| Popis: |
To understand how changes in urban form are related to urban processes driving these changes, increasing use is made of urban growth models. The performance of these models strongly depends on the availability of different types of data. Next to socio-economic data, most of these models require data on topography, road infrastructure, as well as detailed information on land-use/land-cover change. The latter is usually obtained from visual interpretation of historic time series of aerial photographs or satellite imagery, complemented with ancillary information. Recently, Van de Voorde et al. (2011) presented a method for extracting residential and employment related urban land-use patterns from time series of medium-resolution satellite data. The method relies on analysis of the density and the spatial distribution of sealed surface cover within each street block, estimated at sub-pixel level. Based on this method, a framework for calibration of the well-known MOLAND urban growth model was proposed. The approach relies on the comparison of spatial metric values, describing specific characteristics of land-use patterns derived from remote sensing, with metric values obtained from simulated land-use maps (Van de Voorde et al., 2012). Parameters used in the simulation model are tuned in such a way that the simulated patterns of urban growth, as described by the metrics, match the patterns observed in the remote sensing imagery. One of the difficulties in the proposed approach is the uncertainty that is present in the land-use maps obtained through remote sensing, as well as in the estimation of land-use model parameters, and the impact this uncertainty has on the calibration process. This paper focuses on characterizing the uncertainty involved in the remote sensing data processing chain associated with the land-use mapping approach proposed, which consists of two stages: (i) subpixel estimation of sealed surface cover for each urban pixel; (ii) application of a multiple layer perceptron (MLP) approach to infer urban land use from urban form, based on the spatial arrangement of sealed surface cover fractions at street block level. To model uncertainty in the fraction of sealed surface cover, use is made of a first-order autoregressive model, incorporating spatial correlation observed in the fractional errors. To deal with uncertainty associated with the land-use classification, a Bayesian approach is proposed, combining information on the confusion between land-use classes, obtained from the error matrix, with local uncertainty information produced by the MLP classifier. The approach was applied on the central part of the Flanders region (Belgium), covering the cities of Brussels and Antwerp, using a time-series of Landsat/SPOT-HRV data covering the years 1987, 1996, 2005 and 2012. To estimate the impact of uncertainty in both stages of the land-use mapping process, a Monte Carlo simulation was carried out, showing the contribution of uncertainty in sealed surface mapping and in MLP land-use classification, as well as the combined impact of both types of uncertainty on the land-use maps obtained for each period. The ultimate goal of the research is to reduce uncertainty in land-use model calibration by defining a particle filter data assimilation framework, incorporating uncertainty in land-use mapping and land-use model parameter assessment into the calibration process. |
