> This paper discusses how changes in the environment due to indust...
> "Some works of art, even those that do not appear “realistic,” ap...
During his stays in London between the autumn of 1899 and the early...
> "Furthermore, Turner and Monet’s works span the Industrial Revolu...
Aerosols are tiny particles or droplets suspended in the atmosphere...
It is important to benchmark the wavelet method with photographs wh...
This is extreme. 20th/21st century emissions have also been heavily...
> "As a complementary approach, it is also possible to analyze the ...
> “It is clear that industrialization changed the environmental con...
> "Our basic premise is that Impressionism—as developed in the work...
Paintings by Turner and Monet depict trends in 19th century
air pollution
Anna Lea Albright
and Peter Huybers
Edited by William Clark, Harvard University, Cambridge, MA; received November 8, 2022; accepted December 20, 2022
Individual paintings by artists including Vincent van Gogh and Edvard Munch
have been shown to depict specific atmospheric phenomena, raising the question
of whether longer-term environmental change influences stylistic trends in painting.
Anthropogenic aerosol emissions increased to unprecedented levels during the 19th
century as a consequence of the Industrial Revolution, particularly in Western European
cities, leading to an optical environment having less contrast and more intensity.
Here, we show that trends from more figurative to impressionistic representations in
J.M.W. Turner and Claude Monet’s paintings in London and Paris over the 19th
century accurately render physical changes in their local optical environment. In
particular, we demonstrate that changes in local sulfur dioxide emissions are a highly
statistically significant explanatory variable for trends in the contrast and intensity of
Turner, Monet, and others’ works, including after controlling for time trends and
subject matter. Industrialization altered the environmental context in which Turner
and Monet painted, and our results indicate that their paintings capture changes in
the optical environment associated with increasingly polluted atmospheres during the
Industrial Revolution.
air pollution | artwork | environmental reconstruction | atmospheric science
Some works of art, even those that do not appear “realistic,” appear to faithfully record
particular natural phenomena. Edvard Munch’s The Scream (1893), for example, is
argued to depict nacreous clouds (1). Vincent van Gogh’s Moonrise (1889) is dated to
precisely 9:08 p.m. local time on July 13, 1889, using topographic observations, lunar
tables, and letters (2). Nine of Claude Monet’s paintings in his London series are also
dated using solar geometry, with results confirmed by cross-referencing against Monet’s
letters (3). A survey of over 12,000 paintings, moreover, indicates that different schools
reflect local meteorological conditions, such as paler blue skies in the British school than
other contemporaneous European schools (4). Another important example of paintings
depicting the natural environment comes from a set of studies of sunset coloration over
time relative to volcanic eruptions that injected aerosols into the stratosphere (5, 6).
Sunsets seen through an aerosol-laden stratosphere appear redder because of greater
scattering in the limb of the Earth’s atmosphere (7). Across schools of painting, the
red-to-green ratios in sunset paintings from 1500 to 1900 are correlated with independent
proxies of stratospheric aerosol content (5, 6), though difficulty constraining the aerosol
size distribution and solar zenith angle introduces uncertainties to this methodology (8).
Here, we seek to ascertain whether there is a relationship between changes in
atmospheric conditions associated with industrialization and changes in painting style—
primarily that of the British artist Joseph Mallord William Turner (1775 to 1851) and
French artist Claude Monet (1840 to 1926). We focus on Turner and Monet because they
prolifically painted landscapes and cityscapes, often with repeated motifs. Furthermore,
Turner and Monet’s works span the Industrial Revolutions starting in Great Britain in
the late 18th century, a time of unprecedented growth in air pollution (9–11). Over
the course of their careers, Turner and Monet’s painting styles change from sharper to
hazier contours and toward a whiter palette, a progression that is typically characterized
as moving from a more figurative to impressionistic style. We explore the hypothesis
that increasingly impressionistic paintings by Turner, Monet, and several other artists
represent, at least in part, physical changes in atmospheric optical conditions.
Optical Implications of Increasing Aerosol Concentrations
As illustrated in Fig. 1, aerosols absorb and scatter radiation both into and out of a
line of sight. This scattering tends to decrease the contrast between otherwise distinct
objects (12, 13). Edges are used to quantify contrast because they often show the intensity
Individual paintings are known to
depict snapshots of particular
atmospheric phenomena, raising
the possibility that paintings
could also document longer-term
environmental change.
During the Industrial Revolution,
air pollution increased to
unprecedented levels, but these
values remain uncertain given the
lack of widespread, direct
measurements. Here, we show
that stylistic changes from more
figurative to impressionistic
paintings by Turner and Monet
over the 19th century strongly
covary with increasing levels of air
pollution. In particular, stylistic
changes in their work toward
hazier contours and a whiter
color palette are consistent with
the optical changes expected
from higher atmospheric aerosol
concentrations. These results
indicate that Turner and Monet’s
paintings capture elements of the
atmospheric environmental
transformation during the
Industrial Revolution.
Author contributions: A.L.A. and P.H. designed research;
performed research; contributed new analytic tools;
analyzed data; and wrote the paper.
The authors declare no competing interest.
This article is a PNAS Direct Submission.
Copyright © 2023 the Author(s). Published by PNAS.
This open access article is distributed under Creative
Commons Attribution License 4.0 (CC BY).
To whom correspondence may be addressed. Email:
This article contains supporting information online
at http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.
Published January 31, 2023.
PNAS 2023 Vol. 120 No. 6 e2219118120 https://doi.org/10.1073/pnas.2219118120 1 of 8
Downloaded from https://www.pnas.org by on March 11, 2023 from IP address
Fig. 1. Schematic illustrating key processes by which aerosols influence an object’s contrast, intensity, and visibility. A theoretical object (denoted by the gray
disk) reflecting light (black arrows) is visible because of its contrast with the background light (pale blue arrow). Aerosols (navy dots) in the air column scatter
background light into the line of vision (“in-scattering,” highlighted in light yellow), scatter object light out of the line of vision (“out-scattering”), and absorb light.
These optical effects from aerosols lead a viewer to perceive an object as having less-distinct edges (less contrast) and a whiter tint (increased intensity), as
idealized by the images on the left- and right-hand side (Claude Monet’s Houses of Parliament, Effect of Fog, 1899–1904) and described in Methods.
of an object in the foreground relative to that of the background
along nearly equal lines of sight. In order to objectively define
contrast in a manner that adapts to the scale and perspective
of an image, we use a wavelet technique. Wavelet analysis is
selected over Fourier analysis because it allows for quantifying
the local contrast in images (14), and was previously used to
estimate visibility in urban photographic images (15). We use a
Haar wavelet whereby first differences of an image are taken at
various scales (16), ranging from individual pixels to spanning
the height or width of an image. An index of the contrast found
in an image is obtained by computing the 95th percentile of the
wavelet coefficients, w
, normalizing by the median value, w
and taking the logarithm,
contrast index = log(w
). [1]
Normalization accounts for different baseline edge strengths
depending on lighting, scene, and image resolution, and the log-
arithm is suggested by the exponential dependence of contrast on
the extinction coefficient (Methods). SI Appendix, Fig. S1 shows
four example paintings illustrating that the largest gradients relate
to distinct features, such as waves, a bridge, and the hull of ships.
Benchmarking with Photographs. We first demonstrate our
metric for contrast on pairs of photographs taken during clear
and polluted conditions (SI Appendix, Fig. S2). These photo-
graphic pairs involve less artistic interpretation and allow for
benchmarking our technique using better-controlled image
characteristics. Consistent with our expectations, every polluted
photograph has a lower contrast index than its clear-sky counter-
part (SI Appendix, Fig. S2). The mean fractional reduction in the
contrast index from clear-sky to polluted photographs is 19%.
The same techniques used for photographs are next applied to
evaluate trends in contrast in paintings, which are then evaluated
in relation to aerosol emissions over time.
Quantifying Historical Air Pollution. As a proxy for historic
variations in anthropogenic aerosol concentrations, we use a
gridded estimate of annual emissions of sulfur dioxide, SO
(17). The early Industrial Revolution was largely powered by coal
(11, 18), and coal typically contains 1 to 5% sulfur by dry weight
(19). From 1800 to 1850, the United Kingdom emitted nearly
half of global SO
emissions, and the grid box corresponding
to London, known as the “Big Smoke” (9–11), accounts for
approximately 10% of all UK SO
emissions (Fig. 2), despite
accounting for only 1.0% of the area. SI Appendix, Fig. S3
presents qualitative evidence for the optical effects associated
with historical London air pollution captured by sketches and
emissions are only a proxy for changes in atmospheric
environment on account of aerosol concentration and size
distribution at any particular time depending upon factors
including coemissions and local meteorology, e.g., refs. 20
and 21. Detrended British SO
emissions from 1800 to 1850,
spanning Turner’s artistic production, correlate with detrended
black carbon (r = 0.96) and organic carbon emissions (r = 0.95),
indicating that variability in SO
also generally tracks variability
in other aerosol emissions and, thus, total aerosol concentrations.
Later in the 19th century, however, the estimated emissions of
black carbon and organic carbon per unit coal in England begin
to decline (22). In London, in particular, political efforts to
reduce industrial pollution (11), shifts in cooking and heating
sources from coal to gas (18), and a more distributed urban
landscape (9) that was enabled by an expanded railway network
also likely contribute to decreasing peak aerosol concentrations
(10, 18). We thus expect the magnitude of aerosol concentration
associated with a given SO
emission rate to decrease over the
course of the 19th century.
Trends in Contrast in Paintings by Turner,
Monet, and Others
We examine the contrast of 60 oil paintings by Turner spanning
1796 to 1850 and 38 paintings by Monet spanning 1864
to 1901. Across Turner’s works (cataloged in SI Appendix,
Fig. S4), a progression is visually apparent from sharp to hazier
contours, more saturated to pastel-like coloration, and figurative
to impressionistic representation. A similar progression is evident
across Monet’s works (SI Appendix, Fig. S5), with the additional
factor that Monet’s paintings are from two distinct locations.
The first 18 of Monet’s paintings, dating from 1864 to 1872,
depict scenes in or near Paris, and all but one were painted before
Monet’s first visit to London from 1870 to 1871. The latter 20
paintings are from Monet’s 1899 to 1901 visits to London, where
he created serialized views of the House of Parliament, Waterloo
Bridge, and Charing Cross Bridge.
2 of 8 https://doi.org/10.1073/pnas.2219118120 pnas.org
Downloaded from https://www.pnas.org by on March 11, 2023 from IP address
Fig. 2. 19th century sulfur dioxide (SO
) emissions in London and Paris. (A) Time series of emissions (17) in the grid boxes encompassing London (blue) and
Paris (red). Years of paintings by Turner (T), Monet (M), Whistler (W), and Caillebotte, Pissarro, and Morisot (CPM) are indicated by horizontal lines for London
(blue) and Paris (red). Emissions during Monet’s early paintings correspond to those of Turner’s early paintings. (The dotted black line represents mean Parisian
emissions from 1864 to 1872). (B) A geographic distribution of SO
emissions in 1850, highlighting how emissions are concentrated in London (red point in
England) and that emissions in Paris (red point in France around 2
E) trail those in London.
A mixed-effects model is used to evaluate whether local
emissions contribute to variations in contrast across our
collection of Turner and Monet paintings. In our baseline
formulation, we specify fixed effects that capture variations in
contrast according to SO
emissions, year, and subject matter
categories. We also allow for an interaction between year and SO
to account for coemissions involved in producing atmospheric
haze proportionately declining over time (22). Finally, the 98
paintings in our collections are partitioned into three categories,
with 20 clear-sky, 46 cloudy, and 32 dawn or dusk paintings
Our baseline model explains 61% of the variance in the
contrast index (Fig. 3, SI Appendix, Table S1). As expected,
paintings depicting dawn or dusk conditions or cloudy conditions
have a lower contrast index (P < 0.01) relative to clear-sky
conditions. Moreover, the model shows a significant reduction
in contrast in response to increases in SO
emissions (P < 0.01),
whereas the trend across years is indistinguishable from zero. The
interaction effect is also significant (P < 0.01) and is consistent
with the emissions of SO
later in time yielding less change in
the contrast index.
Six other model specifications are also explored that indicate
that the significance of the SO
contribution is robust to
excluding the year term or admitting for quadratic contributions
from SO
, year, or both (SI Appendix, Table S1). Our baseline
formulation is selected from among these models because it bal-
ances simplicity against the major features that we are concerned
with capturing. A means of selecting between models is offered by
the Bayesian information criteria (BIC), which measures model
performance as the difference between a reward term for better
predicting observations and a penalty term based on the number
of parameters that serves to guard against overfitting. A lower
BIC indicates a more apt model. Our baseline specification
gives among the lowest BICs, though admitting for nonlinear
dependencies on year and SO
gives comparable values. The
least apt models, according to BIC, result from excluding SO
The primary reason that SO
, as opposed to year, is inferred
to control contrast relates to the fact that Paris and London
have distinct SO
emission histories (Fig. 2). The magnitude
of SO
emissions in London near the beginning of Turner’s
career in 1796 is similar to the magnitude of the emissions near
the beginning of Monet’s career in Paris in 1864. Monet’s early
paintings in Paris have higher contrast than most of Turner’s
works subsequent to the 1820s, despite coming later, such that
no simple time trend can be fit across these collections (Fig. 3A). If
examined in the context of SO
emissions, however, the contrast
of Monet’s early works overlaps with those of Turner’s, and the
low contrast of Monet’s later works is in accord with the high
emissions in London at the end of the 19th century (Fig. 3B).
Monet and Turner are among the most prolific and iconic
artists whose work spans the industrial era, but paintings by other
artists that depict cityscapes and atmospheric phenomena also
align with our proposed model. Specifically, our model predicts
the contrast found in seven paintings by Gustave Caillebotte
(1848 to 1894), four paintings by Camille Pissarro (1830 to
1903), and one painting by Berthe Morisot (1841 to 1895)
of Paris on the basis of year, local SO
emissions, and subject
matter (Fig. 3). The contrast indices calculated for six Nocturnes
paintings by Whistler in London between 1871 and 1875 are
also predicted by our model. Note that Whistler’s paintings in
less-polluted environments—for example, The Coast of Brittany
(1861) or The Blue Wave Biarritz (1862)—are associated with
substantially greater contrast indices of 2.4 and 2.2, respectively.
Refitting the mixed-effects model to our expanded dataset
including works by Caillebotte, Pissarro, Morisot, and Whistler
leads to conclusions that are consistent with our more limited
analysis of only works by Turner and Monet (Fig. 3). The year
trend inferred from our expanded analysis, however, appears
significant only in the case where SO
is entirely excluded
(SI Appendix, Table S1), further supporting the importance
of SO
Trends in Intensity
As a complementary approach, it is also possible to analyze the
intensity of images across our collection of works. Aerosols scatter
visible light of all wavelengths into the line of sight (23) (Fig. 1),
leading to a whiter tint and increased light intensity during
daytime (13). We examine the relationship between intensity
and SO
emissions using the same mixed-effects methodology
used for contrast and find a significant effect (P < 0.01) of
emissions increasing intensity in our baseline approach
PNAS 2023 Vol. 120 No. 6 e2219118120 https://doi.org/10.1073/pnas.2219118120 3 of 8
Downloaded from https://www.pnas.org by on March 11, 2023 from IP address