IEEE Geoscience and Remote Sensing Magazine - September 2019 - 26

more appropriately. In addition, we recommend the use of
increasingly available ground measurements to further improve this data set.
Our results indicate that, for those studies interested in
using the categories based on PM 2.5 concentrations, the
product can be used directly without many calibrations,
as the categories based on remotely sensed data match
well with those from the ground stations. Although poor
performance was observed when PM 2.5 was lower than
35 μg/m3, using the broader category, the data can also be
used because most areas of China were exposed to PM 2.5
greater than 35 μg/m3. In addition, for studies interested
in the relative values (for example, high value versus low
value) but not the absolute values of PM 2.5 concentration,
the product also works well. For instance, studies that
use the WHO standards of 35 μg/m3 and 70 μg/m3 have
relatively small uncertainties. In addition, spatial-pattern
analyses that focus on regional differentiation and the
spatial interaction between high and low PM 2.5 pollution
levels can directly use this product. However, this product
might not be appropriately used for comparisons between
years due to its large uncertainty.
It is not appropriate to use the absolute values of
PM 2.5 concentration directly if accurate absolute values
are needed in the research. Studies that need to use absolute values of PM 2.5 concentration based on remotely
sensed estimations must calibrate them first. For the 2014
product over China, we suggest increasing the values by
22% when the remotely sensed PM 2.5 concentration is
lower than 35 μg/m3, increasing the values by 27% when
the remotely sensed PM 2.5 concentration ranges from
35 to 70 μg/m3, and increasing the values by 43% when
the remotely sensed PM2.5 concentration is greater than
70 μg/m3. In addition, if a study is concerned with national average values, we suggest increasing the remotely
sensed PM 2.5 concentration by 29%. A point-based correction is also provided in the supplementary materials.
(The supplementary materials are available on IEEE
Xplore.) For the 2015 product, we suggest reducing the
values by 24% when the remotely sensed PM 2.5 concentration is lower than 35 μg/m 3, reducing the values by
4% when the remotely sensed PM2.5 concentration is from
35 to 70 μg/m3, and increasing the values by 20% when
the remotely sensed PM 2.5 concentration is greater than
70 μg/m3. In addition, if the study is concerned with national average values, we suggest reducing the remotely
sensed PM 2.5 concentration by 4%.
There were also some uncertainties that should be noted when considering comparisons in pertinent research.
Absorption of water by particles would affect physical
and chemical properties [20] and, thus, can bring uncertainty to the satellite estimation of PM2.5 concentrations
[21], [22]. The PM2.5 concentration measurements' change
from 35 to 50% relative humidity can lead to an increase
of 5-10% in PM2.5 aerosol-bound water per PM2.5 dryaerosol mass [23]. We also found slopes of approximately
26

0.8 to 0.95 for this article, suggesting that roughly half of
this bias comes from relative-humidity measurement differences rather than a true source of error in either PM2.5
data source.
The impact of ground-truth monitor placement is another important element that needs to be considered when
comparing against satellite-retrieved PM 2.5 measurements.
Satellite PM2.5 measurement provides an estimate over an
area, whereas ground-based monitors provide measurements at a certain discrete location. Such differences can
produce wide variability between the data sets, even in
cases where both are correct. This difference can be particularly relevant for ground-based monitors in heavily
polluted locations, such as those near major roadways and
industrial complexes, where pollution levels are unlikely
to be consistent across the surrounding square kilometers.
The potential effect of such locations strongly suggests
needed future work that detailed urban land-use information could support.
The current PM2.5 concentration data sets took into account many ground monitors from 2008 to 2013 [5]. Those
ground measurements, however, were mainly deployed in
high-income countries (for example, in the United States
and Europe), with few available in low- and middle-income
countries (such as China and India). With increasing concern about PM2.5 pollution, poorer countries have begun
building up their nationwide ground-monitoring networks. For example, China has been working to establish
a ground-monitoring network since 2012 and currently has
more than 1,500 national monitoring stations. Remotely
sensed estimates should take advantage of these newly added ground measurements to further improve the accuracy
of the product.
CONCLUSIONS
Studies on PM2.5 pollution have been conducted in various
fields using the remotely sensed global annual PM2.5 concentration product. However, the uncertainty of the product
still needs to be examined, although it was initially verified
by limited ground measurements outside North America.
This uncertainty may lead to the misuse of the product. For
this research, we collected a total of 1,815 ground-measurement records from China's latest air-quality monitoring network and compared them with the remotely sensed PM2.5
concentration product from both 2014 and 2015. In addition, changes in the remotely sensed PM2.5 concentration
were validated using the ground measurements. Based on
those comparisons, we provided suggestions for the application of the current version of the product in future research
in China.
Remotely sensed PM2.5 concentrations have a strong
linear relationship with those from monitoring stations,
but PM 2.5 concentrations were generally below those
given by ground-based monitoring stations in 2014. In
particular, the remotely sensed PM 2.5 concentrations
were much lower than those for ground monitoring when
IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

SEPTEMBER 2019

IEEE Geoscience and Remote Sensing Magazine - September 2019

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - September 2019