An Evaluation of Four MODIS Collection 6 Aerosol Products in a Humid Subtropical Region

Ming Zhang, Bo Huang, Qingqing HeRemote 2017 · DOI

TL;DR

This study evaluates four MODIS Collection 6 aerosol optical depth (AOD) products in the humid subtropical Pearl River Delta region. It finds that while Terra and Aqua products are similar, the Dark Target (DT) algorithm generally outperforms Deep Blue (DB), though both have limitations in urban areas. Crucially, the paper highlights that all products suffer from low sampling frequencies, especially in this region, which must be considered for quantitative applications.

What Problem Does This Paper Address?

Atmospheric aerosols play a significant role in climate and human health, and satellite remote sensing, particularly from MODIS instruments on Terra and Aqua, has become a vital tool for monitoring their properties. Aerosol Optical Depth (AOD) is a key metric, and MODIS provides several quality-assured AOD products. However, these products are generated using different algorithms (Dark Target - DT, Deep Blue - DB) and resolutions, and their performance can vary significantly depending on the geographical region and surface characteristics. Previous studies have often focused on limited time periods or specific aspects of these products. This paper aims to provide a comprehensive, long-term evaluation of four specific MODIS Collection 6 AOD products within the complex and challenging environment of the Pearl River Delta (PRD), a humid subtropical region, to understand their strengths, weaknesses, and suitability for quantitative analysis.

Methods

The study employs a standard validation approach by comparing satellite-derived AOD products with ground-based measurements from the Aerosol Robotic Network (AERONET). Key methodological aspects include:

  • Spatio-temporal Collocation: Satellite data pixels near AERONET stations were spatially averaged, and AERONET data within a one-hour window around the satellite overpass time were temporally averaged. A minimum of five collocations per site was required.
  • AERONET Data Interpolation: AERONET AOD at 550 nm was derived by interpolating measurements at other wavelengths using a quadratic fit.
  • Statistical Metrics: The evaluation used standard metrics such as R-squared (R²), Root Mean Square Error (RMSE), Relative Mean Bias (RMB), and the percentage of data points falling within the expected error (EE) bounds.
  • Site-Level and Regional Analysis: Comparisons were performed both at individual AERONET sites and across the entire PRD region to capture variations in performance.
  • Analysis of Merging Protocol: The paper specifically investigates the protocol used to merge DT and DB products, examining how it handles data gaps and the contribution of each algorithm.

Datasets

  • MODIS Collection 6 (C6) Aerosol Products:

    • Terra and Aqua MODIS DT 10 km AOD
    • Terra and Aqua MODIS DB 10 km AOD
    • Terra and Aqua MODIS merged DT/DB 10 km AOD (DTDB)
    • Terra and Aqua MODIS DT 3 km AOD These products were obtained for the period 2006-2015. Only high-quality retrievals were used.
  • AERONET Data: Level 2 AOD products from four AERONET stations in the PRD region (HK PolyU, Zhongshan Univ, Kaiping, HK Sheung) were used for validation.

  • MODIS NDVI Data: Monthly vegetation indices (MOD/MYD13C2) were used to understand surface conditions and their influence on algorithm performance, particularly for the DT/DB merging protocol.

Key Insights and Ideas

  • Algorithm Performance Varies with Surface Type: The Dark Target (DT) algorithm, which relies on dark surfaces like vegetation, performs better in vegetated areas. The Deep Blue (DB) algorithm, designed for brighter surfaces, struggles in complex coastal urban environments where surface heterogeneity is high.
  • Urban Areas Pose Challenges for DT: While DT generally outperforms DB, it can overestimate AOD in highly urbanized areas, likely due to the algorithm's sensitivity to surface reflectance variations and shadows in such environments.
  • DTDB Merging Protocol Needs Improvement: The current DTDB merging protocol, which prioritizes DT over DB based on NDVI, is not effective in the PRD. Because the PRD is largely vegetated (high NDVI), the protocol often excludes valid DB retrievals even when DT data is missing, failing to achieve its goal of reducing data gaps.
  • DT 3 km Mimics DT 10 km but with Data Gaps: The finer-resolution DT 3 km product closely follows the DT 10 km product but suffers from significantly reduced spatial coverage over water-contaminated areas.
  • Sampling Frequency is a Major Limitation: Across all MODIS products, the sampling frequency (percentage of valid data) is consistently low, often below 25%, especially in the PRD. This is due to cloud cover (common in humid subtropical regions) and quality assurance filtering. This low sampling is a critical factor for any quantitative analysis.

Key Results

  • Terra and Aqua Similarity: Terra and Aqua MODIS AOD products show quantitatively similar performance, as expected.
  • DT vs. DB Accuracy: The DT 10 km product generally shows better correlation (higher R²) and lower RMSE with AERONET data compared to the DB 10 km product.
  • DB Underestimation: The DB 10 km product tends to underestimate AOD values over the PRD region.
  • DT Overestimation in Urban Areas: The DT 10 km product can overestimate AOD in highly urbanized areas like "Zhongshan Univ."
  • DTDB Coverage: The DTDB 10 km product does not significantly improve spatial coverage over the DT 10 km product in the PRD, as it relies heavily on DT retrievals.
  • DT 3 km Coverage: The DT 3 km product has less spatial coverage than the DT 10 km product, particularly over water.
  • Sampling Frequency: Mean sampling frequencies for all products are typically below 25%.

Limitations and Future Work

  • Limited AERONET Sites: The study acknowledges that the number of AERONET collocations, especially at certain sites, was limited, making it difficult to definitively infer superiority between some products.
  • DT/DB Merging Protocol: The paper explicitly calls for an improvement to the DT/DB merging protocol to better utilize DB retrievals and create more complete AOD datasets in regions like the PRD.
  • Sampling Frequency Mitigation: The authors suggest that for humid subtropical areas with low sampling frequencies, spatiotemporal geostatistical methods could be a promising approach to compensate for data gaps.
  • Algorithm Development: The study's findings are intended to assist users in selecting appropriate datasets and can serve as a reference for future algorithm development and refinement at a regional scale.

Summary updated

Abstract

Moderate resolution imaging spectroradiometer (MODIS) aerosol optical depth (AOD) products have been widely used to characterize the temporal variations and spatial distributions of atmospheric aerosols. In the present study, we evaluate the performance of four Terra and Aqua MODIS Collection 6 (C6) quality assured AOD products in the Pearl River Delta (PRD) region, a humid subtropical region. The 10 km AOD products retrieved by the Dark Target (DT) and Deep Blue (DB) algorithms, the merged DT/DB (DTDB) 10 km product, and the DT 3 km AOD product were obtained for 2006–2015. These products were compared with Aerosol Robotic Network (AERONET) observations, and with each other. The Terra- and Aqua-derived AODs are quantitatively similar. However, there are significant differences among the four AOD products. The DT 10 km product correlates more closely with AERONET AOD observations than does the DB 10 km product. The latter tends to underestimate the AOD, whereas the former typically overestimates it for highly urbanized areas. The DTDB 10 km product is mainly derived from the DT 10 km product; it does not provide a gap-filled data set, because valid DB 10 km retrievals are not included in the merged product even when DT 10 km retrievals are unavailable. Therefore, the DT/DB merging protocol should be improved. The DT 3 km AOD product closely mimics the DT 10 km product; however, it contains fewer data than the DT 10 km product over water-contaminated areas. In addition, although the quality assured AOD products are recommended for use in quantitative applications by the MODIS aerosol science team, the sampling frequency of these products is generally lower than 25%. Thus, the sampling issues of these products should be considered in humid subtropical areas.
BibTeX
@inproceedings{zhang2017evaluation,
  author = {Ming Zhang and Bo Huang and Qingqing He},
  title = {{An Evaluation of Four MODIS Collection 6 Aerosol Products in a Humid Subtropical Region}},
  booktitle = {Remote},
  year = {2017},
  doi = {10.3390/rs9111173},
}
Figure