Estimating installed cooling capacities on city scale

Heating and cooling of buildings is one of the largest final energy uses and largest sources of greenhouse gas emissions. To reduce the impact of heating and cooling on our climate, more efficient strategies are needed. Coupling and centralizing the production of heat and cold in combination with underground seasonal thermal storage (UTES) can significantly reduce CO₂ emissions and costs. To plan and implement such strategies for heating and cooling, information on sources and sinks of heat and cold is essential for local authorities. However, spatial information on the cooling sector is rare and difficult to obtain. Often, the theoretical cooling demand of specific buildings and building types is modeled, but not met by air-conditioning equipment in reality. On the other hand, large-scale cooling demand models, which focus on entire countries, may use data from different countries as proxy or are not applicable below kilometer-scale.

In this study, we present a method to identify air-conditioning equipment on the rooftops of buildings and quantify their cooling capacity. Thus, air-cooled and hybrid evaporative condensers, cooling towers and packaged rooftop units are detected on aerial images. Using manufacturer data, regression analyses are created to estimate the cooling capacity based on the size of the units and the number of condenser fans. The unit locations and all required parameters are obtained by convolutional neural network-based pixel classification models, which are easily executable within a geographical information system (GIS) framework. The approach is successfully evaluated by testing the capability of the detection models and comparing our estimated cooling capacities to the actual installed cooling capacities of air-conditioners for different locations. The detection performance strongly depends on the resolution of the used aerial images. At a resolution of 8 cm/pixel, the model detects 93% of the units and the pixel classification overestimates the relevant parameters for the regression by 0.7%. Using the regression analyses, the overall capacity in the evaluated areas is overestimated by 7-21%. To demonstrate the capability of our approach, we map the cooling capacity of air-conditioners in parts of Manhattan. In the Manhattan financial district alone, a cooling capacity of over 2 GW is estimated, which is equivalent to 1.3% of the summer peak load demand of the energy grid of the entire state of New York.

The presented approach is a fast and easy to conduct method that requires little input data. It can detect individual air conditioners over large areas. The obtained information can support the creation of cooling cadastres and can serve as supplement or validation for other cooling demand models, such as building stock models, or example to include additional building types, such as industrial buildings.