Jump to content


Photo

New Light Pollution Atlas w/o Snow Cover

  • Please log in to reply
62 replies to this topic

#1 DaveL

DaveL

    Explorer 1

  • -----
  • Posts: 78
  • Joined: 30 Sep 2009
  • Loc: Madison, WI, USA

Posted 16 June 2010 - 03:07 AM

I'm sure most of us have used the World Atlas of the Artificial Night Sky Brightness to help us gauge the quality of our skies and to help locate new sites for observing. This atlas is based on Defense Meteorological Satellite Program (DMSP) satellite measurements of light sources on the earth's surface and a model of how these light sources affect the amount of light scattered downward to an observer's eyes (see here for more detail). While the maps are imperfect, they are nevertheless an excellent resource for finding dark skies, particularly for gauging the relative brightness of sites.

This post concerns the possible effect of snow cover on this light pollution atlas. The idea is not new--the authors of the atlas themselves have pointed out the potential impacts of snow cover, as has Tony Flanders in his blog. The satellite measurements used to make the sky atlas were taken in the following three time periods:

1. March 16-23 1996, 2. January 5-14 1997, 3. February 3-12 1997 (from Elvidge,C.D., Baugh, K.E., Dietz, J.B., Bland, T., Sutton, P.C., Kroehl, H.W. 1999. Radiance Calibration of DMSP-OLS Low-light Imaging Data of Human Settlements. Remote Sensing of Environment 68(1), pp. 77-88.)

Snow cover data is available from the National Snow and Ice Data Center (NSIDC) on a weekly basis here. The three weeks that correspond most closely to the three time periods above are:

1. March 18-24 1996, 2. January 6-12 1997, 3. February 3-9 1997

The attached figure shows the number of weeks with snow cover on the ground from the NSIDC data. 100% means that all three weeks had snow cover, 67% means 2 out of three weeks had snow cover, etc. As you can see, much of the northern third of the US had snow cover during the entire period when the light data was taken. This snow cover will dramatically increase the amount of light sensed by the satellite and will thus make the light pollution atlas brighter than it would otherwise be.

How much of an effect does snow cover have on the atlas? It turns out there is additional DMSP satellite data taken from September - November 2001. Except for a relatively small amount of snow centered over northern Wyoming and western South Dakota during the November new moon, this entire period was snow free. (This new data is only available online for the lower 48 states, unfortunately.) In the posts below, I calculate a new light pollution atlas using this 2001 data. First, I try to re-calculate the current atlas with the original 1996/1997 data to make sure the new atlas is a fair comparison.

Attached Files



#2 DaveL

DaveL

    Explorer 1

  • -----
  • Posts: 78
  • Joined: 30 Sep 2009
  • Loc: Madison, WI, USA

Posted 16 June 2010 - 03:09 AM

Here I briefly describe where I obtained the data/model to re-calculate the night sky atlas and how my re-calculated version compares with the original atlas:


The "radiance-calibrated" DMSP satellite data for 1996/1997 are available online here. This DMSP data is special because it includes some observations taken when the satellite's gain setting is reduced so that urban cores are not saturated.

The light pollution model used by Cinzano et al was derived by Roy Garstang in the following two articles:

-Garstang, RH: Model for artificial night-sky illumination, Publications of the Astronomical Society of the Pacific, 98 (601): 364-375, Mar 1986
-Garstang, RH: Night-sky brightness at observatories and sites, Publications of the Astronomical Society of the Pacific, 101 (637): 306-329, Mar 1989

I programmed the Garstang model myself using the parameters given in another Cinzano paper.


A comparison of the original atlas (top) with the atlas re-calculated by me (bottom) is shown in the attached figure. A higher resolution version of my figure can be found here and a higher resolution version of Cinzano's atlas is here. There's very good agreement between the two, but if you look closely you can see differences. I've looked over my code many times and I do not think I made an error. I give some possible reasons for the discrepancy here. Anyway, the differences in these plots are well within the uncertainties involving the assumptions of the model.

Attached Files



#3 DaveL

DaveL

    Explorer 1

  • -----
  • Posts: 78
  • Joined: 30 Sep 2009
  • Loc: Madison, WI, USA

Posted 16 June 2010 - 03:10 AM

The new radiance calibrated satellite data is available here. Like the 1996/1997 data, this data also includes some observations taken when the satellite's gain setting is reduced so that urban cores are not saturated. This data is only available on a map projection that includes the lower 48 states and some surrounding areas (for example, areas in south-central Canada are included but Vancouver, B.C. is cut off).

I used the same model in the previous post to calculate a new Atlas of Artificial Night Sky Brightness for the lower 48 states (see attached). The atlas calculated using the original data is on top, while the new atlas calculated using data from Fall 2001 is on the bottom. (As before, higher resolution versions are available here.)

Looking at the areas from Virginia south along the Atlantic Coast to Florida and then west along the Gulf coast, you see fairly good agreement in the light pollution derived from these two datasets. If you look in the northern US and Canada, on the other hand, the differences are dramatic. For example, in Minnesota, Wisconsin, Quebec and northern New England you tend to be a full light pollution "zone" darker in the 2001 data. In the 1996/1997 dataset, the only area black in the eastern US is a tiny area in the Boundary Waters Canoe Area in northern Minnesota. In the new dataset, this Minnesota region expands and new black regions appear in upper peninsula of Michigan and northern Maine. The black regions in Montana and Idaho are also much bigger. (You can't compare the maps in northwest Washington because Vancouver is not included in the Fall 2001 data.)

Attached Files



#4 DaveL

DaveL

    Explorer 1

  • -----
  • Posts: 78
  • Joined: 30 Sep 2009
  • Loc: Madison, WI, USA

Posted 16 June 2010 - 03:18 AM

To better see the differences between the 1996/1997 and 2001 atlases, you could divide the brightness in 2001 by the brightness in 1996/1997. This plot is very noisy. Therefore I have smoothed the results before dividing (and also compare logarithms). The steps are:

1) For each atlas, take the logarithm base 3 of the brightness values (except for the black/gray transition, the boundary between all color zones in the Cinzano Atlas is a power of three).
2) Smooth the "maps" from (1) (I use a Gaussian kernel with a "standard deviation" of about 18 pixels).
3) Subtract the 1996/1997 smoothed "map" from the 2001 smoothed "map".

The resulting map is attached. Because I take the difference of log base 3, the maps given the change in light pollution "zone" in going from the 1996/1997 to the 2001 atlas. A value of -1 means that you are about one light pollution "zone" (or "color") darker in the 2001 atlas, while a value of +1 means that you are about one light pollution "zone" (or "color") brighter.

As you can see, in Canada and in the northern part of the US, the 2001 map is about one "zone" darker than the 1996/1997 map. Across the southern portion of the US, there tends to be relatively little change in the light pollution "zone". The change in light pollution zone is broadly similar to the snow cover map I showed in the first post, which suggests that snow cover is playing an important role. The differences between the snow and change in zone could be due to any number of things: 1) increase in number of lights over 4-5 years, 2) More foliage on trees in Fall 2001 than Winter 1996 and 1997 (1 and 2 might tend to offset each other), 3) we do not know which of the three periods in 1996/1997 the DSMP satellite had a cloud free overpass (did the satellite 'miss' the week with snow or not?) 4) depth or age of snow (deep and/or new snow is more reflective than a dusting/old snow) 5) growth in natural gas production (for example, western and southwestern Wyoming), 6) inter-calibration issues with the satellites (see here). The last effect would have a uniform effect across the entire map, I believe. Therefore, it cannot account for the huge differences in space that we are seeing.

So...
Once again, the high resolution maps can be found here. If anyone wants the actual brightness data plotted on the maps, let me know. Thanks for reading and I hope you enjoy the new atlas!

-Dave

Attached Files



#5 Tony Flanders

Tony Flanders

    Voyager 1

  • *****
  • Posts: 11137
  • Joined: 18 May 2006
  • Loc: Cambridge, MA, USA

Posted 16 June 2010 - 08:02 AM

Very, very interesting, important work.

In addition to eliminating snow, you have produced two maps separated in time by 4 years, which may be enough to detect changes due to population and streetlight growth. Have you looked at it from that angle?

It's also very helpful to have a map without boundaries overlaid. Most major population centers are on coasts, so the boundaries in Cinzano's released maps tend to blot out most of the interesting detail in places where many of us actually live.

Is any data available after 2001?

#6 Tony Flanders

Tony Flanders

    Voyager 1

  • *****
  • Posts: 11137
  • Joined: 18 May 2006
  • Loc: Cambridge, MA, USA

Posted 16 June 2010 - 09:07 AM

On a personal note, this has changed the zone for most of my customary observing spots. For instance, my country home which used to be in the yellow zone is now in the green. My club's observing field, which used to be on the red/orange border is now deep in the orange.

Going by the new zones, the correlation between the predicted sky brightness for each zone and my own measurements is much better. So is the correlation between the zones and the Bortle descriptions.

This also helps explain why people in the South complain how terrible their red-zone sites are, whereas I have so far said "Oh, the red zone isn't so bad." It's because what I thought was red zone should actually be orange.

#7 Tony Flanders

Tony Flanders

    Voyager 1

  • *****
  • Posts: 11137
  • Joined: 18 May 2006
  • Loc: Cambridge, MA, USA

Posted 16 June 2010 - 09:13 AM

I wonder what caused the huge increases in southeastern Oregon and southeastern Utah. There's kinda like nothing there in either place. Having just been in southeastern Oregon (Steens Mountain) I can vouch that it's still super-dark.

#8 DaveL

DaveL

    Explorer 1

  • -----
  • Posts: 78
  • Joined: 30 Sep 2009
  • Loc: Madison, WI, USA

Posted 16 June 2010 - 11:20 AM

Hi Tony,

I don't believe there is more data that is not saturated in the urban cores. The set of all available data products for this satellite can be found here. There is more recent data if you're not interested in changes in the urban cores.

Detecting population and street light growth is very important but it is probably more difficult to get from this data. The reason is intercalibration. I think I'd expect population/street-lighting growth to be more uniform than the snow effect I emphasized here. If it is more uniform, then it can be hard to disentangle the satellite calibration effect from population/street-lighting growth.

Here are some maps of population change in the US:
http://www.census.go...t/gallery/maps/

There does not appear to be a strong correlation, although the census data is for changes over the last decade.

-Dave

#9 DaveL

DaveL

    Explorer 1

  • -----
  • Posts: 78
  • Joined: 30 Sep 2009
  • Loc: Madison, WI, USA

Posted 16 June 2010 - 11:26 AM

This also helps explain why people in the South complain how terrible their red-zone sites are, whereas I have so far said "Oh, the red zone isn't so bad." It's because what I thought was red zone should actually be orange.



This was my impression too! What motivated this study was a comparison of sites in different parts of the country (not using measurements--just my impression when deep sky observing). I live in Wisconsin, which doesn't have the best skies, but it seemed like my skies were darker than what the light pollution atlas was saying.

#10 DaveL

DaveL

    Explorer 1

  • -----
  • Posts: 78
  • Joined: 30 Sep 2009
  • Loc: Madison, WI, USA

Posted 16 June 2010 - 11:30 AM

I wonder what caused the huge increases in southeastern Oregon and southeastern Utah. There's kinda like nothing there in either place. Having just been in southeastern Oregon (Steens Mountain) I can vouch that it's still super-dark.


I noticed this too because these are some of the darkest ares in the lower 48 (I've always wanted to explore Steens Mountain!). You can see few more light sources in the new maps, but I'm not sure what's going on.

The differences can be relatively big in the dark areas when looking at the ratio of new to old (or similarly for the difference of logarithms). This is because when you add a little bit of light to almost nothing, the ratio of new to old can be very big even if the difference in brightness is very small. So for the areas with very little light pollution to begin with, a difference rather than ratio of brightness may be a more appropriate metric.

-Dave

#11 s58y

s58y

    Fly Me to the Moon

  • -----
  • Posts: 6519
  • Joined: 12 Dec 2004
  • Loc: Eastern NY

Posted 16 June 2010 - 12:11 PM

Great job. Thanks.

#12 star drop

star drop

    contra contrail

  • *****
  • Moderators
  • Posts: 70303
  • Joined: 02 Feb 2008
  • Loc: Snow Plop, WNY

Posted 16 June 2010 - 12:56 PM

Now all we have to do is to keep that pesky snow away. Any ideas?

#13 Michael Cook

Michael Cook

    Sputnik

  • *****
  • Posts: 25
  • Joined: 30 May 2008
  • Loc: Newcastle, ON, Canada

Posted 16 June 2010 - 01:35 PM

I work for a municipality, and this additional analysis is very useful as we continue to evolve our outdoor lighting policies to address light pollution.

Thanks.

#14 Tony Flanders

Tony Flanders

    Voyager 1

  • *****
  • Posts: 11137
  • Joined: 18 May 2006
  • Loc: Cambridge, MA, USA

Posted 16 June 2010 - 01:48 PM

Now all we have to do is to keep that pesky snow away.


I was thinking this over, and concluded that while snow certainly increases skyglow, it probably doesn't increase it as much as it affects the satellite data.

Remember, these satellites are not measuring skyglow; they're measuring the light that goes out into space, which is just a crude proxy. Insofar as light goes out into space, it's not coming back into your telescope.

For instance, a spotlight shining straight up will send the satellite sensors wild while producing very little skyglow. Turn that same spotlight horizontal, so it's scattering through miles of dirty, ground-level air, and it will produce huge skyglow while being almost invisible to the satellite.

Imagine a city lit with fully shielded, down-pointing streetlights. These are good in every way, producing little skyglow and also almost invisible to the satellite.

Now let's say that there's a fresh snow. All those down-pointing streetlights suddenly become highly visible to the satellite, because most of their light is now reflected upward. But because the light source (the snow) is at ground level, the ability of the light to travel horizontally and generate skyglow is limited.

#15 DaveL

DaveL

    Explorer 1

  • -----
  • Posts: 78
  • Joined: 30 Sep 2009
  • Loc: Madison, WI, USA

Posted 16 June 2010 - 03:46 PM

I think Tony's right.

Within the context of the model used to calculate the light pollution atlas, you can even quantify the relative contributions of reflected light and direct light to the total light pollution. The effect depends on distance from the source.

In the attached plot, I show brightness as a function of distance from a point source of light (purple). The y-axis tells how the light polution zone changes relative to the zone at the origin (more negative implies darker). The red and blue curves show the contribution of reflected and direct to the light polution. Near the source they are comparable, but as one travels further from the source the direct contribution dominates:

At 15km from the source, reflected light = 25% of total
At 50km from the source, reflected light = 19% of total
At 100km from the source, reflected light = 13% of total

These numbers depend on assumptions in the model and so should not be taken as precisely accurate. Nevertheless, I'm sure the basic result that reflected light is most important near the source still holds. So, as Tony stated, the satellite is overestimating the effect of snow. This is especially true when you are relatively far from major light sources.

-Dave

(Two side comments:
1. This is a log plot, so even though the red plus the blue curve add up to the purple, it is not apparent on the plot.
2. Technically this is not a point source, but a source with a size of about 0.5-1.0km.)

Attached Files



#16 DaveL

DaveL

    Explorer 1

  • -----
  • Posts: 78
  • Joined: 30 Sep 2009
  • Loc: Madison, WI, USA

Posted 16 June 2010 - 05:20 PM

I've just realized that my post directly above is not really the right way to look at the problem. It does correctly give the relative roles of reflected versus direct light (when there's no snow), but how this translates to the snow problem and the satellite bias is not clear from the above graph. I'll have the revised figure soon, but I have to go.

The short answer is:
when there is snow on the ground the satellite over-estimates the light pollution by a factor of 1.6 near the source, a factor of 3 at 50km, a factor of 4 at 170km etc. (This is if you simply apply the light pollution model without changing the parameter related to the reflectivity of the surface.)

-Dave

#17 DaveL

DaveL

    Explorer 1

  • -----
  • Posts: 78
  • Joined: 30 Sep 2009
  • Loc: Madison, WI, USA

Posted 17 June 2010 - 12:37 PM

I now have results for how snow cover affects the satellite compared to how it affects "reality" (i.e. the light pollution model but properly taking into account the reflectivity of snow). The original light polution model has the albedo of the ground as a parameter, so changing this from the default setting to a snow-like setting will give the effect of snow.

The albedo of snow is:
Fresh snow, dry = 0.85
Fresh snow, wet = 0.80
Old snow, dry, clean = 0.70
Old snow, wet, clean = 0.60
Old snow, wet, medium dirty = 0.50
Old snow, wet, heavily dirty = 0.40

I used the new wet snow (0.80) and the old, wet, clean snow (0.60). The effect of snow in "reality" with these parameter settings is given by the purple and blue curves. (In my previous post I used an albedo of 0.85.) The snow effect is largest closest to the source, and decreases to about 25% larger at a distance of 300km.

If the satellite predominantly measures light directed nearly straight up, then it is measuring mostly reflected light. To make the light pollution map from a satellite, you use this "straight up" light as a proxy for the light in all directions. If you assume the wrong surface albedo (which is what you do if you apply an asphalt/concrete albedo to snow covered ground (as in the original atlas)), then you essentially multiply all the light by a certain factor independent of direction. This is unlike reality, where snow only changes the reflected component. The factor that you wrongly apply to all light is simply the albedo of snow divided by default, asphalt/concrete albedo (=0.15). The effect of snow wrongly estimated from the satellite is shown by the red and orange lines.

So as Tony said, the original light pollution atlas is not very useful even when there is snow on the ground. For example, if the snow albedo is 0.6, the orignal light pollution atlas increases the light pollution by a factor of 4 everywhere (= 1.3 light zones). In reality, the effect of snow is more like the blue curve, which is a factor of 1.7 at 30km (= 0.5 light zones) and smaller for greater distnces. If, on the other hand, you do take into account the snow albedo, then the satellite data shoud correctly recover the blue and purple curves.

-Dave

(For a point source, the blue and purple curves would go up to the orange and red at very close distances. The source here has a finite size.)

Attached Files



#18 DaveL

DaveL

    Explorer 1

  • -----
  • Posts: 78
  • Joined: 30 Sep 2009
  • Loc: Madison, WI, USA

Posted 17 June 2010 - 01:20 PM

Here I compare Tony's Sky Quality Meter data in the Boston area with the light pollution atlas using the 1996/1997 data and the 2001 data. Tony's data is from his blog entry entitled the Ground Truth for the Light Pollution Atlas.

The x-axis in the figure is the distance from Cambridge Common and the y-axis is the brightness in magnitudes per square acrsec (bigger means darker). The black line is the actual measurements taken by Tony with a Sky Quality Meter (SQM-L). The red and the blue lines are from my calculations of the atlas using the 1996/1997 data (which is contaminated with snow cover) and the new 2001 data.

As Tony mentioned above, the new atlas based on the 2001 data is in better agreement with real observations. Nevertheless, the 2001 atlas still has a tendency to over-estimate the brightness (i.e. the blue line is too low compared to the black line). Also note the jumps and plateaus in the real data which are not in the atlases (see Tony's blog).

The discrepency could be caused by a variety of different things: 1) the day Tony took measurements could have been especially clear of aerosols. Therefore, if we change the aerosol parameter in the model, we could reproduce the data, 2) other parameters in the model just need to be "tuned", 3) the assumed relationship between the satellite brightness and the light sources needs to be modified or 4) something is more fundamentally wrong with the model.

-Dave

Attached Files



#19 Tony Flanders

Tony Flanders

    Voyager 1

  • *****
  • Posts: 11137
  • Joined: 18 May 2006
  • Loc: Cambridge, MA, USA

Posted 17 June 2010 - 02:52 PM

The discrepancy could be caused by a variety of different things: 1) the day Tony took measurements could have been especially clear of aerosols ... 2) other parameters in the model just need to be "tuned", 3) the assumed relationship between the satellite brightness and the light sources needs to be modified or 4) something is more fundamentally wrong with the model.


Yet another possibility -- something is fundamentally wrong with the SQM-L. Remember, this is not a sophisticated scientific instrument. It's a clever, cheap gizmo that has been tuned to some semblance of scientific accuracy by seat-of-the-pants methods. I think that two different SQM-L measurements are comparable to each other, but you can't necessarily expect them to agree with results from scientific istruments. Among other issues, the SQM-L is well-known not to have the same spectral response as a CCD fitted with a V-band filter.

#20 Phillip Creed

Phillip Creed

    Idiot Seeking Village

  • *****
  • Moderators
  • Posts: 2086
  • Joined: 25 Jul 2006
  • Loc: Canton, OH

Posted 17 June 2010 - 06:10 PM

Dave,

First off,

:bow: :bow: :bow:

That's a lot of hard work that went into this. This is a long-overdue refinement of the light pollution map. Regardless of where your new model still has lingering issues, one thing's certain, at least IMO--it's more accurate than the previous one.

I frequent many sites in Ohio and noticed that a lot them in the "yellow" zones were actually quite dark. The areas around Mohican State Park and, in NW Ohio, the area around Harrison Lake always seemed like they were darker than their "yellow zone" would indicate based on observations I'd had there.

Many sites in SE Ohio ("green" on the old map) were good enough to rival some "blue" zone sites in western PA that I'd previously used. According to the new data, some of these sites, notably the Zaleski State Forest / Lake Hope State Park area are borderline blue/green, and I've found this matches previous observations of sky darkness.

Further refinement, as you've already pointed out, is necessary. But a job well done, nonetheless. The 2001 data and accompanying image much more closely resemble sky conditions I've seen in the areas that I've frequented.

One thing I did note was that Cherry Springs, according to the 2001 data, has gone from "blue" to "dark gray". I'd have to heartily concur. I use a "dark gray" site frequently, Calhoun County Park, WV, and on nights of similar transparency, they're very close.

Clear Skies,
Phil

#21 George N

George N

    Mercury-Atlas

  • *****
  • Posts: 2729
  • Joined: 19 May 2006
  • Loc: Binghamton & Indian Lake NY

Posted 18 June 2010 - 11:32 AM

The new map looks more accurate to what I’ve experienced living in rural NY, with a camp in the central Adirondacks, plus frequent trips to Cherry Springs and once-a-year to Stellafane.

Snow is bad news for observational astronomy! I’ve been getting about .3 to .5 brighter SQM readings with snow cover, and qualitatively the sky just looks brighter – not to mention the annoying light from the ground while trying to observe.

#22 DaveL

DaveL

    Explorer 1

  • -----
  • Posts: 78
  • Joined: 30 Sep 2009
  • Loc: Madison, WI, USA

Posted 19 June 2010 - 01:02 AM

I'm posting because I have discovered a relatively minor problem with the 2001 DSMP satellite data used to make the new atlas. Apparently, in some locations the light sources in the 2001 data are displaced from where they should be. The maximum displacement is about 4 to 5 km in the east/west direction and 2 to 3 km in the north/south direction. I have determined the origin of this problem is the DSMP satellite data online because there are also independent population and land-use data in the same location as the DSMP satellite data. These are on the same projection/grid and there is no horizontal displacement in these datasets. I dicovered this problem by overlaying the maps in google earth and zooming in on small towns.

I think I might be able to come up with an algorithm to fix the problem.

This doesn't affect the what we have talking about above, and you would never notice it if you didn't zoom in on google maps. Nevertheless I'm sorry I didn't check things out more closely before I posted. :(

-Dave

#23 DaveL

DaveL

    Explorer 1

  • -----
  • Posts: 78
  • Joined: 30 Sep 2009
  • Loc: Madison, WI, USA

Posted 20 June 2010 - 04:21 AM

I've come up with an algorithm to fix the displacement errors in the 2001 light data. The new data is now available on the webpage. The fix is not perfect but it's definitely better than before. Most people will not notice, unless you overlay the maps in google earth and zoom way in.

The adjustment is not uniform across the map and some area have no adjustment. The biggest adjustment is about 5 pixels (the size of a pixel is 1/120 degrees). I'll add more detail on the adjustment on the web site soon.

-Dave

#24 DaveL

DaveL

    Explorer 1

  • -----
  • Posts: 78
  • Joined: 30 Sep 2009
  • Loc: Madison, WI, USA

Posted 21 June 2010 - 01:17 AM

You can read about the small position errors in the 2001 light data and how I corrected for them here. None of this effects the brightness of the lights, etc.

The corrections are small (I thought about reposting the figures in this thread, but I can't see a difference after I apply the corrections). You will only notice the improvement when you compare the light pollution atlas to small towns when zoomed in in Google Earth or something similar.

-Dave

#25 csa/montana

csa/montana

    Den Mama

  • *****
  • Posts: 86090
  • Joined: 14 May 2005
  • Loc: montana

Posted 21 June 2010 - 09:32 AM

Dave, thanks for all the work you have put into this! It is greatly appreciated!






Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics