FireNet: Dense Forecasting of Wildfire Smoke Particulate Matter Using Sparsity Invariant Convolutional Neural Networks

Renhao Wang 1, Ashutosh Bhudia 1, Brandon Dos Remedios 1, Minnie Teng 1, Raymond Ng 1
1University of British Columbia,

NeurIPS 2020 ML for Public Health [Best Paper Award]
arXiv BibTeX

Sample forecasts 24 hours in advance. We accurately capture highs and lows of PM2.5 in correspondence with the Pacific Northwest fire season (beginning in April, peaking in July and August, and ending in October). Despite lacking reliable ground truth on over 99% of regions, FireNet captures complex and diverse PM2.5 falloff patterns and interactions between smoke dispersion from various points of ignition.

Abstract

Accurate forecasts of fine particulate matter (PM 2.5) from wildfire smoke are crucial to safeguarding cardiopulmonary public health. Existing forecasting systems are trained on sparse and inaccurate ground truths, and do not take sufficient advantage of important spatial inductive biases. In this work, we present a convolutional neural network which preserves sparsity invariance throughout, and leverages multitask learning to perform dense forecasts of PM 2.5values. We demonstrate that our model outperforms two existing smoke forecasting systems during the 2018 and 2019 wildfire season in British Columbia, Canada, predicting PM 2.5 at a grid resolution of 10 km, 24 hours in advance with high fidelity. Most interestingly, our model also generalizes to meaningful smoke dispersion patterns despite training with irregularly distributed ground truth PM 2.5 values available in only 0.5% of grid cells.



Model Architecture





We use a standard sparsity invariant CNN backbone to featurize sparse spatial maps of wildfire-related data. We define two branches which learn to output the noisy PM2.5 values of existing forecasting frameworks BlueSky and FireWork. A third branch regresses our full density PM2.5 values. Taken together, the model is trained end-to-end via a combination of these losses.