Internet Education and Research Laboratory (intERLab), Asian Institute of Technology, Pathum Thani, Thailand.
Faculty of Public Health, Thammasat University, Pathum Thani, Thailand.
Environ Pollut. 2023 Dec 1;338:122701. doi: 10.1016/j.envpol.2023.122701. Epub 2023 Oct 5.
The widespread adoption of Internet of Things (IoT) sensors has revolutionized our understanding of the formation and mitigation of air pollution, significantly improving the accuracy of predictions related to air quality and emission sources. This study demonstrates the use of IoT air quality sensors to detect forest fire incidents by focusing on an area affected by forest fires in Tak Province, Thailand, from January to May 2021. We employed PM and carbon monoxide measurements from IoT sensors for forest fire detection and utilized the number of hotspots reported through satellite and human observations to identify forest fire incidents. Our data analysis revealed three distinct periods with forest fires and three periods without fires (non-forest fires). For model training, two forest fire and non-forest fire periods were selected and the remaining periods were set aside for validation. J48, a computer algorithm that helps make decisions by organizing information into a tree-like structure based on key characteristics, was used to construct the decision-tree model. Our model achieved an accuracy rate of 72% when classifying forest fire incidents using the training data and a solid accuracy of 69% on the validation data. In addition, we investigated the dispersion of PM plumes using a regression model. Notably, our findings highlighted the robust explanatory power of the lag time in PM, for predicting PM, in the next 15 min. Our analysis highlights the potential of IoT-based air quality sensors to enhance forest fire detection and predict pollution plume dispersion once fires are detected.
物联网 (IoT) 传感器的广泛采用彻底改变了我们对空气污染形成和缓解的理解，极大地提高了与空气质量和排放源相关的预测准确性。本研究通过关注泰国塔克拉省 2021 年 1 月至 5 月期间受森林火灾影响的地区，展示了使用物联网空气质量传感器来检测森林火灾事件。我们利用物联网传感器测量的 PM 和一氧化碳来检测森林火灾，并利用卫星和人工观测报告的热点数量来识别森林火灾事件。我们的数据分析揭示了三个明显的森林火灾期和三个无火灾期（非森林火灾期）。为了进行模型训练，选择了两个森林火灾和非森林火灾期，其余的时期则留作验证。J48 是一种计算机算法，它通过根据关键特征将信息组织到树状结构中来帮助做出决策，我们使用 J48 构建决策树模型。我们的模型在使用训练数据对森林火灾事件进行分类时达到了 72%的准确率，在验证数据上的准确率也相当稳定，为 69%。此外，我们还使用回归模型研究了 PM 羽流的扩散情况。值得注意的是，我们的研究结果突出了滞后时间在 PM 预测中的强大解释能力，可用于预测下一个 15 分钟的 PM 值。我们的分析强调了基于物联网的空气质量传感器在增强森林火灾检测和预测火灾发生后污染羽流扩散方面的潜力。
