Urban climate monitoring network design: Existing issues and a cluster-based solution

Dense sensor networks are being built to collect urban climate information in global cities, yet the performance of network design has rarely been accessed. Existing studies have three major issues: 1) focusing on summertime and lack of seasonal variation, 2) overlooking meteorological variables other than air temperature, 3) not incorporating the current network into the future design. In this study, we proposed a cluster-based network design for monitoring meteorological variables within the urban canopy layer, and examined its applications in Beijing and Hong Kong by using weather simulation data as ground truth. The clustering analysis separates the city into groups with similar urban climate characteristics, and consequently one sensor is sufficient to collect representative meteorological information for each group. Results show a robust design strategy is to train the cluster model with multiple meteorological variables that contain seasonal variations. Utilizing the cluster-based design strategy, we optimize the current network by rearranging sensor locations. For the study period, the sampling error of meteorological variables by the rearranged network is 22.7% smaller in Beijing and 10.7% smaller in Hong Kong than that by the current network. With a sampling ratio of 6.3%, the expanded monitoring network has a mean bias of 0.58 °C (0.44 °C) for representing the air temperature variability across Beijing (Hong Kong). The proposed method is not sensitive to the cluster models, background climate, and resolution of the weather simulation data. The design strategy thus can be applied to other cities for establishing dense urban climate monitoring networks.