Spatial prediction of urban–rural temperatures using statistical methods

Spatial information on climatic characteristics is beneficial in e.g. regional planning, building construction and urban ecology. The possibility to spatially predict urban–rural temperatures with statistical techniques and small sample sizes was investigated in Turku, SW Finland. Temperature observations from 36 stationary weather stations over a period of 6 years were used in the analyses. Geographical information system (GIS) data on urban land use, hydrology and topography served as explanatory variables. The utilized statistical techniques were generalized linear model and boosted regression tree method. The results demonstrate that temperature variables can be robustly predicted with relatively small sample sizes (n ≈ 20–40). The variability in the temperature data was explained satisfactorily with few accessible GIS variables. Statistically based spatial modelling provides a cost-efficient approach to predict temperature variables on a regional scale. Spatial modelling may aid also in gaining novel insights into the causes and impacts of temperature variability in extensive urbanized areas.