The goal of KI-SAN is to develop, validate, and integrate an AI-supported procedure for the building-specific detection of energy renovation states at the neighborhood and city levels. To this end, we train Convolutional Neural Networks (CNNs) on time-series aerial imagery to enable the scalable detection of renovations on roofs and facades. Based on our results, heat and cooling demand analyses for municipal planning processes will be scientifically grounded and refined.

• To what extent can energy renovation measures on roofs and facades be detected solely from time-series aerial imaging data?
• How sufficient is the analysis of roof areas from orthoimage data to reliably estimate the energy renovation status of buildings?
• How does the inclusion of these derived renovation states affect the accuracy of building- and neighborhood-level heat and cooling demand calculations?

We analyze building roofs and facades from oblique aerial imagery, as the renovation status of these components have a significant impact on energy demand. By comparing images taken in different years, we can infer energy renovations and their timing. From this, we determine the correlation degree between roof and facade renovations and investigate the extent to which orthorectified aerial images (i.e. images taken from right above and limited to roof views, often available as Open Data) alone suffice for assessing the energy status of a building. Our methodology will be integrated into the SimStadt simulation platform following successful evaluation, significantly improving the accuracy of existing workflows for building energy demand under consideration of various building renovation measures calculations within the tool.

• Development and validation of AI models for detecting roof and facade renovations from oblique and orthorectified aerial imagery.
• Standardized representation of renovation status in a CityGML/Energy-ADE-compliant data model and implementation of an open API interface for integration into SimStadt.