School of Chemical Sciences , University of Auckland , Auckland 1010 , New Zealand.
School of Environment , University of Auckland , Auckland 1010 , New Zealand.
ACS Sens. 2018 Apr 27;3(4):832-843. doi: 10.1021/acssensors.8b00074. Epub 2018 Mar 15.
We provide a simple, remote, continuous calibration technique suitable for application in a hierarchical network featuring a few well-maintained, high-quality instruments ("proxies") and a larger number of low-cost devices. The ideas are grounded in a clear definition of the purpose of a low-cost network, defined here as providing reliable information on air quality at small spatiotemporal scales. The technique assumes linearity of the sensor signal. It derives running slope and offset estimates by matching mean and standard deviations of the sensor data to values derived from proxies over the same time. The idea is extremely simple: choose an appropriate proxy and an averaging-time that is sufficiently long to remove the influence of short-term fluctuations but sufficiently short that it preserves the regular diurnal variations. The use of running statistical measures rather than cross-correlation of sites means that the method is robust against periods of missing data. Ideas are first developed using simulated data and then demonstrated using field data, at hourly and 1 min time-scales, from a real network of low-cost semiconductor-based sensors. Despite the almost naïve simplicity of the method, it was robust for both drift detection and calibration correction applications. We discuss the use of generally available geographic and environmental data as well as microscale land-use regression as means to enhance the proxy estimates and to generalize the ideas to other pollutants with high spatial variability, such as nitrogen dioxide and particulates. These improvements can also be used to minimize the required number of proxy sites.
我们提供了一种简单、远程、连续的校准技术，适用于具有少数维护良好、高质量仪器（“代理”）和大量低成本设备的层次网络。这些想法基于对低成本网络目的的明确定义，这里定义为在小时空尺度上提供可靠的空气质量信息。该技术假设传感器信号具有线性。它通过将传感器数据的平均值和标准偏差与同一时间段内代理数据得出的值进行匹配，来估计传感器的斜率和偏移量。该方法非常简单：选择合适的代理和足够长的平均时间，以消除短期波动的影响，但又要足够短以保留规律的日变化。使用运行的统计措施而不是站点之间的互相关，意味着该方法对缺失数据的时期具有鲁棒性。该方法首先使用模拟数据进行开发，然后使用现场数据进行演示，现场数据来自真实的低成本基于半导体的传感器网络，时间尺度为每小时和 1 分钟。尽管该方法几乎是天真的简单，但它对于漂移检测和校准校正应用都具有鲁棒性。我们讨论了使用一般可用的地理和环境数据以及微观尺度的土地利用回归作为增强代理估计和将这些想法推广到其他具有高空间变异性的污染物（如二氧化氮和颗粒物）的方法。这些改进还可用于最小化所需代理站点的数量。
