Foreword
This work is part of a contract between ESB and INRA. It is one of the thematic applications based on the Soil Geographical Data Base of Europe at scale 1:1 Million. It should be considered as an intermediate step towards a "state-of-the-art erosion modelling at the European scale", subsequent to the work of Van der Knijff et al. (2000) and prior to the initiation of the PESERA project.
This work was coordinated by Yves Le Bissonnais*, with the collaboration of Joël Daroussin, Anthony Lecour, Dominique King , Marcel Jamagne, Jean-Jacques Lambert, and Christine Le Bas
* Institut National de la Recherche Agronomique,
Centre de recherche d'Orléans, Science du Sol,
Avenue de la Pomme de Pin,
B.P. 20 619, Ardon,
45 166 OLIVET cedex. - FRANCE
fax : 33 (0)2 38 41 78 69
Email : yves.le-bissonnais@orleans.inra.fr
Summary
The goal of this work was to develop
and apply a methodology based on present knowledge and available data for
the assessment of soil erosion risk at the European scale. Factors influencing
erosion have been graded for the diverse geographical situations existing
in Europe and erosion mechanisms have been expressed with the help of experimental
and expert-defined empirical rules.
Land cover and crust formation on
cultivated soils were considered as key factors influencing runoff and
erosion risk. A soil geographical database has been created for Europe,
and a model of erosion risk has been developed using a Geographical Information
System (GIS). The model uses empirical rules to combine data on land use
(CORINE Land Cover database), soil crusting susceptibility, soil erodibility
(determined by pedotransfer rules from the Soil Geographical Data Base
of Europe at scale 1:1 Million), relief (USGS HYDRO1K digital elevation
model), and meteorological data at a 1 x 1 k m pixel size (Space Applications
Institute, Ispra Joint Research Center). Spatial units for the presentation
of results are defined using either administrative units or watershed catchment
units.
1. Definitions and data bases
Erosion risk assessment on such a large and variable territory involves choosing an approach, i.e. a type of model, and selecting relevant parameters. The approach used in this work is clearly different from expert assessment as performed by De Ploey (1989) or for the GLASOD project (Oldeman et al., 1991; Van Lynden, 1995). The present work is based on a modelling approach using a hierarchical multifactorial classification. The method was previously developed at the French scale (Le Bissonnais et al., 2000). It is designed to assess average seasonal erosion risk at a regional scale. The model is based on the premise that soil erosion occurs when water that cannot infiltrate into the soil becomes surface runoff and moves soil downslope. A soil becomes unable to absorb water either when the rainfall intensity exceeds surface infiltration capacity (Hortonian runoff), or when the rain falls onto a saturated surface because of antecedent wet conditions or an underlying water table (saturation runoff). These two types of runoff generally occur in different environments: bare crusting soils for the first one and humid areas for the second one, though they also may be combined in some cases. Once runoff is initiated on a cultivated field, various forms of erosion are likely to occur, showing various combinations in space and time: sheet hillslope erosion, parallel linear erosion, and gullying erosion. The model developed here attempts to provide a global assessment of these various upstream erosion forms.
As suggested by these definitions,
erosion processes result from numerous interacting factors. Modelling is
difficult because of the complexity of these interactions. Four main factors
are generally considered: land cover, soil, topography, and climate.
We used the best available information
layers covering the whole Europe (fig.):
(i) the 250 m resolution raster version of the CORINE Land Cover database
at scale 1: 100,000 (source: SAI/JRC Ispra) (EEC, 1993); (ii) the Soil
Geographical Data Base of Europe at scale 1:1,000,000 (source: ESB/JRC
Ispra) (Jamagne et al., 1995); (iii) the 30 arc seconds (~1 x 1 km) resolution
HYDRO1K raster digital elevation model (source: USGS HYDRO1K); and (iv)
25 years of daily meteorological data at 50 km resolution (source: SAI/JRC
Ispra).
The CORINE Land Cover database provided land cover data (EEC, 1993). Information was collected between 1985 and 1992. The initial classification included 44 classes. Original land cover classes have been reclassified into 9 classes, each corresponding to a specific behaviour towards soil erosion. The 9 classes are described as follows :
Soil crusting and soil erodibility parameters are derived from soil names (third level FAO legend), soil dominant and secondary surface textural class, parent material class (third level), using chained pedotransfer rules (Bouma and Van Lanen, 1986; Daroussin and King, 1996). However, some basic information necessary for estimating these parameters is absent from the soil data base (carbon an clay contents, clay mineralogy, cation content, etc.). In addition, Soil Mapping Units (SMU) are very heterogeneous at this scale, and the dominant soil type may represent less than 50% of the unit area in some cases. Therefore, pedotranfer rules are kept simple and the resulting parameters are to be considered very approximate (see also Use of pedotransfer rules for European Soil Database interpretations). Moreover, maps of soil characteristics, whether infered by pedotransfer rules or not, are allways accompanied by a so called "purity" map to warn the map reader that the view he is looking at does not show exactly the information that is available in the database (see Understanding purity maps for a full demonstration of this concept).
Crusting and erodibility parameters are based on the combination of a common physico-chemical parameter (1) and two specific textural/parent material parameters (2 & 3). These 3 intermediate parameters are themselves established from pedotransfer rules.
(1) The physico-chemical parameter (5 classes) is derived
from the soil name information (third level) by taking into account the
positive or negative effect of organic matter, carbonates, cations, and
other pedogenetic characteristics.
The pedotransfer rule for determining
this physico-chemical soil property is based on the following main principals:
(3) The erodibility textural/parent material parameter is based on a combination of the dominant soil surface texture and the type of parent material. Massive rocks like granite or limestone are classified in low erodibility, whereas loose rocks like sand or molasse are classified in high erodibility. Erodibility is also high for coarse and medium textures.
We used a gridded digital elevation
model (DEM) covering Europe with a cell resolution of 30 x 30 arc seconds
(~1 x 1 km). The DEM determined the basic cell resolution for all other
information layers. This resolution is the finest available to date covering
the whole of Europe, however a finer DEM would clearly lead to a better
assessment of slopes. The average slope of each cell (in percentage) was
calculated from the greatest elevation difference between it and its 8
neighbouring cells. Slopes were classified into 8 classes, the limits of
which were defined according to field experience or values to be found
in the literature. Class limits were adjusted in order to take into account
the smoothing effect in slope assessment due to the coarse cell resolution
of the DEM. Slope classes used are the following :
0-1% >1-2%
>2-5% >5-10% >10-15%
>15-30% >30-75% >75%
Rain is the main factor for water erosion
and its erosive effect is related to its amount and intensity; hence,
an attempt has been made to combine both parameters to characterize erosivity
for each season. Basic climate data come from SAI/JRC Ispra, 25 years of
daily meteorological data covering Europe at a resolution of 50 x 50 km
cells. Mean monthly rainfall was calculated for each cell. These values
were reclassified into five classes for each season (Winter: December to
February, Spring: March to May, Summer: June to August, Autumn: September
to November).
The mean seasonal frequency of daily
rainfall above 40 mm was used as an indicator of rainfall intensity. This
choice is a satisfactory balance between an acceptable precision and a
sufficient discrimination, allowing expression of rainfall intensity in
the Mediterranean zone. Two thresholds of frequency giving a good differentiation
of the main climatic areas in Europe were empirically determined. We combined
the five total rainfall classes and the three intensity classes: medium
and high intensity classes increase the total rainfall class by one or
two class levels, respectively. This combination resulted in five classes
of climate erosivity.
A "decision tree" type model was developed. The method combines parameters according to logical operations. A specific decision tree was built for each of the six first land use classes corresponding to cultivated land or land with natural vegetation, thereby accounting for the various erosion types defined above. Erosion risk was not considered in this study for the three other classes (bare land, water, and urban areas). Parameters were ranked and weighted using present knowledge on the different erosion types. Each combination of parameters was ranked from 1 to 5 indicating the relative soil erosion risk (Fig.). This processing was performed on data layers rasterized to 1 x 1 km cells corresponding to the DEM resolution.
Factor ranking and combinations
The choice of a decision tree type model requires ranking of each parameter. We gave priority to the factors which can be modified by human activities. The entry order was the following: land use, crusting, slope, soil erodibility (Fig.). These factors were linked successively and 5 classes (very low, low, medium, high, very high) were defined from literature, experiments and expert analysis, giving the agro-pedo-topographical sensitivity to erosion of each pixel. This sensitivity to erosion was then combined with the climate erosivity for each season to provide the overall seasonal erosion risk. For the medium climate erosivity class (3), the class of erosion risk is the class of sensitivity to erosion. For the higher and lower climate erosivity classes, the class of sensitivity is respectively increased or decreased to give the erosion risk class. Moreover, in summer arable land and heterogeneous agricultural land erosion risks are decreased by 1 class in order to account for a more developped crop cover during this season. The main rules included in the expert system are the following:
The results obtained after the application
of the model at 1 x 1 km cell were then aggregated according to two types
of spatial units: (i) catchments, which are the natural spatial units for
runoff and erosion processes; (ii) administrative units, which allow the
comparison with other information sources available on the same spatial
unit basis and correspond to decision units. Automatic spatial aggregation
of risk is done with a decision rule that takes into account the surface
ratio of each risk class within each aggregation unit. Arbitrary thresholds
are used in order to give a global risk class for each aggregation unit,
from the surface percentages of the five risk classes (see apendix).
The same method is used for spatial aggregation of soil sensitivity to
erosion.
3. Errors assessment and results validation
At this stage, the validation was done
by local experts by qualitative assessment based on their experience. However,
at such a scale, it is necessary to use error assessment and validation
procedures. Several ways to validate can be identified: (i) a first approach
would be to estimate error propagation during the various data treatments
in the geographical information system. This aims to estimate errors (uncertainty)
of each information layer and evaluate their impact through processing
of the decision tree model. This estimation of uncertainty was done for
soil data (not presented here); (ii) a second approach would test the method
on areas that can be considered as reference sectors as they provide more
accurate data. This approach is currently being implemented in the Normandy
region in France. Preliminary results show good agreement between these
approaches; (iii) a third approach would be to compare the results from
this study with actual erosion records or with data from river sediments
monitoring, etc. Such data are not available for the whole Europe, however
it has been possible to use the mudflow records in France during a ten
year period (1985-1995) for comparison with our erosion risk classification
at the county level (Le Bissonnais et al., 2000). The correlation coefficient
between the number of recorded mudflood events and the class of erosion
risk was highly significant at p=0.01 (r=0.19 for n=3406). However, these
phenomena are not strictly comparable: erosion may occur without leading
to mudflow and conversely mudflow may result from landslides or other processes.
In any case, this is the only validation by independent data that can be
performed for the moment.
4. Discussion
The modelling approach presented here is very simple and versatile: it can accomodate heterogeneous data resolution and quality; it does not require the use of parameters that are not available at national scale, as does the USLE model. This new approach is much more precise and accurate than the CORINE erosion model. Both approaches are based on a decision tree, but the latter uses a single decision tree and takes into account only 2 land use classes, 3 climate classes and 4 slope classes. In addition, the CORINE erosion model does not take crusting into account. However, the disadvantages of qualitative methods remain in our model. In particular, the final information is provided on a 5 class scale of risk and it is not possible to link these classes to quantitative values of erosion, nor is it possible to assess the errors associated with the results.
However, errors and uncertainty associated
to the results are much more dependent on the resolution and quality of
the input data than on the model itself, because the model is based on
very simple and global assumptions accepted by all experts. Comparison
between this model and the Kirkby model for France using the same input
data showed good agreement (Kirkby et al., 2000). The main source of uncertainty
in this model is certainly related to the parameters obtained from the
soil database, i.e. susceptibility to crusting and erodibility, because
both the spatial resolution and accuracy of these data are very low.
The aggregation of individual cell
results into larger spatial units partly compensates for the uncertainty
related to local assessment of erosion risk. The disadvantage is a smoothing
effect that may erase some small or sparse high risk areas.
5. Conclusion and perspectives
The methodology presented here allowed to generate a single homogeneous map of erosion risk at the European scale that makes it possible to compare between regions. The decision tree type model considers different types of erosion depending on land use. The production of seasonal maps shows the importance of the seasonal effect on erosion. The aggregation according to different spatial units makes it possible to adapt the results to different users needs. Finally, the model is easy to modify in terms of the rules and to update with new data.
On the other hand, this survey is somehow limited by the accuracy of some fundamental data such as soil data. This explains why in some cases, crusting and erodibility estimates are rather unreliable. Another limitation is the 1 x 1 km grid resolution for the DEM, which does not result in accurate assessment of slope values in areas of gentle relief or short hillslopes. In addition, CORINE Land Cover and agricultural statistics data should be updated.
This work could be improved and developed
with respect to some guidelines: (i) a more precise analysis in some areas,
making use of more accurate data, such as a soil map at 1: 250,000 scale
or a 250 x 250 m or finer grid resolution DEM. This methodology was used
with success for France at a 250 x 250 m resolution (Le Bissonnais et al.,
2000). Results are available on the INRA web site (http://erosion.orleans.inra.fr/).
The database built for this work at the European level will be used with
a physically-based model (Kirkby et al., 2000) in the framework of the
ongoing EU funded PESERA (Pan European Soil Erosion Risk Assessment) project.
Bibliography
Bouma J., Van Lanen H.A.J., 1986. Transfert functions and threshold values : from soil characteristics to land qualities. In proceedings of the international workshop on quantified land evaluation procedures. 27/04 - 2/05/1986, Washington DC. USA, 106-110.Daroussin J., King D., 1996. Pedotransfer rules database to interpret the Soil Geographical Database of Europe for environmental purposes. In: The use of pedotransfer in soil hydrology research in Europe. Workshop proceedings. Orléans, France. 10- 12 octobre 1996, p 25 - 40.
De Ploey J., 1989. Soil erosion map of western Europe. Bublished by CATENA. Laboratory of Experimental Geomorphology, Leuven, Belgium.
EEC, 1992. CORINE soil erosion risk and important land resources in the southern regions of the European Communities. Brussels. 97 p + cartes.
EEC, 1993. CORINE Land Cover. Guide technique, Brussels. 144 p.
Jamagne M., Hardy R., King D., Bornand M., 1995. La base de données géographique des sols de France. Etude et Gestion des Sols, 2, 3, 153-172.
Kirkby M.J, Y. Le Bissonnais, T.J. Coulthard, J. Daroussin, M.D. McMahon, 2000. The development of land quality indicators for soil degradation by water erosion. Agriculture, Ecosystems and Environment, 81, 125-135.
Le Bissonnais Y., C. Montier, M. Jamagne, J. Daroussin, D. King , 2000. Mapping erosion risk for cultivated soil in France), Catena, in press.
Ministry of Environment, 1996. Unpublished. Les "coulées de boue" liées à l'érosion des terres agricoles. Dossiers et cartes nationaux, dossiers et cartes régionaux.
Oldeman, L.R., Hakkeling, R.T.A. and Sombroek, W.G. (1991). World Map of the Status of Human-Induced Soil Degradation, with Explanatory Note (second revised edition) - ISRIC, Wageningen; UNEP, Nairobi.
Van Lynden, G.W.J. (1995). European soil resources. Nature and Environment No. 71. Council of Europe, Strasbourg.
Yassoglou et al. (1998). Soil Erosion in Europe. Soil erosion working group, ESB/JRC internal report.
Rules used for aggregating the cell by cell
results into natural or administrative spatial units:
| Priority order | Aggregated erosion risk class |
|
|
|
|
or class 5 = 0% or class 4+5 < 3% or class 3+4+5 < 5% |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
or 2% < class 4+5 < 9% or 4% < class 3+4+5 < 13% |
|
|
|
or 8% < class 4+5 < 15% or 12% < class 3+4+5 < 21% |
|
|
|
or 14% < class 4+5 < 23% or 20%< class 3+4+5 < 31% |
|
|
|
or class 4+5 > 22% or class 3+4+5 >30% |
Examples: