EGU25-490, updated on 14 Mar 2025
https://doi.org/10.5194/egusphere-egu25-490
EGU General Assembly 2025
© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.
Oral | Monday, 28 Apr, 11:25–11:35 (CEST)
 
Room 2.31
Prediction Air Polution due to Wildfire in Kathmandu Valley: Remote Sensing and Machine Learning Techniques
Sajesh Kuikel, Saugat Sapkota, Dipesh Kuinkel, Him Kiran Paudel, Khagendra Prasad Joshi, Suresh Marahatta, Deepak Aryal, and Binod Pokharel
Sajesh Kuikel et al.
  • Central Department of Hydrology and Meteorology, Kathmandu, Tribhuvan University, Nepal (sajesh.kuikel@cdhm.tu.edu.np)

The Central Himalaya faces significant air pollution challenges, with nearly half of the days each year in Kathmandu Valley surpassing the PM2.5 national air quality guideline of 40 µg/m³. Wildfire smoke, especially during the pre-monsoon season, is a major contributor to these polluted days in the valley and is also a key driver of air pollution in high-altitude regions of the Himalayas. To identify the presence of wildfire smoke in the valley, we utilized multiple datasets, including in-situ observations, Himawari satellite Aerosol Optical Depth (AOD) data, and satellite imagery. Between 2018 and 2023, we identified 114 days that met our criteria for being classified as smoke days during the pre-monsoon season. Due to the lack of in-situ observation data prior to 2018, we utilized PM2.5 data from the Copernicus Atmosphere Monitoring Service (CAMS), MODIS AOD and wildfire data to classify smoke days for earlier years. Using in total of nine variables, we trained a Random Forest Classifier model on the previously categorized dataset, our model performed with outstanding accuracy (0.91), where AOD in nearby regions (~150km) was found to be the most significant parameter, followed by number of wildfires occured in the past three days. In total, 213 days were classified as wildfire smoke days in Kathmandu Valley from 2003 to 2023, with 2021 recording the highest number of smoke days and 2009 with the highest amout of PM2.5 due to the smoke. Additionally, these wildfire smoke days do not folow any trend but were strongly correlated with wildfire occurrences in nearby regions. The Machine Learning model further highlighted the high correlation between wildfire numbers in surrounding areas and the presence of high air pollution in the valley. This research contributes to policymaking on air pollution and enhances preparedness for extreme pollution events in Kathmandu Valley, ultimately helping to protect public health and well-being.

How to cite: Kuikel, S., Sapkota, S., Kuinkel, D., Paudel, H. K., Joshi, K. P., Marahatta, S., Aryal, D., and Pokharel, B.: Prediction Air Polution due to Wildfire in Kathmandu Valley: Remote Sensing and Machine Learning Techniques, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-490, https://doi.org/10.5194/egusphere-egu25-490, 2025.