Assessing uncertainty dimensions in land-use change models: using swap and multiplicative error models for injecting attribute and positional errors in spatial data

Assessing errors and uncertainties in land-use change (LUCH) models is important because these are an unavoidable part of geospatial data and spatially explicit models. Urban planners, decision-makers, and managers need to carefully consider the effects of errors and uncertainties within the LUCH maps produced from LUCH models. There are two common types of error in spatial data: (1) attribute error (e.g. error in data of a categorical nature) and (2) positional error (e.g. error in data of a continuous nature). This article proposes two statistical approaches (swap and multiplicative error models) to inject random errors corresponding to both types of error. This enables us to assess various dimensions of uncertainties in urban land-use change simulated maps obtained by artificial neural networks (ANNs) and spatial logistic regression (SLR) models. The effects of uncertainty dimensions are examined by comparing urban change simulated maps with the reference LUCH map for the Muskegon River Watershed (MRW) area of the USA. The results of data uncertainty scenarios show that attribute errors lead to larger uncertainty in outcome in comparison with that caused by positional error. The model parameter uncertainty scenarios show that ANNs can handle the attribute and positional errors in the training run better than the SLR model. ANNs are able to detect general LUCH patterns in the disturbed data with greater accuracy due to the iteration run (termed the cycle). The resulting uncertainty tables indicate that data uncertainties lead to greater uncertainty in LUCH models in comparison with those caused by model parameter uncertainty. Hence, data uncertainty should be more carefully dealt with, minimizing its occurrence, and paying more attention to its effects on the final products of LUCH models.