guest ::
| Home > Publications database > Large-scale subsurface characterization using image classification ofmulti-configuration electromagnetic induction data assisted by direct soilsampling |
| Home > Publications database > Large-scale subsurface characterization using image classification ofmulti-configuration electromagnetic induction data assisted by direct soilsampling |
| Conference Presentation (After Call) | FZJ-2018-02461 |
; ; ; ; ; ;
2018
Please use a persistent id in citations: http://hdl.handle.net/2128/18200
Abstract: An appropriate characterization of the shallow subsurface with information on the spatial variability of soil propertiesand layering is vital in hydrological modelling. For areas up to several hectares, such subsurface structurescan be investigated with a combination of soil sampling and non-invasive geophysical electromagnetic inductionEMI measurements within a quantitative inversion framework. At larger scales, however, the collection and analysisof ground truth and EMI data with high spatial resolution is challenging due to limited field access and theneed to acquire measurements in a relatively short timeframe to avoid artifacts associated with dynamic changesin soil properties affecting EMI measurements. In this study, we used an image classification method to analyzehigh-resolution multi-configuration EMI measurements and characterize patterns of soil structural organization(layering and texture) in an agricultural area of 102 ha near Selhausen (Germany). The area consists of 51 agriculturalfields managed with a regular crop rotation. Measurements were collected between April and December2016 within a few days after harvest of each field. EMI data were automatically filtered, temperature corrected,and interpolated onto a common grid of 1 m resolution. The apparent electrical conductivity (ECa) maps indicatedfour main sub-areas with characteristic subsurface heterogeneity. Further small-scale geomorphological structuresas well as anthropogenic activities such as soil management and buried drainage networks could be identified bycomparing the ECa maps with soil maps, historical aerial photos, morphometric data, and satellite images. Todelineate areas with similar subsurface structures, we stacked the ECa maps obtained with different coil configurationsin a multiband image and applied supervised and unsupervised image classification methodologies. Bothshowed good results in reconstructing the subsurface structure that is associated with patterns in plant productivity.At our site, the supervised methodology proved more efficient in classifying the whole study area. In a secondstep, we selected one hundred soil sampling locations within the study area and, in February 2017, we obtained soilprofile description with type, depth, thickness, and texture of all soil horizons up to 2 m depth. Using this groundtruth data, it was possible to assign a typical soil profile with soil textural information to each of the classes obtainedfrom the classification of EMI data. The proposed methodology was effective in producing a high resolutionsubsurface model in a large and complex study area that extends well beyond the field scale. Consequently, thismethodology can represent an added value in various applications such as hydrological and agronomic modellingas well as precision agriculture.
|
The record appears in these collections: |
PDF
PDF (PDFA)