In this study the infrastructure development and operation of an urban human comfort monitoring networks and information system in Szeged (Hungary) and Novi Sad (Serbia) are discussed. The selection of the representative sites of the networks is based primarily on the pattern of the Local Climate Zones (LCZ) in and around the cities. After the processing of the incoming data (air temperature and relative humidity, as well as global radiation and wind speed) a human comfort index (Psychologically Equivalent Temperature) is calculated from the four parameters with a neural network method, then the measured and calculated parameters interpolated linearly into a regular grid with 500 m resolution. As a public information maps and graphs about the thermal and human comfort conditions in the cities appear in 10-minute time steps as a real-time visualisation on the internet. As the preliminary case studies show the largest intra-urban thermal differences between the LCZ areas occurred in the nocturnal hours reaching even 5ºC in early spring. In the spatial distribution of human comfort conditions there are distinct differences in the strength of the loading between the neighbourhoods during the daytime. Overall, it can be stated that the monitoring networks are able to provide beneficial information for urban climate research and for the wider audience, specialist and non-specialist, too. They record data with proper spatial and temporal resolution and the accuracy of the sensors is satisfactory. Based on our evaluation the site selection was successful, as the temperature has different characteristics at sites with differently classified environment. The planned operation time of networks is minimum five years so in the case of this period the available long data series will provide an opportunity to perform spatially and temporally very detailed climatological investigations in relation to urban environment.

Large cities with high population density are inherently vulnerable to severe weather, such as severe rainstorm, lightning, and tornado. In recent years, urban inundations caused by localized severe rainstorm are becoming a matter of wider concern in Japan. Such severe rainfall has a small horizontal scale, random occurrence, and develops without being triggered by any organized cloud system, thereby making it difficult to prompt warning of such event using existing nowcast system. On the other hand, an X-band multi-paramerer (MP) radar network is increasing its members in Japan with purpose to determine the mechanism of local severe weather and to develop the algorithm for estimating various rainfall and microphysical parameters from radar measurements. Meanwhile, Tokyo Metropolitan Area, where approximately 30 million people live at, has a high risk of inland flooding because of the large asphalt pavement ration and the closely spaced concrete buildings it contains. The early warning of localized severe rainstorm for flood fighting and risk avoidance is one of the important roles for the network. This study focused on rain drops hanging in the air, utilized the time lapse before they reach the ground to improve the accuracy of nowcast for localized severe rainstorm. The method is inspired from the RadVil model proposed by Boudevillain et al. (2006), which used vertically integrated liquid water content (VIL) in the rainfall forecasting model. VIL is a convenient parameter that includes vertical information on total amount of water above a point but does not need to consider the microphysical processes of rainfall. However, when using the rain data measured by conventional radar, the forecasting performance is not stable since the relationship between VIL and rainfall rate is too much complex and variable in place and time. In this study, authors derived VIL values from X-band dual polarization radar network and introduced a mechanism to adjust the relationship between VIL and rainfall rate. In addition, an e-mail alert system is designed to deliver the warning message of severe rainstorm to check the availability of the model during social experiment. As the result, VIL nowcast performed better than the classical rainfall rate nowcast model, and obtained positive evaluations from more than 70 percent users.

Atmospheric concentration of passive scalars such as greenhouse gases or non-reactive pollutants is mainly driven by two processes: (i) the emission at the surface (i.e. the amount of gas/pollutant released into the atmosphere from sources), and (ii) the atmospheric transport (i.e. the wind-driven advection, the temperature and the turbulent transport within the Planetary Boundary Layer). Since atmospheric concentrations of air pollutants directly affect human health and wellness in urban areas, they are one of the most important environmental issues that public administrations have to deal with. This is generally accomplished by controlling pollutant emissions after imposing restrictions, such as traffic control or low emission zones, or supporting improvements (e.g. in buildings energy efficiency or in urban planning design through more green areas). However, the effect of such policies on actual concentrations is not straightforward, since concentrations are mediated by atmospheric processes: for example, favourable atmospheric conditions, such as high wind speed and strong turbulence conditions, may lead to even higher concentration reductions than a drastic emission abatement.

Carefully taking into account both surface emissions and atmospheric dynamics, 3-D atmospheric models coupled to chemical models may provide a comprehensive framework to forecast air pollution levels. However, their practical usefulness is limited by their inherent complexity, lack of detailed space/time-resolved emission inventory, and lack of validation data on transport, dispersion and, finally, emissions amount since urban monitoring networks of chemical and meteorological stations alone are not capable to disentangle the interrelation between emissions and concentrations, so these emissions basically remain unknown.

In this work, a purely data-driven approach based on CO2 emission measurements (a reliable proxy of anthropogenic pollutant emissions) performed by an eddy covariance station located in the city centre of Florence (Italy) was developed. Data were collected for an 8-year (2005-2012) period with a 30-min resolution, representing a comprehensive dataset of city-level surface emissions, atmospheric concentrations and turbulent transport processes and providing the basis to disentangle interrelations between such variables within a non-linear framework. The analysis was based on unsupervised neural networks, used to cluster data into self-organizing maps (SOMs), and supervised networks, to develop predictive models.

Three datasets were analyzed: (i) diurnal, (ii) nocturnal, and (iii) daily overall. As a result, CO2 mean concentrations proved to be mostly affected by the atmospheric conditions, which explain 80.03 (daily) and 94.67% (nocturnal) concentrations, markedly through the wind (47.08 and 48.25%, respectively), while a marginal role is played by urban emissions (about 20%). Also in the diurnal case, atmospheric conditions explained the most (93.21%) concentrations, markedly friction velocity (62.58%) and temperature (30.63%), while emissions remained poorly significant (5.33%).

Applied for CO2, this method can be extended to other non-reactive species once eddy covariance measurements would be available. Summarizing, this level of information may be useful when combined to a meteorological forecast to easily estimate the impact of potential restrictions, thus becoming an air quality control tool for administrative planners to monitor and predict pollutant concentrations and assess the impact of mitigation actions.

A forecasting system composed of the Weather Research and Forecasting (WRF) model coupled with a single-layer urban canopy parameterization scheme is implemented to perform high-resolution forecasts of the human thermal comfort inside the urban area of Trento, a medium-sized city located in the Italian part of the Alps. Simulations with the WRF model are routinely performed at 1-km resolution over the Province of Trento, to provide suitable meteorological upper boundary conditions to the urban canopy model. The single-layer urban parameterization scheme is used to downscale the WRF forecasts inside the urban area, taking into account the local characteristics of the city, and to calculate indexes for the bio-meteorological assessment of the thermal environment. Gridded maps of urban canopy parameters (UCPs) and anthropogenic heat emissions are utilized as input for the urban canopy model. UCPs are obtained with GIS techniques from LIDAR maps with an horizontal resolution of 1 m, while anthropogenic heat emissions from detailed energy consumption and vehicular traffic data. Results from the modeling system are validated against measurements from different field experiments. The validation involves both the forecasts with the WRF model, which are particularly challenging in a region characterized by highly complex terrain, and the thermal field inside the city calculated by the urban parameterization.