About understanding limitations and challenges in satellite data (product) quality
Introduction
Satellites provide a very reliable tool to observe and monitor many aspects related to Earth System Sciences – for example cloud coverage, vegetation cover types, surface temperature, atmospheric composition, glacier and snow cover extent and many more. Satellites allow us to gather observations in areas where weather and daylight conditions severely hamper human activities – such as the polar regions or mountains, or in remote areas that are logistically difficult to reach or to cover with an observing system on the ground, e.g. the open ocean or regions covered by rain forest. In addition, satellites allow us to observe the same area repeatedly or decades without human intervention on the ground – by e.g. the installation of a ground-based instrument network. Current records of sea-surface temperature, cloud coverage, soil moisture and several other geophysical quantities date back to the early 1980s or even before that - often with global coverage.
The quality and reliability of satellite observations depends on many factors. Using such observations for monitoring and research purposes requires a sound understanding of the observational capabilities and limitations-of-use. For this understanding, one needs to take into account the type of the satellite observations which depends i) on the orbit characteristics, and ii) on the sensing characteristics. One needs to understand that most satellite observations are influenced to some degree by the atmosphere. Even when the observable is a part of the (upper) atmosphere itself, e.g. the stratospheric ozone concentration, the influence of the atmosphere on the satellite observation cannot be neglected. Furthermore, satellite sensors very rarely measure the quantity of interest directly. They measure voltages, strengths of electric and/or magnetic fields, or they count photons, to mention a few. Hence, the measurement of a satellite sensor needs to be converted into the desired geophysical quantity, which are subsequently provided as so-called satellite data products. This conversion may involve processing chains of various complexity, various sets of calibration coefficients, assumptions and geophysical models that translate between the actually measured quantity, e.g. a microwave brightness temperature, and the geophysical quantity, e.g. the near-surface soil moisture content or sea-ice concentration on the polar oceans. Finally, one needs to understand that the measurement of a satellite sensor very often represents the conditions of a finite volume or area. The dimension of the sampled volume or area depends on orbit and sensing characteristics (see above, further explained in the respective sections).
All the above is to emphasize that the usage of satellite observations and/or satellite data products requires knowledge of the uncertainty of the observations / the data products. While more and more satellite data products include information about the uncertainties in the data files themselves, a large part of this information has to be taken from reports that are delivered together with the satellite data (products). Here, in addition, one has to distinguish between different levels (or kinds) of uncertainties. One kind results from the satellite measurements themselves and the conversion process between these measurements and the geophysical products. Another kind of uncertainty results from the validation of the satellite data products against independent measurements of the same geophysical quantity, i.e. an inter-comparison between the land surface temperature retrieved from a satellite measurement and the land surface temperature measured in situ on the ground. Such inter-comparisons quantify the difference between the satellite data product and the in situ measurement and the resulting difference is commonly termed error. Users are advised to check the so-called Algorithm Theoretical Basis Document (ATBD), if it exists, to learn about the retrieval steps, assumptions, auxiliary data and retrieval uncertainties, see e.g. ESA-CCI soil moisture satellite data product or EUMETSAT OSI SAF sea ice satellite data products. For an understanding of the error, users are advised to look at the respective reports (see end of section about “Representativity & Fit-for-purpose”)
Orbits
Most satellite sensors with relevance for Earth System Sciences are operating from, basically, three different kinds of orbits. One is the so-called geo-stationary orbit. The second one is the so-called GPS-orbit which is not further detailed in this contribution because the fraction of satellite data products that are based on satellites in this orbit is yet quite small. The third one is the so-called Low-Earth Orbit (LEO).
Satellites in the geostationary orbit are stationary above the Earth’s surface at or close to the equator at an altitude of approximately 35 780 km. These satellites remain at their position and revolve together with the Earth around its axis. This orbit is mostly used for so-called “weather satellites”. They have been providing maps of the cloud conditions at regular time intervals of 30 minutes or even 15 minutes since the 1970s. The high temporal sampling paired with the stationarity is the main advantage of these satellites because they allow continuous monitoring of the development of clouds and weather systems. The disadvantage of these satellites is that the sensors carried – mostly operating in the visible and infrared portion of the electromagnetic spectrum – do not permit to observe the polar regions. Furthermore, incidence angles at the surface decrease poleward, causing an expansion of the field-of-view (or air volume) sampled which has consequences for the representativity of the measurement. Also, the atmospheric influence on the observations carried out by a geostationary satellite increases poleward because the path through the atmosphere increases in length. Both these effects increase the uncertainty of the measurements made by sensors aboard this kind of satellite polewards.
Satellites in LEO fly around the Earth in different constellations. These satellites are the working horse for Earth System Sciences. Their orbits permit to completely cover the polar regions with satellite observations within one day – depending on the sensing characteristics (see next Section). LEO satellites typically fly around the Earth between 13 and 16 times per day. The number of orbits depends on the altitude above ground, which for most LEO satellites is between 600 km and 800 km, and the inclination (the overflight angle of the satellite with respect to the equator). The inclination determines together with the width of the area observed across the satellite track – the so-called swath width - the most poleward latitude at which the respective satellite can observe. The satellite track along which observations are performed shifts from one orbit to the next. LEO satellites are often sun-synchronous, i.e. they fly over the same spot on Earth into the same direction (i.e. northward – aka ascending, or southward – aka descending) at the same local time, e.g. always at noon. Local overpass times of the ascending and the descending orbit of one LEO satellite differ by about 12 hours. The advantage of being sun-synchronous is that for observations targeting the diurnal cycle of a phenomenon or geophysical quantity a constellation of two LEO satellites permits to observe at four different local times. Placing two LEO satellites at a local overpass time of their ascending orbits at noon and, e.g., noon + 4 hours, results in observations at four different local times. One of the disadvantages is that observations targeting surface properties at the same local time could be negatively influenced by diurnal cycles in cloud coverage.
Sensing characteristics
The sensing characteristics of a satellite sensor are determined by the part of the electromagnetic spectrum within which the sensor is designed to perform measurements, by the targeted geophysical quantity that is going to be derived from the satellite measurements and the requirements with respect to the uncertainty, and technical constraints. This contribution will not deal with the latter.
One can distinguish between two kinds of satellite sensors, active and passive sensors. Passive sensors require an external energy source, i.e. a body, surface or material that emits electromagnetic radiation. This can be the sun or the Earth itself. In case the external source is the sun, satellite sensors measure reflected or attenuated sunlight – mostly in the visible frequency and partly also the near-infrared range. In case the external source is the Earth, satellite sensors measure emitted thermal infrared or microwave radiation. Most (if not all) such satellite sensors are so-called imagers, providing an image with sizes between several tens to a few thousands of kilometers, or so-called sounders, providing a vertical profile through the atmosphere. Active sensors emit an electromagnetic signal by themselves, which is reflected, scattered, and attenuated by the Earth and/or in the atmosphere. Common imaging active sensors are scatterometer and synthetic aperture radar (SAR). Both types operate in the microwave frequency range of the electromagnetic spectrum; image sizes are similar to the above-mentioned passive sensors. Another type of active sensor are the so-called altimeters. These operate in the visible (laser altimeter or LiDAR) or microwave (radar altimeter) frequency range of the electromagnetic spectrum. By measuring their altitude above whatever surface directly underneath the satellite track these sensors provide high-resolution surface elevation measurements. Such sensors do not provide an image, hence the spatial coverage of altimeter observations is very sparse.
Both types of satellite sensors (passive and active) are subject to atmospheric influence, which causes uncertainties in both, the satellite observation itself and the geophysical quantity derived. In the visible and infrared frequency ranges, clouds inhibit satellite observations of the Earth’s surface often completely. Sensors designed to observe the Earth’s surface therefore mostly avoid frequencies or wavelengths for which the atmospheric influence is particularly large or the atmosphere is even opaque. This applies especially to infrared and microwave satellite sensors . However, despite choosing special frequencies or wavelengths, very often some atmospheric influence remains that needs to be removed before the retrieval of the targeted geophysical quantity can take place. Such a removal is commonly based on additional assumptions and uses auxiliary data which also have their uncertainties, e.g., data from atmospheric re-analyses. For the understanding of the uncertainties of a satellite data product it is therefore quite important to learn about the processing steps, assumptions, and data involved in any correction or removal of the atmospheric influence (e.g., from the ATBD, see above). The same applies to satellite sensors designed to observe the Earth’s atmosphere.
Field-of-view and areal coverage
The field-of-view of a satellite sensor – also called spatial resolution - is the area observed by one sensing element of the sensor. The field-of-view (FoV) size ranges from less than one meter for some satellite sensors operating in the visible frequency range to several tens of kilometers for low-frequency passive microwave satellite sensors. Small FoV sizes are realized with detectors or antennas that are composed of many small cells like for SAR sensors, while large FoV sizes are common for passive microwave sensors that often operate with one or a set of parabolic dish antennas.
The FoV (size) of a satellite sensor is determined by the part of the electromagnetic spectrum / frequency used, the targeted geophysical quantity and the type of sensor – including its sensing characteristics. For the same type of sensor, for instance an active microwave sensor or an altimeter, a substantial range of FoV sizes can exist. FoV sizes differ by more than one order of magnitude between laser and radar altimeters or between the lowest and the highest frequency of commonly used multi-frequency passive microwave sensors. Differences in FoV size are even larger in the visible frequency range (< 1 m to ~250 m) and for active microwave sensors (~3 m to ~25 km). However, the fine FoV sizes mentioned above come with the drawback that the image that is acquired during one satellite overpass as a whole often is quite small. Satellite images with < 1 m spatial resolution are generally not larger than 10 km x 10 km while satellite images with ~250 m spatial resolution can cover 2000 km x 2000 km or even more. This has consequences for the application areas of satellite observations. There is a trade-off between areal coverage and spatial resolution.
Note that the FoV size and the spatial resolution must not be mixed with the grid resolution often mentioned in the context of satellite data products. The grid resolution is the dimension of the grid cells of the geographic grid into which the satellite observations are mapped to make them more useful for the user community. The most striking example here are satellite passive microwave sensors whose FoV at the surface is of elliptical shape and where adjacent FoVs often overlap each other. Satellite data products relying on such satellite observations, e.g. the sea-ice concentration, are provided with a specific grid resolution (often 25 km x 25 km) while the FoV sizes of the satellite observations used to retrieve the sea-ice concentration are often considerably larger. The remapping or gridding of the native satellite observations onto a regular geographic grid also comes with an uncertainty, which is not quantified that often yet and is often not provided together with satellite data products. This so-called “smearing uncertainty” or “sampling uncertainty” can be substantial (see [@lavergneetal2019], [@kolbeetal2024]).
Representativity / Fit-for-Purpose
Targeted phenomena or geophysical quantities vary on different spatial and temporal scales. Capturing these variations across the full range of scales with satellite observations is a challenge. Satellite sensors permitting to carry out observations with daily, all-weather, daylight independent global coverage often fail to resolve small-scale (say at 1 km) spatial variations because such global observations are often only possible at comparably coarse spatial resolution. Conversely, satellite sensors that permit to resolve such small-scale spatial variations of a geophysical quantity or phenomenon often lack global coverage. Hence, different types of satellite sensors offer a different degree of being fit-for-purpose. Users need to take this into account when searching for satellite data or satellite data products with the aim to study a specific phenomenon or geophysical quantity.
Another aspect that needs to be mentioned in this context is the way how satellite observations are converted into a geophysical quantity that is distributed in a satellite data product. The way how the conversion is done often depends on the requirements of the satellite data product. For instance, satellite data products that are supposed to be global but are based on satellite observations of an altimeter face the challenge that altimeters only provide measurements underneath the satellite track with very sparse spatial coverage. In order to fulfil the requirement of a satellite data product with global coverage one way to go is to compute monthly mean values of the targeted geophysical quantity gridded into a comparably coarse scale geographic grid. Such products then provide the mean, eventually also the median of the geophysical quantity, sometimes together with the standard deviation resulting from the averaging process. Such products can only provide a first-order approximation of the true conditions and are neither representative of the true variation of the targeted geophysical quantity nor are they fit-for-purpose studying changes in the spatiotemporal distribution of this quantity at the sampling the satellite sensor is originally providing. This is another aspect, users need to take into account when selecting a satellite data product for a certain research or application purpose.
Another area where the above-mentioned aspect of representativity plays an important role is the evaluation of satellite data products (see e.g. [@langsdaleetal2025]). It is common to evaluate satellite data products with in-situ measurements. These are, however, local and at best have been collected along one or more transects within the satellite sensor’s FoV. Often, such best practices are applied, however, only during validation experiments specifically dedicated to evaluate a particular satellite data product. A large fraction of the in-situ or other measurements used to evaluate satellite data products is originating from other sources of measurements, existing in-situ networks or stations. While all these contribute considerable to the total amount of in-situ measurements used for the evaluation it often remains unclear how representative these measurements are with respect to the spatial and temporal scales that are represented by the satellite data product. Users of satellite data products are therefore advised to check the respective reports in which information about the evaluation procedures and data is given, i.e. either the product user guide (PUG) or manual (PUM) or reports named like product validation report (PVR) or product validation and intercomparison report (PVIR) (see e.g. ESA-CCI soil moisture satellite data product or EUMETSAT OSI SAF sea ice satellite data products.
Some final remarks
This contribution is a first step to communicate some of the challenges that are awaiting for satellite remote sensing data product users. There is a lot more to write. Issues related to the different levels with which satellite data and satellite data products are distributed (from Level-0 to Level-4), issues related to the multitude of different sensing techniques used - imaging, altimeter, profiling, sounding, radio occultation, limb sounding, to mention a few, or issues about the different degrees of maturity of satellite data products, all these - and others - were not yet treated in this contribution. Also, this contribution is comparably short with respect to the different topics described as there is more to say about the different contributions to uncertainties and the different ways how a geophysical product is derived from satellite remote sensing observations.