Urban sensing and weather prediction: Can iot devices be used to improve weather prediction in cities?

Abstract

Data assimilation (DA) in numerical weather prediction (NWP) has relied on observations sourced from variety of sources like radiosondes, dropwindsondes, meteorological sensors and remote sensing through satellites. The sensors used for collecting meteorological variables such as wind speed, wind direction, temperature, relative humidity are designed to provide high quality observations. Additionally, a set of defined instructions must be followed for installation of such systems, with the presumption that the sensors will behave as defined in the user m anual. Due to the aforementioned r easons, firstly, these systems are expensive to install and operate, and secondly, they cannot be installed at a large scale within cities, thus requiring alternative ways of sensing environmental conditions within such areas. In recent times, Internet of Things (IOT), Big Data and Cloud computing has been attracting considerable attention. This has been made possible by the availability of high speed connectivity, ease of access to high quality computing resources at pay-per-use basis and the increased emphasis on informed decision making. Due to such developments, there has been increased diffusion of smart systems equipped with miniaturized sensors allowing such systems to adapt to their environmental conditions e.g., smart air-conditioners controlling the indoor conditions in response to the changing ambient environment. These disruptive technologies have led to the development of platforms such as Tulip1, Array of Things2 and Dryp3, that rely on alternative sensing methods and technologies. Historically, weather and climate predictions has been carried out by the national weather agencies. These agencies use weather models together with the data from their observation networks to provide weather forecasts for public use on time scales ranging from daily, yearly to decadal. Despite the considerable advancements in modelling and computing systems, the weather forecasts from agencies can only attain spatial resolution of 1-10km. Due to the cost and computational systems needed to run such models, there has been a growing interest in combining big data and machine learning to aid in localizing the predictions performed by these models. In the current study, we investigate the implications of combining observations from Array of Things (Catlett et al., 2017) network installed in City of Chicago with the model outputs from Conformal Cubic Atmosphere Model (CCAM). To that end, the study utilizes modelled and observed air temperature over a one month period. The preliminary results showed a good correlation between the modelled and observed air temperature. However, the sensors mounted on AOT node use different sensing techniques to measure air temperature. Due to these differences, a considerable spread exists in the air temperature observed by the different sensors mounted on a single AOT node. This observed spread in the air temperature underlines the need of caution when using data from IOT devices. Further evaluation against the data from a co-located meteorological sensor may elucidate the implications of this spread when data from IOT devices is combined with the models.