Have roofs in Berlin become greener? Evaluation of Berlin's green roof subsidy program performance using geodata and deep learning

Abstract

Green roofs are one of the most widely applied blue-green infrastructure in urban regions to serve several purposes moving towards climate change mitigation and urban adaptation. Their large-scale adoption is critical in enhancing resilience against urban hazards, such as urban flooding, urban heat island effects, and biodiversity loss. Currently, the most popular policy format to encourage their roll-out is subsidy programs. However, the success of such programs is oftentimes evaluated based on siloed governmental data, local evaluation reports, and non-recurrent monitoring campaigns, which may become inconsistent and incomparable across temporal scales and different geographical regions. Due to the lack of open data, complementary metadata, and standard quantitative evaluation tools, monitoring and consistently comparing the effectiveness of different green roof incentivization policies is a challenge in practice. This lack of data and high cost of frequent large-scale monitoring campaigns also hinders city-wide spatial distribution analysis of green roofs and identification of green roof development potential, which could support policymakers in devising effective and sustainable urban management strategies.Moving towards an automated frequent monitoring of green roof development, previous work by Wu and Biljecki developed “Roofpedia”, an open-source deep learning algorithm for green roof mapping and urban sustainability evaluation using satellite imagery. In this work, we validate Roofpedia and evaluate its accuracy in automatically identifying and classifying green roofs from satellite images with public ground truth data in Berlin, Germany. Furthermore, we develop a Berlin-based case study where Roofpedia is applied using geospatial data across temporal scales to assess the efficacy of Berlin’s green roofing subsidy program "GründachPLUS", which has provided 2.7 Million Euros of funding for green roof construction since 2019. We first retrieve open-access orthoimagery data, then extract green roof coverages in Berlin across two temporal steps (i.e., before and after subsidy program instigation), and finally evaluate how effectively and promptly the subsidy program fostered the development of green roofs. This study contributes a Machine Learning-based add-on to the current evaluation protocol of the Berlin municipality, which is implemented via threshold-based spectral analysis. We analyze the spatial distribution of green roofs and provide insights into further green roof potentials in the city of Berlin, by identifying interesting hotspots for future green roof development. Upon imagery availability, this automated assessment may be extended to multiple cities to enable comparative studies of various green roofing incentivization policies and offer a transferrable and scalable policy evaluation framework.