Digital Soil Mapping with Limited Data

Digital Soil Mapping (DSM) can be defined as the creation and population of spatial soil information systems by numerical models inferring the spatial and temporal variations of soil types and soil properties from soil observation and knowledge and from related environmental variables. DSM is now moving toward the operational production of soil maps thanks to a set of researches that have been carried out for the last fifteen years. These researches dealt with various topics: the production and processing of soil covariates, the collection of soil data, the development of numerical models of soil prediction, the evaluation of the quality of digital soil maps and the representation of digital soil maps. The recent advances and open questions within each of these topics are successively examined.

The emergence of DSM as a credible alternative to fulfill the increasing worldwide demand in spatial soil data is conditioned to its ability to (i) increase spatial resolutions and enlarge extents and (ii) deliver a relevant information. The former challenge requires to develop a specific spatial data infrastructure for Digital Soil Mapping, to grasp Digital Soil Mapping onto existing soil survey programs and to develop soil spatial inference systems. The latter challenge requires to map soil function and threats (and not only “primary” soil properties), to develop a framework for the accuracy assessment of DSM products and to introduce the time dimension.

This chapter reviews some hardware and software for digital soil mapping. By

hardware

we mean various kinds of sensor and instrument which can give us better soil and

scorpan

data, and by

software

we mean mathematical or statistical models that can improve our spatial predictions. There are two approaches for the development of hardware for acquiring soil information: the top-down, and the bottom-up. The top-down approach asks which technologies are available and which variables can we measure that are related to

scorpan

factors. The bottom-up approach starts from a problem that we systematically analyse so as to identify the information that is needed to solve it. We then tackle the technical problems of collecting this information, and only at the end move to developing the field technology. We evaluate various software approaches to improve spatial prediction of soil properties or soil classes. Finally, the implication of using data-mining tools for the production of digital soil maps is discussed.

The question has to be asked why – given the substantial advances in quantitative techniques over the years – ‘full’ Digital Soil Mapping has not been mainstreamed further and harnessed to the problems soil information can help address. This paper suggests some reasons for a slow adoption, causes for optimism for a wider adoption than at present and – using a case study – demonstrates the ease of further development at national scale. Finally, we propose how a major effort of digital soil mapping could support development in Africa, outlining the opportunities and obstacles that await contributors.

In this Chapter we describe our use of digital soil mapping estimates as input to traditional field soil survey in California, U.S.A. We also describe the development of these raster soil property models as stand-alone products, and practical implications of their use. This Chapter deals with application of existing digital soil mapping tools in active soil surveys, rather than research of new methods.

The soil survey program in the United States is nearing completion of “once-over” coverage of the nation. Many potential soil survey users in the remaining unmapped areas expect to use traditional polygon-based soil maps.

Soil-landscape models based on field point data have been developed in support of selected soil survey projects. We expand on our previous models in a test area that has existing point data and polygon soil mapping. New soil-forming factor covariates (IFSAR elevation data and ASTER satellite images) are used to derive the models. Minor improvements in the model estimates were obtained. Then significant variables from these models are used to test the feasibility of the creation of field soil survey office tools.

We feel that raster soil-landscape models are a developmental product of soil survey. They are just becoming useful as pre-mapping estimates of the spatial distribution of some individual soil properties. The explicit estimation of all significant soil properties based on a suite of individual models is still to be developed. This is required before informed land management decisions can be based on digital soil mapping.

Since natural resource management methods and regulations are coordinated locally, regionally, and nationally, standards for the creation and implementation of these models are required for consistent and coordinated outputs within a nation.

Little effort has been made to link soil mapping and soil data density to a nation’s welfare. Soil map density in 31 European countries and 44 low and middle income countries is linked to Gross Domestic Product (GDP) per capita and the number of soil scientists per country. National coverage of exploratory soil maps (>1:250 000) is generally higher in the poorest countries and decreases with increasing GDP per capita, whereas the national coverage of detailed soil maps (<1:50 000) tends to increase with increasing GDP. GDP is larger in countries with more soil scientists per unit area, likewise, the number of soil scientists increases with increasing GDP. More soil scientists per ha of agricultural land was found to be related to higher crop yields. Obviously, there are many confounding and interacting factors but this analysis illustrates how proxies for soil map density can be used; it is suggested that appropriate indicators should also be developed for spatial data infrastructures and digital soil maps to demonstrate their effectiveness for society and human welfare.

This chapter introduces the concepts of data

rescue

of legacy soil surveys, here defined as a simple conversion to archival format by scanning or direct entry into a database, and data

renewal

, here defined as the process of bringing these surveys up to modern standards by taking advantage of technological and conceptual advances in geoinformation technology. This is especially important in areas with sparse soil data infrastructure, as it is both more likely that the data will be lost and less likely that a new survey can be commissioned. Digital Soil Mapping (DSM) techniques, although designed for new surveys, can play an important role in data rescue and renewal, in particular as geodetic control for a GIS coverage, as a medium-resolution elevation model (DEM) and derived terrain parameters to adjust terrain-related boundaries, and synoptic satellite imagery to adjust vegetation or landuse-related boundaries. The semantic issues raised by soil-landscape modelling within DSM are especially important for data renewal and integration with supplementary surveys. As with DSM in general, a data renewal exercise may require cultural and institutional change in traditional soil survey organization.

Digital Soil Mapping to capture or determine categorical or property information has undergone a tremendous increase in capability and application during the past decade. Many successful technologies have been developed through research activities worldwide, including generalized linear models, classification and regression trees, neural networks, fuzzy systems, expert systems, and geostatistical methods and applications. These technologies have matured beyond a research activity and have potential for use by soil scientists to more accurately, consistently, and efficiently define soil categories and soil properties based on digital proxies to soil-forming factors. These applications for producing soil maps are now poised to become production tools to either update older soil survey information or to produce soil information on previously unmapped areas.

As these technologies move into the mainstream for producing soil survey information, there are challenges that must be overcome. The community of soil scientists and soil classifiers engaged in producing soil information must become familiar with the technologies and their potential uses and limitations. More importantly, the users of soil survey information must be convinced of the relevance and applicability of maps and data that appear different from the “traditional” products with which they have become familiar. New challenges include developing acceptable standards and procedures for the production and quality control and interpretation of the information that relates to agricultural, engineering, forestry and other soil-landscape uses.

Brazilian soils are generally acidic with low base saturation and low plant available potassium (K). Potassium fertilizers play important role in production costs and farmers receive no governmental subsidies. Strategies are needed to improve potassium fertilizer delivery to different regions in Brazil and to establish affordable prices and balanced potassium consumption. For such strategy, it is necessary to take into account the different soil classes with its varying K levels. The purpose of this study was to map soil K in Brazil considering the different biomes and applying techniques to reduce problems caused by limited soil profile data. A soil profile data set was constructed from the soil archives of Embrapa Soils, Rio de Janeiro, Brazil. Descriptive statistics was performed on K levels in different soil classes and biomes. The different soil K levels were grouped in intervals and mapped using ArcGIS 9.1 tools from ESRI. Brazil’s soil map and biome map at 1:5,000,000 scale were used in the geoprocessing. Our results showed that mapping soil K levels based on soil survey reports at the regional scale is difficult because of limitations in georeferencing and spatial distribution of soil profiles. However, this mapping would help fertilizer distribution planning in Brazil.

Traditional soil surveys follow a specific methodology to identify, characterize, and fit mapping units in a classification system and to spatialize them in order to produce soil maps. The need for observation and characterization on field, associated with the physical and chemical analyses, makes the surveys expensive and therefore scarce. The low number of surveys stimulated the development of models for digital soil mapping, whose results proved to be possible to predict and spatialize many soil characteristics. However, conventional soil surveys remain important as a basis for the development of digital soil mapping models, setting a reason to continue the development of methodologies to improve the conventional surveys. Technologies like GPS and GIS contribute to make field observation and soil sampling more objective and make the mapping process and the production of hardmaps easier and faster. The objective of this study was to develop methodologies to integrate cartographic base elements with field work, using GIS and GPS in an area corresponding to 20 topographic charts in scale 1:50,000 in the State of Rio Grande do Sul, Southern Brazil, to obtain soil mapping based on the Brazilian Soil Classification System. The result obtained was a georeferenced digitized soil map, continuous for the whole region, free of inconsistency among neighbor map sheets and with attributes associated with the mapping units. These characteristics allow the use and application of the soil map for many purposes like zoning, diagnosis, suitability analysis as well as serving as a basis to the development of models for digital soil mapping.

This chapter considers both opportunities and constraints to applied, operational digital soil mapping (DSM) from the points of view of a) availability of suitable input data layers, b) protocols available for DSM, c) quality of input data layers and resultant output maps and d) other efforts required to build predictive mapping capacity and apply it effectively.

Many potential DSM practitioners are discouraged by the real or perceived lack of availability of suitable input data layers to support DSM, particularly in regions with weakly developed spatial data infrastructures. Solutions to addressing problems of limited or sparse spatial data sets are identified for input layers derived from digital elevationmodels (DEM’s), remotely sensed imagery and available secondary source maps.

A variety of protocols for producing predictive soil maps are discussed under the general headings of unsupervised, supervised and knowledge-based (or heuristic) approaches. These key protocol activities support the ability to make maps of the spatial distribution of soil classes or attributes by developing predictive relations between spatially distributed input variables or classes and the desired output classes. Different strategies are reviewed to acquire and formalize tacit knowledge embodied in soil-landform conceptual models and to capture this tacit knowledge as quantitative rules.

A considerable amount of resistance to DSM arises from real or perceived concerns about the quality of the resulting maps in comparison to existing maps produced using traditional mapping methods. Quality, defined as the ability of a map or product to correctly predict the characteristics of the landscape at particular points or within particular small areas, is discussed as are suitable approaches for evaluating and reporting it.

The capacity to applyDSMroutinely and operationally requires additional support in the form of training, access to suitable tools and software and access to suitable input data. Approaches to developing support for DSM from decision makers and funding agencies in the face of institutional and discipline resistance to embracing new technologies are identified, specifically incremental projects with clearly defined goals and testable measures of success. Finally, it is noted that perhaps the biggest hurdle to building capacity is our own hesitancy to believe in ourselves and to dream big and try big. It is hoped that this chapter will encourage individualswith an interest in applying new predictive mapping techniques to embrace change and to try to create useful, operationalmaps for large areas in their own regions of interest.

The re-use of legacy soil data together with increasing numbers of environmental co-variables becomes increasingly more interesting in digital soil mapping at intermediate scales, in areas with limited data. This poses important issues regarding the reliability of these data as well as of the final product of mapping. It also requires that the data and the manner in which they are (re-)used do not have a negative influence on the quality of the mapping product. Existing quality management approaches in soil mapping emphasise the producer perspective. In addition, rather than being preventive in nature they mainly rely on detection of defects in end-product testing. A shift is required from a focus on the quality of the end-product of mapping to quality control of the mapping process itself. The development of a framework for soil data quality management is proposed in this chapter.

Land evaluation is the prediction of land performance over time under specific uses. These predictions are then used to guide strategic land-use decisions. Modern land evaluation has a 30-year history, yet the results are generally accepted to be disappointing. Land users and planners are inclined to ignore land evaluators, reflecting the poor quality and low relevance of many actual land evaluations, as well as poor communication with users. The main objective of this research was to improve use and usefulness of information for rural land use decisions based on an operational demand-driven approach for land evaluation with case studies in Santa Catarina State, Brazil. First, the use and usefulness of soil surveys and land evaluation reports to land use planners were described and quantified and the relation between latent demand and actual supply was observed. Then, the farmers’ decision environment and its implications for land evaluation were studied. These were the basis for the subsequent steps of this research. Next, the applicability of a data-intensive distributed environmental model (AgNPS) in a relatively data-poor environment was evaluated. This model and other tools for visualization of scenarios were used with community participation, to test their effects on collective understanding of shared environmental problems. Finally, the potential of a participatory approach for integrating risk analysis into decision making for rural land use was evaluated. This research showed that a demand-driven approach to make the information more relevant and useful to rural decision makers for land use planning is possible in practice and should be further explored, but its effectiveness needs time to be confirmed. Applying the proposed approach, new demands were raised and considering that the number of soil scientists and financial resources are scarce in the region, digital soil mapping based on existing data emerges as a potential alternative to help to answer to the increasing rural decision makers demands.

This paper discusses the potential of soil diffuse reflectance spectroscopy (DRS) for rapid and cheap soil analysis and its application to digital soil mapping. We consider both visible-near infrared (vis-NIR) and mid infrared (mid-IR) spectroscopy, the use of multivariate calibrations, the development of soil spectral libraries and the cost and benefits of soil DRS. Finally, we conclude with some thoughts on the potential use of the techniques for digital soil mapping and soil science generally.

One of the fundamental parameters in the soil formation equation is that relating to the parent material from which the soils have been derived. Such information is typically derived from geological surveys and paper maps. However, an increasing propensity to directly produce digital geological maps and associated data bases means that a far greater range of information can be made available to assist the soil scientist in mapping and predicting soil characteristics. Such geo-information typically can include, detailed lithological parameters, geochemistry of soils and sediments, engineering parameters and remotely sensed information. In this paper we describe on-going work at the British Geological Survey in which we are actively developing a national digital parent material map and property data base at a scale of 1:50 000. The main aim in doing this is to support the development of national soil data sets at a similar scale by those responsible for soil survey in the UK. However, our experience to date suggests that an adoption of similar strategies in regions and countries with sparse, soil orientated, data infrastructures could be of considerable value. For example many countries have, or are receiving, aid in support of the development and licensing of mineral resources (i.e. Madagascar, Afghanistan and Mauritania) which include not only significant improvements in geological mapping and associated GIS infrastructure, but also remote sensing and geochemical survey.

Developments in GIS based technology have greatly aided the routine production of three-dimensional geological maps. Similarly the continued development of airborne remote sensing, geophysics and infrared measurement now provide tools that can assist in the mapping of soil structure and properties rapidly in 2D, 3D and even 4D. Whilst the combined use of such techniques have grown popular for performing site investigations and developing conceptual models of contaminated sites their use in determining and mapping soil has been restricted.

In this paper, we describe ongoing work at the British Geological Survey in which we have combined a variety of remote sensing, soil, geological and geophysical survey techniques to assist in the production of site specific, 3D digital soil models and geological maps. We were particularly interested in investigating (a) to what extent do methodological differences between the UK’s soil and geological communities hinder the development of an integrated near surface model (b) whether technologies to map geology in 3D can be used to develop spatial models of the soil; and (c) can technologies used in digital soil mapping assist in reducing uncertainties associated with such models at a range of scales.

To date we have found clear evidence that differences in terminology do hinder the development of linked models of the near surface environment; but that such differences can be resolved by dialog between field surveyors from each discipline at an early stage in the process. The GSI3D software used in this work performed well in this, relatively simple usage and a successful 3D model of the Brakenhurst surface environment was obtained. However our attempt to use digital soil mapping techniques was compromised by the relatively poor contrast in soil properties across this specific site. Further investigations across representative soil landscapes are being carried out that should address this issue and provide more insight into the adoption of digital soil mapping techniques at a local scale.

We propose that Landsat remotely sensed spectral data represent useful environmental covariates for digitally mapping soil distribution on the landscape, especially in arid and semiarid areas. Based on the common conceptual model that unique soils are the products of unique sets of soil-forming factors, Landsat spectral data can represent environmental covariates for vegetation (e.g., normalized difference vegetation index, fractional vegetation cover) and parent material and/or soil (e.g., band ratios diagnostic for gypsic and calcareous materials). In areas with sufficient relief, topographic data (e.g., slope, compound topographic index) derived from digital elevation models (DEMs) can be combined with Landsat-derived data to quantitatively model soil distribution on the landscape. These digital data can by analyzed using commercially available image processing software. Various classification and analysis methods (e.g., optimum index factor; principle component analysis; unsupervised and supervised classification) can be used to recognize meaningful soil-landscape patterns. . Training sites can be selected from existing soil surveys or from areas that have actual field data collection points. Accuracy assessment with independent field observation can be performed, and various classification methods can be used to generate estimates of prediction error. Landsat scenes are spatially explicit, physical representations of environmental covariates on the land surface. While the 30-m spatial resolution and fairly coarse spectral resolution may limit some applications, the wide availability and low expense should facilitate the utility of Landsat spectral data in digital soil mapping.

This paper compares two approaches for upscaling the Aisne (a 7,536 km

2

French department) soil database from the initial 1:25,000 nominal scale to the 1:250,000 target scale. Soil features are represented at the nominal scale, whereas pedolandscapes, which are a combination of soil-forming factors and soil variables, are required at the target scale. Because the initial soil database does not contain soil forming factor information, data on pedogenesis have to be added to the initial database. Based on the assumption that most of lithographic layers are horizontal in the area, only landform attributes are chosen to represent the soil-forming factors.

Two different approaches are used to map the final pedolandscapes. The first one, called the bottom-up approach consists of classifying the soil and the landform attributes together for defining taxonomic units, which then undergo generalisation of their contours to result in pedolandscape mapping units.

The second approach, called a top-down approach, consists of classifying and then mapping the landform units in order to delineate the pedolandscapes. In this paper, we focus only on the pedolandscape delineation for the target scale. The results of the two methodologies are compared to contours manually drafted by soil surveyors. The final discussion analyses the impact of taking the very detailed soil database in the Digital Soil Mapping process into account, and to give advice for digital soil mapping with limited input data.

In this chapter we present an automated approach to correct the delineation of nominal soil and environmental datasets based on auxiliary metric attributes, aiming to enhance positional accuracy. The detection of uncertainties is based on different spatial and non-spatial approaches. The methodological framework mainly consists of nearest neighbour approaches and comprises supervised feature selection, different ensemble classification techniques, as well as spatial and non-spatial smoothing and generalization approaches. The method is described and applied to an artificial dataset as well as a 1:50 000 German soil map and a 1:1 000 000 geological map of the Republic of Niger.

As the relationship between soils and landscape within the context of soil formation is well known, predictive relationships between soils and soil formation factors can be established by regression techniques, relating soil and terrain attributes to occurrence of soil classes. This study proposes the production of maps using logistic regression on soil and terrain information from a pilot area to reproduce the original map and predict soil distribution in other similar landscapes in three study areas (Ibibubá Municipality, Sentinela do Sul Municipality, and Arroio Portão Watershed) in map scales from 1:30,000 to 1:50,000 and located in three ecological regions in Southern Brazil (Planalto, Encosta da Serra do Sudeste, and Depressão Central, respectively). By using logistic regressions for digital soil mapping, the method predicts the occurrence of soil units based on reference soil maps (produced by conventional methods), and on several parameters derived from a USGS SDTS-SRTM DEM, namely slope gradient, profile curvature, planar curvature, curvature, flow direction, flow accumulation, flow length, Stream Power Index (SPI), and Topographic Wetness Index (TWI). Results show that parameters such as elevation, curvature, SPI, TWI, and distance to streams are more frequently selected as parameters for predicting the occurrence of soil classes, with overall percent correct from 61% to 71%, and Kappa Index from 36% to 54% when the maps produced are compared with the original soil maps with a simplified legend (which simulate the production of soil maps with smaller scales that the original soil map). The prediction of soil map units using logistic regressions generated reliable soil maps, and the method appears to deserve more research effort, given the reliability and low cost of the resulting information.

Digital soil mapping requires two basic pieces of information: spatial information on the environmental conditions which co-vary with the soil conditions and the information on relationship between the set of environment covariates and soil conditions. The former falls into the category of GIS/remote sensing analysis. The latter is often obtained through extensive field sampling. Extensive field sampling is very labor intensive and costly. It is particularly problematic for areas with limited data. This chapter explores a purposive sampling approach to improve the efficiency of field sampling for digital soil mapping. We believe that unique soil conditions (soil types or soil properties) can be associated with unique combination (configuration) of environmental conditions. We used the fuzzy c-means classification to identify these unique combinations and their spatial locations. Field sampling efforts were then allocated to investigate the soil at the typical locations of these combinations for establishing the relationships between soil conditions and environmental conditions. The established relationships were then used to map the spatial distribution of soil conditions. A case study in China using this approach showed that this approach was effective for digital soil mapping with limited data.

Direct assessment of land degradation globally is constrained by limited spatial data~– soil data in particular (see also Chapter 7). As a proxy, biomass has been adopted as an integrated measure of productivity; its deviance from the norm may indicate land degradation or improvement. Biomass can be assessed by remote sensing of the normalized difference vegetation index (NDVI); norms may be established according to climate, soil, terrain and land use. As a pilot for a

Global Assessment of Land Degradation and Improvement

, spatial patterns and temporal trends of green biomass across Kenya were analysed using 23 years of fortnightly NOAA-AVHRR NDVI data and CRU TS 2.1 station-observed monthly precipitation. Trends of various biomass indicators and climate variables were determined by regression at annual intervals and mapped to depict spatial changes. In Kenya over the period of 1981–2003, biomass increased over about 80% of the land area and decreased over 20%. Most of the decrease has been across the more-productive areas – cropland in the high-rainfall zones. To assess whether this trend represents land degradation or declining rainfall, we calculated rain-use efficiency, the ratio between green biomass (NDVI) and rainfall. Combined trends of biomass and rain-use efficiency may be a more robust indicator of land degradation in areas where productivity is limited by rainfall. Thus defined, degrading areas occupy 17% of the country: most extensively in the drylands around Lake Turkana and the marginal cropland in Eastern Province.

The purpose of this study was to evaluate spatial-temporal dynamics of land cover of the Campo Grande and Santa Cruz Administration Regions, both in Rio de Janeiro city. LANDSAT5-TM images from 1984, 1994 and 1999 were used to create land cover maps. A Geographical Information System was used for integrating information into a cohesive and easy to consult cartographic base and database. Matrices were generated by applying Markov’s Chains, which allow to describe, model and predict transitions of the land cover. This study examined the relationship between land use change, soil orders and carbon stock in the top 10 cm. It was possible to observe the land cover dynamics, with the conversion of agriculture, anthropogenic area and exposed soil in urban areas, especially in the period 1994 – 1999. Using secondary data, from soil survey reports, and combine it with the land cover maps in the temporal series, it was possible to observe a potentiality of this approach in soil properties-landscape modeling. The main finding of this study was that land use change is a dynamic process, and the use of soil properties based on secondary data – soil survey reports – can helps environmental planning, but the accuracy depends of the quality and the spatial data distribution. So, its stress that is important to planning the soils surveys for the good data exploitation for future projects.

Spatial variability of soil properties strongly influenced by human activity is not well documented by most soil surveys. Soil tests performed at farmers’ request represent a large capital of soil information. In France, the results of a large part of these soil tests are continuously gathered in a unique database, the national soil testing database (named BDAT). The aims of the project were to analyse the evolution of soil features within discrete entities over successive time periods and to test the potential of the BDAT for soil dynamic monitoring. Two illustrations are shown: spatial variability of soil pH at national scale, and evolution of soil phosphorus content at regional scale. A validation by census data on agricultural systems was also tested. Taking into account sampling and statistical bias, databases such as the BDAT appear to be relevant tools for soil properties monitoring and can be helpful for digital soil mapping.

Australian agencies are starting to provide online access to soil information through the Australian Soil Resource Information System (ASRIS – www.asris.csiro.au). ASRIS has been designed to integrate soil information collected using both conventional and digital methods. Here we review our experience in developing the system and focus on the importance of good standards for data collection and exchange. There is a clear need for an international standard (in the form of a GML schema) to enable efficient exchange of soil data. We also comment on the problem of market failure and its affect on investment in soil information.

The National Soil Resources Institute has a considerable amount of legacy data in the form of auger bore observations and detailed soil maps. Both have limitations due to inconsistencies in mapping, extent and spatial distribution of the data. Expert knowledge and quality assessment of the inference model can be used to analyse the available training data as well as the resulting map to identify shortcomings. Expert knowledge will identify soils which are either under predicted or missing from the training dataset, whereas the quality assessment will identify soils and landscape units that are missing from the training data. In addition, the methodology provides the means to assess accurately the number and locations of any additional samples required. Using this framework, legacy data can be a valuable source of information in Digital Soil Mapping.

Soil acidification represents a serious problem in mountainous areas of the Czech Republic. It is mainly caused by acid parent materials, high precipitation, the type of vegetation, and acid deposition. These factors act in different combinations and result in different soil conditions. The aim of this chapter is to distinguish areas in the Jizera Mountains with different levels of soil acidification and sensitivity using fuzzy classification. A set of 98 sampling sites was analysed and sampling density was approximately one site per 2 km

2

. Samples were collected from surface organic horizons (O), depth ranged from 4 to 22 cm depending on site conditions. Soil analysis included active and exchangeable soil pH, total content of C, N, and S, pseudototal content of Ca and Mg (after aqua regia digestion), and the ratio of absorbances of soil sodium pyrophosphate extract at the wavelengths of 400 and 600 nm as indicator of humus quality A

400

/A

600

). Moreover, concentrations of exchangeable Al in KCl extract and organically bound Al in Na

4

P

2

O

7

extract were determined. Soil classes were calculated using fuzzy

k

-means method with extragrades. Five classes were selected. The first class with high exchangeable Al content, high S and N, and low Ca, represents the area that was most affected by the acid deposition. The second class with the lowest pH represents strongly acid soils that have very high sensitivity to acidification, but with smaller acid deposition. The third class with high Ca content includes the areas that were limed in the past. The fourth class includes principally the sites with the highest S and N deposition that are populated by grass. The fifth class includes the areas with high Mg content; its distribution corresponds to beech forests that have more favourable effects on soils than spruce forests. Fuzzy classification distinguished soils with strongest sensitivity to acidification. Positive effect of beech forest, grass cover, and liming on surface organic soil horizons is shown.

Soil data and reliable soil maps are imperative for environmental management, conservation and policy. Data from historical point surveys, e.g. experiment site data and farmers fields can serve this purpose. However, legacy soil information is not necessarily collected for spatial analysis and mapping such that the data may not have immediately useful geo-references. Methods are required to utilise these historical soil databases so that we can produce quantitative maps of soil properties to assess spatial and temporal trends but also to assess where future sampling is required. This paper discusses two such databases: the Representative Soil Sampling Scheme which has monitored the agricultural soil in England and Wales from 1969 to 2003 (between 400 and 900 bulked soil samples were taken annually from different agricultural fields); and the former State Chemistry Laboratory, Victoria, Australia where between 1973 and 1994 approximately 80,000 soil samples were submitted for analysis by farmers. Previous statistical analyses have been performed using administrative regions (with sharp boundaries) for both databases, which are largely unrelated to natural features. For a more detailed spatial analysis that can be linked to climate and terrain attributes, gradual variation of these soil properties should be described. Geostatistical techniques such as ordinary kriging are suited to this. This paper describes the format of the databases and initial approaches as to how they can be used for digital soil mapping. For this paper we have selected soil pH to illustrate the analyses for both databases.

Typical examples of azonal soils are Fluvisols and Gleysols that occur around watercourses; they are bound to the alluvial part of landscape and have a characteristic spatial manifestation. They are good examples of a strong relationship between landform and soil. Here we wish to verify the efficacy of different relief characteristics derived from a digital terrain model (DTM) for the delineation of hydromorphic soils around small watercourses. The study is focused on choosing the most appropriate terrain attributes and their combinations. The study area consists of a small 83 km

2

catchment. A DTM with 10 m by 10 m pixels was derived from contours with a 2 m vertical interval. Three methods were compared: (1) combination of drainage area and slope curvature, (2) compound topographic index (CTI) and (3) combination of drainage area and height above the watercourse. The success of methods was verified by comparison of the width of estimated alluvial soils and alluvial soils extent delineated in detail soil map. Detailed comparison of the maps created showed discontinuities in predicted alluvial plain. The delineation based on compound topographic index provided was the worst. The alluvial plain was strongly underestimated (on average by 43%). Discontinuities of the alluvial plain were very frequent. Steep valley bottoms around smaller watercourses, that cause relatively low CTI values even near the watercourses, are the reason of this failure. The third method that was supported by the assumption that alluvial soil can be present only at some level above the watercourse with consideration of the size of the watercourse was the most successful. The extent of alluvial soils was underestimated by less than 22% and there were no discontinuities in the alluvial plain delineations. This study shows that terrain attributes can be a useful aid for delineation of soils strongly related to terrain.

The region near the Solimões river in the Brazilian Central Amazon receives much attention because of oil and gas transport from the Urucu river Province to the refinery in Manaus. Information about soil characteristics and its spatial distribution is important to allow secure intervention in the case of an accident (oil spill). The objectives of this chapter is to present the methodology used to built a soil digital data base of this region combining soil surveys that are mainly available as printed maps at different scales. First, the soil maps were scanned and vectorized and the soil units were identified. All information was put in a digital soil database, with scales from 1:250,000 to 1:10,000. The predominant soils near the borders of Solimões River are Eutric Fluvisols and Eutric Gleysols, whereas in the terra firme predominate yellow Ferralsols, Acrisols and Plinthosols occur. Some Podzols are found scattered in the area, normally at the base of short valleys. Anthrosols with rich antropic horizons also occur, which are called

Terra Preta de Índio

. A soil digital database using this approach to collect all information can be used to plan, monitor and reduce the impacts caused by the petroleum exploitation. It is also useful for land use planners in this region.

The lack of soil maps in Brittany in the north west of France, leads to an approach based on the inference of soilscape units which can be delimited and characterised with relatively fewer field observations than conventional survey. Whereas geology and landform are generally used data to map soilscape units, natural and agricultural landscapes indicate relevant information on soils within them. Remote sensing is obviously the main source of data to map landscape units at regional scale, but one must look carefully how to analyse landscape units, including soil properties, without simply focusing on land-use class. The proposed method for landscape classification is based on a specific classification system developed at regional and local scales, including the role of landscape patterns using object-oriented classification. Post-classification processing is then developed to generalise the results and define mixed landscapes. Finally fusion techniques are tested to examine the probability of common soilscape boundaries arising from different environmental factors (geology, elevation, landscape).

This paper applied pedological mapping in an experimental center of “APTA-Frutas” in Jundiaí, São Paulo, Brazil, using morphometric parameters and GIS tools. The aim of this work was to obtain a preliminary legend of a soil map and to compare the preliminary map with maps made by the traditional soil survey methods. The area has 59 hectares and is located at a mountainous relief in the Atlantic Plateau. The original soil map of this area was made at 1:10 000. A digital elevation model (DEM) was generated with 4 m spatial resolution based on a topographical map at 1:10 000 scale, where the level curves are equidistant at 5 m. Based on the DEM we generated altitude, curvature and slope maps. In order to map the hydromorphic soils it was generated a buffer around the hydrography. We also calculated frequency distribution graphics of altitude, curvature and slope maps. After the interpretation of the frequency distribution, we defined classes to predict the soils types. The curvature map was divided into two class intervals (< or =0 and >0), the altitude map was divided into four class intervals (690–703, 704–714, 715–730, and 731–757 m), and the slope map was divided into four class intervals (0–9, 10–19, 20–44, and 45–72%). The maps were reclassified and converted to shape files. The shape files were intersected with the others to generate the final preliminary soil map. The methodology was adequate for the preliminary mapping of some types of soils.

The present study aimed to improve the recognition of patterns of soils organization in mountainous tropical landscapes, hence helping soil surveys. The study area is located in the northwest Rio de Janeiro State, with a total area of approximately 16.470,ha. In this concern, geomorphometric features that define the geomorphic signature of the soil-landscape, were used. Geomorphometric features includes: elevation, relative elevation, aspect, curvature, curvature plane, curvature profile, slope, flow direction, flow accumulation and drainage’s Euclidian distance, being all these features obtained by geoprocessing techniques. Almost all attributes were obtained from a digital elevation model and, therefore, the primary elevation data were obtained from the topographic maps. Through these geomorphometric attributes, a geomorphometric signature of the landscape was elaborated, and the particularities of each soil-landscape unit improved the supervised classification. The results showed the feasibility of using geomorphometric attributes to perform a supervised classification, using either neural networks or a maximum likelihood algorithm for soil-landscapes classification of mountainous tropical areas. In addition, we showed that geoprocessing techniques used to extract geomorphometrics attributes can subsidize soil surveys, making soil mapping faster and less biased by subjectivity.

Creating detailed soil maps is an expensive and time consuming task that most developing nations cannot afford. In recent years, there has been a significant shift towards digital representation of soil maps and environmental variables and the associated activity of predictive soil mapping, where statistical analysis is used to create predictive models of soil properties. Predictive soil mapping requires less human intervention than traditional soil mapping techniques, and relies more on computers to create models that can predict variation of soil properties. This paper reports on a multi-disciplinary collaborative project applying advanced data-mining techniques to predictive soil modelling for Honduras. Gaussian process models are applied to map continuous soil variables of texture and pH in Honduras at a spatial resolution of 1,km, using 2472 sites with soil sample data and 32 terrain, climate, vegetation and geology related variables. Using split sample validation, 45% of variability in soil pH was explained, 17% in clay content and 24% in sand content. The principle variables that the models selected were climate related. Gaussian process models are shown to be powerful approaches to digital soil mapping, especially when multiple explanatory variables are available. The reported work leverages the knowledge of the soil science and computer science communities, and creates a model that contributes to the state of the art for predictive soil mapping.

A soil database for Rio de Janeiro State was collated in Access, for a project on quantifying the magnitude, spatial distribution and organic carbon in the soils of Rio de Janeiro State (Projeto Carbono_RJ). The main activities were the search, selection, analysis and review of the data for each soil profile already described in the study area, the georeferencing of each soil profile (when spatial coordinates were not available) and the input of new soil profiles into a new interface. The Rio de Janeiro soil dataset now contains 731 soil profiles, 2744 soil horizons, and 48 soil attributes usually described at the soil survey process. From this soil dataset, only 431 soil profiles that were adequately geo-located have been used in this application. The dataset contains limited data for bulk density and hydraulic soil properties, among others. From this dataset, quantitative modelling and digital soil mapping have been completed experimentally at 90 m resolution, using soil data and predictor variables, such as satellite images, lithology, a prior soil map and a DEM and its derivates. This dataset, which is one of the more complete soil datasets in Brazil, is being used as a testbed for learning and teaching DSM, using a variety of methods based on the

scorpan

model (Embrapa, 2006). In the first instance, the soil dataset was used to predict soil classes at the Order level of the Brazilian Soil Classification System – SiBCS (Embrapa, 2006). Five models were built and their results were compared and mapped.