This document discusses a project called AutoMAPticS that aims to improve soil mapping in Portugal using digital soil mapping techniques. The project uses artificial neural networks to predict soil classes in unmapped areas of Portugal based on relationships learned from existing soil maps and landscape data. The goals are to complete soil map coverage in Portugal at a scale of 1:100,000 and harmonize different regional soil classifications to improve transnational data integration. The document provides background on soil mapping efforts in Portugal and challenges with existing maps. It also describes how artificial neural networks can be trained on landscape and soil data to generate new digital soil maps.
Salient Features of India constitution especially power and functions
7th euregeo volume_1 164_165
1. 146
7th
EUREGEO 2012 Session 2 - Poster
SOIL MAPS – TURNING OLD INTO NEW
INÊS FONSECA (1)
; RICARDO BRASIL (1)
; JORGE ROCHA (1)
; SÉRGIO FREIRE (2)
AND JOSÉ
TENEDÓRIO (2)
(1) Centro de Estudos Geográficos, Edf. Fac. Letras, Universidade de Lisboa, Alameda da Universidade,
1600-214 Lisboa, Portugal. i.fonseca@campus.ul.pt
(2) e-Geo, Centro de Estudos de Geografia e Planeamento Regional, Faculdade de Ciências Sociais e
Humanas da Universidade Nova de Lisboa, Av. de Berna 26-C, 1069-061 Lisboa, Portugal.
KEY WORDS: Soil digital mapping, Artificial
Intelligence, Soil Survey.
INTRODUCTION
Soils have long been recognized to be an
important non-renewable resource crucial for
human activities (Potocnik & Dimas, 2005) and
for modulating global change. Yet little account is
taken of soil properties and their spatial variation
in planning, hindering the assessment of soils and
their ability to support a range of services, from food
production, to biodiversity and pollution buffering.
Whilst soil surveys have been carried out in
many countries, the area coverage and scale
of resulting soil maps are far from being ideal for
planning applications at national level (Dobos et
al., 2006). Additionally, the lack of consistency
between soil classifications and legends across
countries contributes towards a slow progression
in integrating soil datasets, even in Europe (ESBN,
2005).
Geographical Information Systems (GIS)
together with artificial intelligence (AI) maximize the
information content of existing soil maps by learning
the rules that have, more or less explicitly, led to
the mapping of soil classes across the landscape,
and then use that knowledge to predict the spatial
distribution of soil classes or properties at similar
resolutions.
Thus, one of the main objectives of AutoMAPticS
(Automatic Mapping of Soils), a portuguese project
carried out at national level and based on the
development of artificial neural network (ANN)
models, is to predict soil classes in (i) currently
unmapped areas of Portugal, and (ii) harmonise soil
legends across regions with distinct soil mapping
classifications, using Portuguese and Spanish
soil spatial datasets to a) improve the level of
transnational data integration and b) assess existing
data.
SOIL MAPPING
Portugal, like most EU member states, only has
a fraction of its territory covered with soil maps at
semi-detailed or reconnaissance scales (McBratney
et al., 2003). Whilst 55% of continental Portugal
has soil maps at 1:50000 produced by traditional
methods of soil survey before the 1970s, only
about 40% of the territory has more recent soil
map coverage at 1:100000 with some degree of
overlap (Figure 1). Thus, not only the published
coverage remains incomplete, but there are also
significant problems with the existing cartography.
There is a lack of cartographic uniformity between
the different regions: (1) scales are different, (2)
four different taxonomic systems were used, and
(3) the framework behind the mapping of soil units
at the two scales is different: the 1:100000 maps
have a physiographic basis whereas the 1:50000
maps have a taxonomic basis. Moreover, using
taxonomy as the basis of map design often results
in high intra-unit variability of soil properties (Mulla
& McBratney, 2000) and limited correlation between
soil type and soil hydrologic parameters (Western
& Grayson, 2000). Therefore, only 43% of the area
of Portugal has high standards of soil cartography.
Figure 1 – Scale and legends of regional soil maps of
continental Portugal.
2. 147
7th
EUREGEO 2012 Session 2 - Poster
The Portuguese soil map coverage provides
an outstanding opportunity for advancing Digital
Soil Mapping (DSM) because (1) half the country
has high quality soil-map coverage for which a
very large number of geo-referenced soil samples
were collected, (2) there are geo-referenced soil
profile data at a number of sites throughout the
unmapped area, and (3) there are also soil maps
at different scales. Additionally, as soil units are
not coterminous with political soil maps, Spanish
soil spatial datasets were collated to allow a better
integration of the Portuguese soil cartography at an
international level.
SPATIAL MODELLING OF SOILS USING GIS &
ARTIFICIAL INTELLIGENCE
With the technologies available today, such
as GIS, which allow storage and retrieval of large
amounts of spatial data, including of those which are
considered to be the main factors of soil formation
and development (Irvin et al., 1997; MacMillan et
al., 2000), combined with advances in Artificial
Neural Networks (ANN) and Fuzzy Logic (FL), it has
become possible to map the spatial distribution of
soils in a cheaper, more consistent and flexible way,
using surrogate landscape data. Thus, ANN models
are used to translate knowledge of soil-survey,
which is encapsulated in existing soil maps, into
relationships between soils and landscape features
in order to predict soil types in unmapped areas for
which the landscape characteristics are known. In
this way, first-order digital soil maps are used to
generate third-order ones (Behrens & Scholten,
2006).
During the training phase, the ANN learns about
regularities present in the datasets (soil classes
and landscape features, i.e. digital terrain data, land
cover, lithology) and, based on these regularities
constructs rules (Tso & Mather, 2001) that can be
extended to the unmapped areas.
The ANN is trained by presenting it a number of
different examples of the same object (i.e. different
landscape-feature vectors for a certain soil type)
and this methodology is applied to each pilot region
in Portugal and Spain and at different resolutions.
FUTURE PERSPECTIVES AND NEEDS
Processes involved in soil development do not
usually operate on a discrete level but produce a
continuum of change (McBratney & Odeh, 1997).
Also, according to the same authors, imprecision and
uncertainty are inherent parts of natural systems.
Fuzzy Logic is an approach that can address the
continuous nature of soil in DSM. There is a natural
synergy between ANNs and FL that makes their
hybridization powerful for DSM, which will be later
developed in this project.
Results obtained using both methodologies will
be compared and validated using existing maps
and soil profile data, and the best model will be
used to map soil classes across areas which are
currently lacking spatial soil data. Thus, the main
aim of AutoMAPticS is to complete the Portuguese
soil map coverage at 1:100000.
ACKNOWLEDGEMENTS
AutoMAPticS is supported by PTDC/CS-
GEO/111929/2009, a grant from FCT – Foundation
for Science and Technology (Portugal), and the
principal investigator is employed under the FCT
Science 2008 Programme.
REFERENCES
BEHRENS T. & SCHOLTEN T. (2006) Digital soil mapping
in Germany – a review. J. Plant Nutr. Soil Sci., 169,
434-443.
DOBOS E., CARRÉ F., HENGL T., REUTER H.I. & TÓTH
G. (2006) Digital Soil Mapping as a Support to Pro-
duction of Functional Maps. EUR 22123 EN. Office for
Official Publications of the European Communities,
Luxemburg, 68pp.
ESBN (EUROPEAN SOIL BUREAU NETWORK) (2005)
Soil Atlas of Europe. European Commission, Office
for Official Publications of the European Communi-
ties, L-2995 Luxemburg, 128pp.
IRVIN B.J., VENTURA S.J. & SLATER B.K. (1997) Fuzzy
and isodata classification of landform elements from
digital terrain data in Pleasant Valley, Wisconsin. Ge-
oderma, 77, 137-154.
MACMILLAN R.A., PETTAPIECE W.W., NOLAN S.C. &
GODDARD T.W. (2000) A generic procedure for au-
tomatically segmenting landforms into landform ele-
ments using DEMs, heuristic rules and fuzzy logic.
Fuzzy Sets & Systems, 113, 81-109.
MCBRATNEY A.B., MENDONÇA SANTOS M.L., MINA-
SNY B. (2003) On digital soil mapping. Geoderma,
117, 3-52.
MCBRATNEY A.B. & ODEH I.O.A. (1997) Application of
fuzzy sets in soil science: fuzzy logic, fuzzy measur-
ments and fuzzy decisions. Geoderma, 77, 85-113.
MULLA D.J. & MCBRATNEY A.B. (2000) Soil spatial va-
riability. In M.E. Sumner (Ed.) Handbook of Soil Scien-
ce, A 321-352. CRC Press, Boca Raton.
POTOCNIK J. & DIMAS S. (2005) Preface. In Soil Atlas
of Europe. European Soil Bureau Network. Europe-
an Commission, Office for Official Publications of the
European Communities, L-2995 Luxemburg, 128pp.
TSO B. & MATHER P.M. (2001) Classification Methods
for Remotely Sensed Data. Taylor and Francis, Lon-
don, 332pp.
WESTERN A. & GRAYSON R. (2000) Soil moisture and
runoff processes at Tarrawarra. In R. Grayson & G.
Blöschl (Eds) Spatial Patterns in Catchment Hydro-
logy: Observations and Modelling. Cambridge Univer-
sity Press, Cambridge, 209-246.