This is an exploratory overview of current and upcoming sources of data, processing pipelines and data products. It is aimed to offer non GIS experts an exploration of the unfolding revolution of earth observation, with an emphasis on development. See footer for license and contributors.
Satellite platforms are getting better and cheaper to build and launch. New platforms have higher resolution, collect images with greater frequency and collect new types of data, allowing new insights.
A typical serviced satellite image (bought or downloaded) includes several processing steps. From the raw image (Level 0) band images are combined to produce a color image in calibrated units (called Level 1 processing), and the image is geolocated and ortho-rectified following an elevation model of the terrain or Ground Control Points (GCP) with a certain Accuracy, under a geodetic reference frame (E.g. Mercator). For big areas, several captures need to be stitched together. This product is usually referred as Level 2 or similar. In many cases, the end user often only sees the final map image, either in a report, or as an interactive map on the web, like the maps in this page.
An important factor regarding the use of satellite images is the License of the data. It is important to understand the limitations, reusability and shareability of the data. For open data, like Landsat, there are no restriction of use. For a typical high-resolution image purchase, usage is restricted to a project, and only rendered maps are publishable. Sometimes tracing rights for open platforms is granted, but License terms vary widely on a user or usage base.
Typically serviced data come from filtering the requested locations and selecting available times from the satellite archives. Sometimes new images need to be collected (e.g. for monitoring, disaster recovery or similar). In these cases prices and quality are very different since it's subject to orbital, cloud coverage or other constraints.
Notes on the table: Satellite refers to the main optical instrument onboard. Operational indicates satellite status. Open Data refers to no data costs or license restrictions to download and use ortho-rectified data. Spatial resolution is ground sampling at nadir for (pansharpened) data. Revisit data is time between orbital overpass from equatorial ground. Cost refers to order of magnitude cost for archived data. Typically cost can vary depending on resolution, instrument, processing, minimum area, license and distributions rights, or other factors.
While there is still a premium for the highest resolution imagery (<0.50m), medium to low resolution is suitable for many applications, and increasingly affordable or available at no cost.
The imagery samples below can help select the right imagery for your project.
In an urban environment like Fortaleza, free Landsat imagery (15m) shows the extent of the city. Medium resolution imagery like RapidEye (5m) shows roads. With Worldview 3 (.31m) and Pleiades (.5m) images you can count beach umbrellas.
Resolution refers to the size one pixel in the image has on the ground. For example 15 meters means that one pixel on the image corresponds to a square of 15 by 15 meters on the ground.
Pansharpening. Often satellites take images of lower resolution in many bands (e.g. Multispectral red, green and blue for a color image), and then have another channel with much wider spectral range (i.e. "panchromatic") with finer resolution. By means of a mathematical relations, it is possible to pan-sharpen the multispectral image with the panchromatic resolution.
Accuracy. Resolution refers to the size of the pixel. Accuracy refers to the possible difference between the reported locations on the images and their actual location in the real world. A usual standard is a 90% likelihood within a Circular Area around the location (CE90). To ensure accuracy, satellite images are ortho-rectified for orography with either an elevation model, or using Ground Control Points (GCP).
Increased capture capacity for large satellites and new microsatellite platforms offer frequent revisit rates and more extensive image capture, allowing high frequency monitoring of changes on earth. More regular data makes it easier to study change over time, to track the progress of projects, and to “rewind” to the days leading up to a disaster.
Planet Labs high frequency imagery over Canada below shows the progress of the growing season, week by week.
Many modern satellites collect infrared and ultraviolet light not visible to the human eye. This information can be used to measure vegetation health, to monitor biomass, or to track forest fires. Newer laser and Radar-based sensors precisely scan the earth, eliminating weather dependencies, producing 3D earth models, and enabling better change detection.
In this imagery over rural Uganda, infrared sensors augment imagery to highlight vegetation (False Color) and to measure the health of that vegetation (NDVI). These images show a range of imagery resolutions from different growing seasons.
The combination of new sensors, deeper archives, and better computing ability make it possible to draw sophisticated information from satellite imagery. Development partners can now process imagery on the cloud and don't need to invest in expensive hardware and software.
Many satellites capture wavelengths such as infrared or ultraviolet light that can help to understand characteristics of the surface substances. Different objects reflect these light frequencies in different ways. Common satellite analysis techniques combine visible and invisible light bands (called "spectral bands") to characterize it. For example, spectral bands covering frequencies in the middle-infrared wavelength regions are responsive to moisture content in vegetation, forest canopy and soil, while near infrared wavelength regions tend to emphasize vegetation health and— at a courser scale— overall biomass.
Taken together, the spectral response of each band creates a unique signature, referred to as its spectral curve. The information it communicates can help to separate different types of land cover or derive information about the conditions within a single type of land cover.
A simple form of image interpretation involves assigning invisible light bands to Red, Green, and Blue channels to create a false color image to highlight certain characteristics. False color images can distinguish muddy water from muddy land, pinpoint fires, or clearly show the extent of a growing city. This example combines the moisture sensitivity of Mid-Infrared reflectance with the vegetation sensitivity of Near-Infrared to highlight agricultural fields.
There are several well-known combinations that are optimized to provide maximum contrast between categories of interest under various use cases.
| Combination Name | Red Wavelength | Green Wavelength | Blue Wavelength | LS 8 Bands |
|---|---|---|---|---|
| Natural Color (actual RGB) | 0.64-0.67µm | 0.53-0.59µm | 0.45-0.51µm | 4 3 2 |
| False Color (urban) | 2.11-2.29µm | 1.57-1.65µm | 0.64-0.67µm | 7 6 4 |
| Color Infrared (vegetation) | 0.85-0.88µm | 0.64-0.67µm | 0.53-0.59µm | 5 4 3 |
| Agriculture | 0.85-0.88µm | 0.64-0.67µm | 0.53-0.59µm | 5 4 3 |
| Atmospheric Penetration | 2.11-2.29µm | 1.57-1.65µm | 0.85-0.88µm | 7 6 5 |
| Healthy Vegetation | 0.85-0.88µm | 1.57-1.65µm | 0.45-0.51µm | 5 6 2 |
| Land/Water | 0.85-0.88µm | 1.57-1.65µm | 0.64-0.67µm | 5 6 4 |
| Natural With Atmospheric Removal | 2.11-2.29µm | 0.85-0.88µm | 0.53–0.59µm | 7 5 3 |
| Shortwave Infrared | 2.11-2.29µm | 0.85-0.88µm | 0.53–0.59µm | 7 5 3 |
| Vegetation Analysis | 1.57-1.65µm | 0.85-0.88µm | 0.64-0.67µm | 6 5 4 |
In addition to False Color composition, spectral bands can be combined mathematically to emphasize a particular set of characteristics. These techniques draw from all relevant bands, rather than the three band limits of false color analysis, to draw out very specific characteristics. These techniques involve greater processing of raw data in order to leverage all available data to cluster signals and minimize noise. While false color can be used to identify agricultural fields, mathematic transformations can detect the health of each field. While false color can distinguish mud from water, mathematic transformations can tell you how wet the mud is.
Multiplying the data from spectral bands in different combinations produces an index that can be used to compare every point in an image on the same scale. The most common indexes measure vegetation health and distinguish between types of plants, such as Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI).
Principle Component Analysis (PCA) decorrelates the data within each spectral band, such that the most common characteristics of all bands are placed in the highest category and less common characteristics are placed in lower categories until all variance is explained. It is invaluable for exploration of data and landscape characteristics, simultaneously drawing attention to the most noteworthy and best-hidden features in a scene.
KT-Transformations are a type of PCA, which, rather than being data-driven, combine information from multiple bands using specially developed coefficients to create the biophysically meaningful variables of Brightness, Greenness, and Wetness- the essential components of a landscape.
Advanced analytical techniques go beyond inter-band math. Advanced techniques compare data pixel by pixel, compare imagery taken at different times or different angles, or combining multiple source datasets. These techniques often result in data products like 3D terrain maps, the footprints of buildings, or the surface type of the road network. These techniques are most computationally intensive and require significant processing resources. The availability of elastic cloud computing has made these techniques more practical for development organizations.
Elevation and Surface models are constructed by calculating the offset in radar or image-based data acquired from different angles. 3D terrain models are used to measure landslide or flood risk. 3D urban models are used to measure and manage vertical growth of cities.
Analysis of imagery from different time periods to detect and understand changes. Examples include urban growth, deforestation, ice melting, landslide detection, or tracking urban conflict.
Analysis of imagery to extract and identify features. These can include man-made objects (buildings, roads, structures) and natural objects (Ex: land use, water extent, crop areas). Once extracted this items can be sorted, counted, and analyzed.
Conventional models of identifying, buying and delivering imagery are giving way to new cloud based selection and delivery models. Users can search through archival content from multiple suppliers to find the best suited imagery, buy the imagery through an enterprise subscription and have it delivered online. New delivery models are not limited to raw imagery. As the value of raw pixels declines, there is greater innovation and competition to provide data products derived from imagery.
Partnership purchase - Providers are considering new pricing options that make older imagery available for free or at significant discounts. For example, discussions are underway to streamline new licensing models that allow products licensed to the development partner to be accessed by the development partner's clients in country.
Imagery "rental" - As providers offer analytical and processing services they will create business and pricing models to accompany these services. In some cases analytical tools may come at a premium. Eventually these services will drop in cost. Providers will offer analytical products derived from imagery for less than the cost of the raw imagery itself.
New open satellite systems - Several new satellite systems are coming online, offering open information for development use. E.g.:
ESA’s Copernicus system will deploy six Sentinel satellites between 2014 and 2021 to monitor land, ocean, emergency response, atmosphere, security and climate change.
NASA’s satellite launches last year with data to come online with information on climate change. Data from the Global Precipitation Measurement Core Observatory, Orbiting Carbon Observatory-2, and Soil Moisture Active Passive are now coming online.
Mosaics blend together imagery from multiple scenes into a single, cloudfree seam-less image.
Hosted imagery and mosaics are tiled basemaps that can be easily integrated into web applications
Cloud Hosting and Analytics services allow users to run heavy computation across many images and only download the end product
Hosted catalogs allow users across an organization to browse and re-download any imagery you've already purchased
Unified search platforms provide results from multiple vendors and platforms, allowing choice of best available option.
More regular imagery makes it possible to monitor the progress of infrastructure projects. High resolution imagery and video can help to measure traffic on roads and road quality. Satellite data and analysis can provide the underlying data to inform infrasturcture investments.
As exemplified in the maps below, in countries lacking infrastructure maps, imagery could offer a cheap way to map the national road network. The highest commercial resolution imagery today (0.3 meters) allows for the detection of narrow, unpaved footpaths. These become invisible at 1.5 meters, at which point the gridded residential roads are the smallest observable feature. This grid remains recognizable at 2.5 meters, but is more likely to confuse with the surrounding vegetation. At 5-meter resolution, even the relatively large secondary road which cuts through the neighborhood becomes difficult to reliably distinguish, making the single primary road which travels the length of Palawan the only visible linear feature.
Due to the massive size and limited accessibility of forests, satellites have long been an indispensable tool for forest management. Satellite imagery derived forest maps measure increases in greenhouse gas emissions from deforestation, detect illegal logging activities, evaluate the availability and use of wood as cooking fuel, and assess the environmental impact of infrastructure projects such as road construction.
In agriculture, satellite imagery measures agricultural productivity. Imagery shows crop type and area of coverage, estimates rainfall, measures the effectiveness of irrigation systems, and evaluates the risk of droughts and floods. Satellite data is used for everything from developing insurance products for small farmers and early warning systems to detect decreasing agricultural production.
High frequency multispectral imagery creates significant opportunities for Forest and Agriculture projects. Regular imaging of rural areas in developing countries will yield comparable data at regular intervals unlocking new types of analysis of forest change and trends in agricultural production. More immediate information on crop health is critical to enable warning systems for potential drops in production in food sectors. Automated change detection will be invaluable in monitoring huge swaths of forest and agricultural areas.
Disaster events tend to affect large areas. The impact of such events can disrupt the normal functioning and organization of local authorities. Remote sensing information is immensely useful to gain oversight of the current situation after an event and to identify areas that are most affected. In the agriculture sector, remote sensing data is increasingly used for estimating crop yields to develop early warning systems for famines. Satellite image based coastline displacement and river bank erosion monitoring is also an important instrument to measure vulnerability of urban areas to flooding, storm surges, and tsunamis. While urban land cover maps are used to estimate impervious surfaces to track and manage water runoff patterns and sewer systems. Satellite imagery contributing to the mapping of informal settlements also helps access vulnerabilities.
Daily imaging will make it possible to prepare for, and track disasters as they occur. To "rewind" to immediately before a disaster, and support the recovery process. Time is a critical factor for disaster response. The quicker images can be distributed to measure impact and develop response strategies. Recent events such as the Nepal earthquake show the importance and usefulness of near real time satellite imagery. Having images taken shortly after disaster events can be immensely helpful to assist disaster response on the ground. Such images inform damage assessments and provide information on what infrastructure still exists for rescue efforts to reach affected areas. Better distribution approaches will be critical in allowing satellite data and analysis to get to responders as quickly as possible. Furthermore, before disasters happen, satellite images can provide means to reduce the vulnerabilities, and prepare for different scenarios. E.g. high resolution digital elevation maps are critical for evaluating the risk of floods, landslides, and earthquakes.