Geographical information modelling for land resource survey

摘要: The increasing popularity of geographical information systems (GIS) has at least three major implications for land resources survey. Firstly, GIS allows alternative and richer representation spatial phenomena than is possible with the traditional paper map. Secondly, digital technology improved accessibility ancillary data, such as elevation models remotely sensed imagery, possibilities incorporating these into target database production. Thirdly, owing to greater distance between data producers consumers there a need uncertainty analysis. However, partly due disciplinary gaps, introduction not resulted in thorough adjustment survey methods. Against this background, overall objective study was explore demonstrate utility new concepts tools within context pedological agronomical surveys. To end, research conducted on interface five fields study: geographic theory, resource survey, remote sensing, statistics fuzzy set theory. A demonstration site chosen around village Alora southern Spain. Fuzzy theory provides formalism deal classes that are indistinct result vague class intensions. sets characterised by membership functions assign real numbers from interval [0, 1] elements, thereby indicating grade set. When used classify attribute linked geometrical presence dependence among elements ensures they form spatially contiguous regions. These can be interpreted objects indeterminate boundaries or objects. thus adds conventional conceptual assume either discrete continuous fields. This thesis includes two case studies use acquisition querying information. first explored c -means clustering derived model represent transition zones soil-landscape model. Validity evaluation resulting terrain descriptions based coefficient determination regressing topsoil clay grades. Vaguely bounded regions were more closely related observed variation content () crisply units soil (). second involved uncertain data. It explains differences fuzziness stochastic basis an example query concerning loss forest ease access. Relationships probabilities memberships established using linguistic probability qualifier (high probability) expectation function defined travel time. processing compared crisp processing. response contained because, unlike response, it indicated degree which individual locations matched selection criteria. In typically involves collecting small sample precisely measured primary well larger even exhaustive secondary Soil surveyors often rely relationships image interpretation enable efficient mapping properties. Yet, generally fail communicate about knowledge methods employed deriving map statements their content. thesis, methodological framework formulated demonstrated takes advantage interactively formalise stepwise inductive learning relationships. examines topology record potential part links hierarchically nested corresponding distinct formation regimes. applied similar areas facilitate restricting lower level visualisation create images (e.g. perspective views) illustrating landscape configuration expected support different analysing describing relation description, including those requiring models. though, only object Satellite sensing become important tool cover mapping, providing attractive supplement relatively inefficient ground common approach extract imagery probabilistic classification multispectral Additional incorporated classification, translating Bayesian prior each type. particularly advantageous spectral overlap classes, i.e. when unequivocal assignment alone impossible. demonstrates procedure iteratively estimate regional pertaining stratification. method incorporation additional process without requirement known probabilities. project Landsat TM 1984 1995. Image stratification geological area. Overall accuracy 76% 90% (1984) 64% 69% (1995) employing estimated fact any description limited implies never completely certain. error inaccuracy contributes significantly uncertainty. Usually, datasets global indices see above). Error modelling, other hand, indication distribution inaccuracies given. approaches analysis imagery. local conditional pixels' intermediate results indicate magnitude implication change detection comparing independently classified images. shortcoming implicitly assumes neighbouring pixels independent. Moreover, does make full available reference ignores component. consider nor assumption independent obviously impedes proper assessment uncertainty, joint several taken together. Therefore, geostatistical methods, exploit rather ignoring it. how above updated conditioning sampled locations. Stochastic simulation generate 500 equally probable maps, uncertainties regarding extent olive orchards could inferred. Future challenges include quality aspects datasets. present analysis, so that, example, precision fitness addressed. Other extensions work concern inclusion third dimension modelling temporal aspects.