How much do we know about the extent and location, the type and time-frame of past and on-going Earth cover changes? Often surprisingly little.
It is well known that the collection of pointwise ground data for mapping/monitoring purposes is typically tedious, expensive, and either difficult or impossible to gather at several locations. Spaceborne remote sensing is the only reliable system to collect systematic data, with high temporal frequency, over large areas. Therefore, it can be considered as the ideal tool to observe the spatial and temporal aspects of land cover changes. Since the early seventies, operational spaceborne sensors have been continuously collecting images of the Earth: in 40 years huge amounts of data have been stored in the archives of national and international space agencies. Unfortunately these data are still nearly unexploited! In order to transform this huge amount of multi-temporal multi-source data into information, automated data understanding systems are needed.
Conventional remote sensing change detection techniques do offer the capability to capture the location of the change, but not the type. This is because the processing is performed either on non-calibrated data (hence without any physical meaning) or at calibrated signal level (reflectance and/or reflectivity). In both cases the type of change can be exclusively obtained by visually interpreting the single images or by analysing them in digital way by means of deterministic/probabilistic approaches.
If change detection is applied at semantic level (category/class), not only the location of the change but also the type can be directly and straightforward assigned. This assumes that remote sensing data, prior to the change detection step, are transformed into a common basis (i.e. semantic level) by means of automatic prior knowledge approaches. In summary, the system is based on rules designed to mimic well-known spectral/interferometric signatures of target land covers. Its output map consists of spectral/interferometric layers provided with a symbolic meaning.
The proposed change detection technique is automatic, fast, robust, objective and reliable. Moreover it allows to compare data acquired by recent and past remote sensing sensors, enabling in this way to detect changes happened over long time spans using the same sensor
or by combining different sensors, Landsat-5 TM imagery with interferometric ALOS PALSAR data, as illustrated in the figure below.
F. Holecz, M. Barbieri, A. Cantone, P. Pasquali, and S. Monaco, Synergetic use of ALOS PALSAR, ENVISAT ASAR, and Landsat TM/ETM+ data for land cover and change mapping, JAXA Kyoto and Carbon Initiative, Tokyo, 2009. F. Holecz, M. Barbieri, A. Cantone, P. Pasquali, and S. Monaco, Synergetic Use of multi-temporal ALOS PALSAR and ENVISAT ASAR data for Topographic/Land Cover Mapping and Monitoring at National Scale in Africa, IGARSS Symposium, Cape Town, 2009.