A Methodological Study to a Multifaceted Approach to the Classification of Recent and Historical Fluvial Structures in the Alluvial Plain of the Red River Delta, Vietnam

Start date: 9 November, 2014 End date: 11 December, 2014 Project type: Master's Thesis (prior to 2018) Project code: A27608 Countries: Vietnam Institutions: University of Copenhagen (UCPH), Denmark Grant recipient: Andreas Elmelund Hass and Andreas Hvam Hoffmann Total grant: 31,000 DKK



A suspected coherence between arsenic contamination of aquifers in the alluvial plain of northern Vietnam and the prehistoric courses of the Red River originates within the theoretical fields of geochemistry and fluvial morphology. Arsenic is known to cause, if exposed to on a regular basis, chronic poisoning, skin traumas, skin cancer and risk of developing cancer in internal organs. The release of arsenic into groundwater is believed to be caused by the oxidation of organic carbon, coupled to the reductive dissolution of arsenic-bearing iron oxides which have been brought along in abundance via suspended sediment-load in rivers. Shortly, this is the context for this thesis. By investigating the historic courses of the Red River, interpreted satellite images constitutes an invaluable overview since structures of prehistoric fluvial dynamics is very apparent in these, which can likely lead to reconstruction of prehistoric river courses. If then being able classify these structures this would provide the means of prioritizing areas that might be suitable for extraction of safer groundwater for drinking. Unfortunately manual digitizing of both water filled and overgrown historic and present structures from images or maps is laborious and subjective. To circumvent this, we have developed a methodology and thereto an automated procedure, for the recognition of paleohydrological structures in satellite images capturing the alluvial plain of the Red River. By using remote sensing classification software, eCognition (ver. 9.1), as a tool to perform Geographic Object Based Image Analysis (GEOBIA), we have incorporated classification of values of not only spectral, but also of shape and contextual information. This is key, since the structures that needs classification implies a wide range of spectral behaviour and thus the classification have to primarily be based on the distinct shape that traces of fluvial activity leaves in the landscape. Our primary classification was performed on a Landsat 7 ETM+ scene from November 2000 recording a low-flood situation. The partial results indicate that including the shape index and asymmetry of image objects is crucial to classify these present and historic fluvial structures. To fully comprehend the dynamics that has formed the alluvial plain of the Red River, temporal analysis of Corona and multiple Landsat sensors, have also been established. This provides overview of the last 50 years, but by producing a historical assessment of old maps and literary sources, insight has been gained into the dynamic history dating several hundreds of years back. Compiling all this diverse information and additional soil measurements, provided by GEUS, has enabled the creation of a 3D conceptual model of the evolution of the Red River floodplain. Geographically transcending the methodology to the alluvial plain of the Irrawaddy River in Burma further confirms the applicability of the approach that GEOBIA and our methodology offers.

Additional measures of the nature of involved sediments and arsenic concentrations would provide adjustment and validation to the relation between paleohydrology and heightened levels of arsenic contamination which would be a much welcomed perspective for this study.