A GIS analyst's desk at dusk, two monitors showing a Sentinel-2 tile of the Mekong Delta beside a printed land cover map, a handheld GPS unit and a cold cup of tea at the edge of the keyboard.
How do I choose satellite imagery for a project in Asia?
Start with the question, not the catalogue. Write down two numbers: the size of the smallest feature you must see, and the longest gap between observations you can tolerate. A rice mapping project that needs field boundaries might require 10 metre pixels and a pass every week or two. A monsoon flood assessment might accept 20 metre pixels but need a new image within 24 hours of the peak.
Those two numbers eliminate most of the catalogue immediately. Optical sensors cannot see through cloud, and the Asian monsoon makes that a hard constraint from June to September across South and Southeast Asia. If your project window falls inside the wet season, either plan for radar or plan for gaps.
Licences matter as much as pixels. Landsat and Sentinel-2 are free and open, which means you can publish derived maps without negotiation. Some national datasets are free for research but restricted for commercial use, and a few require a request form and a waiting period. Read the terms before you build a workflow around a source.
Finally, check the archive depth. A sensor launched last year gives you a baseline of months, not decades. For change detection over ten or twenty years, the long Landsat record is usually the only option, even when a newer sensor offers sharper pixels. The Ground Truth Asia editorial team covers this trade-off in detail, and their material on choosing satellite data for Asia is a reasonable place to see how practitioners weigh resolution against archive length.
A practical order of operations: define the feature size, define the revisit, filter by licence, then filter by archive. Most projects end up with two datasets rather than one, a coarse long record for context and a finer recent record for detail.
Where can I download free satellite images of Asia?
Landsat is the oldest continuous record. The USGS EarthExplorer portal distributes the full archive at no cost, with 30 metre multispectral bands and a 16 day revisit for a single satellite. Combining Landsat 8 and 9 shortens the effective interval to eight days.
Sentinel-2, run by the European Space Agency, offers 10 metre bands in visible and near infrared, a five day revisit with both satellites, and global coverage. Access runs through the Copernicus Data Space Ecosystem, which replaced the older SciHub interface. For most Asian land projects, Sentinel-2 is the default first choice.
Radar is the wet season answer. Sentinel-1 provides C-band imagery that penetrates cloud and works at night, at 10 metre resolution and a 12 day revisit. It is harder to interpret than optical data, and flood mapping with it is a well-established but specialised skill.
National portals fill the gaps. Japan distributes ALOS data through JAXA, India through Bhuvan and the ISRO data archive, China through the China Centre for Resources Satellite Data and Application, South Korea through the Korea Aerospace Research Institute, and Thailand through GISTDA. Access conditions vary widely, from fully open to case-by-case review.
For weather rather than land, Himawari-8 and Himawari-9 deliver full-disk images every ten minutes over Asia and the western Pacific. That cadence is what makes real-time monsoon tracking possible, and the data is free through JMA and NOAA distribution channels.
Which Asian countries fly their own Earth observation satellites?
Japan operates a long line of missions, including ALOS-2 for radar and the GCOM series for climate variables. India runs the IRS programme, one of the largest national Earth observation fleets in the world, with Cartosat for high-resolution stereo and Resourcesat for moderate-resolution multispectral work.
China has built a substantial civilian and commercial constellation, including the Gaofen series under the CHEOS programme, with resolutions ranging from tens of metres down to sub-metre in the commercial tier. South Korea flies KOMPSAT satellites, several of which carry high-resolution optical sensors. Thailand operates THEOS-1 and THEOS-2, and has positioned itself as a regional data provider through GISTDA.
Smaller programmes exist across the region, often as technology demonstrators or as hosted payloads. The practical consequence for a project is that national data is rarely as easy to obtain as Landsat or Sentinel, but it can offer resolution or coverage that the open programmes do not.
What do resolution and revisit actually mean?
Resolution is usually quoted as ground sample distance, the size of one pixel on the ground. A 10 metre sensor cannot reliably detect a feature smaller than roughly two or three pixels across, so 10 metre data is useful for fields, roads and large buildings, not for individual trees or vehicles.
Spatial resolution is only one of several. Spectral resolution describes how many wavelength bands the sensor records and how narrow they are. Temporal resolution is the revisit interval. Radiometric resolution is the number of bits used per pixel, which sets how finely brightness differences are recorded. A dataset can be strong in one and weak in another.
Revisit is often quoted as a best case. A satellite with a five day revisit passes over a given point every five days, but only captures it if the sensor is switched on, the swath covers it, and the sky is clear. In practice, effective revisit over monsoon Asia is far longer than the nominal figure, and any project plan should be tested against actual scene availability rather than the specification sheet.
How do methods and ground truth change the choice?
Classification accuracy depends on training data, and training data depends on field visits. The term ground truth refers to observations collected on the ground that are used to check what a map claims. A land cover map of the Mekong Delta is only as good as the rice, orchard and aquaculture points someone recorded on a handheld GPS.
This is where dataset choice loops back to logistics. A sensor with a long archive and frequent passes gives you more chances to match a field visit to a cloud-free image. A sensor with very high resolution gives you sharper training polygons but fewer dates to choose from.
Free tools shape the choice too. QGIS handles the desktop workflow, and Google Earth Engine removes the download bottleneck by letting you process archives in place. Both are widely taught in the region, and both work better with open datasets than with restricted ones.
The community around this work meets annually. The Asian Conference on Remote Sensing has run for more than four decades, with the 45th edition in Colombo in November 2024 and the 46th in Makassar in October 2025, drawing several hundred participants and a few hundred papers each year. Proceedings from past conferences are archived by the Asian Association on Remote Sensing, and they are a useful record of which datasets have actually worked for which Asian problems.
A short decision sequence
Write the feature size. Write the maximum acceptable revisit. Decide whether cloud is a blocker for your season. Check the licence against your intended use. Check the archive length against your study period. Then pick two datasets, one coarse and long, one fine and recent, and test both against a known location before committing.
Most failed projects do not fail because the wrong sensor was chosen. They fail because the sensor was chosen before the question was written down.