ArtISS (Artificial Impervious Surfaces detection with Snow-featured satellite imagery) was a study to improve detection of artificial impervious surfaces using satellite images acquired during winter or night.
Artificial impervious surfaces are the example of extreme human pression the environment: a antuarl land cover is being transformed into a concrete, asphalt, housing. When soil become sealed, a local thermal and humidity conditions changes, impacting biotic and abiotic nature. Scientist around the world develop methods for mapping impervious surfaces locally, regionally and globally. Satellite remote sensing is one of the most effective way for detecting this specific type of land cover.
ArtISS project verified whether the appearance of snow cover on satellite images help to improve the accuracy of detection of impervious areas, or not. Typically, such data are considered less valuable and thus excluded from the analysis (similarly to an observations contaminated by clouds). The project assumed that the presence of snow cover changes the spectral properties of land surface, increasing the spectral contrast between impermeable areas and the background (other types of land cover). The hypothesis was verified positively.
The analysis showed that the detection performed based snow-free imagery was more accurate than the detection with snow-featuring imagery. However, when snow-free and snow-featuring data merged into a single dataset, the results was the most accurate among all obtained. It means that combining snow-free and snow-featuring data into multi-temporal dataset improves the reliability of impervious area detection (maps are more accurate). The accuracy achieved was usually greater than 93%, in the worst case amounting to 89%. There was no difference in accuracy between dense urban fabric and rural regions.
It was also observed that the high accuracy was not only the result of snow cover presence, but also it was impacted by the advance in image classification methods. The approach applied in the project was an ensemble classification with ‘boosting’ method. Such classification features high autonomy and optimize itself during a classification process. It motivated us to carry out the experiment, in which data from one region of the world (eg. China) were classified by reference data (training) in another region of the world (eg. Canada). In that way versatility of the ‘boosting’ method was evaluated. The experiment showed that the ‘classification with remote training points’ resulted with maps with an accuracy ranging from 62% to 95% (in most cases more than 80%). This means that the method is very perspective to be applied for fast (thus cost effective) mapping of the impervious areas around the world.
The project also experimented with an alternative method for the detection of impervious areas: satellite observations made at night, in the visible range of electromagnetic radiation (‘night lights of the Earth’). In that case, it was assumed that the location of anthropogenic light coincides with the location of impermeable areas. To verify that assumption we used a photograph of Berlin made by astronauts onboard the International Space Station (ISS). It was the very first time the ISS photographs were used to determine the extent of impervious areas.
Classification of the image showed that the observations of this type allows for achieving 83% accuracy in detection of the impervious surfaces. Previous maps which also used night imagery produced maps with only 40% accuracy. Unlike snow-featuring imagery, the nighttime data can be obtained world-wide, any time of the year, and its classification is much faster. This suggests that data from the International Space Station (or collected with planned NightSat mission) may significantly improve the reliability of the global coverage maps of impervious areas.
||March 5, 2013 – January 4, 2016
||National Science Centre, Poland
||dr Andrzej Z. Kotarba (CBK PAN)