A geo-information theoretical approach to inductive erosion modelling based on terrain mapping units

Three main aspects of the research, namely the concept of object orientation, the development of an Inductive Erosion Model (IEM) and the development of a framework for handling uncertainty in the data or information resulting from a GIS are interwoven in this thesis. The first and the second aspect of the thesis discuss simultaneously the application of a terrain mapping unit (TA" in hierarhical observational procedures and an IEM in a GIS environment. These aspects were aimed at providing an alternative solution to the traditional approach to data acquisition, data capture and producing aggregated information for a GIS.The third aspect discusses the application of standard deviation, probability of misclassification, membership degree and plausibility reasoning for handling error and uncertainty associated with data inputs and information outputs handled by a GIS in general and into and from the Indonesian Field Engineering Design Plan (FEDP) in particular. It is aimed mainly at establishing a framework for representing uncertainty in geographical data manipulation. GIS logical models, the characteristics of logical GIS models, types of uncertainty including error due to variability, imprecision, ambiguity and a proposed conceptual framework based on the concept of certainty factors are discussed.The research involved the establishment of stable basic mapping units that allow the definition of repeatable and hierarchical observational procedures. This solution was addressed especially to the situation when sophisticated software and good quality data are not available. In this research, TMUs are defined as areas with a particular combination of geology, geomorphology, morphometry and soil characteristics, usually obtained by interpretation of aerial photo or SPOT images. Terrain areas having similar relief characteristics are identified, delineated and verified in the field. The delineated TMUs represent natural divisions of the terrain often with distinct boundaries.Attributes associated with the established TMUs were selected and used to clasify TMUs. A classification hierarchy of TMU was established in the fight of object oriented modelling including abstraction, inheritance, aggregation and association of terrain objects. The hierarchy has three levels, namely level +1 (superclass level refered to as TMU), level 0 (class level refered to as sub TMU) and level -1 (elementary object refered to as subsub TMU). A lower level in the classification hierarchy represents more refined or specialised information.The well known deductive erosion model, the Universal Sod Loss Equation (USLE) is incomplete in predicting spatial erosion processes. More sophisticated models (i.e. CREAMS, ANSWERS, EPIC, WEPP, GAMES) have failed to account for the complexity of erosion processes and there are no means for validation of model predictions. An alternative to the problem is suggested through an inductive (bottom-up) approach. This approach involves an Inductive Erosion Model (IEM), which was built on observations including dynamic (resilience) and static (inertia) site specific erosion influencing factors in one or more sample areas, made on site at the farmer's field level which is the best functional unit to describe erosion class at local level. An IEM model therefore is region specific. Once an IEM is built and tested for each type of TMU then it can be incorporated within the GIS environment as an acceptable means to predict safely the severity of sod erosion for the entire study area. Erosion severity classes predicted by an IEM are considered as active or dynamic attributes of the established TMUs. By definition TMU provides inherently erosion influencing factors, so called terrain characteristics including morphometry, geology, soil and ground cover. An IEM is intended to predict homogeneous erosion severity classes, related to TMUs at different aggregation or hierarchical levels. The aggregation levels are related to point observations, farmers field level (FFL) and larger parts of the terrain. The discussion of this aspect is focused on the role of the TMU in the observational procedure providing input for an IEM.The established hierarchical mapping units served as a basis for inductive erosion modelling, incorporating expert knowledge-based inference rules. The inductive erosion modelling followed a multi-scale approach and was implemented in a GIS environment. Application of the concepts of regionalization, observed pattern, and decision rules in predicting and modelling purposes are discussed. At regional level patterns associated with the main erosive processes such as sheet, rill, gully and ravine features are generally still identifiable on the aerial photos at scale 1 : 50 000. However, more detailed information on these types of active process at local level can be obtained only by more detailed study, i.e., erosion study at the FFL. In this regard, the FFL is considered as a suitable basic functional unit to describe erosion at local level.Instead of using probability reasoning, which must follow statistical constraints, production rules allow the introduction of a Certainty Factor (CF) for handling both uncertainty in data, models and the resulting information. The C17 can be obtained as a subjective judgment made by experts and comes naturally to experts either in inferring underlying processes or estimating quality of data and models being used. With special reference to the situation when all procedures and techniques for determining probability and obtaining quantitative information particularly in data poor environment are unlikely to be performed, this study demonstrated sufficiently the application of the concept of CF.In the fight of evidence theory, an IEM for predicting erosion severity at a specific TMU was built as a function of various certainty factors of spatial erosion influencing factors. The certainty factor has a value between -1 and +1 and its value indicates the estimated change in belief of allocation of a TMU to a particular erosion class as evidence (from maps, air photos, field observations etc.) is gathered, for each contributing factor. The erosion severity class to which a TMU is finally allocated is the one with overall certainty factor closest to +1. It is proposed as a method of handling uncertain information caused by incompleteness such as inferences established and derived by experts from a set of observations including the effect of causal relationships among various uncertain evidences.