Monitoring Old Industrial Activities from 1930 to Nowadays for Planning Brownfields and Vacant Lots Redevelopment
Lefebvre, Antoine; Sannier, Christophe

One of the most challenging issues in the next decade's urbanization is the planning of brownfields or vacant lots redevelopment. Renewal of the old industrial sites is considered as a solution to limit the urban sprawl and build a denser and more functional urban pattern. Nevertheless these planning projects need a particular attention on the land background. For example, these areas can be polluted by the storage or production of a previous industrial activity and then present a risk for citizens in the future. Depending to the location, retrieving land background can be a hard task because the availability of information on past activities can be not sufficient or nonexistent. In this context, remote sensing data is considered as an alternative way to reconstruct land history and provide historical information of urban particularly relevant for town planners.

This paper focuses on monitoring old industrial sites from 1930 to nowadays in Lille agglomeration (Nord Pas de Calais). Nord Pas de Calais was the most industrialized French province in the last century and still keeps a crippling legacy. From 1850 to 1970, Lille agglomeration (composed of 3 cities: Lille, Roubaix and Tourcoing) concentrated the most important textile factories and indirectly numerous chemical, raw material storage industries. Since 70's, these industries phased out due to a loss of profitability. The development of urban areas, which has been mainly conducted by the industrialization, is currently facing planning limitations due to land pollution and neglected buildings going to rack to ruin. Aerial photographs from 1930 to nowadays have periodically covered the agglomeration of Lille. These acquisitions have been done at a scale around 1:8'000 and provide an average resolution about 50 cm. Data then constitute a time series with a 20-year time step and very high spatial resolution, which is particularly suitable to monitor urban environments.

The proposed method is composed of 2 steps: (1) the identification of old industrial sites at each date and (2) the time series smoothing. The identification of the factories relies on their specific architecture of the 19th century. These differentiate themselves from other urban structures by their saw-tooth roof. It consists on a series of windows and bricks, which was design to optimize direct sunlight lighting. From an aerial photograph, this structure represents a particular pattern composed of parallel elements. This combination of items, entirely related to an anisotropic texture, can be characterized by its spatial arrangement. Many texture techniques provide a pertinent way to identify anisotropic texture in remote sensing images. One of the most relevant topics is the characterization of vine-plots where techniques based on Gabor, Fourier, wavelet transform and co-occurrence analysis have been tested. In this context, our method is based on a wavelet analysis, which is particularly adapted to noisy data. It is based on the maximization of a criterion that deals with the coefficients enclosed in the different bands of a wavelet decomposition of the original image. More precisely, it searches for the orientation that best concentrates the energy of the coefficients in a single direction.
Due to the noise and the different nature of input data, the energy of the wavelet coefficients cannot be directly compared date-to-date. The time series analysis aims to smooth these latter. We assume that urban changes can be modeled as an abrupt and irreversible evolution in multi-temporal energies. Then, this step consists in fitting a Heaviside step function. It ensures a homogeneous energy value for unchanged periods and highlights precisely a change as a discontinuity in the time-series. This enables to extract, for each pixel, a profile of change which dates the evolution of the old industrial sites to another land use.

Results performed on a study area about 10 square km bring out the efficiency of the method to monitor old industrial activities. It detects factories from 1930 to nowadays and identifies ones, which changed to another land use. It enables to reconstruct the land background of brownfields and vacant lots but also provides previous land use information of current housing. This method constitutes a reliable tool to extract historical information in urban environment and can have a key role to support local administration, which is currently undertaking redevelopment plans.