wavelength range can be used to estimate the iron quantity. In order to determine the most suitable technique for the final iron map-