LabHub

Blog

AI Satellites & Earth Observation 2026 Complete Guide - Planet Labs · Maxar Intelligence · Capella Space · ICEYE · Spire Global · BlackSky · Albedo · SkyFi · Sentinel · Korean KARI · Japanese JAXA + Synspective Deep Dive

한국어English日本語

Intro — In May 2026, Earth Observation is a "Daily-Refreshed Digital Twin"

Ten years ago, satellite imagery was a static product that arrived in days or weeks. As of May 2026, that paradigm has been completely demolished. Planet Labs’ SuperDove constellation of 200+ satellites images nearly all of Earth’s landmass every day at 3-meter resolution, Maxar Intelligence delivers near-hourly 30 cm optical with three WorldView Legion satellites and WorldView-3, Capella Space and ICEYE bypass clouds and night at 50 cm SAR, and Spire Global’s 100+ satellite constellation collects AIS, ADS-B, and GNSS-RO signals in real time.

This guide is not marketing material — it organizes the operators who actually launch, collect, sell, and analyze data as of May 2026. It walks through every satellite modality (optical, SAR, RF, hyperspectral, thermal), the analytics SaaS players (Planet Insights Platform, Maxar Esri Foundation, EarthDaily, Orbital Insight), EO foundation models like IBM-NASA Prithvi and Clay, and the Korea–Japan track that runs from KOMPSAT through JAXA ALOS to Synspective.

The EO 2026 Stack — Six Modalities

First, the big picture. As of May 2026, commercial EO splits across six modalities.

  1. Optical (electro-optical): visible + NIR. Sub-3 m is standard, 10 cm class now exists.
  2. SAR (Synthetic Aperture Radar): active microwave. Indifferent to clouds and night.
  3. RF (Radio Frequency): passive reception of ground/sea signals. SIGINT and maritime tracking.
  4. Hyperspectral: 100+ bands. Minerals, methane, crop stress.
  5. Thermal IR: wildfires, heat islands, factory activity.
  6. Meteorological / atmospheric: geostationary and GNSS-RO. Weather and climate.

Each modality has separate operators, layered with analytics SaaS and free government data. We walk through them in order.

Optical EO — Planet Labs’ Dove · SuperDove · Pelican · Tanager Lineup

The largest constellation in commercial optical EO is Planet Labs PBC’s Dove series. As of May 2026 the lineup is:

A typical API call looks like:

import os
import requests
from datetime import datetime, timedelta

API_KEY = os.environ["PLANET_API_KEY"]
URL = "https://api.planet.com/data/v1/quick-search"

search = {
    "item_types": ["PSScene"],
    "filter": {
        "type": "AndFilter",
        "config": [
            {
                "type": "GeometryFilter",
                "field_name": "geometry",
                "config": {
                    "type": "Polygon",
                    "coordinates": [[
                        [126.9, 37.5], [127.1, 37.5],
                        [127.1, 37.6], [126.9, 37.6],
                        [126.9, 37.5]
                    ]]
                }
            },
            {
                "type": "DateRangeFilter",
                "field_name": "acquired",
                "config": {
                    "gte": (datetime.utcnow() - timedelta(days=7)).isoformat() + "Z",
                    "lte": datetime.utcnow().isoformat() + "Z"
                }
            },
            {"type": "RangeFilter", "field_name": "cloud_cover", "config": {"lte": 0.1}}
        ]
    }
}

resp = requests.post(URL, auth=(API_KEY, ""), json=search)
for item in resp.json()["features"][:5]:
    print(item["id"], item["properties"]["acquired"], item["properties"]["cloud_cover"])

Planet’s strength is daily refresh. As of 2026 Planet is still nearly the only operator that can compare the same farm, port, or construction site against itself on the same pixel grid, every day.

Maxar Intelligence — 30 cm Resolution and WorldView Legion

Maxar, the long-time leader in commercial EO, split in September 2024 between its space-systems arm (the manufacturer that built WorldView) and its intelligence arm. Maxar Intelligence now runs the satellites and sells the imagery, while Maxar Space Systems focuses on building hardware.

As of May 2026, Maxar Intelligence’s lineup is:

Demand for Maxar imagery surged after the Ukraine war as governments and media moved in. Maxar holds the largest share of the U.S. NRO EOCL (Electro-Optical Commercial Layer) contract. Enterprises typically subscribe via SecureWatch or Esri integration.

BlackSky Gen-3 — 35 cm and Hourly Revisit

BlackSky went public in 2021 via SPAC. As of May 2026 BlackSky operates 14 Gen-2 satellites and began launching Gen-3 in 2026.

BlackSky differentiates on frequency. It concedes 30 cm to Maxar but aims to hit the same spot more often. In the defense and intelligence market it competes most directly with Maxar.

Albedo · Satellogic · Pixxel — Next-Gen Optical Challengers

Three companies are squeezing into the optical EO market from different angles.

Albedo’s 10 cm resolution redraws the line for commercial EO in 2026. It became viable only after the U.S. NOAA licensing guidelines relaxed from 25 cm to 10 cm.

SAR — The 50 cm Race Between Capella Space and ICEYE

SAR is the modality that runs in parallel with optical. Active microwave sensors observe the surface regardless of clouds, night, or haze. As of May 2026, four operators are competing.

SAR imagery is cognitively different from optical. Pixels encode grayscale intensity, but inside that signal lie surface roughness, moisture, and metallicity. Change detection (InSAR interferometry) can detect millimeter-scale ground deformation.

import sarsen   # open-source SAR processing
from sarsen import apps

# Download a Sentinel-1 GRD scene and apply terrain correction
apps.terrain_correction(
    input_path="S1A_IW_GRDH_20260501.zip",
    dem_path="copernicus_dem.tif",
    output_path="s1_corrected.tif",
    interp_method="bilinear",
)

ICEYE in Ukraine — Commercial SAR Validated for Military Use

Since 2022, ICEYE became the first operator to directly supply SAR data to the Ukrainian government. With 50 cm resolution it tracked Russian assets under cloud and at night, becoming the first proof point that commercial SAR is good enough for military operations.

In 2024 ICEYE transferred full ownership of one satellite to Ukraine. After that, Poland, Spain, and others signed government SAR contracts with ICEYE. As of May 2026, ICEYE has additional contracts with six NATO members.

Synspective StriX and the Japanese SAR Track

Synspective is a University of Tokyo spinout that has been launching the StriX series since 2020. As of May 2026, the lineup is:

JAXA’s ALOS-2 (PALSAR-2) provides complementary L-band SAR. JAXA is preparing ALOS-4 for launch in 2026, which is central to Japan’s EO self-sufficiency policy.

Umbra Lab — A 16 cm Record and an Open Catalog

Umbra Lab has been the fastest-growing SAR operator since its first launch in 2021. As of May 2026, it operates 9 satellites and holds the 16 cm commercial SAR resolution record.

Umbra differentiates on two axes.

Umbra is also a winner of the U.S. NRO Stratum 2 SAR contract — meaning NRO directly procures from Umbra as a core commercial SAR supplier.

Sentinel-1·2·3·5p·6 — ESA’s Public EO Backbone

ESA’s Copernicus program has been releasing Sentinel data for free since 2014. As of May 2026, the lineup is:

Sentinel-1 and Sentinel-2 are effectively the baseline of commercial EO. 80% of academic research in agriculture, disaster, and urban monitoring starts from Sentinel-2.

NASA Landsat 8·9 and Landsat Next

The NASA-USGS Landsat program has provided the longest continuous satellite record since 1972. As of May 2026, the operational fleet is:

Landsat is, together with Sentinel-2, one of the two pillars of free public data. It is the only dataset with 50+ years of continuous time series, making it the standard for climate, forest, and urban-change research.

RF SIGINT — Spire Global · HawkEye 360 · Unseenlabs

Beyond optical and SAR, a separate track has emerged for collecting RF signals from satellites.

RF constellations see what optical cannot. Fishing vessels turning off GPS (IUU fishing), ships with disabled AIS transponders (dark vessels), and nighttime communications geolocation all show up on RF. By 2025, HawkEye 360 data became a standard input to U.S. maritime security policy.

Spire Global — A Full Stack of AIS · ADS-B · GNSS-RO

Spire Global’s 100+ LEMUR constellation collects three data types simultaneously:

Spire’s GNSS-RO data is ingested directly by weather agencies like NOAA and ECMWF to improve numerical forecasts. As of May 2026, Spire is the largest supplier under the NOAA Commercial Weather Data Program.

import requests

token = "YOUR_SPIRE_TOKEN"
headers = {"Authorization": f"Bearer {token}"}

# AIS messages for a specific time window and bounding box
params = {
    "received_before": "2026-05-16T00:00:00Z",
    "received_after": "2026-05-15T23:00:00Z",
    "bbox": "125,33,130,38",  # southern Korean peninsula
    "limit": 100,
}

resp = requests.get(
    "https://ais.spire.com/messages",
    headers=headers,
    params=params,
)
for msg in resp.json()["data"]:
    print(msg["mmsi"], msg["timestamp"], msg["latitude"], msg["longitude"])

Hyperspectral EO — Pixxel · Wyvern · HySpecIQ · OroraTech

Hyperspectral imagers slice the spectrum from visible to shortwave infrared into 100+ bands and identify the spectral signature of materials. As of May 2026 four operators compete.

Hyperspectral tells you what something is, which optical and SAR can’t. Methane leaks, crop nitrogen deficits, mineral classes, and even algae species can be identified by spectral signature.

Methane and GHG Monitoring — Tanager · MethaneSAT · GHGSat · Carbon Mapper

In response to the climate crisis, methane leak monitoring became the hottest EO application after 2024. As of May 2026 the lineup is:

The loss of MethaneSAT was a big event in the EO community in 2025. Tanager + Carbon Mapper and GHGSat are now filling that gap. With tighter methane regulations from the U.S. EPA and the EU, commercial demand has also grown rapidly.

Wildfire and Thermal IR — OroraTech · NASA FIRMS · CalFire AI

Wildfire detection and spread prediction became a major EO use case after the 2024 Canada and Hawaii fires and the 2025 California and LA fires.

import requests

# NASA FIRMS real-time fire pixels (VIIRS 24h)
MAP_KEY = "YOUR_FIRMS_MAP_KEY"
url = (
    f"https://firms.modaps.eosdis.nasa.gov/api/area/csv/{MAP_KEY}/"
    f"VIIRS_SNPP_NRT/USA/1"
)
resp = requests.get(url)
print(resp.text[:500])

Geostationary Weather Satellites — GOES-T · Himawari · GEO-KOMPSAT

Separate from LEO EO constellations, geostationary (GEO) weather satellites form the backbone of global weather forecasting.

These six satellites produce the 10-minute global weather imagery, the primary source for tracking particulate pollution, typhoons, and wildfires.

JAXA ALOS · Himawari · GOSAT — Japan’s EO Lineup

JAXA (Japan Aerospace Exploration Agency) operates the following EO satellites as of May 2026.

Japan’s EO policy is self-sufficiency with limited commercialization. Some ALOS data is sold commercially via RESTEC, but government use is prioritized.

Korea’s KOMPSAT and KARI Lineup

As of May 2026, the KOMPSAT (multipurpose satellite) series run by KARI is as follows.

KOMPSAT imagery is commercially sold worldwide by SI Imaging Services (SIIS). The most frequent Korean coverage comes from KOMPSAT-7 (30 cm) and KOMPSAT-3A (55 cm).

Korean EO Startups — Telepix · SIIS · Rise · Nara Space

A wave of Korean EO startups emerged after 2020.

The Korean government targets 100+ private EO satellites by 2030 in addition to KARI. The direction is expansion from KOMPSAT-only reliance to a private constellation track.

Japanese EO Startups — Synspective · Axelspace · iQPS · Sky Perfect

Outside JAXA, Japan’s private EO scene is also diverse.

The strength of Japan’s EO startup ecosystem is the depth of SAR constellations. Synspective + iQPS combined will exceed 60 SAR satellites by 2030 — surpassing ICEYE in size.

Analytics SaaS — Planet Insights · Orbital Insight · Descartes Labs

A separate track sits above the satellites: operators that don’t own satellites but sell insights derived from imagery.

Analytics SaaS is moving EO imagery toward an "API-first" world. SkyFi in particular is the first marketplace where you can order a satellite scene with a credit card.

Open Data + STAC — Microsoft Planetary Computer · AWS Open Data · Earth Engine

Beyond commercial EO, the volume and tooling of free public EO data has exploded.

The STAC + Planetary Computer workflow looks like:

import pystac_client
import planetary_computer

catalog = pystac_client.Client.open(
    "https://planetarycomputer.microsoft.com/api/stac/v1",
    modifier=planetary_computer.sign_inplace,
)

search = catalog.search(
    collections=["sentinel-2-l2a"],
    bbox=[126.9, 37.5, 127.1, 37.6],
    datetime="2026-05-01/2026-05-15",
    query={"eo:cloud_cover": {"lt": 10}},
)

items = list(search.items())
print(f"{len(items)} scenes")
for item in items[:3]:
    print(item.id, item.datetime, item.properties["eo:cloud_cover"])

EO Foundation Models — IBM-NASA Prithvi · Clay · Satlas · NVIDIA Earth-2

Across 2024–2025, Earth observation foundation models arrived in earnest.

A Prithvi loading snippet:

from huggingface_hub import hf_hub_download
import torch
from einops import rearrange

ckpt = hf_hub_download(
    repo_id="ibm-nasa-geospatial/Prithvi-EO-2.0-600M",
    filename="Prithvi_EO_V2_600M.pt",
)

model = torch.load(ckpt, map_location="cpu", weights_only=False)
model.eval()

# Input: (B, T, C, H, W) = batch, time, channels, height, width
# Prithvi-2.0: 6 channels (HLS B02,B03,B04,B05,B06,B07), 224x224
x = torch.randn(1, 4, 6, 224, 224)
with torch.no_grad():
    feats = model.encoder(x)
print(feats.shape)  # learned token sequence

The Ukraine War and Commercial EO Going Military

Russia’s invasion of Ukraine in February 2022 instantly proved the military value of commercial EO. Over the past four years, several shifts followed.

As of May 2026, the U.S. defense and intelligence budget allocates 5x more to commercial EO than a decade ago. The era of government-only satellites is over.

AI Use Cases — Crops · Dark Vessels · Insurance · Minerals

EO + AI applications keep accelerating in 2026.

Each domain has its own AI model, but they’re converging on a pattern of Sentinel-2 + commercial EO + time-series transformer.

A Roadmap — From Zero to EO Productionization

When you first adopt EO, this sequence is safest.

  1. Start with free data: Sentinel-2 and Landsat 8/9. Search and download via STAC + Planetary Computer.
  2. Learn AI libraries: TorchGeo, raster-vision, eo-learn. PyTorch-based.
  3. Adopt foundation models: fine-tune Prithvi or Clay for downstream tasks.
  4. Order commercial imagery when needed: trial SkyFi (consumer) or Planet/Maxar (enterprise).
  5. Trial one SAR operator if needed: ICEYE, Capella, Umbra, or Synspective. A separate learning curve.
  6. Evaluate analytics SaaS: Orbital Insight, Ursa Space, EarthDaily. Faster and cheaper than running in-house.
  7. Self-operation comes last: launching your own satellites is unrealistic unless you are a government-scale player like KOMPSAT, JAXA, or KARI.

EO is not a game of launching satellites, but a game of data, models, and domain knowledge. When you adopt it, free data and open models can take you very far.

Wrap-Up — In May 2026, EO is "Automation, Not Observation"

The conclusion of this guide is simple. Earth observation in 2026 is no longer a game of data collection (observation) — it is a game of analytics automation. With Planet’s 200, ICEYE’s 30, and Spire’s 100 satellites, data is already abundant. The question is what you can automatically extract from it.

The biggest shift is the arrival of EO foundation models. Open models like Prithvi and Clay are speeding up downstream task delivery, and NVIDIA Earth-2 is now serious about weather and climate digital twins. Within five years, "EO foundation model" will be standard infrastructure, much like LLMs.

The other shift is the military validation of commercial EO. The Ukraine war proved that commercial satellites can be equal to — or faster than — government ones as a data source. The U.S. defense budget reflects this, expanding EOCL and Stratum contracts every year.

Don’t agonize over tool choice. With free data (Sentinel + Landsat) and open models (Prithvi + Clay), 90% of use cases are accessible. Start there.

References

Comments

No comments yet.

Sign in to leave a comment