{
  "id": "landsat-program",
  "title": "Landsat Program",
  "lang": "en",
  "source_lang": "en",
  "summary": "The Landsat Program is a United States satellite initiative providing a continuous, openly accessible record of Earth’s land surface since 1972.",
  "category": "technology",
  "subcategory": "transport",
  "type": "concept",
  "aliases": [
    "Landsat",
    "Landsat satellite program",
    "Earth Resources Technology Satellite program",
    "ERTS program"
  ],
  "generated_at": "2026-11-03",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/landsat-program/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/landsat-program/",
      "json": "https://aiwiki.page/zh/landsat-program.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "The Landsat Program is a series of civilian Earth-observation satellite missions jointly managed by the National Aeronautics and Space Administration (NASA) and the United States Geological Survey (USGS). Beginning with the launch of Landsat 1 on July 23, 1972, it established the longest continuous satellite record of Earth’s land surface. Its images measure reflected sunlight and, on most missions, emitted thermal radiation, enabling changes in vegetation, water, settlements, ice, and exposed rock to be studied over periods ranging from days to decades.\n\nLandsat occupies a distinctive position in remote sensing: its observations cover large areas systematically while retaining enough spatial detail to distinguish agricultural fields, forest clearings, river channels, and urban neighborhoods. The program’s scientific value rests not simply on individual images but on repeat observations, carefully characterized instruments, a preserved archive, and a free and open data policy introduced in 2008. Together, these features have made Landsat a foundation for environmental research, resource inventories, and public-sector mapping.\n\nThe program has also shaped how satellite observations become usable evidence. Successive spacecraft have changed instruments and measurement capabilities, while ground processing has sought to maintain comparability across generations. Understanding Landsat therefore requires attention to its institutional history, orbital design, sensors, data products, applications, and the limits of interpreting a long but imperfect observational record."
    },
    {
      "heading": "Origins, institutions, and the transition to public infrastructure",
      "text": "Landsat emerged from the convergence of space technology and demands for better information about natural resources. During the 1960s, photographs from crewed space missions demonstrated that broad patterns of geology, agriculture, and settlement could be recognized from orbit. Scientists and resource agencies proposed a dedicated, repetitive observing system rather than relying on occasional photographs. In 1966, Secretary of the Interior Stewart Udall announced an initiative to develop an Earth-resources satellite capability. NASA subsequently developed the spacecraft and launch system that became the Earth Resources Technology Satellite, or ERTS.\n\nThe program differed from both weather satellites and reconnaissance systems. Weather satellites emphasized frequent observations over very large areas, while reconnaissance missions served security requirements and generally restricted access to their imagery. ERTS was designed to provide civilian observations suitable for mapping land resources. Its measurements could support applications across national boundaries, making questions of data distribution, international ground stations, and scientific participation important from the outset.\n\nNASA launched ERTS-1 in 1972. The mission’s name changed to Landsat 1 in 1975, around the time Landsat 2 was launched. NASA’s role centered on research, instrument and spacecraft development, launch, and initial evaluation. The Department of the Interior, principally through the USGS, developed the long-term archive and many of the services through which observations reached users. The Earth Resources Observation and Science Center near Sioux Falls, South Dakota, became the central U.S. facility for preserving and distributing Landsat data.\n\nInstitutional arrangements changed substantially during the 1980s. The Land Remote-Sensing Commercialization Act of 1984 promoted commercial operation and distribution, and the Earth Observation Satellite Company, commonly known as EOSAT, assumed responsibilities for Landsat 4 and 5 operations and data marketing. Charging substantial prices for individual scenes limited the feasibility of studies requiring hundreds or thousands of images. The commercial model also exposed a structural difficulty: maintaining a continuous national observing capability did not necessarily align with the revenue obtainable from selling its measurements.\n\nThe Land Remote Sensing Policy Act of 1992 redirected the program toward federal stewardship and continuity. Subsequent arrangements evolved into the NASA–USGS partnership associated with later missions. Landsat thus became an example of scientific infrastructure whose value depends on sustained operation and long-term preservation rather than on the commercial value of isolated observations. International cooperation remained essential: receiving stations outside the United States collected imagery that later helped recover gaps in the central archive. The program’s institutional history explains why continuity, open access, and archive management are treated as core mission requirements rather than secondary services."
    },
    {
      "heading": "Spacecraft generations and continuity of the record",
      "text": "The first three Landsat satellites established the basic utility of repeated multispectral land observation. Landsat 1 launched in 1972, Landsat 2 in 1975, and Landsat 3 in 1978. Their Multispectral Scanner instruments recorded broad visible and near-infrared bands, revealing distinctions between vegetation, water, soil, and built surfaces that were not readily apparent in ordinary photographs. Landsat 1 operated until January 1978, substantially exceeding its original design life. These early missions created a baseline for environmental changes that later became prominent scientific and policy concerns.\n\nLandsat 4, launched on July 16, 1982, introduced the Thematic Mapper, a major improvement in spectral coverage and spatial resolution. Landsat 5 followed on March 1, 1984. Its unexpectedly long service became crucial to continuity: it supplied observations for nearly three decades and was decommissioned in June 2013. Landsat 5 also demonstrated the difficulty of planning a permanent observational record around spacecraft with finite design lives. Components, communications links, and onboard storage could constrain operations even when an imaging instrument remained scientifically valuable.\n\nLandsat 6 was launched on October 5, 1993, but failed to reach orbit. It was intended to carry an Enhanced Thematic Mapper that included a higher-resolution panchromatic band. The loss left Landsat 5 as the principal operational source for much of the 1990s. Landsat 7, launched on April 15, 1999, restored an overlapping mission and carried the Enhanced Thematic Mapper Plus. Its 15-meter panchromatic imagery and improved calibration supported both visual interpretation and quantitative analysis.\n\nOn May 31, 2003, Landsat 7’s scan line corrector failed. This mechanism had compensated for spacecraft motion during scanning. Images acquired afterward contained wedge-shaped gaps, amounting to roughly 22 percent of a standard scene outside the central gap-free portion. The remaining measurements retained their radiometric and geometric utility, so operations continued. Analysts could use the valid observations directly or combine several dates, but filling gaps could not recreate conditions that had never been measured.\n\nLandsat 8 launched on February 11, 2013, initially under the name Landsat Data Continuity Mission. Landsat 9 followed on September 27, 2021. Both use separate instruments for reflected solar radiation and thermal emission. Their overlap provides opportunities to check measurements against one another and increases the frequency of observations. Across the series, continuity has therefore meant more than launching replacement satellites: it has required retaining comparable spectral bands, arranging overlapping measurements, recovering historical holdings, and documenting instrument differences. The archive is continuous in its historical span, but not uniformly complete for every location and date."
    },
    {
      "heading": "Orbital geometry, coverage, and spatial sampling",
      "text": "Landsat satellites operate in near-polar, Sun-synchronous orbits. Such an orbit is designed so that the orbital plane gradually precesses around Earth, keeping daytime observations near a consistent local solar time. This reduces, but does not eliminate, differences in illumination between acquisitions. Comparable lighting is particularly useful when measuring vegetation change or comparing surface reflectance across years. Seasonal changes in solar elevation, terrain shadows, and atmospheric conditions still require careful treatment.\n\nThe earliest missions orbited at approximately 917 kilometers and followed an 18-day repeat cycle. Landsat 4 and later standard mapping missions used an altitude of about 705 kilometers and a 16-day repeat cycle. Landsat 8 and 9 were placed in the same orbital system with an eight-day offset. Their nominal combined opportunity to observe a given location is therefore eight days, although neighboring swaths create additional opportunities in overlap regions. Higher latitudes generally have more overlap because orbital ground tracks converge toward the poles.\n\nA typical modern Landsat swath is about 185 kilometers wide. Images are indexed using the Worldwide Reference System: the original WRS-1 serves Landsat 1–3, while WRS-2 serves later missions. A path identifies an orbital ground track, and a row identifies a position along it. This system allows a user to request approximately the same geographic footprint repeatedly without having to specify a new boundary for every acquisition. The scene grid is an organizational convention, however; environmental features rarely coincide with its boundaries.\n\nSeveral different meanings of resolution must be distinguished. Spatial resolution describes the size of the ground area represented by a measurement. Spectral resolution concerns the placement and width of wavelength bands. Temporal resolution concerns repeat observations, while radiometric resolution describes how finely measured signal levels are quantized. Improving one does not automatically improve the others. A 15-meter panchromatic band offers finer spatial detail than a 30-meter multispectral band, but it integrates light over a broad wavelength interval and cannot replace the spectral information in several separate bands.\n\nPixel spacing also should not be confused with the physical resolving power of an instrument. Optical response, detector dimensions, spacecraft motion, and resampling all influence how sharply boundaries appear. Thermal observations distributed on a 30-meter grid, for example, do not thereby acquire the native spatial detail of 30-meter reflected-light measurements. Similarly, a scheduled overpass does not guarantee a usable surface observation: clouds can obscure the ground for successive cycles. These distinctions are essential when choosing Landsat for tasks involving small features, rapidly changing events, or persistently cloudy regions."
    },
    {
      "heading": "Instruments and the physical basis of measurement",
      "text": "Landsat instruments separate electromagnetic radiation into wavelength bands that respond differently to surface materials. Visible bands record contrasts associated with color, pigments, and brightness. Near-infrared radiation is strongly reflected by healthy leaf structures but absorbed by water. Shortwave infrared bands respond to vegetation moisture, soil properties, and some mineral absorption features. Thermal infrared instruments measure radiation emitted by the surface and atmosphere. The scientific value lies in combining these signals rather than treating any single band as a direct label for a land-cover type.\n\nThe Multispectral Scanner, or MSS, was an especially influential early instrument. On Landsat 1 and 2 it recorded four bands spanning green, red, and near-infrared wavelengths. Its nominal ground resolution is generally described as approximately 80 meters, although its original sampling geometry was not a simple square pixel; standard archived products commonly use a 60-meter grid. A rotating mirror swept the instrument’s field of view across the ground. The MSS demonstrated that digital multispectral scanning could provide repeatable regional measurements, despite having originally accompanied a camera-based Return Beam Vidicon system.\n\nThe Thematic Mapper expanded coverage to seven bands. Reflective bands had 30-meter spatial resolution, while its thermal band had a native resolution of 120 meters. Two shortwave infrared bands greatly strengthened studies of vegetation moisture, burned areas, geology, and snow. Landsat 7’s Enhanced Thematic Mapper Plus retained broadly comparable multispectral coverage, added a 15-meter panchromatic band, and provided thermal measurements at 60-meter native resolution. These instruments used cross-track scanning, often called a whiskbroom design.\n\nLandsat 8 and 9 employ the Operational Land Imager and Thermal Infrared Sensor, designated OLI and TIRS, with OLI-2 and TIRS-2 on Landsat 9. They use pushbroom imaging: long detector arrays observe across the swath while orbital motion supplies the along-track dimension. Longer observation times per ground element improve signal-to-noise ratio, but the large number of detectors requires careful characterization so that differences among them do not create striping or artificial spatial patterns.\n\nOLI provides nine bands, including 30-meter multispectral bands, a 15-meter panchromatic band, a coastal/aerosol band, and a cirrus-detection band near 1.37 micrometers. TIRS measures two thermal bands at 100-meter native resolution. Its observations can support estimates of surface temperature, but temperature retrieval also requires atmospheric information and an estimate of surface emissivity. Landsat 8’s thermal measurements were affected by stray light originating outside the intended field of view; correction methods and instrument characterization improved the record. This illustrates why measured digital values must be interpreted through documented calibration, not assumed to be direct physical quantities."
    },
    {
      "heading": "From raw observations to consistent scientific products",
      "text": "The first stage of Landsat processing converts instrument telemetry into organized image measurements. Detector responses must be calibrated, defective measurements identified, and observations assigned to geographic positions. Geometric processing accounts for orbital position, spacecraft attitude, viewing geometry, and terrain. Well-controlled products use ground reference information and a digital elevation model to correct displacement caused by topography. Without this orthorectification, a mountain slope can appear shifted relative to a map or another image, creating false evidence of change.\n\nRadiometric processing distinguishes several physical quantities. At-sensor spectral radiance describes radiation reaching the instrument. Top-of-atmosphere reflectance normalizes reflected-solar measurements for factors such as incoming solar energy and Sun angle, but it still contains atmospheric effects. Surface reflectance attempts to remove the influence of atmospheric scattering and absorption. These corrections depend on estimates of aerosols, water vapor, ozone, and other properties; they are consequently less reliable under some conditions, including heavy haze or complex cloud edges.\n\nThermal processing follows a related but distinct chain. Calibrated thermal radiance can be expressed as brightness temperature using an inverse Planck relationship. Brightness temperature is not necessarily the physical land-surface temperature because the atmosphere modifies the signal and real surfaces do not emit as perfect blackbodies. Surface-temperature products therefore incorporate atmospheric corrections and emissivity information. Uncertainty can be appreciable over mixed surfaces, near clouds, or where emissivity inputs do not adequately describe changing land cover.\n\nThe USGS organizes standardized holdings into Landsat Collections. Collection 1 established a more consistent archive-wide framework, and Collection 2 further improved geometric referencing and provided globally available Level-2 products for supported instruments and acquisition periods. Level-1 products principally supply calibrated, geometrically processed observations; Level-2 products include surface reflectance and, where supported, surface temperature. Availability is not identical across every instrument generation: the early MSS record does not have the same standard Level-2 product coverage as later sensors.\n\nQuality information is integral to these products. Quality-assessment bands identify conditions such as clouds, cloud shadows, snow, saturation, and missing observations. A valid-looking pixel may still be inappropriate for a particular analysis, so filtering is usually necessary before constructing a time series. Product metadata also provide scaling factors, processing versions, acquisition geometry, and other information needed to interpret stored values correctly.\n\nConsistency across decades requires more than applying the same calculation to every image. Sensors have different spectral response functions, and changes in atmospheric correction or geometric reference data can introduce systematic differences. Cross-calibration, overlap comparisons, and explicit treatment of measurement uncertainty help separate these effects from environmental change. Reprocessing an archive can therefore improve historical measurements without changing the original observations themselves; preserving both the raw record and its processing history is a fundamental scientific responsibility."
    },
    {
      "heading": "Open access, archiving, and large-scale computation",
      "text": "In 2008, the USGS began implementing free electronic access to the Landsat archive. This change transformed the economics of using the record. Under a per-scene purchasing system, a regional analysis covering many years could require a large data-acquisition budget before interpretation began. With unrestricted downloads, researchers could instead examine all suitable observations, test alternative methods, and repeat analyses as improved products became available. The policy made exhaustive time-series analysis practical rather than exceptional.\n\nThe shift also changed the unit of scientific work. Earlier studies frequently compared a small number of carefully selected images, such as one scene before forest clearing and another afterward. Open access encouraged analysis of every available clear observation. This made it possible to estimate seasonal behavior, identify the approximate timing of disturbance, and distinguish an enduring trend from an unusual single acquisition. A cloudy or imperfect scene could still contribute useful pixels rather than being discarded as an indivisible unit.\n\nArchive completeness nevertheless remained a separate challenge. Historically, some observations were received and retained only by international ground stations. The Landsat Global Archive Consolidation initiative sought to bring these holdings into the USGS archive, recovering data that otherwise might have become inaccessible as storage media aged or receiving facilities closed. Recovered scenes can fill important temporal and geographic gaps, particularly in periods when direct downlink, onboard storage, or acquisition planning limited central holdings.\n\nModern distribution increasingly supports computation close to the data. Cloud-hosted archives, cloud-optimized file formats, and catalog services let software read selected bands or geographic windows without downloading every complete scene. An application programming interface can locate observations by time, location, and quality attributes, while distributed processing systems can build continental mosaics or analyze millions of pixels through time. The computational burden shifts from finding and purchasing imagery toward quality control, reproducible methods, and managing processing costs.\n\nTools based on machine learning have benefited from this accessibility, but algorithms do not remove the need for independent validation. Training labels may be concentrated in easily surveyed regions, and apparent classification performance can decline when a model is applied to different landscapes or seasons. Large-scale processing also makes small systematic errors consequential: an imperfect cloud mask repeated across millions of observations can generate a misleading regional pattern.\n\nOpen access therefore encompasses more than a zero purchase price. It depends on persistent archives, understandable metadata, stable product definitions, documentation, and sufficient computing capacity. Landsat’s experience helped establish the expectation that publicly funded Earth observations can support many unanticipated uses when users receive the underlying measurements rather than only finished maps."
    },
    {
      "heading": "Agriculture, forests, and ecosystem change",
      "text": "Agricultural applications exploit Landsat’s ability to observe fields repeatedly at a scale finer than most regional crop patterns. Visible, near-infrared, and shortwave infrared measurements reveal differences in planting dates, canopy development, moisture conditions, and harvest timing. A single acquisition may not distinguish crops with similar appearances, but their seasonal trajectories often differ. Combining imagery with field observations and agricultural records supports crop classification, cultivated-area estimates, and studies of management changes.\n\nA widely used calculation is the Normalized Difference Vegetation Index, or NDVI:\n\n\\[\\mathrm{NDVI}=\\frac{\\rho_{\\mathrm{NIR}}-\\rho_{\\mathrm{red}}} {\\rho_{\\mathrm{NIR}}+\\rho_{\\mathrm{red}}},\\]\n\nwhere \\(\\rho_{\\mathrm{NIR}}\\) and \\(\\rho_{\\mathrm{red}}\\) are reflectances in near-infrared and red bands. The contrast arises because chlorophyll absorbs red light while leaf structures strongly scatter near-infrared radiation. NDVI is useful for tracking green vegetation, but it is not a direct measurement of crop yield or plant health. Soil background, atmospheric contamination, canopy structure, and saturation in dense vegetation affect its interpretation.\n\nThermal and reflective measurements together support estimates of evapotranspiration, the combined transfer of water to the atmosphere through evaporation and plant transpiration. Surface energy-balance models relate net radiation to sensible heat, ground heat, and latent heat. Landsat’s field-scale detail can reveal differences in water consumption among neighboring agricultural parcels. These estimates are relevant to irrigation accounting, but they depend on meteorological inputs and model assumptions; the satellite does not directly measure the volume of water withdrawn from a well or canal.\n\nIn forestry, Landsat time series identify abrupt canopy disturbance from logging, fire, or storms and slower changes associated with regrowth or declining canopy condition. The distinction between forest cover and forest use is important. An area temporarily cleared for timber production may remain classified as forest land in an inventory, while satellite measurements record the immediate loss of tree cover. Global tree-cover loss products derived from Landsat therefore require careful interpretation and are not interchangeable with every legal or ecological definition of deforestation.\n\nThe long archive is particularly valuable for studying ecosystem trajectories. Repeated observations can distinguish a recently disturbed forest from one undergoing decades of recovery, or persistent agricultural abandonment from a temporary fallow period. They can also document changes in wetlands, grasslands, and coastal vegetation. Yet spectral recovery need not imply ecological recovery: a site may become green again while retaining different species, lower structural complexity, or altered soil conditions.\n\nRelationships to biodiversity are consequently indirect. Landsat can map habitat extent, fragmentation, disturbance, and some vegetation properties, but it usually cannot identify individual animal species or many understory plants. Ecological interpretation gains strength when imagery is combined with field surveys, airborne measurements, and explicit knowledge of the organisms or processes being studied."
    },
    {
      "heading": "Water, urban environments, and environmental hazards",
      "text": "Water mapping is one of Landsat’s most extensive applications because water generally absorbs near-infrared and shortwave infrared radiation strongly. Repeated observations can distinguish persistent lakes and rivers from seasonal inundation, reservoirs, floodplains, and intermittently wet land. The long archive allows changes in water occurrence to be separated from conditions on a single date. One influential global analysis published in 2016 used more than three million Landsat images acquired between 1984 and 2015 to map the distribution and persistence of surface water.\n\nSuch measurements document reservoir filling, lake contraction, river migration, and changes associated with drought or water management. They do not directly establish the cause of a change. A shrinking lake may reflect altered precipitation, increased evaporation, upstream abstraction, or several interacting influences. Turbid water, floating vegetation, shadows, and narrow channels can also complicate classification. Connections to groundwater require additional evidence because ordinary optical imagery does not directly measure subsurface water storage.\n\nUrban studies use Landsat to reconstruct the expansion of built-up land, changes in vegetation, and the conversion of agricultural or natural surfaces. A 30-meter pixel often contains a mixture of roofs, roads, trees, and soil, so urban classifications commonly describe dominant cover or estimate fractional components. These observations support research on urbanization and provide spatial context for urban planning, but population, building occupancy, and social conditions must come from other sources.\n\nThermal observations reveal differences in land-surface temperature associated with vegetation, moisture, construction materials, and exposure to sunlight. They are widely used to investigate the surface urban heat island. Surface temperature, however, differs from near-surface air temperature, and Landsat’s daytime overpass captures only one part of the daily heating cycle. A hot roof in an image is not a direct measure of indoor temperature or human heat exposure. Those questions require meteorological and other observations.\n\nFor wildfires, shortwave infrared and near-infrared measurements help delineate burned areas and characterize changes in vegetation and surface condition. The normalized burn ratio and its pre-fire-to-post-fire difference are commonly used indicators. Field-based calibration is needed when interpreting these differences as ecological burn severity. Landsat also provides information after floods, volcanic eruptions, landslides, and coastal storms, although cloud cover and revisit timing may prevent observation during the event itself.\n\nThe program contributes to climate-change research through records of glacier margins, changing snow cover, permafrost-related surface disturbance, vegetation change, and land-use transitions. These are observations of particular components of the climate system, not a complete global climate record. Their greatest strength often lies in connecting broad climatic changes to landscape-scale responses while preserving the spatial detail needed to distinguish neighboring watersheds, slopes, fields, or settlements."
    },
    {
      "heading": "Observational limits and the design of future continuity",
      "text": "Landsat’s principal limitations arise from the same design compromises that make systematic global coverage feasible. Moderate-resolution imagery cannot resolve every small field, narrow stream, building, or fragmented habitat. Optical instruments cannot see the land through thick cloud, and cloud shadows may obscure additional areas. Snow and low Sun angles complicate observations at high latitudes, while humid tropical regions can have long intervals without clear views. A nominal repeat interval is therefore not the same as the effective interval between usable measurements.\n\nTemporal sampling also affects conclusions about change. If a forest disturbance occurs between two clear acquisitions, its date can usually be bounded but not necessarily identified precisely. Short floods, brief crop stages, or rapid fire behavior may be missed entirely. Combining Landsat with other systems can reduce these gaps. Synthetic-aperture radar is especially complementary because it operates independently of daylight and can observe through clouds, although its measurements respond to surface structure and moisture in ways that differ fundamentally from optical reflectance.\n\nCross-mission integration is another route to more frequent coverage. The European Sentinel-2 satellites provide broadly comparable visible, near-infrared, and shortwave infrared observations, with some bands at finer spatial resolution. The Harmonized Landsat and Sentinel-2 project adjusts selected observations to common conventions for spatial gridding, atmospheric correction, spectral response, and viewing geometry. Harmonization improves interoperability but does not make the instruments identical. Different overpass times, bandpasses, and observation geometries remain relevant for some applications.\n\nInterpreting long records also requires attention to changing acquisition practices. Early archives contain geographic and seasonal gaps that later systematic acquisition cannot retroactively repair. Instrument failures, orbital changes, inconsistent atmospheric conditions, and varying ground control may introduce artifacts. Statistical change detection must therefore test whether an apparent shift corresponds to an actual surface change, a change in the measurement system, or both. Independent observations and overlapping missions are particularly valuable for resolving this ambiguity.\n\nNASA and the USGS have developed Landsat Next as a successor concept intended to expand the program’s capabilities while retaining compatibility with its historical record. The design announced in 2022 envisaged a constellation of three spacecraft, a combined six-day repeat interval, and 26 spectral bands. Proposed spatial sampling ranged from 10 to 20 meters for many reflected-light bands and 60 meters for thermal and atmospheric bands. Additional bands were intended to improve observations of water quality, vegetation processes, minerals, snow, and surface temperature. These are design objectives rather than properties of the historical Landsat fleet.\n\nExpanded capability creates its own continuity problem: new measurements must be tied to established observations without discarding the advantages of improved instruments. This requires overlap, laboratory and on-orbit characterization, stable geographic referencing, and processing methods that document uncertainty. Long-term preservation must also accommodate greater data volumes and revised products while retaining access to earlier versions needed to reproduce published analyses."
    }
  ],
  "links": [
    "remote-sensing",
    "infrastructure",
    "united-states",
    "signal-to-noise-ratio",
    "temperature",
    "calibration",
    "orthorectification",
    "measurement-uncertainty",
    "application-programming-interface",
    "machine-learning",
    "evapotranspiration",
    "irrigation",
    "forestry",
    "ecosystem",
    "wetland",
    "biodiversity",
    "groundwater",
    "urbanization",
    "urban-planning",
    "climate-change",
    "synthetic-aperture-radar"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}