Geodaysit 2023

Open multitemporal Earth Observation data for land surface albedo estimation in urban areas
06-16, 09:30–09:45 (Europe/London), Sala Videoconferenza @ PoliBa

Carlo Barletta, Alessandra Capolupo, Eufemia Tarantino

Nowadays, data in an open format, easily accessible and characterized by the fact that they can be freely used and shared by anyone and for any purpose, play an important role due to the social and economic impact they can produce, such as, for instance, the possibility of fostering the development of new services based on them, as well as the transparency and the democratic and participatory processes in public policies. In the field of geographic information and Earth Observation (EO), the satellite images collected by Landsat and Sentinel initiatives are the most typical example of open data. The former, provided by National Aeronautics and Space Administration (NASA) and United States Geological Survey (USGS), have a geometric resolution of 30m and have been accessible for decades, whereas the latter, released by the European Union's Copernicus program, have an accuracy of up to 10m and have been available since 2015. According to the literature, both of them are useful for investigating and monitoring natural resources as well as environmental phenomena that occur on the Earth's surface, allowing for the assessment of numerous surface environmental variables on a local and regional scale. Among these, the land surface albedo, which represents the capability of a surface to reflect incident solar radiation, is a useful parameter for climatic and hydrological studies, both in urban and rural contexts. Moreover, the growing attention to the effects of climate change and urbanization on the environment and territory, such as, for example, the Urban Heat Island (UHI) phenomenon, desertification, and drought, makes it necessary for these aforementioned sources of information to be freely and easily available to citizens, researchers and decision-makers.

The objective of this study is to estimate the broadband land surface albedo and its spatial and temporal variability using accessible data from the Landsat 8 and Sentinel-2 satellites over two separate study areas: the city of Bari, in Southern Italy, and the city of Berlin, in North-eastern Germany. Because these two pilot sites have such disparate geomorphological features, they allow generalizing of the research conclusion independent of environmental context. For this purpose, various Landsat 8 and Sentinel-2 satellite images, very close for acquisition time and date, and collected in different seasons, from 2018 to 2019, were used. Furthermore, the performance of the two implemented algorithms, namely the Silva et al. approach for Landsat 8 data and the Bonafoni et al. technique for Sentinel-2 data was assessed and statistically compared. Urban Atlas 2018 land use/land cover (LU/LC) class vector data, provided in an open format by the Copernicus land monitoring service, were used to better explore the variability of the albedo within each case study. These data were processed in the Google Earth Engine (GEE) platform, which is free-to-use for research and non-commercial use, and consists of an integrated data catalogue mainly composed by open raster and vector data, e.g. Landsat and Sentinel images. Such catalogue, daily updated, is directly connected with the interactive programming environment, on which it is possible to process satellite images by developing own codes in JavaScript or Python languages. Most of its available tools are in open-source format. The statistical analysis, on the other hand, was carried out using the free and open-source R environment.

For both case studies, the investigation revealed that the Landsat 8 approach produced somewhat higher mean albedo values than the Sentinel-2 methodology. So far, the statistical comparison indicated that, for the Bari location, all of the returned Landsat 8 and Sentinel-2 albedo maps were strongly correlated, with a correlation coefficient (ρ) higher than 0.84; for Berlin, instead, a medium-high correlation was discovered (ρ > 0.78). Additionally, for both sites, the findings appear to be more correlated when spring and summer scenarios are considered rather than other seasons. Indeed, the correlation between Landsat 8 and Sentinel 2 images appears to follow the same seasonal pattern, though more satellite images from more years should be investigated for a more accurate interpretation. The dependability of the two approaches will be evaluated in the future through the collection of ground control points in field data campaigns. These new data will enable the most accurate findings to be detected and the other methods to be calibrated to increase their reliability.