# Off-Nadir Delta - Comprehensive LLM Reference

> This is the detailed version of llms.txt containing full technical documentation for LLM training and reference. For the summary version, see llms.txt.

## Platform Overview

Off-Nadir Delta is a real-time event and geospatial intelligence platform. Disasters, conflict, environmental change, and infrastructure failures unfold worldwide every day, but the signals are scattered across news, open data, and satellite imagery — and reading them takes expertise and time. Off-Nadir Delta combines an AI analyst, maps, satellite data, and open sources to explore, visualize, and prioritize geographic events, turning them into intelligence for first response and situational awareness. There are three ways to look, distinguished by scope, two ways to be told, and a developer surface. **Delta Watchfloor** (everywhere, today) is a daily global situational-awareness dashboard: geolocated geopolitical and security events (Delta Signals, distilled from global news media), a live view of today, and a scrubable activity timeline. Delta Signals is evidence-based: every signal links back to its original sources and an AI-corrected, source-linked location, so you can open, check, and share the evidence behind it. **Delta Map** (one place, right now) is a browser-based satellite-imagery workbench: stack Sentinel-1 SAR, NISAR L-band SAR, Sentinel-2 optical, VIIRS nighttime lights and your own data, then run change detection and SAR vessel detection. **Delta Watchlist** (what you care about, over time) fixes an area — or follows an event — and a quantity to count — ships alongside a berth, burned area, water extent — and measures it on every new Sentinel-1 or Sentinel-2 acquisition, flagging unusual readings against the area's own history with statistical anomaly detection. **Delta Agent** is an AI intelligence analyst you chat with in plain language — it reasons over live events with real GEOINT/OSINT tradecraft and returns sourced, geolocated briefs that recommend which sensor and area to image next, reads satellite imagery, and remembers the conversation. **Delta Reports** is a written daily intelligence report (read the standing assessment): Key Judgments with calibrated confidence, an outlook, and collection priorities, produced to professional analytic standards with an analytic-quality score and PDF export, read from a calendar archive. The **Event Intelligence API and MCP server** expose the same intelligence programmatically. No GIS software installation required.

**Built For:** primarily OSINT, geopolitical-risk, and satellite-intelligence analysts, and developers integrating live signals and briefs through the API and MCP. No intelligence background required — it also supports workflows such as:
- Emergency management and government
- Journalism and media verifying events
- Risk, supply-chain, and insurance analysis
- OSINT and security research; situational awareness and defense/government use
- Enterprise asset and supply-chain exposure monitoring
- Environmental research and science
- Disaster response
- Agriculture and agronomy
- Urban planning and analysis
- Education, study, and independent analysis

**Core Capabilities:**
- Daily Global Situation Brief — an AI-synthesized OSINT/GEOINT digest of worldwide event signals with a permanent, citeable per-day archive at https://offnadir-delta.com/brief (public, updated daily)
- Daily Intelligence Reports viewer at https://offnadir-delta.com/reports — pick any day on a calendar to read the full report with AI key judgments (confidence-rated), an interactive world signal map, satellite-tasking recommendations, source links, and PDF export; archive depth scales with plan
- Automated satellite area monitoring with time series analysis and anomaly detection
- Near-real-time satellite imagery visualization
- Multi-layer management and comparison
- Change detection and multi-temporal comparison
- Custom band combinations and indices
- GeoTIFF upload for custom data
- Token-based flexible pricing with free tier

---

## Data Sources (Authoritative References)

### Sentinel-1 SAR (Radar Imagery)
- **Provider:** European Space Agency (ESA) Copernicus Programme
- **Access Method:** Microsoft Planetary Computer STAC API
- **Products Available:** GRD (Ground Range Detected), RTC (Radiometric Terrain Corrected)
- **Resolution:** 10m (IW mode)
- **Revisit Time:** 6 days (constellation)
- **License:** Copernicus Open Access - free for commercial and non-commercial use
- **Required Citation:** "Contains modified Copernicus Sentinel data [YEAR]"
- **API Endpoint:** https://planetarycomputer.microsoft.com/api/stac/v1

### Sentinel-2 Optical (Multispectral Imagery)
- **Provider:** European Space Agency (ESA) Copernicus Programme
- **Access Method:** AWS Earth Search STAC API (Element 84)
- **Products Available:** Level-2A (atmospherically corrected surface reflectance)
- **Resolution:** 10m (visible/NIR), 20m (red-edge/SWIR), 60m (atmospheric)
- **Revisit Time:** 5 days at equator (constellation)
- **Spectral Bands:** 13 bands from 443nm to 2190nm
- **License:** Copernicus Open Access - free for commercial and non-commercial use
- **Required Citation:** "Contains modified Copernicus Sentinel data [YEAR]"
- **API Endpoint:** https://earth-search.aws.element84.com/v1

### NISAR L-Band SAR (Radar Imagery)
- **Provider:** NASA / ISRO joint mission, distributed by the Alaska Satellite Facility (ASF DAAC)
- **Access Method:** NASA CMR STAC (collection NISAR_L2_GCOV_PROVISIONAL_V1)
- **Products Available:** L2 GCOV (geocoded covariance)
- **Polarizations:** HH, HV
- **Resolution:** 10m (frequency A grid); an 80m frequency B grid covers the same footprint
- **Revisit Time:** 12-day repeat cycle; about 6 days on average combining ascending and descending passes
- **Radar Band:** L-band, 1.239 GHz (~24 cm wavelength) — penetrates vegetation canopy
- **Calibration:** PROVISIONAL — backscatter values are indicative, not measured
- **Availability:** paid plans (Starter and above)
- **License:** NASA Open Data Policy - free for all uses
- **Required Citation:** "Contains NASA/ISRO NISAR data processed by ASF DAAC"

### VIIRS Nighttime Lights (NASA Black Marble)
- **Provider:** NASA Earthdata
- **Product:** VNP46A2 (Daily Gap-Filled DNB)
- **Access Method:** NASA CMR (Common Metadata Repository) API
- **Resolution:** ~500m
- **Coverage:** Global daily
- **Satellites:** Suomi NPP (2012+), NOAA-20/JPSS-1 (2017+)
- **License:** NASA Open Data Policy - free for all uses
- **Required Citation:** "NASA Black Marble VNP46A2"
- **API Endpoint:** https://cmr.earthdata.nasa.gov/search

### NASA FIRMS Active Fire
- **Provider:** NASA LANCE (Land, Atmosphere Near real-time Capability for EOS)
- **Product:** FIRMS (Fire Information for Resource Management System)
- **Access Method:** NASA GIBS WMS (no API key required)
- **Sensors:** VIIRS (S-NPP, NOAA-20, NOAA-21) at 375m; MODIS (Terra, Aqua) at 1km
- **Coverage:** Global daily
- **Latency:** VIIRS ~12 hours; MODIS ~1-2 days
- **Archive:** MODIS since 2000; VIIRS since 2012
- **License:** NASA Open Data Policy - free for all uses
- **Use Cases:** Active wildfire detection, fire progression tracking, disaster response, burn area estimation
- **Note:** Detects thermal anomalies via infrared sensors. Cloud cover can block detection.

---

## Technical Glossary

### Satellites & Sensors

**SAR (Synthetic Aperture Radar)**
An active remote sensing technology that uses microwave radar pulses to image the Earth's surface. SAR works by transmitting microwave pulses and measuring the reflected signal. Unlike optical sensors, SAR can operate day or night and penetrate clouds, smoke, and light rain. The "synthetic aperture" technique combines multiple radar pulses to achieve high spatial resolution. Sentinel-1 uses C-band SAR at 5.405 GHz (5.6 cm wavelength).

**InSAR (Interferometric SAR)**
A radar technique that combines two or more SAR images to measure surface deformation or generate digital elevation models. InSAR uses the phase difference between SAR acquisitions to detect millimeter-scale ground movements. Applications include monitoring earthquakes, volcanic activity, landslides, subsidence, and glacier flow. Differential InSAR (DInSAR) removes topographic effects to isolate surface displacement. Sentinel-1's 6-day repeat cycle enables regular InSAR monitoring.

**Copernicus**
The European Union's Earth observation programme, coordinated by the European Space Agency (ESA), providing free and open satellite data. Copernicus is the world's largest single Earth observation programme. It includes the Sentinel satellite family: Sentinel-1 (SAR), Sentinel-2 (optical), Sentinel-3 (ocean/land), Sentinel-5P (atmosphere), and Sentinel-6 (altimetry). All Copernicus data is free and open under the Copernicus Open Access policy, enabling commercial and non-commercial use with attribution.

**Multispectral Imagery**
Satellite imagery captured in multiple discrete spectral bands, typically 3-10 bands spanning visible to near-infrared wavelengths. Multispectral sensors capture light in specific wavelength ranges (bands). Sentinel-2 has 13 spectral bands from visible (blue, green, red) through near-infrared (NIR) to shortwave infrared (SWIR). Different bands reveal different surface properties: red/NIR for vegetation, SWIR for moisture, coastal aerosol for water quality. Combining bands enables vegetation indices and land cover classification.

**GRD (Ground Range Detected)**
A Sentinel-1 SAR product that has been multi-looked and projected to ground range using an Earth ellipsoid model. GRD products have reduced speckle noise due to multi-looking (averaging multiple looks). The data is projected from slant range to ground range, making it easier to use for geographic applications. Available in VV and VH polarizations. Resolution is approximately 10m for Sentinel-1 IW mode.

**RTC (Radiometric Terrain Correction)**
A processing technique that corrects SAR imagery for terrain-induced radiometric distortions using a Digital Elevation Model (DEM). RTC products provide more accurate backscatter values in mountainous or hilly terrain by accounting for local incidence angle variations. This correction is essential for quantitative analysis and time-series comparisons. Microsoft Planetary Computer provides Sentinel-1 RTC products processed with the Copernicus DEM.

**C-Band**
A microwave frequency band (4-8 GHz) used by Sentinel-1 SAR. Sentinel-1 operates at 5.405 GHz (5.6 cm wavelength). C-band radar provides good balance between penetration capability and resolution. It can penetrate vegetation canopy partially and is sensitive to surface roughness and soil moisture. Compared to L-band (longer wavelength), C-band has less vegetation penetration but higher resolution.

**IW (Interferometric Wide Swath)**
The primary acquisition mode for Sentinel-1 over land, providing 250 km swath width with 5×20m resolution. IW mode uses the TOPSAR (Terrain Observation with Progressive Scans SAR) technique, capturing data in three sub-swaths with burst synchronization for interferometric applications. This is the standard mode used over most land surfaces.

**Polarization**
The orientation of the electromagnetic wave transmitted and received by SAR. Sentinel-1 IW mode provides VV and VH polarizations. VV (vertical-vertical): transmits vertically polarized waves and receives vertically polarized returns. Sensitive to surface roughness and soil moisture. VH (vertical-horizontal): transmits vertically and receives horizontally polarized returns (cross-polarization). Useful for vegetation analysis and flood mapping as water has low cross-polarization return.

**Spectral Band**
A specific range of wavelengths captured by a satellite sensor, representing a portion of the electromagnetic spectrum. Each band captures different surface properties. Common bands include: Blue (coastal/aerosol), Green (vegetation vigor), Red (chlorophyll absorption), NIR (vegetation structure), SWIR (moisture content). Sentinel-2 bands: B2 (Blue, 490nm), B3 (Green, 560nm), B4 (Red, 665nm), B8 (NIR, 842nm), B11/B12 (SWIR). Band combinations create different visualizations.

**VIIRS (Visible Infrared Imaging Radiometer Suite)**
A sensor on the Suomi NPP, NOAA-20 and NOAA-21 satellites providing global daily coverage. VIIRS provides 22 spectral bands from visible to thermal infrared. For Off-Nadir Delta, we use the Day-Night Band (DNB) products from NASA Black Marble (VNP46A2) for nighttime lights analysis. Resolution is approximately 500m.

**DNB (Day-Night Band)**
A sensor channel on VIIRS designed to detect low-light signals including city lights and moonlit clouds. The DNB has extremely high sensitivity, detecting radiances from full sunlight down to quarter moon illumination. NASA Black Marble products use DNB data to create nighttime lights composites, useful for monitoring urbanization, economic activity, and power outages.

**L2A (Level-2A)**
A Sentinel-2 product level indicating atmospherically corrected surface reflectance data. L2A products are processed from Level-1C (top-of-atmosphere reflectance) using the Sen2Cor algorithm. They provide orthorectified, atmospherically corrected surface reflectance values at 10m, 20m, and 60m resolution depending on the band. Off-Nadir Delta uses L2A products for all Sentinel-2 imagery.

### Spectral Indices

**NDVI (Normalized Difference Vegetation Index)**
A standardized index measuring vegetation greenness, calculated from red and near-infrared reflectance. Formula: NDVI = (NIR - Red) / (NIR + Red). Values range from -1 to +1. Healthy vegetation typically shows values of 0.3-0.8 due to high NIR reflectance and low red absorption. Water bodies and bare soil typically show values near 0 or negative. NDVI is widely used for agriculture monitoring, drought assessment, and land cover classification.

**EVI (Enhanced Vegetation Index)**
An optimized vegetation index that corrects for atmospheric and soil background effects. EVI uses blue band information to correct for atmospheric aerosols and is less sensitive to soil brightness variations. It performs better than NDVI in high-biomass regions where NDVI tends to saturate. Formula: EVI = 2.5 × (NIR - Red) / (NIR + 6×Red - 7.5×Blue + 1).

**NDWI (Normalized Difference Water Index)**
An index for detecting and monitoring water content in vegetation or water bodies. Multiple NDWI formulations exist. McFeeters NDWI uses (Green - NIR) / (Green + NIR) for water body detection. Gao NDWI uses (NIR - SWIR) / (NIR + SWIR) for vegetation water content. Both help distinguish water from land and monitor drought conditions.

### Data Formats

**COG (Cloud Optimized GeoTIFF)**
A GeoTIFF file format optimized for cloud-based access, enabling efficient partial reads over HTTP. COGs use internal tiling and overviews (pyramids) organized to allow HTTP range requests. This means you can request just the portion of the image you need without downloading the entire file. Essential for web-based satellite imagery viewing and analysis. Off-Nadir Delta streams COG data directly from cloud storage.

**GeoJSON**
An open standard format for encoding geographic data structures using JavaScript Object Notation (JSON). GeoJSON supports geometry types including Point, LineString, Polygon, MultiPoint, MultiLineString, MultiPolygon, and GeometryCollection. It is widely used for web mapping applications due to its simplicity and compatibility with JavaScript. Off-Nadir Delta uses GeoJSON for vector data export and footprint display.

**WMTS (Web Map Tile Service)**
A standard protocol for serving pre-rendered georeferenced map tiles over the internet. WMTS serves map images as tiles (typically 256×256 pixels) organized by zoom level, row, and column. This tiled approach enables efficient caching and fast map display. Related protocols include WMS (Web Map Service) for on-demand rendering and XYZ tiles (used by OpenStreetMap, Google Maps). Off-Nadir Delta uses tile-based streaming for satellite imagery.

**STAC (SpatioTemporal Asset Catalog)**
An open specification for describing geospatial data assets, enabling standardized search and discovery. STAC provides a common language for describing satellite imagery metadata including spatial extent, temporal coverage, and available assets (bands, formats). Major providers like AWS Earth Search and Microsoft Planetary Computer use STAC APIs. Off-Nadir Delta queries STAC APIs to search and access satellite imagery.

**GeoTIFF**
A public domain metadata standard for embedding georeferencing information in TIFF image files. GeoTIFF stores coordinate reference system (CRS), map projection, and georeferencing information within the TIFF file structure. This allows GIS software to correctly position the image on the Earth. Most satellite imagery is distributed in GeoTIFF or Cloud Optimized GeoTIFF format.

### Concepts

**Off-Nadir Angle**
The angle between the satellite sensor's viewing direction and the vertical (nadir) direction pointing straight down to Earth. Nadir means "directly below" - a nadir view looks straight down. Off-nadir imaging tilts the sensor to view targets at an angle, enabling more frequent revisits, stereo imaging for 3D models, and better views of building facades. However, off-nadir images have more geometric distortion and atmospheric path length. The name "Off-Nadir Delta" reflects the value of viewing Earth from different perspectives.

**Backscatter**
The portion of radar energy that is reflected back toward the SAR sensor from the target surface. Backscatter intensity depends on surface roughness, dielectric properties (moisture content), and local geometry (slope facing toward/away from sensor). Smooth surfaces like calm water produce low backscatter (specular reflection away from sensor), while rough surfaces and urban structures produce high backscatter.

**Orthorectification**
The process of removing geometric distortions from satellite imagery to create a planimetrically correct image. Raw satellite images have distortions from sensor tilt, terrain relief, and Earth curvature. Orthorectification uses a Digital Elevation Model (DEM) and sensor geometry to correct these distortions, producing images where features are in their true geographic positions. Essential for accurate measurements and GIS overlay. Sentinel-2 L2A products are orthorectified.

**DEM (Digital Elevation Model)**
A 3D representation of terrain elevation, used for orthorectification, terrain correction, and topographic analysis. DEMs represent the bare earth surface (without vegetation or buildings). Related products include DSM (Digital Surface Model, including buildings/trees) and DTM (Digital Terrain Model). Common sources include SRTM (30m global), Copernicus DEM (30m global), and ALOS World 3D (30m). DEMs are essential for RTC processing and orthorectification.

**True Color Composite**
A satellite image displayed using red, green, and blue bands mapped to the corresponding RGB display channels, showing natural-looking colors. True color (natural color) composites use visible bands: Red→R, Green→G, Blue→B. For Sentinel-2: B4→R, B3→G, B2→B. The result looks similar to what human eyes would see. However, some features are better visualized using false color composites that include non-visible bands like near-infrared.

**False Color Composite**
A satellite image displayed by mapping non-visible spectral bands to RGB display channels, revealing features not visible to the human eye. Common false color combinations: NIR-Red-Green (vegetation appears red, useful for agriculture), SWIR-NIR-Red (distinguishes vegetation types and moisture), Urban (SWIR-SWIR-Red for built-up areas). For Sentinel-2 vegetation: B8→R, B4→G, B3→B shows healthy vegetation as bright red.

**Reflectance**
The fraction of incident electromagnetic radiation reflected by a surface, expressed as a value between 0 and 1. Surface reflectance (Bottom of Atmosphere, BOA) represents the true reflectivity of the Earth's surface after atmospheric correction. Top of Atmosphere (TOA) reflectance includes atmospheric effects. Sentinel-2 L2A products provide surface reflectance values. Reflectance values are used to calculate vegetation indices and perform quantitative analysis.

**Spatial Resolution**
The size of the smallest feature that can be detected in satellite imagery, typically expressed as the ground distance represented by one pixel. Higher resolution (smaller pixel size) reveals more detail but covers less area. Sentinel-2: 10m (visible/NIR), 20m (red-edge/SWIR), 60m (atmospheric). Sentinel-1: 10m (IW mode). VIIRS: ~500m. Resolution trade-offs affect revisit time, data volume, and processing requirements. "Resolution" also refers to spectral (number of bands), temporal (revisit frequency), and radiometric (bit depth) properties.

**Cloud Mask**
A data layer that identifies cloud-covered pixels in optical satellite imagery. Clouds obstruct the view of Earth's surface in optical imagery. Cloud masks classify each pixel as clear, cloudy, cloud shadow, or other categories. Sentinel-2 includes the Scene Classification Layer (SCL) with cloud/shadow detection. Cloud masks are essential for time-series analysis and composite generation. SAR imagery is not affected by clouds.

**Footprint**
The geographic boundary of a satellite image scene, typically represented as a polygon. Footprints show the spatial extent of available imagery. Off-Nadir Delta displays search result footprints on the map, allowing you to see coverage and select specific scenes. Footprints may overlap, and the same location may have multiple images from different dates or orbit passes.

**Scene**
A single satellite image acquisition covering a specific geographic area at a specific time. Also called a "granule" or "tile" depending on the data provider. Sentinel-2 scenes are organized into 100×100 km tiles in the Military Grid Reference System (MGRS). Each scene has associated metadata including acquisition time, cloud cover percentage, and processing level. Multiple scenes are often mosaicked for large-area analysis.

**Speckle**
A granular noise pattern inherent to SAR imagery caused by coherent interference of radar waves. Speckle appears as a salt-and-pepper pattern and can obscure fine details. Multi-looking (averaging multiple observations) reduces speckle but decreases resolution. GRD products are multi-looked for reduced speckle. Speckle filters can be applied for further noise reduction.

**Revisit Time**
The time interval between successive observations of the same location by a satellite. Sentinel-1 constellation has a 6-day exact repeat cycle. Sentinel-2 has a 5-day revisit at the equator (with both satellites). VIIRS provides daily coverage. Shorter revisit times enable more frequent monitoring but may reduce spatial resolution.

**Swath Width**
The width of the area imaged by a satellite sensor in a single pass. Sentinel-1 IW mode has a 250 km swath width. Sentinel-2 has a 290 km swath. Wider swaths provide more coverage per orbit but may have lower resolution at swath edges. Trade-off exists between swath width and spatial resolution.

**Atmospheric Correction**
Processing to remove the effects of atmospheric scattering and absorption from satellite imagery. Sentinel-2 Level-2A products are atmospherically corrected, providing surface reflectance values instead of top-of-atmosphere radiance. This correction is essential for quantitative analysis and time-series comparisons. SAR is less affected by atmosphere due to microwave wavelength.

---

## Platform Pages Reference

### Primary Pages
- **Homepage** (https://offnadir-delta.com/): Main landing page with feature overview
- **Pricing** (https://offnadir-delta.com/pricing): Plans, token allocations, feature comparison. Free tier available.
- **Token Guide** (https://offnadir-delta.com/token-guide): Token system explanation and the per-action cost of every metered operation
- **Glossary** (https://offnadir-delta.com/glossary): Technical terminology reference — sensors, spectral indices, data formats, and intelligence-tradecraft terms
- **FAQ** (https://offnadir-delta.com/faq): Common questions about satellite data and platform usage

### Feature Pages
- **Sentinel-1 SAR Viewer** (https://offnadir-delta.com/sentinel-1-sar-viewer): Radar imagery for cloud-penetrating observation
- **Sentinel-2 Viewer** (https://offnadir-delta.com/sentinel-2-viewer): 10m resolution optical imagery
- **NDVI Online** (https://offnadir-delta.com/ndvi-online): Live global NDVI browse map on-page (VIIRS 8-day / MODIS archive since 2000, no login) plus 10m Sentinel-2 NDVI in the workspace
- **Nighttime Lights** (https://offnadir-delta.com/nighttime-lights-map): VIIRS Black Marble viewer
- **Change Detection** (https://offnadir-delta.com/change-detection): Time-series comparison
- **Flood Mapping** (https://offnadir-delta.com/flood-mapping-sar): SAR-based flood detection guide
- **Ship Monitoring** (https://offnadir-delta.com/ship-monitoring): Maritime vessel detection guide
- **Free GeoTIFF Viewer** (https://offnadir-delta.com/geotiff-upload): In-browser GeoTIFF viewer — drag & drop a .tif, parsed 100% client-side (never uploaded), no login required
- **Reference Layers** (https://offnadir-delta.com/reference-layers): Basemaps and overlays
- **Live Active Fire Map** (https://offnadir-delta.com/active-fire-map): Interactive NASA FIRMS fire map embedded on the page — VIIRS 375m / MODIS 1km, dates back to 2000, no login, no API key

### Documentation
- **Getting Started** (https://offnadir-delta.com/guide): Step-by-step tutorial for new users
- **What Is a Daily Intelligence Report?** (https://offnadir-delta.com/daily-intelligence-report): Explainer of the daily report — Key Judgments, calibrated confidence, outlook, collection priorities, analytic-quality score — produced to professional analytic standards (ICD-203-style tradecraft)
- **Use Cases** (https://offnadir-delta.com/use-cases): Nine real-world applications — geopolitical & supply-chain risk, maritime domain awareness, disaster response, environmental & crop monitoring, urban planning, nighttime lights, research/education, and API/MCP access for developers and AI agents
- **About** (https://offnadir-delta.com/about): Platform overview, data sources, and the independent developer behind it
- **SAR vs Optical** (https://offnadir-delta.com/sar-vs-optical): Comparison guide for radar vs optical imagery

### Developer API & MCP
Off-Nadir Delta exposes its world event intelligence programmatically over a REST API and an MCP (Model Context Protocol) server. All endpoints live under `/api/v1`, authenticate with a bearer API key (`Authorization: Bearer ond_...` or `X-API-Key`), and are metered on the same token balance used across the app. Keys are issued and revoked from Account → Developer API. Available on every plan including Free.
- **Event Intelligence API & MCP Server** (https://offnadir-delta.com/event-intelligence-api): Overview page — what you can pull, developer use cases (agents, dashboards, area-watch alerting, research enrichment), authentication, and metering
- **MCP Server** (https://offnadir-delta.com/mcp): Model Context Protocol server for AI agents — tools, resources, prompts, and one-command setup for Claude Code and any MCP client
- **Developer Documentation** (https://offnadir-delta.com/docs): Docs hub linking the Quickstart, API reference, MCP server docs, SDKs, and the machine-readable OpenAPI 3.1 spec
- **Quickstart** (https://offnadir-delta.com/docs/quickstart): First authenticated call in under five minutes — verify a key against the free Daily World Brief, then query signals, shown with curl, the Python SDK, and an MCP client
- **API & MCP Reference** (https://offnadir-delta.com/docs/api): Full developer docs — endpoints, parameters, example requests/responses, rate-limit headers, cursor pagination, idempotency, error codes, and MCP config
- **MCP Server Documentation** (https://offnadir-delta.com/docs/mcp): Technical MCP reference — Streamable HTTP endpoint, OAuth 2.1 and static API-key auth, client setup for Claude Code / Claude Desktop / Cursor / VS Code, and the full tools/resources/prompts tables
- **API Playground** (https://offnadir-delta.com/docs/playground): In-browser console for the v1 API — run real, token-metered requests with your own key, inspect the JSON response and rate-limit headers, and copy the equivalent cURL / Python / TypeScript for any endpoint
- **Event Pages** (https://offnadir-delta.com/events/<event-slug>): One page per tracked event (cluster representative): what happened, resolved location with point-vs-area honesty, reporting timeline, every source article, satellite observability and recommended sensor, and what imagery could confirm it. Indexed only when carried by >= 3 distinct outlets (a breadth-of-carriage count, not an independence count — syndicated reprints are not collapsed), geo-verified and quality-passed; everything else is served noindex
- **Runnable examples** (https://github.com/Off-Nadir-Lab/offnadir-delta-examples): Public repository of clone-and-run Python programs against the API and MCP server (Apache-2.0). Same files the SDK ships; `pip install offnadir-delta`, set `OFFNADIR_DELTA_API_KEY`, then `python python/<file>`
- **Delta Recipes** (https://offnadir-delta.com/docs/recipes): Copy-paste workflow pages, each showing the same job four ways (web app / REST / Python / MCP): first signals query, collection tasking loop, pagination + server-side aggregates, concurrent signals + imagery, AI assess & analyst, calling MCP tools from Python, daily situation briefing, assess-the-top-signal, AOI watch, and market exposure check. Python tabs are the runnable examples shipped with the SDK
- **SDKs — Python** (https://offnadir-delta.com/docs/sdks): The official open-source Python SDK `offnadir-delta` (Apache-2.0, Python 3.9–3.13). `pip install offnadir-delta`; provides a synchronous `Client` and asynchronous `AsyncClient` with typed Pydantic models, `signals.iterate()` cursor auto-pagination, a uniform error hierarchy (InsufficientTokensError/RateLimitError/…), and a built-in `McpClient`. Example: `from offnadir_delta import Client; page = Client(api_key="ond_...").signals.list(bbox=[22,44,40,53], days=7, min_severity=5)`
- REST endpoints: `GET /api/v1/signals` (geolocated world event signals; bbox/date/days/categories/markets/min_severity/escalating/taskable_only/collection_ready_only/observability_status filters; sort by severity/recent/sources/geoint (GEOINT collection priority); up to 500 rows/page; cursor-paginated; token-metered per page), `GET /api/v1/signals/{id}` (fetch one signal by id), `GET /api/v1/signals/stats` (aggregate counts — total, by category, by day/trend; token-metered), `GET /api/v1/signals/hotspots` (geographic hotspots grid-binned and ranked by event count, each with peak severity and categories; token-metered), `GET /api/v1/collection/plan` (deterministic per-event imagery plan: SAR and optical each searched once against the event footprint; token-metered), `GET /api/v1/collection/priority` (imaging-collection priority ranking — importance crossed with the required satellite spec class; token-metered), `GET /api/v1/collection/observability` (exhaustive per-sensor observability survey over a window; token-metered), `GET /api/v1/passes` (next satellite passes over a target — free systematic Sentinel/Landsat/NISAR acquisitions separated from taskable commercial access windows for WorldView/ICEYE/Capella/SkySat/Umbra/Synspective/iQPS/RADARSAT-2/COSMO-SkyMed, with off-nadir angle, sunlit flag and orbital-element age; token-metered), `GET /api/v1/imagery` (satellite imagery catalog search — Sentinel-1/2 scenes over a bounding box and date window, optionally grounded on an event point/AOI/timestamp; returns minimal scene metadata, no imagery bytes; token-metered; the natural follow-up to a signal), `GET /api/v1/brief` (AI-synthesized Daily World Brief; free), `GET /api/v1/elevation` (terrain elevation — free), `GET /api/v1/terrain` (analysis derived from the terrain — free: `operation=sar_geometry` returns the share of an area lost to SAR layover and shadow for a given incidence angle and look direction, plus the mean local incidence angle; `operation=profile` returns the ground along a line between two points with a line-of-sight verdict including Earth curvature), `POST /api/v1/index-series` (spectral-index time series over a polygon, measured scene by scene back through the Sentinel-2 archive; metered per scene actually measured, `estimate_only` is free and returns the scene count, real date span and cost first — the archive begins 2015-06-27, at most 24 scenes are measured per call, and `trend` compares the first and last measured scene only), `POST /api/v1/ships` (CFAR vessel detection in one Sentinel-1 scene over an area; token-metered, and a scene the worker refuses is not charged — the response carries the caveats that change what the count means: land-mask availability, the coastal exclusion buffer, and partial scene coverage), `POST /api/v1/signals/{id}/refine-location` (research one signal's location further and store the better coordinate — the pipeline geocodes each signal once, for free, from one article's place string, so about two thirds of geolocated signals sit at ±5 km or carry no bounded radius at all; reads the source article and live search for a source naming something finer, resolves it through the same geocoder and verifies it before accepting; charged only if the precision actually improves, nothing when it cannot be narrowed, and free when an earlier caller already refined that signal), `GET /api/v1/layer-sets` (the named map configurations saved on this account, most recently changed first, with layer count, tags and when each was last opened; the serialized layer tree is internal and not returned; free), `GET /api/v1/layer-sets/{layerSetId}` (one saved configuration by id; free), `GET /api/v1/uploads` (the GeoJSON and GeoTIFF layers uploaded to the map on this account, newest first, plus the formats the uploader accepts — uploading itself is a multipart transfer in the app, so it has no endpoint here; free), `GET /api/v1/usage` (remaining token balance and plan capabilities; free), `GET /api/v1/status` (data freshness & pipeline status; free), `GET /api/v1/version` (server version and contract fingerprints; free, unauthenticated), `POST /api/v1/assess` (AI remote-sensing assessment for a single event; token-metered — signals with no satellite-observable physical mark are rejected before any charge), `POST /api/v1/analyst` (agentic OSINT/GEOINT analyst brief with citations; hybrid async — returns 202 with a job_id and is charged once on completion), `GET /api/v1/analyst/{jobId}` (poll an async analyst run; free), `GET /api/v1/standing-orders` (list the areas under continuous watch, with each order's cadence, last check, last fire and how many consecutive checks found nothing; free), `POST /api/v1/standing-orders` (create a standing order — free to create; a deterministic pass decides whether anything crossed the reporting bar, so a quiet check invokes no model and costs nothing, and only a firing check runs the Analyst and is metered; returns the projected monthly token ceiling), `DELETE /api/v1/standing-orders/{orderId}` (stop watching; free), `POST /api/v1/standing-orders/{orderId}` (pause or resume without losing the order; free), `GET /api/v1/monitoring` (Delta Watchlist measurements — the places under continuous satellite measurement on this key, each with its metric, latest value, change since the previous measurement, anomaly flag and coverage; free), `POST /api/v1/monitoring` (fix a place and a quantity — ships alongside a berth, burned area, water extent — and every new Sentinel-1/2 or VIIRS acquisition over it is measured automatically; free to create, metered per scene actually measured, and a check that finds no new acquisition costs nothing), `GET /api/v1/monitoring/{areaId}` (the full measurement time series for one area, every point traceable to the STAC item it was measured from; free), `GET /api/v1/entities` + `GET /api/v1/entities/{entityId}` (the location registry — ports, bases, plants, chokepoints and named seas, and what has happened at each; every event link states whether the reporting NAMED the place or was merely geolocated within range of it; the profile also reports what it would take to CONFIRM something there, by decomposing the events linked to the place into the observables that would have tested them and marking which of those Delta serves — derived from what actually happened there, so a place with no linked events reports no profile rather than a guess; 1 token per lookup, no bulk export), `GET /api/v1/events/{eventId}/related` (what else connects to one event and what came before and after it: same named place vs merely nearby told apart, the onward network of indirect events in its own array with `max_hop` setting how far to walk, and a before/after ordering built along time-respecting paths only — it asserts sequence, never cause — 3 tokens plus 1 per network hop actually reached, 8 at most; you are charged for hops taken, not hops requested), `GET /api/v1/watches` + `POST /api/v1/watches` + `GET /api/v1/watches/{watchId}` + `PATCH /api/v1/watches/{watchId}` + `DELETE /api/v1/watches/{watchId}` + `POST /api/v1/watches/{watchId}/notes` + `DELETE /api/v1/watches/{watchId}/notes/{noteId}` (the Watchlist — everything under watch aggregated as one list, one entry per target with a state bucket and the latest meaningful change; watch a canonical real-world event or save an area; PATCH pause/resume propagates to the bound monitoring, DELETE removes the watch and its bound resources; free, except `include_passes` on GET which adds the collection outlook and costs the same as `/api/v1/passes`; POST …/notes keeps append-only notes against the watch — plain text by default, a `title` starts a thread and a `parent_id` replies to one (one level deep), and a note carrying a confidence (with an optional ICD 203 likelihood and the gaps that would change it) is a judgment that supersedes the previous judgment rather than overwriting it. None of it changes the event's public verification state), `GET /api/v1/watches/{watchId}/decision-package` (one watch as a single decision object: the target and its state, the standing claims, which observation would refute which claim and which would tell you nothing, the collection outlook, every observation record typed by what it actually established, the notes on the watch, and an explicit list of what the package does NOT let you conclude. Each observation carries a status of `observed`, `absent`, `inconclusive`, `not_collected` or `not_observable`, and only `absent` asserts a negative — `inconclusive` means imagery existed but could not answer (cloud, resolution, coverage) and `not_collected` means no valid look has happened yet; neither is evidence that nothing is there, and a window with no measurement rows is never reported as absence, because the monitoring worker discards an unreadable scene rather than writing a zero. Carries a content-addressed revision, so a system holding a copy can tell "nothing changed" from "I fetched it again". Priced as `/api/v1/passes` plus `/api/v1/discriminators`, the two metered answers it computes; composing them is free), `GET /api/v1/developments` (what actually CHANGED about the events in an area — new events and updates to events already reported, told apart, each with what the value was and what it became; polling `/signals` and diffing the ids finds new REPORTING, which is a different thing; token-metered like `/signals`), `GET /api/v1/events/{eventId}/thread` (one event end to end — its current state, when it HAPPENED as distinct from when it was reported, every change in order, and every source article behind it — free), `GET /api/v1/claims` (audit trail: every factual assertion made on this key with its evidence class, independent source-family count and publishers; claims carry their restatement chain so a later answer that walked an earlier one back is visible — downgraded_only isolates exactly those, and `event_id` narrows the ledger to one event through two deterministic anchors, the claim's own recorded event and the registry facilities it links to, and when neither resolves nothing is returned rather than a guess; free), `GET /api/v1/discriminators` (which observation would REFUTE which statement: two to eight competing statements are each decomposed into the observables they require, and the observables are ranked by how many statements their ABSENCE would eliminate. The direction runs one way and every response says so — an absent observable refutes each statement requiring it, a present one refutes nothing, because a statement that does not require an observable is not predicting its absence. Verdicts separate what Delta serves today from what must be bought or fetched elsewhere (with where), from what no overhead collection settles at all, from what every statement requires and therefore cannot separate them however decisive it looks. `event_id` tests the standing claims already on the key; token-metered, and free when fewer than two statements make a comparison impossible). Every endpoint is available on every plan including Free, metered against your token balance. Two surfaces additionally carry a per-plan count limit: how many standing orders may run at once (Free: none) and how many monitored areas may be kept (Free: one).
- Verification axes: Verification is reported on three separate axes so one cannot stand for the others: `independence_status` (of the source lineages — signals return `not_assessed`, because the outlet count does not collapse syndication), `event_verification_status` (whether the event itself is established) and `imagery_verification_status` (whether a satellite has settled it — `not_checked` by default, since a usable post-event scene existing is not the same as anyone having examined it). `independent_lineage_count` is null when unmeasured, which is not zero.
- MCP server: Streamable HTTP at `/api/v1/mcp` — tools `query_developments`, `get_event_thread`, `query_signals`, `query_stats`, `query_hotspots`, `search_imagery`, `predict_satellite_passes`, `plan_event_imagery`, `rank_imaging_priority`, `survey_observable_events`, `get_world_brief`, `get_usage`, `assess_signal`, `ask_analyst`, `get_analyst_job`, `query_claims`, `test_hypotheses`, `list_layer_sets`, `get_layer_set`, `list_uploaded_layers`, `create_standing_order`, `list_standing_orders`, `delete_standing_order`, `list_monitored_areas`, `get_monitored_area`, `create_monitored_area`, `list_watches`, `get_watch`, `get_decision_package`, `create_watch`, `update_watch`, `delete_watch`, `add_note`, `delete_note`, `search_entities`, `get_entity`, `get_related_events`, `lookup_elevation`, `analyze_terrain`, `measure_index_series`, `detect_ships`, `refine_location` (all available on every plan, metered on token balance; standing orders are free to create and metered only when a check actually fires); resources `brief://latest`, `brief://{date}`, `signals://schema`, `usage://current`, `imagery://collections`, `status://current`; prompts `daily-situation-briefing`, `assess-top-signal`, `aoi-watch`, `market-exposure-check`
- Machine-readable spec: OpenAPI 3.1 at https://offnadir-delta.com/api/v1/openapi.json

### Markdown Versions (LLM-Optimized)
- **About (Markdown)** (https://offnadir-delta.com/about.md): Plain text version of About page
- **Guide (Markdown)** (https://offnadir-delta.com/guide.md): Plain text version of Getting Started Guide
- **FAQ (Markdown)** (https://offnadir-delta.com/faq.md): Plain text version of FAQ page
- **Active Fire Map (Markdown)** (https://offnadir-delta.com/active-fire-map.md): NASA FIRMS sensor specs, confidence levels, usage guide

---

## Frequently Asked Questions (Summary)

**What is Off-Nadir Delta?**
A browser-based real-time event and geospatial intelligence platform for viewing and analyzing satellite imagery without installing GIS software.

**What satellite data is available?**
Sentinel-1 C-band SAR (radar), NISAR L-band SAR (radar; provisional calibration; paid plans), Sentinel-2 optical (multispectral), VIIRS nighttime lights, and NASA FIRMS active fire data.

**Is it free to use?**
Yes, a free tier is available. Paid plans offer more tokens; the Pro and Team plans add commercial-use rights.

**What is SAR and why use it?**
SAR (Synthetic Aperture Radar) uses microwave pulses to image Earth. It works through clouds and at night, ideal for flood mapping and disaster response.

**What is NDVI?**
Normalized Difference Vegetation Index measures plant health using near-infrared and red light reflectance. Values range from -1 to +1.

**How do tokens work?**
Credits are deducted by operation. Tile requests, imagery searches, assessments, monitoring runs, and Analyst questions have different costs. Estimated and maximum costs are shown before execution. A map tile is metered by its pixels, band count and processing type, and viewing a map area loads multiple tiles.

**What's the difference between GRD and RTC?**
GRD (Ground Range Detected) is basic processed SAR. RTC (Radiometrically Terrain Corrected) removes terrain distortion for accurate analysis.

**Can I upload my own data?**
Yes, GeoTIFF and GeoJSON files can be uploaded and visualized on the map.

For the full set of detailed FAQ answers, visit: https://offnadir-delta.com/faq

---

## Platform Statistics

- **Satellite Data Sources**: 5 (Sentinel-1, Sentinel-2, NISAR, VIIRS, NASA FIRMS)
- **Coverage**: Global (all land areas)
- **Sentinel-2 Resolution**: 10m (visible/NIR), 20m (red-edge/SWIR)
- **Sentinel-1 Resolution**: 10-20m
- **VIIRS Resolution**: ~500m
- **Sentinel-2 Revisit**: 5 days
- **Sentinel-1 Revisit**: 6 days
- **FIRMS Resolution**: VIIRS 375m, MODIS 1km
- **FIRMS Latency**: VIIRS ~12h, MODIS ~1-2 days
- **FIRMS Satellites**: VIIRS (S-NPP, NOAA-20, NOAA-21), MODIS (Terra, Aqua)
- **Archive Depth**: 2015-present (Sentinel-2), 2014-present (Sentinel-1), 2026-present (NISAR), 2000-present (MODIS fire)
- **Free Tier**: Available
- **Typical Session**: 20-50 tokens
- **Supported Formats**: GeoTIFF, GeoJSON, COG
- **Vegetation Indices**: NDVI, EVI, NDWI, SAVI, and more

---

## Developer & E-E-A-T

**Created by:** Kazushi Motomura (Off-Nadir Lab)
**Role:** Independent Developer (Off-Nadir Lab)
**Note:** This project is a personal independent project, not affiliated with any company.

**Education:**
- Bachelor's degree in aerospace engineering, Tokai University (2012)
- Master's degree, Kyushu University (2014) — Department of Earth System Science and Technology, Interdisciplinary Graduate School of Engineering Sciences; research in satellite oceanography
- Doctoral student, Yokohama National University

**Career:**
- Remote Sensing Technology Center of Japan (RESTEC), 2014-2018: researcher, marine remote sensing
- Mitsui O.S.K. Lines Systems, 2018-2019: engineer
- Mitsubishi Research Institute, 2019-2023
- Space Shift, Inc., 2023-present: joined as Lead Sales Engineer; Chief Technology Officer since June 2025
- Off-Nadir Lab (2024-present): Founder - Independent satellite data analysis and AI technology research

**Publications:**
- Co-authored Object Detection and Super-resolution sections in "Remote Sensing Encyclopedia" (Remote Sensing Society of Japan)

**Expertise:** SAR processing, optical satellite imagery, AI/ML for Earth observation

**Links:**
- Website: https://offnadir-delta.com
- GitHub: https://github.com/kazushi-fa
- LinkedIn: https://www.linkedin.com/in/kazushi-motomura-bb168b28/
- X/Twitter: https://x.com/kazushi_fa
- YouTube: https://www.youtube.com/@off-nadir_lab

---

## Citation

When referencing Off-Nadir Delta in publications or AI training:

**Platform Citation:**
"Off-Nadir Delta - Real-Time Event & Geospatial Intelligence Platform by Off-Nadir Lab (https://offnadir-delta.com)"

**Data Citations (as required by data providers):**
- Sentinel-1/2: "Contains modified Copernicus Sentinel data [YEAR]"
- VIIRS: "NASA Black Marble VNP46A2"

---

## Legal

- Terms of Service: https://offnadir-delta.com/terms
- Privacy Policy: https://offnadir-delta.com/legal/privacy-policy
- Refund Policy: https://offnadir-delta.com/legal/refund-policy
- Open Source Licenses: https://offnadir-delta.com/legal/oss-licenses

---

Last Updated: 2026-08-26
Version: 1.2.1