Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Assessing the Potential of PlanetScope Satellite Imagery to Estimate Particulate Matter Oxidative Potential

Ian Hough, Loïc Argentier, Ziyang Jiang, Tongshu Zheng, Mike Bergin, David Carlson, Jean-Luc Jaffrezo, Jocelyn Chanussot, Gaëlle Uzu

TL;DR

This study investigates whether high-resolution PlanetScope satellite imagery can be used to estimate particulate matter oxidative potential (OP), measured by OPAA and OPDTT, in Grenoble, France. The authors develop a deep learning pipeline combining a CNN-based image encoder (ResNet50) with a multilayer perceptron, optionally augmented by meteorological data, to predict daily OP at 1 km resolution. Across multiple modeling paradigms—baseline meteorology-only, transfer learning with ImageNet features, fine-tuning, and contrastive learning—the best OPAA performance reached about $R^2 \approx 0.63$ and OPDTT about $R^2 \approx 0.48$, with image-only signals still explaining substantial variance; PM10 predictions followed a similar but weaker pattern. The results suggest a feasible, low-cost approach to broaden OP monitoring, though validation in other regions and larger datasets is needed to confirm generalizability and quantify spatial variability.

Abstract

Oxidative potential (OP), which measures particulate matter's (PM) capacity to induce oxidative stress in the lungs, is increasingly recognized as an indicator of PM toxicity. Since OP is not routinely monitored, it can be challenging to estimate exposure and health impacts. Remote sensing data are commonly used to estimate PM mass concentration, but have never been used to estimate OP. In this study, we evaluate the potential of satellite images to estimate OP as measured by acellular ascorbic acid (OP AA) and dithiothreitol (OP DTT) assays of 24-hour PM10 sampled periodically over five years at three locations around Grenoble, France. We use a deep convolutional neural network to extract features of daily 3 m/pixel PlanetScope satellite images and train a multilayer perceptron to estimate OP at a 1 km spatial resolution based on the image features and common meteorological variables. The model captures more than half of the variation in OP AA and almost half of the variation in OP DTT (test set R2 = 0.62 and 0.48, respectively), with relative mean absolute error (MAE) of about 32%. Using only satellite images, the model still captures about half of the variation in OP AA and one third of the variation in OP DTT (test set R2 = 0.49 and 0.36, respectively) with relative MAE of about 37%. If confirmed in other areas, our approach could represent a low-cost method for expanding the temporal or spatial coverage of OP estimates.

Assessing the Potential of PlanetScope Satellite Imagery to Estimate Particulate Matter Oxidative Potential

TL;DR

This study investigates whether high-resolution PlanetScope satellite imagery can be used to estimate particulate matter oxidative potential (OP), measured by OPAA and OPDTT, in Grenoble, France. The authors develop a deep learning pipeline combining a CNN-based image encoder (ResNet50) with a multilayer perceptron, optionally augmented by meteorological data, to predict daily OP at 1 km resolution. Across multiple modeling paradigms—baseline meteorology-only, transfer learning with ImageNet features, fine-tuning, and contrastive learning—the best OPAA performance reached about and OPDTT about , with image-only signals still explaining substantial variance; PM10 predictions followed a similar but weaker pattern. The results suggest a feasible, low-cost approach to broaden OP monitoring, though validation in other regions and larger datasets is needed to confirm generalizability and quantify spatial variability.

Abstract

Oxidative potential (OP), which measures particulate matter's (PM) capacity to induce oxidative stress in the lungs, is increasingly recognized as an indicator of PM toxicity. Since OP is not routinely monitored, it can be challenging to estimate exposure and health impacts. Remote sensing data are commonly used to estimate PM mass concentration, but have never been used to estimate OP. In this study, we evaluate the potential of satellite images to estimate OP as measured by acellular ascorbic acid (OP AA) and dithiothreitol (OP DTT) assays of 24-hour PM10 sampled periodically over five years at three locations around Grenoble, France. We use a deep convolutional neural network to extract features of daily 3 m/pixel PlanetScope satellite images and train a multilayer perceptron to estimate OP at a 1 km spatial resolution based on the image features and common meteorological variables. The model captures more than half of the variation in OP AA and almost half of the variation in OP DTT (test set R2 = 0.62 and 0.48, respectively), with relative mean absolute error (MAE) of about 32%. Using only satellite images, the model still captures about half of the variation in OP AA and one third of the variation in OP DTT (test set R2 = 0.49 and 0.36, respectively) with relative MAE of about 37%. If confirmed in other areas, our approach could represent a low-cost method for expanding the temporal or spatial coverage of OP estimates.
Paper Structure (35 sections, 3 equations, 18 figures, 8 tables)

This paper contains 35 sections, 3 equations, 18 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Materials
  3. Study area
  4. Particulate matter and oxidative potential
  5. Meteorological data
  6. Satellite data
  7. Methods
  8. Network architecture
  9. Experimental setup
  10. Baseline model
  11. Random features
  12. Transfer learning
  13. Fine-tuning
  14. Contrastive learning
  15. Further experiments
  16. ...and 20 more sections

Figures (18)

  • Figure 1: Study area and air quality monitoring stations. UC: urban center. UB: urban background. SU: suburban. Inset shows study area location in southeastern France.
  • Figure 2: Number of satellite images and air quality observations
  • Figure 3: Example satellite images illustrating that visual haziness is not consistently correlated with air quality. Top row: days with low low PM10, OPAA, and OPDTT at stations UC (A), SU (B), and UB (C). Bottom row (D-E): days with high low PM10, OPAA, and OPDTT.
  • Figure 4: Network architecture. Inputs and outputs are labelled by size (Height x Width x Channels).
  • Figure 5: Observed PM10 (A), OPAA (B), and OPDTT (C) from 2017-02-08 to 2018-03-11
  • ...and 13 more figures