iCity 2: Space4iCity

The development of urban digital twins creates geo-data spaces that optimize a wide range of planning, decision-making and economic processes or make them possible in the first place. Official geoinformation such as the official real estate cadastre information system ALKIS forms an essential basis for this. In recent years, immense amounts of sensor data from local measurements with a very high temporal resolution have been added for monitoring urban space, for example in the areas of the environment (air quality, urban climate) and traffic. At the same time, ESA's Copernicus program provides satellite-based earth observation data in ever higher spatial and temporal resolution as open data.

Although much of this data is also relevant for municipal applications, it is rarely used in this context. This is generally due to the significantly lower spatial resolution of earth observation data compared to aircraft and drone-based data collection. On the other hand, this data has a much higher temporal resolution in comparison, as satellites cover an area about once a week, whereas aerial campaigns are carried out at intervals of several years.

The scientific objectives and questions can be summarized as follows:

• Is it possible to increase the spatial resolution of satellite-based measurements of surface temperature using AI tools and ground-based measurements?
• Is AI capable of recognizing and characterizing green spaces in urban areas?
• Can earth observation data improve the database of an urban digital twin in such a way that CFD simulations of the urban climate deliver significantly better results?

In order to use earth observation data for municipal applications, the problem of the comparatively low spatial resolution of satellite data must be solved. To this end, deep learning methods are to be developed and neural networks trained in order to realize a spatial disaggregation of the data. The sensor data already available in iCity will be used for training and validation. The concept is to be tested using the two use cases "urban heat islands" and "urban greenery". The information extracted from the earth observation data is to be used in an urban climate simulation in order to evaluate the added value of the data for forecasting the urban climate.

In order to answer these scientific questions, the following results are to be achieved:

• The compilation of satellite time series data of the greater Stuttgart area: Landsat (thermal band) and Sentinel-2 (bands B, G, R, NIR) and processing of the data
• Training and evaluation of neural networks for the disaggregation of surface temperature data (Landsat + Sentinel-2)
• Training and evaluation of neural networks for the characterization of vegetation in urban areas (Sentinel-2)
• Integration of the data into an urban digital twin and comparative simulation of the urban climate with and without information extracted from earth observation data