The Potential Of Remote Sensing For Monitoring Rural Land Use Changes And Their Effects On Soil Conditions

Share on facebook
Share on twitter
Share on pinterest


Remote sensing is traditionally used for monitoring changes of land cover and land use, whereas the direct detection and quantification of soil conditions as function of these changes is still considered to be in a pre-operational stage. Nevertheless, soils form an important element for remote sensing of terrestrial ecosystems. 

Soils in the context of environmental remote sensing

Because some parts of the solar radiation penetrate even dense plant canopies, soils intercept much of the solar energy incident on such surfaces, thus taking an important role at the interface of clamato-logical and bio-physical land processes driven by the absorption and reflection of solar radiation. Soils also influence the reflectance of composite land sur-faces, particularly in regions with sparse vegetation cover.

The nutrients contained in soil substrates and their water buffering capacity constitute important re-sources for the growth of agricultural crops and natural vegetation. Conversely, the degradation of soils has an enormous effect on the regeneration and further development of the permanent vegetation cover. Resulting changes in soil properties control-ling soil and sediment buffering capacity, with sub-sequent changes of the mobility of chemicals contaminants., also produce effects on flora, fauna and human health.

Land degradation processes, which imply a reduction of the potential productivity of the land e.g., soil degradation and accelerated erosion, reduction of the quantity and diversity of natural vegetation., are widespread in Europe. They result from a long his-tory of human pressure upon land resources as well as from interactions between varying climatic characteristics and ecologically unbalanced human interventions and are consequently often related to rural land use changes. Although ecological processes, related to these changes and their implications for the future development of ecosystems, are intensely studied at many field sites, it is not fully clear how findings from field studies at patch-scale can be extrapolated and upscaled to relatively large areas. However, it is believed that remote sensing approaches can significantly contribute to solving this problem, as they hold the potential to perceive, map and monitor operationally at regional scales the ex-tension and effects of land degradation and how they might develop and evolve with time. Both mapping and monitoring represent important pre-requisites for defining and implementing political decisions and development plans for the protection and sustained use of land resources Hill et al., 1995b..

However, although it is agreed that remote sensing provides a convenient source of information, the problem is that the data collected by these instruments usually do not directly correspond to the information needed. Therefore, the detected signal, resulting from the interaction of electromagnetic energy with remote objects, must be interpreted in terms of the properties of these remote objects Verstraete, 1994..

Reflective properties of soils

The observation and mapping of soil conditions through optical remote sensing is restricted to sur- face reflectance that can be retrieved from the radiation measured by a sensor system. Because the characteristics of radiation reflected from a material are a function of its physical chemical properties, the observation of land surface re-flectance principally carries information on the properties and the state of the topsoil. This means in turn that only effects can be observed and mapped which cause significant changes to specific surface characteristics.


The spectral reflectance of soils is a cumulative property which derives from the inherent spectral behavior of heterogeneous combinations of minerals, organic matter and soil water absorptions and reflectance changes in the 400–2500 nm range of the reflected electromagnetic radiation provide diagnostic features, which can be used to identify directly important mineralogical constituents of soils such as iron oxides e.g., hematite, limonite, goethite. clay minerals e.g., illite, kaolinite, montmorillonite. or carbonate e.g., calcite, dolomite. These diagnostic features result from electronic transitions and molecular vibrations in the crystal lattice of minerals. For instance, transition metals often cause electronic processes resulting in absorption features at short wave-lengths. At longer wavelengths, minerals bearing hydroxyl and carbonate groups show absorption features because of molecular vibrations of OHy e.g., clay minerals at 2200 nm. and CO 32y e.g., calcite at 2300 nm. groups.

Furthermore, soil mineralogy is often interrelated with soil texture and organic matter content, and soil reflectance, in addition to specific absorption features, is largely characterized by the spectral re-flectance continuum, i.e., the overall spectral shape and the albedo. Numerous studies describe the relative contributions of various parameters such as organic matter, soil moisture, particle size distribution, soil structure, iron oxides and soil mineralogy to the above-mentioned reflectance continuum determined five distinct soil reflectance types which can be identified by curve shape and the presence or absence of spectral absorption features related to above-mentioned soil constituents.

Although there exists a large variability in the spectral response of common soil types, there is clear evidence that most of the known soil spectra resemble one of these five reflectance type curves.

Both progressive and regressive pedogenesis cause alterations to the soil surface, which to a certain extent are spectrally detectable and fit into the typology of Fig. 1. For most soils of the Mediterranean e.g., cambisols, fluvisols, luvisols, vertisols, regosols., the authors could show that the brunification or rubefication and the organic matter content of the topsoil provide the most important diagnostic features for a spectral identification of developed soil substrates. Fig. 1 shows that reflectance spectra of such soils generally correspond to spectral types C and D, and spectral measurements of undeveloped or disturbed soils e.g., leptosols. frequently resemble type B and exhibit spectral absorption features which are characteristic for the mineralogy of the parent material. These principles seem to provide a widely applicable framework for relating spectrally detectable surface phenomena to soil conditions, thus satisfying an important requirement for the successful application of remote sensing techniques.

Use of remote sensing for land use inventory and land cover mapping—traditional approaches

In the past decade, satellite remote sensing has developed rapidly and it has become a practical tool for monitoring the environment and assessing natural resources in a number of applications. Operational to

pre-operational remote sensing programs for land use and crop inventory as well as for land cover mapping have been implemented at a national level in Europe.

Operational remote sensing data is of particular importance in supra-national environmental and agri-cultural inventories established by the EU. In this context, especially the European Commission’s Co-ordination of Information on the Environment CORINE. and Monitoring Agriculture with Remote Sensing MARS. projects have attracted the attention of the international user community.

Under CORINE, the land cover of the EU member states is being mapped from earth observation satellite images at an original scale of 1:100 000 aims to support the CEC Directorate General for Agriculture DG VI. and the CEC Statistical Office EUROSTAT. in the following three major fields 

  • The improvement of agricultural statistics through the use of remote sensing techniques.


  • The use of remote sensing and related techniques to support the implementation of the Common Agricultural Policy.
  • To support transfer of methods to other geographical regions outside the EU and the adaptation of

these methods to new themes.

The above-mentioned projects rely essentially on use of remotely sensed data for land color mapping and statistical inventories, which according to Hill 1993a. implies a number of significant differences.


Land color mapping from satellites aims to pro-duce cartographic documents and the main emphasis is consequently placed on high accuracy of local mapping. This requires very precise interpretation of each single pixel.

Satellite-based statistical inventory uses statistical techniques for improving the precision and extension of traditional i.e., ground based. surface area estimates through the use of space images. Ground observations from specifically designed area frames are used to trigger the automatic classification of the imagery. Ground data and classified images are then used in so-called regression estimates, which are considered capable of providing significant spatial extensions to conventional ground surveys. In regression estimates local image classification accuracy i.e., local mapping precision. is no longer of primary importance, because the degree of correlation between ground surveyed and image classified cover proportions in the sampling units determines the quality of the inventory.

Within this framework, automatic land cover mapping by supervised or unsupervised classification techniques Thomas et al., 1987. has been tradition-ally considered a primary technique for obtaining objective maps of the earth’s surface. It is acknowledged that broad land cover types can sometimes be satisfactorily mapped at a local level with single date images. However, the similarity in spectral re-flectance properties of natural surfaces impedes consistent identification and mapping of a wider range of cover classes, such as agricultural crops or natural vegetation communities. In addition, the spectral confusion between cover types varies throughout the year: classes which appear very similar in spring may become separable at earlier or later stages of the annual cycle. It is therefore expected that multi-temporal approaches, implying repeated satellite observations at different dates, will provide the means for obtaining more detailed results at regional levels, particularly under the variable environmental conditions of Europe.

 gives a typical example for the production of accurate land cover maps and statistical inventories with multi-temporal earth observation satellite images.

However, besides well-known methodological problems i.e., spectral class confusion as function of image acquisition dates and phenological conditions., classification approaches cannot really provide objective land cover maps because of the required human intervention in highly interactive subprocesses such as spectral class definition. Furthermore, because of normal variations of the annual phenological cycle, spectral class descriptors also cannot be directly applied to comparable scenes from other years. New sets of training statistics must be developed for each new satellite image, and this training is tedious, time consuming and requires the services of an experienced image analyst. Objective monitoring is almost excluded because of the difficulty in com-paring the updated mapping results with reference documents from earlier dates. Mapping of soil features is even more difficult because multi-spectral classification is not able to handle efficiently second order spectral variations, such as soil moisture differences, partial vegetation cover or illumination effects.

Another important, widely used approach to multi-temporal analysis of land cover and environmental changes are remotely sensed indices, of which the major part is related to the characterization of the vegetation cover. Because plants have a distinct spectral signature with low reflectance in the visible 400–700 nm. and very high reflectance in the near infrared region 700–1200 nm. of the solar spectrum, attempts have been made to exploit this spectral contrast for identifying the presence of green vegetation and evaluating some characteristics e.g., cover and biomass. through various so-called vegetation indices, such as the Normalized Difference Vegetation Index NDVI. Although images of vegetation indices exhibit significant structure and are well acknowledged within the remote sensing user com-munities, much concern has been expressed about their sensitivity to the influences of atmosphere, illumination and observation geometry and to the background reflectance of soils and parent material. In fact, a study by Price 1993. revealed that the NDVI, calculated for a large range of laboratory-measured soil spectra 564 samples., produced a statistical average value of 0.207 with a variability of “0.1. These values also cover most of the value range of the NDVI, which is considered characteristic for sparse vegetation covers, e.g., of semiarid regions. The fact that band ratios and NDVI are not optimized for detecting vegetation in case of spectrally variable soils was also confirmed through the examination of Landsat-TM images of arid regions.

Alternative indices have been proposed which should better account for reflectance contributions from background substrates and which are less sensitive to atmospheric and illumination conditions However, as the soil corrected indices implicitly or explicitly employ ‘average’ soil spectra, it is still difficult to compensate fully the effects that result from spatially variable substrate reflectance.

New methods for monitoring land cover changes and their effects on soil conditions

Although, under ideal conditions, even subtle spectral features can be identified and classified, the major obstacle for using remote sensing data for land cover and soil observation is the difficulty in consistently interpreting surface spectral characteristics under a wide range of environmental conditions. Additionally, most land surfaces are covered with varying amounts of vegetation so that remotely sensed spectra usually represent a mixture of soil and vegetation components.

Conceptual framework for monitoring changes of land color and soil state

Data about the detector sensitivity i.e., calibration coefficients. allow the encoded digital numbers i.e., image grey values. of remotely sensed images to be converted into values of radiance, and radiative transfer calculations can be used to correct for atmospheric effects, such that the surface-reflected radiation is restored from the remotely sensed signal. Dividing this term by the downward solar irradiance provides the important primary parameter of ‘bidi-reactional’ reflectance.

Although of relevance to the global radiation budget, albedo and reflectance changes per se are not direct indicators of land surface processes, in particular when spatially complex areas like Europe are considered. As already noted, a simple change in albedo reflectance does not necessarily imply relevant changes. It is therefore not sufficient simply to map albedo reflectance or its changes over time, but to infer the environmental impact of reflectance bed changes by characterizing their physical nature in terms of land surface conditions. Thus, albedo and reflectance changes must be analyzed in comparison with the multi-spectral characteristics of a range of

known surface materials. Consequently, an appropriate scene model which can be used to convert multi-spectral reflectance into thematic information Fig. 3. is required. Here, the question is not so much to identify a particular surface type, but to characterize it A variety of methods have been proposed which range from empirical spectral indices to the design and inversion of physically-based models. Although the applicability of the various approaches depends on the nature and accuracy of the desired information and the availability of resources e.g., sensor characteristics., an important prerequisite for their operational use is that they must satisfy specific requirements in terms of standardization and that they must be transferable to different regions.

Both the development of suitable indices and their interpretation in the thematic context of land use and degradation monitoring require a conceptual frame-work which allows the drawing of concise conclusions about the land surface conditions. Although these underlying concepts vary as a function of regional ecosystem characteristics i.e., depending on physiographic conditions such as parent material, aridity, etc.., the results from different regions can be consistently evaluated on a higher hierarchical level by using both image-derived and ancillary in-formation layers implementing a GIS based ecological stratification. Important conclusions will nevertheless depend on the capacity to analyze multi-annual time series, and it is for this reason that retrospective studies are so important with respect to developing methods for continuous monitoring of environmental changes.

Spectral mixture analysis (SMA)

These conceptual considerations require information extraction methods which provide largely unbiased estimates for green vegetation cover, permit the identification of soil related spectral information and allow sufficient standardization for multi-temporal monitoring. It has already been argued that traditional multi-spectral classification approaches as well as most vegetation indices are not ideally suited to fulfil these requirements. Because the inversion of physically-based bidirectional re-flectance models against satellite data is not feasible with currently available data sets, attention is drawn to suitable semi-empirical models.

One of the most promising approaches has be-come known as ‘Spectral Mixture Analysis’. Spectral mixture analysis SMA. assumes that most of the spectral variation in multi-spectral images is caused by mixtures of a limited number of surface materials, and it attempts to model the multispectral reflectance as a mixture of representative ‘prototype’ spectra, the so-called ‘spectral endmembers’ i.e., vegetation, soil and bedrock components, ‘shade’ as an illumination component, etc. Linear mixing within the foot-print of a multichannel spectrometer or imaging sensor further assumes that the surface components are large or opaque enough to allow photons to interact with only one component, i.e., the radiative transfer processes are additive. Spectra can then be ‘unmixed’ by inverting the linear mixing equation:

A practical example of the application of this methodology to the mapping of soil degradation in semi-natural ecosystems of the Mediterranean is de-tailed by Hill et al., 1995a,c and is outlined here. The presented approach is principally based on the definition of soil condition as a function of the mixing ratio between developed soil and the related parent material, which have to be spectrally distinct. If this is the case, the spectral characterization of these mixtures is expected to provide the basis to obtain maps of soil condition from remote sensing images. In a first step, spectral unmixing was applied to a representative number of soil reflectance spectra taken at 99 field locations in the test site. According to taxonomic criteria, soil depth and surface constituents, these soils had been assigned to the following soil condition levels: I. fair; II. degraded; III. severely eroded; IV. exposed parent material 

Subsequently a ‘four endmembers’ mixing model Fig. 5. was defined, which was applied to model the soil state of complete AVIRIS imaging spectrometer and Landsat-TM frames of the test site. It was thereby assumed that the camisole spectrum EM2 represents predominantly the spectral response of well-developed soils in the area, whereas degraded soils spectrally respond by increasing fractions of the related parent rock endmembers EM3r4, photosynthetically active vegetation in the image data. An additional ‘artificial’ endmember ‘shade’ was introduced to isolate the influence of shading and shadows which relate to vegetation and soil rock roughness elements, topography and solar elevation. Its spectral characteristics were approximated with a continuous spectrum of zero reflectance.

All  these  reference  spectra were  taken from a library of high spectral resolution Spectro radiometric measurements built up according to the above-mentioned spectral library concept. The application of this endmember set to the analysis of the image data of course requires prior atmospheric correction. As an intermediate result abundance images for each endmember can be calculated with the above-de-scribed SMA algorithm. After removal of the shade fraction and masking of pixels with more than 50% vegetation cover, thus normalizing and rescaling the soil rock dominated pixels i.e., vegetation cover 50%., the relative abundances of developed soils and parent material provide a measure of soil condition in terms of erosional degradation. A soil condition map was then obtained through a combination of cluster analysis and automatic classification of the renormalized fraction images in the presented example, the method was also as applicable to the high spectral resolution imagery of imaging spectrometers e.g., AVIRIS, MIVIS, as to Landsat-TM images, because the major spectral contrast within the endmember set results less from the over-all spectral continuum and more from narrow absorption features.

3.3. Remote detection of the impact of soil contaminants

Apart from changes to topsoil properties related to soil erosion, which can be spectrally mapped under specific conditions, rural soils are often affected by the deposition of soil contaminants such as pesticides and heavy metals and these are not directly detectable with remote sensing methods. Further, as a combination of contamination, land use and climatic changes, soil properties such as pH, redox potential, exchange capacities, microbial activity, organic matter and salt content may change resulting in a reduction of soil productivity. Again, with the exception of the last two parameters, no direct spectral detectability of these effects is possible.

However, it is known that vegetation reflectance is affected by physiological stress as induced by soil contaminants and can thus be considered a secondary indicator of changing soil properties. These reflectance changes are extremely subtle and thus in turn require high spectral resolution measurements at the highest possible radiometric accuracy, which implies the use of imaging spectrometry.

Although the spectral reflectance of fresh vegetation was characterized more than 30 years ago by Gates et al. 1965. and the first high spectral resolution imagers were used for plant stress detection in the early eighties by Collins et al. 1983., the application of imaging spectrometry to the retrieval of quantitative information from leaf canopies is still young and has to be considered experimental Goetz, 1991. Nevertheless, promising examples for the detection of vegetation stress, e.g., from heavy metal contamination of soils covering natural ore deposits and former waste dumps show the potential of imaging spectrometry for the detection of soil contaminants.

In a recent study, Rothfuss 1994. demonstrated a clear correlation between spectral anomalies of agricultural vegetation rape, rye, maize. grown on the area of a former waste deposit, and heavy metal anomalies. The spectral anomalies were derived and mapped from atmospherically corrected, high spatial resolution airborne GER 63-band imaging spectrometer data.

Geochemical sampling and analysis of heavy metals was performed, both of the soil and of the plants, over the former waste deposit which was closed and filled in 1949. The concentrations of Lead approx. 500–3250 ppm., Zinc approx. 1500–2500 ppm. and Copper 500–1000 ppm. were clearly increased in the soil, whereas in the plants only Zinc was concentrated to a critical level 140–170 ppm of dry substance.

Four spectral anomaly classes representing different degrees of anomaly were defined according to an increase of reflectance ranging from 3% in the VIS 400–700 nm. rSWIR 2000–2500 nm. to 5% in the NIR 700–1300 nm. spectral regions. These in-creases of reflectance were correlated with an in-crease of Zinc concentration in the plants and in turn with lower contents of chlorophyll and carotenoids. The classification was performed according to the range of standard deviation of the image pixels relative to the four anomaly classes. In the case of multiple assignments, a match score, based on the average deviation of the pixel from the matched anomaly spectra, was used as a criterion in the class assignment. The mapped vegetation spectral anomalies Fig. 7. show the same distribution pattern as those obtained from geochemical and geophysical field surveys.

The results of this study demonstrate the potential of hyperspectral remote sensing data to contribute to classification of spectral anomalies of rape and rye grown on a filled waste deposit. The classification was performed according to the spectral match of GER 63 band imaging spectrometer pixels with heavy metal affected reference spectra. The highest spectral anomalies, marked with the white pointers, are related to Zinc anomalies. The factor of the grey scale is related to the standard deviation of pixel and reference spectra Roth fuss, 1994..

the survey of spatially distributed soil contaminants under vegetation cover using the spectral response of the vegetation as a secondary indicator. Especially over agricultural areas with their relative uniform vegetation cover, the use of this technique for recon-naissance and monitoring of questionable contamination areas may help to reduce and optimize cost intensive geochemical geophysical field surveys. In the future, new airborne hyperspectral instruments like DAIS Digital Airborne Imaging Spectrometer, operator DLR, Germany. and MIVIS Multispectral Infrared and Visible Imaging Spectrometer, operator CNR–LARA, Italy., both including also thermal infrared channels Bechtel and Sommer, 1994., are expected to improve the analytical capabilities of remote sensing.

Present work perspectives

Regional patterns of land cover, soil degradation, erosion damage and vegetation change can be reliably mapped and monitored with earth observation satellites and airborne systems. The presented new techniques, which are primarily based on linear spectral mixture analysis and hyperspectral data, provide results at a level of precision and spatial detail which, because of methodological and financial constraints, are difficult to obtain by conventional remote sensing or mapping approaches. Further verification experiments under variable environmental conditions are still required to draw a final conclusion on the future operational applicability of these new methods. Some of these studies have already been initiated, e.g., Demon 2 and MEDALUS III within the 4th framework programme 1994–1998. of the European Commission. If results confirm the conclusions drawn from the first pilot studies, these new approaches may contribute substantially to environmental monitoring in Europe.

To better understand the processes of land degradation in a spatial context, it is essential to monitor soil conditions and the disturbance regime of plant communities over time, including their successional recovery. The available data archives Landsat MSS since 1976, TM since 1983. are important for retrospective studies which may provide the key for understanding the present situation, and for determining the frequency of regular monitoring.Shows a flow chart for a proposed schematic interpretation of remote sensing data in a future operational environmental monitoring system. The proposed scheme consists of several standardized modules characterized by the ellipsoids in .. Radiometrically rectified satellite images provide the primary parameter _ for a given control site, followed by spectral mixture analysis SMA. to convert the multispectral surface reflectance into soil and vegetation related information layers. The ‘synoptic interpretation’ module uses the intermediate information layers. soil condition index, projected foliage cover. for computing an index of degradation and susceptibility to land degradation processes, and the comparison of susceptibility indices from different years will provide evidence of either degradation, stability or recovery on a regional scale. However, an important objective of the thematic interpretation is to separate the rhythmic phenological changes of growth and senescence from episodic alterations introduced by climate or human induced intervention. This also avoids direct comparison of images from different seasons to exclude artefacts from phenological effects, and to minimize radiometric distortions which result from bidirectional effects and illumination differences.


It has been illustrated how remotely sensed primary parameters, such as the spectral surface reflectance, can be converted into a standardized characterization of soil and vegetation conditions. In this context, the term ‘thematic concept’ has been introduced, which is the conceptual background for identifying functional links between surface re-flectance, vegetation and soil characteristics. Such a concept has to be primarily based on research in geosciences and ecology including careful validation based upon field data. It should be valid not only for specific local environmental conditions, but also for regions, being possibly subdivided according to specific physiographic characteristics e.g., bioclimate, lithology, topography, etc.


The key requirements for environmental research and monitoring can be summarized as follows: a. Remote sensing approaches need to be based on primary parameters, being related to physicochemical properties of surface material. b. Techniques need to be portable to different regions with comparable conditions and repeatable independently of the individual analyst in order to allow for comparable analysis, including those of a time series. c. Data collected by remote sensing normally do not directly correspond to the information needed for predictive models. The information therefore has to be thematically interpreted. d. ‘Thematic concepts’ are needed to help understand the conceptual background for identifying functional links between surface re-flectance and vegetation and soil characteristics. The strength of remote sensing is rather the potential to quantitatively map and monitor changes than to provide input parameters to predictive models. f. Time series analysis may be important to validate and ‘calibrate’ models.