INTERANNUAL VARIATIONS IN THE RETREAT OF THE NOTHERN
SEASONAL CAP OF MARS USING COMPUTER VISION.
P. J. Acharya1 (pruthvi1@yorku.ca), I. B. Smith1 (ibsmith@yorku.ca), W. Calvin2 (wcalvin@unr.edu),1York
University, Toronto, Ontario, Canada, 2Nevada University, Reno, Nevada

Introduction: One of the largest features on Mars
is the Northern Seasonal Polar Cap (NSPC), with can
cover up to 12% of the surface seasonally [1]. The
NSPC condenses up to 25% of the atmosphere during
wintertime and releases it during springtime [2],
leading to a major change in the global atmospheric
pressure and composition. Intense atmospheric
activities are often seen originating near the cap’s edge
during springtime [3,4]. These activities have been
seen to reach very close to the equator, indicating a
large amount of material being moved in and out of the
polar region. For these reasons, it’s critical to
understand the NPSC’s recession to understand the
past, present, and future Martian climate.
The focus of this work was to update and extend
the previously published results [4] using an automated
approach. We were able to track the cap’s edge at a
higher temporal resolution than previously possible [4]
and were able to add new MYs that haven’t been
analyzed previously. With the enhanced dataset, we
determined the seasonal variation in the recession rate
for each year, interannual observations, and midseason changes that were previously undetected.
MARCI Observations: We used large-scale visual
observations from the Mars Color Imager (MARCI)
from MY 29 to MY 35. Each observation is a mosaic
of 12-13 images taken within 24 hours, each image
covers ~27.7⁰ of the northern polar region (Fig 1A) [4].
For each MY, we tracked the cap until Ls = 70⁰; around
this time the cap’s edge reaches the polar layered
deposits and it’s difficult to distinguish between the
seasonal and residual ice.
Automated Tracking: Using Python and an opensourced computer vision library called OpenCV, we
automated outlining the cap and fitting an ellipse of
best fit. Our method relied on using the Hue,
Saturation, Value (HSV) color format, where
important properties of color are decoupled into their
own channels. Using this format, we empirically
determined the best upper and lower H, S, and V
bounds that best outlined the seasonal cap and ignored
the water-ice clouds (Figs 1A, B). Then we fitted an
initial ellipse best fit (Fig 1B) and used a simple
filtering algorithm to remove any outlier outline points,
which were caused by imaging artifacts (Figs 1D) and
dust storms (Fig 1e), to get a filtered ellipse fit. We

record the average axis, and the area of the ellipse fits
while using over 99% of the MARCI observations. Our
new method was significantly faster and produced a
higher temporal resolution curve compared to the
previous approaches using the same MARCI dataset
(Figs 2B, C, D) [3,4].

Figure 1: MARCI mosaic at MY 29 Ls = 24.9⁰. A. Full color
MARCI mosaic. B. MARCI mosaic with the initial ellipse fit
(Blue) and outline points (Purple). C. MARCI mosaic with
the filtered ellipse (White), outlines points that passed the
filtering algorithm (Green), and points that failed (Red). D.
MARCI mosaic during an imaging artifact at MY 29 Ls =
56.9°. E. MARCI the mosaic at MY 30 Ls = 27.7⁰ during a
dust storm.

Validation: We validated our method by
comparing results to those from Calvin et al. (2015) by
looking at the outline (Fig 2A) and recession rates for
MY 29 to MY 31 (Figs 2B-2D). Both methods
produced similar outlines; however, Calvin et al.
(2015)’s outline is smoother than our outline.
Additionally, both methods produced similar seasonal
recession rates and latitude extent, except for MY 29
where our method is ~2.5° larger than Calvin et al.
(2015)’s observations. When we compare our results to
Cantor et al. (2010), who used similar MARCI
observations and methods to find the seasonal
recession rates for MYs 28 and 29, our method closely
matches their results (Fig 2B).

Sudden Latitude Increasing Event: We find that
MY 30 experiences a sudden latitude increase
beginning Ls = ~44⁰, where the cap shrinks by ~0.8° by
Ls = ~45⁰. Following this, the cap returns to its linear
fit by Ls = ~51⁰; however, the recession rate is ~55%
faster than before this event (Fig 4A). In addition, there
are dust storms seen moving southwards just before
this increase (Fig 4B). It should be noticed that we
believe that this is the second sudden increasing event,
as near the start of the observations we noticed that
cap’s latitude is converging towards the linear fit.
Additionally, there are intense storms seen over the
entire region; however, due to the lack of observations
we are unable to see the initial increase, so our
interpretation will focus on the latter increase.
Figure 2: A. Comparison between our results and those from
Cantor et al. (2010) and Calvin et al. (2015) for a MARCI
mosaic at MY 29 at Ls 62.5. A: (1) our outline in green, (2)
Calvin et al. (2015)’s outline in blue, (3) Calvin et al.
(2015)’s outline in purple, (4) our circle of best fit using the
HSV method in black, (5) our ellipse of best fit using the HSV
method in orange. B: Comparisons of the three survey’s
latitudinal extents over MY 29. Calvin et al. (2015) in blue;
Cantor et al. (2010) in green); our ellipse of best fit in
orange; our circle of best fit in purple. C and D: Same as B
but for MY 30 and MY 31.

Results: We used a line of best fit, a 7th order
polynomial fit and its first derivative to analyze the
multi-year average and individual MY recession
curves. We found that most MYs behaved very similar
to each other, with an average recession rate of
~0.24⁰/Ls. Most MYs diverged from the linear rate
between Ls = ~35⁰ and Ls = ~50⁰, transiting from the
annual minimum to maximum recession rate (Figs 3C,
D). Also, the latitude recession rate is slower early in
the season and faster later in the season.

Figure 3: Comparison of all MYs from Ls = 0⁰ to Ls =
70⁰A. Area change (in km2). B. Average axis change (in °N).
C. First derivative for the 7th order area fit (in km2/Ls). D.
First derivative for the 7th order average axis fit (in °/Ls).

Figure 4: A. shows the latitude vs Ls for MY 29 with the
linear fit before (Blue) and after (Orange) sudden increase.
B. shows the latitude vs Ls for MY 30 with the linear fit
before (Blue) and after (Orange) sudden increase. C. shows
a dust storm occurring at MY 29 Ls = 51.1⁰ around the time
of the increase. D. shows a dust storm occurring at MY 30 Ls
= 43.6⁰ around the time of the increase.

We believe that the sudden increase may be caused
by a major sublimation event and katabatic winds
flowing away from the pole (Fig 5b). After this,
recession halts because the cap’s edge is sufficiently
far away from the Sun’s maximum latitude reach and
won’t resume until the Sun catches up to the cap’s
edge (Fig 5c). Once the Sun reaches the cap’s edge,
recession resumes but it is enhanced by katabatic
winds, which are stronger than before and known to
increase sublimation rates [5,6,7,8] (Fig 5d).

Figure 5: Timeline of the sudden latitude increasing event.
The blue circle represents the seasonal cap, the red circle
represents the maximum latitude where the cap is stable or
the maximum latitude reach of the Sun, and the black arrows
represent the katabatic winds. A. Cap receding solely due to
solar radiation and closely following the maximum stable
latitude. B. A major sublimation event decreases the size of
the cap and seasonal katabatic winds formed during this
time, causing the storms seen around this time. C. Recession
halts because the cap is well within the stable latitude region
and katabatic winds are sufficient to continue recession. D.
Maximum stable latitude reaches the cap’s edge and
recession resumes but is enhanced by katabatic winds.

Effects of the Global Dust Storms: The seasonal
cap following the MY 28 and MY 34 global dust
storms deviated significantly from the MY that did not
follow the storms. MY 29’s cap was significantly
smaller, and the area recession rate was significantly
slower, until Ls = 40°, than the other MYs. MY 35’s
cap was slightly larger and receded slightly faster
throughout the entire observational period. We believe
that the abnormal behavior of the seasonal cap
following the global storm is due to the difference in
the behavior of the storm.
The MY 28 storm first appears at Ls = ~262° [9],
just as the NPSC reached maximum extent [1], and the
storm has mostly died out by Ls = ~322° [9]. On the
other hand, the MY 34 storm first appears much
earlier, around Ls = ~185° [9], before the NPSC began
to form [9], and quickly died out by Ls = ~220° [9],
about midway through the formation period [1].
For the early MY 28 global storm, we have
hypothesized three explanations by which the global
storm can cause a smaller NPSC. The polar vortex may
be displaced during the storm [1], this would cause
parts of the NPSC to sublimate faster than others [1],
resulting in a smaller NPSC following the storm.
Another possible explanation is that as the storm dies
out, dust is deposited on the surface, lowering the
surface albedo, and the cap sublimates faster, resulting

in a smaller cap. There could be a combination of the
previously discussed explanations; asymmetries
originally arose from the displaced polar vortex could
reduce the surface albedo at the margins of the cap,
enhancing sublimation and accelerating recession prior
to the start of our observations.
To test this, we measured the displacement of the
center of the ellipse with respect to the geographical
north pole (Fig 6A), the eccentricity of the ellipse (Fig
6B), azimuth of the ellipse (Fig 6C), and the average
brightness of the outer edge of the NPSC (Fig 6D) for
all MYs. It’s expected that early in the season; MY
29’s cap should have a large displacement, a highly
eccentric cap, and/or be a darker cap than other MYs.
Contrary to our prediction, prior to Ls = 40°, MY 29
has a smaller displacement and the outer edge of the
NPSC is the brightest. Interesting, during this period,
MY 29 is the most eccentric cap and the largest
azimuth value, suggesting some sort of polar vortex
displacement. However, these together put our
explanation to question.
Considering MY 34, because the storm died
quickly, there was ~100 Ls between the ending of the
storm and the start of the recession. This gives time for
the polar vortex to correct itself and the atmospheric
dust levels to return to background levels. Also,
previous work showed that during the MY 28 global
dust storm, the average temperature decreased by
~10°K [10]. Assuming similar conditions during the
MY 34 storm, we hypothesize that a cooler surface is
present following the MY 34 global storm, enhancing
the cold trapping rate, and resulting in a larger cap.
The larger cap would sublimate faster early in the
season from lower latitude, where the sun is the most
intense. This agrees with our observations for MY 35
which has the highest early-season recession rate.

Figure 6: Ellipse fit parameters to test the moving polar
vortex’s effect on the NPSC retreat, with a moving average
of 10 to reduce noise. A: Displacement of the ellipse fit with
respect to the geographical north pole vs Ls. B: Eccentricity
vs Ls. C: Azimuth angle vs Ls, negative values represent
longitude values west of 0°E. D: Mean brightness of the
outer 3° vs Ls.

Future Work: Going forward, we plan to develop
a dynamic color detector that can better accommodate
regional dust storm; current static bound fails during
those storms. We also plan on further testing our
explanation for the sudden latitude increasing event
and effect of the global dust storm using numerical
models and new observations. In addition, we plan on
using our method on the Southern Seasonal Polar Cap
using similar MARCI mosaics and comparing the two
polar caps.
Acknowledgments: We gratefully acknowledge
the support of the NSERC Discovery grant to IBS and
from the Technology for Exo-Planetary Science
(TEPS) CREATE grant to PA
References: [1] Piqueux, S. et al. (2015) Icarus,
251, 164–180. [2] Tillman, J. E.et al. (1993) JGR
98(E6), 10963. [3] Cantor, B. A.et al. (2010) Icarus,
61–81. [4] Calvin, W. M. et al. (2015) Icarus181–190.
[5] Smith, I. B. et al. (2013) JGR 1835–1857. [6]
Smith, I. B. and Spiga, A. (2018) Icarus, 308, 188–196.
[7] Bramson et al. (2019 JGR 124, 1020–104. [8]
Spiga, A. and Forget, F. (2009) JGR, 114. [9]
Wolkenberg, P. et al. (2020) JGR, 125(3). [10] Plesa,
A.-C. et al. (2016) JGR, 121(10), 2166–2175.