Humboldt 250 Symposium 2019


New satellite observations for developing cloud climatologies at both high spatial and high temporal resolution for biogeographical applications

This page provides an expanded summary of a talk prepared for the August 2019 Humboldt 250 Symposium and the Second LatinAmerican Biogeography Congress to be held in Quito,Ecuador between Aug 6-9 2019.   Because the talk is allocated 12 minutes (plus 3 for questions) it is useful to have and online supplement to help with explaining various concepts and to illustrate more permanently the material that is presented in the talk.

The actual talk (when completed) will be here in a Google Photos version.

Material in support of the talk is given below.

One line summary of why this is important to biogeographers:   Any species distribution modeling work must incorporate satellite-based climatologies; Worldclim-based climatologies alone will not produce accurate results at higher spatial resolutions in the tropics.

The new GOES satellite series

This talk was originally motivated by the availability of observations from a new series of Geostationary Operational Environmental Satellites (GOES) being placed in geostationary orbit by the National Oceanic and Atmospheric Administration (NOAA).  Two such satellites are now covering the Americas and the eastern Pacific Ocean region; a similar satellite operated by Japan sits over Indonesia and covers the western Pacific.  It will be a few years more before the European community launches comparable satellites to cover Africa and Europe and also south Asia and the Indian Ocean.  Older satellites exist today and provide global geostationary coverage, it is just that technology continues to improve.  An example of the difference in spatial resolution between the old and new GOES satellites is in Fig 1.

GOES 14 vrs GOES16 Mar 17 2014 2019

Fig 1.  An image from the previous GOES-east satellite at 1745 UTC image on March 17 2014 (left) and one from the new GOES-East satellite on March 17 2019 (right).  Area is the south coast of Peru from the Paracas Peninsula (upper left).  Image on left has been contrast and brightness adjusted to be closer to the image on right.  Click on the image to see detailed view and compare the resolution of coastal and land surface features.

Despite the clear improvement in resolution offered by the new GOES imagery, it is still not quite as sharp as current MODIS-type imagery provided by the NASA research satellites like Aqua and Terra satellites – imagery that has been available for more than 16 years (see Fig. 2).  The difference of course, is in the time resolution of the imagery – daily from each satellite for the MODIS daytime imagery and about 70 times each day (for daytime visible imagery) from the new GOES.

MODIS GOES16 comparison Mar 17 2019

Comparison between MODIS imagery from the Aqua satellite (left) and from the new GOES-East on March 17 2019. Paracas Peninsula is shown – as in Fig 1, and the images have been enlarged to show same area (approximate because of different look angles – Aqua is polar orbiting and GOES is geostationary).  The MODIS imagery, aside from the color depiction, has higher resolution (250m versus 500m for GOES).  Click to enlarge.


Visualization of satellite data

Although perhaps not quantitatively useful, it is helpful to know how to visualize satellite imagery to understand better the physical processes contributing to the climate of any region, but especially tropical regions.

The new GOES imagery is now being redistributed via many websites.  However, most of these focus on the north American region, and few sites offer the spectrum of full resolution (both in space and time) imagery for the entire satellite view.  The site we have used is from the NASA Marshall Space Flight Center:

Another site has full disk, full resolution imagery for the past few weeks:

Individual images from a geostationary satellite are interesting, but much of the value of such images comes from animating a sequence of such images – or otherwise manipulating many of them.  Here we show examples of such manipulations.

Individual image inspection


animation of images to show cloud evolution






averaging images to show cloud motion and topographic effects

NOV11 MED all avg

varying the threshold to extract clouds of differing brightness (reflectance)

The threshold value of pixel brightness can be varied to detect clouds of different reflectivity.  This is initially unsettling because we would normally hope that there is one unique threshold value to “optimally” detect clouds in a satellite image.





This slideshow requires JavaScript.

extracting the diurnal cycle from many images at different times

By averaging morning and afternoon images separately one can see any diurnal changes that might be expected between morning and afternoon cloudiness.  The images below show such an averaging for Nov 11 2018.  The ovals and arrows highlight some of the more apparent changes between the morning and afternoon cloudiness.  Afternoon clouds are enhanced over and in the lee of significant topography in Cuba, Jamaica, Haiti and the Dominican Republic.  Cloud streets (arrows), due to daytime heating of islands under the northeasterly flow, are more evident in the afternoon downwind of Bahamian islands that are nearly flat.





This slideshow requires JavaScript.

Why no nighttime cloud products?

Although some of the satellite channels can detect clouds (or more precisely the infrared radiation emitted by them) at night and are very useful for meteorological forecasting applications, their use for basic cloud detection over land is limited.  This is because the radiation from the cloud tops is related to their temperature and in many cases at night the surface temperature an be colder than low clouds.  And where the elevation is high, such as over the altiplano and other high elevation terrain, the temperature at night can be comparable to clouds at 6km elevation over the ocean or over the Amazon basin.  Thus at night you cannot easily distinguish cold land surfaces from a low- or even mid-level clouds – at least in a strictly objective manner.

Where it is important to know the diurnal cycle of cloudiness

There are a few regions where knowing the diurnal cycle of the cloudiness is really important.  Foremost of these are the coastal regions where low stratus frequently occurs.  Such regions, shown schematically in fig cc, include the Pacific coast of south america, the atlantic coast of southern africa and the west coast of the USA and NW Mexico.  Some coastal areas of NW Africa, the northern Arabian Sea, and a variety of archipelago (Canary and Hawaiian Islands, Azores, Cape Verde Islands, Galapagos and a variety of others).


Fig ddd.  Regions where low stratus cloud impacts the coastal vegetation.  1) Hawaiian Islands, 2) coastal California and Baja California, 3) Galapagos Islands, 4) coastal Ecuador, Peru and Chile, 5) Macaronesia (Azores, Canary, Madeira and Cape Verde Islands), 6) Angola, Namibia and western South Africa, and 7) Socotra and coastal Oman.  Many other cloud-impacted locations can be seen (click on image for larger view).  This is an average of about 8 years of twice daily imagery mosaics at 5km resolution.  Brighter areas are more cloudiness, darker are less.  Ice and snow artifacts are apparent at higher latitudes.

These regions are impacted by stratus because the coastal topography extends to or above the layer of stratus and this leads to fog drip on windward slopes as well as reducing solar radiation to the underlying surface.  Similar impacts occur in many other parts of the tropics and subtropics where cumulus clouds are prominent.  Inspection of Fig ddd reveals such regions.

The climatological products described by Wilson and Jetz (2016) and Douglas et al (2016) are based on MODIS polar orbiting satellite data. Douglas et al (2016) showed that some information on the diurnal cycle can be obtained by processing separately the sun-synchronous Aqua and Terra satellite observations (3 hr separation in time) and differencing the results.  Unfortunately the satellite observations are near 1030 and 1330 local time and while they provide cloudiness tendencies around mid-day, this is a relatively poor depiction of the entire diurnal cycle.  Hence the need for GOES imagery to develop better estimates of the diurnal cycle of cloudiness.

GOES imagery for the diurnal cycle of cloudiness

Because frequent GOES imagery has been available since the late 1970’s, it has been used to develop estimates of the diurnal cycle of cloudiness.  This is most common with infrared imagery, despite its lower spatial resolution, since it is available at all hours.  Visible imagery has been used much less, since it is subject to artifacts due to snow and ice, sunglint over calm waters and the varying reflectance off of clouds due to the varying sun angle throughout the day.  Tall clouds can also shadow other clouds, especially when the sun angle is low.

The Chilean coast during the boreal summer

A good example of the strong diurnal variation of coastal cloudiness is that seen along the coast of Chile in the summer months (Dec-March approximately).  Imagery from only Dec 28 2018 through the end of March 2019 was used to develop averages of cloudiness for two-hour periods throughout the day.  In principle the averages could be made every 10 minutes, but to improve the signal-to-noise for the very short period (3 months) of available observations we averaged imagery over two hour periods (usually 8 images).  Images were initially available every 15 min from GOES16; they only became available every 10 minutes due to improved ground-station processing capability.


The results of the cloudiness for the two-hour periods is shown in Fig. vvv.  During the first hours of the morning the low stratus extend far inland, especially along low-elevation drainages.  There is a rapid diminishing of the stratus thereafter and the minimum cloudiness overall is late in the afternoon.  One can see that some areas have frequent cloudiness even late in the afternoon.  These locations are along the coast, against steep topography.

An average of all of the times is shown below:

MEAN all times 160 DEC28-MAR31

Mean of all images shown in Fig ccc.  Approximately 660 images at each of the 6 time periods was used.


That these “climatologies” are reasonable one can compare with multi-year MODIS-based climatologies (using a different threshold value).  Below are the Terra (1030 local time) and Aqua (1330 local time) average cloudiness for the period Nov-April.  The domain shown is somewhat larger than the GOES domain, and the map projection is slightly different.  An energetic student could remedy both of these deficiencies.


The 1030 Local time average is similar to the mean GOES-based cloudiness for the period 1500-1645 UTC (shown above in Fig vvv), and shows less inland penetration of cloudiness than the first two periods of GOES-based mean cloudiness.  The Aqua mean (right) is similar to the GOES average cloudiness for the period 1700-1845 UTC in Fig vvv.

Given the differences in the averaging period, there is excellent agreement between the GOES and MODIS cloud climatologies, especially along the coast where the impact of cloudiness on the underlying vegetation is critical.

Mean cloudiness along the Tropical Andes and its diurnal variation: boreal summer

Approximately three months of 15 minute (now it is available every 10 minutes) imagery from late December 2018 to the end of March 2019  was used to produce a boreal summer mean cloudiness “climatology” for the Andean region from northern Colombia to northern Argentina.  The focus was to show the broad diurnal variation in daytime cloudiness.  Given that only three months of data was used, the imagery was grouped into periods representing morning (1230-1515 UTC), midday (1530-1815 UTC) and afternoon (1830-2115 UTC) hours.  Although the hours selected for each period were consistent, because the locations do not lie on exactly the same longitude line, the diurnal changes are only approximately the same.

The focus here is on the broad patterns first; these are shown using the 1km pixel resolution imagery (the ‘Medium” resolution setting on the NASA Marshall webpage menu).  Because there is currently no well-established historical archive of imagery available online, the procedure to build a climatology is tedious, involving daily downloads of the individual images.  Even sorting the images, required for obtaining the diurnal cycle means,  is tedious, since the downloaded files do not have a date and UTC time.  A better procedure will be developed, but for the demonstration objectives of this talk and webpage, the “brute-force” procedure was adequate.  Other uncertainties in the development of consistent climatologies using visible imagery are of greater importance, such as the variable solar illumination angle throughout the day, cloud shadows – especially during the early morning and late afternoon hours, and contamination from bright land surfaces (salt flats, snow and ice).

Mean cloudiness along the Andes during January-March 2019

(merged Figure)