Geostatistical and Artificial Neuronal Networks maps of the Texture of the soils of Geneva Canton
| dc.contributor.author | Maignan, Michel | |
| dc.contributor.author | Kanevski, Mikhail | |
| dc.contributor.author | Celardin, F. | |
| dc.contributor.author | Besson, A. | |
| dc.contributor.editor | Minier, Philippe | |
| dc.contributor.editor | Susini, Alberto | |
| dc.date.accessioned | 2019-09-16T09:34:01Z | |
| dc.date.available | 2019-09-16T09:34:01Z | |
| dc.date.issued | 2004 | |
| dc.description.abstract | The mapping of soil texture categories of the Canton of Geneva, and the methodological and computational approaches in order to achieve it are presented. Earlier maps based on the same database were established by bilinear interpolation using UNIRAS-UNIMAP software. The present study aims to finer and more reliable representations with the additional possibility of obtaining a textural class map which is better adapted to practical agronomic assessments . The database includes 2616 geo-referenced samples covering the totality of cultivated land area. The clay (Ø<0.002 mm), silt (0.002<Ø<0.05 mm) and sand (0.05<Ø<2.0 mm) fractions were determined by the pipette method. The spatial correlations found are satisfactory and allowed to establish kriging maps, as well as ANN MLP Artificial Neuronal Networks Maps by Multilayer Perceptron approach. The individual particle size component maps (%clay, %silt, %sand) were combined and normalized in order to construct a textural class map according to the Swiss system. The present study has shown the existence of spatial correlations with a range of approximately 500 meters, for the three textural components. Mapping by kriging (optimal in the sense of minimization of variance) for each of these variables delivered comprehensive maps, where no real inadequate result occur. These maps are useful tools for field scale pedological observations. Addition of the three independently obtained kriging maps gave site specific results very close to 100%. The subsequent normalisation to 100% required less than 1% adjustment. Another important practical conclusion is that mapping directly by interpolating textural class information is nonsensical since the categories expressed as variables are non additive. From a computational point of view, this work confirms the efficiency of the Neuronal Network Residual Kriging NNRK in smoothing the large tendencies and in avoiding some problems of non stationarity when applied to Multilayer Perceptron MLP mapping. Further investigations are in progress in order to improve the mapping results and in order to export the Geostat Office results on a GIS Map, inclusive of the lakeshore, and the district borders of the communes. | de |
| dc.description.uri | http://enviroinfo.eu/sites/default/files/pdfs/vol110/0388.pdf | de |
| dc.identifier.uri | https://dl.gi.de/handle/20.500.12116/27167 | |
| dc.publisher | Editions du Tricorne | |
| dc.relation.ispartof | Sh@ring – EnviroInfo 2004 | |
| dc.relation.ispartofseries | EnviroInfo | |
| dc.title | Geostatistical and Artificial Neuronal Networks maps of the Texture of the soils of Geneva Canton | de |
| dc.type | Text/Conference Paper | |
| gi.citation.publisherPlace | Geneva | |
| gi.conference.date | 2004 | |
| gi.conference.location | Geneva | |
| gi.conference.sessiontitle | Posters |