The application of a visual data mining framework to determine soil, climate and land use relationships

In this research study, the methodology of action research dynamics and a case study was employed in constructing a visual data mining framework for the processing and analysis of geographic land-use data in an agricultural context. The geographic data was made up of a digital elevation model (DEM), soil and land use profiles that were juxtaposed with previously captured climatic data from fixed weather stations in Australia. In this pilot study, monthly rainfall profiles for a selected study area were used to identify areas of soil variability. The rainfall was sampled for the beginning (April) of the rainy season for the known ‘drought’ year 2002 for the South West of Western Australia. The components of the processing framework were a set of software tools such as ArcGis, QuantumGIS and the Microsoft Access database as part of the pre-processing layer. In addition, the GRASS software package was used for producing the map overlays. Evaluation was carried out using techniques of visual data mining to detect the patterns of soil types found for the cropping land use. This was supported by analysis using WEKA and Microsoft Excel for validation. The results suggest that agriculture in these areas of high soil variability need to be managed differently to the more consistent cropping areas. Although this processing framework was used to analyse soil and rainfall climate data pertaining to agriculture in Western Australia; it is easily applicable to other datasets of a similar attribution in different areas.