Copernicus Sentinel Data & Philippine EO Ecosystem

CoPhil EO AI/ML Training - Day 1, Session 1

Stylianos Kotsopoulos

EU-Philippines CoPhil Programme

Welcome

Course Introduction

4-Day Advanced Training

• AI/ML for Earth Observation
• Philippine EO Professionals
• Focus: DRR, CCA, NRM
• Online format

Today’s Goals

• Understand Copernicus data
• Explore Philippine EO ecosystem
• Learn AI/ML fundamentals
• Hands-on Python and GEE

Welcome participants to the 4-day advanced training. Emphasize that this is part of the EU-Philippines partnership and will provide practical skills for disaster risk reduction, climate change adaptation, and natural resource management.

EU Global Gateway Initiative

• EU-Philippines space cooperation flagship
• Building strong partnerships
• Smart, clean, secure digital links
• Strengthening health, education, research systems globally

The Global Gateway strategy represents the EU’s commitment to building partnerships that boost smart, clean and secure infrastructure globally. CoPhil is a unique flagship initiative within this framework.

CoPhil Programme Overview

Mission

Support Philippine Space Agency (PhilSA) and DOST to improve use of Earth Observation data for:

• Disaster Risk Reduction (DRR)
• Climate Change Adaptation (CCA)
• Natural Resource Management (NRM)

Key Outputs

• Copernicus Mirror Site
• Digital Space Campus
• Capacity building
• Pilot services

CoPhil is an EU-funded Technical Assistance programme positioning the Philippines as a pioneer in the EU’s international cooperation on Copernicus.

PhilSA & DOST Partnership

Philippine Space Agency

• Established 2019
• Central civilian space agency
• Space+ Data Dashboard
• Co-chair of CoPhil

Department of Science and Technology

• ASTI AI initiatives
• SkAI-Pinas program
• National AI investments
• Co-chair of CoPhil

Both PhilSA and DOST are co-chairs of the CoPhil programme, demonstrating strong national commitment to building EO and AI capacity.

Session 1 Roadmap

1. Copernicus Programme Overview
2. Sentinel-1 Mission (SAR)
3. Sentinel-2 Mission (Optical)
4. Data Access Methods
5. Philippine EO Ecosystem
6. CoPhil Infrastructure

Duration: 2 hours

This session provides the foundation for understanding where EO data comes from and how Philippine agencies complement Copernicus data.

Part 1: The Copernicus Programme

What is Copernicus?

Europe’s Eyes on Earth

• EU flagship Earth Observation program
• Family of Sentinel satellites
• Free and open data policy
• Operational since 2014

“Looking at our planet and its environment for the benefit of all European citizens”

Copernicus is the European Union’s Earth Observation program, providing free and open satellite data globally. It comprises a fleet of Sentinel satellites designed for environmental monitoring.

Copernicus Programme Architecture

Copernicus Programme Structure showing Space Component, Services, and End Users

This diagram shows the complete Copernicus architecture: the space component with Sentinel missions, in-situ data sources, the six core services (CAMS, CMEMS, Land, Climate, Emergency, Security), and the various end-user categories.

The Sentinel Family

Sentinel-1 (SAR)

• C-band radar imaging
• All-weather, day/night
• 1A operational, 1C launched Dec 2024
• 6-day repeat (dual constellation)

Sentinel-2 (Optical)

• Multispectral imaging
• 13 spectral bands
• 2A, 2B, 2C operational Jan 2025
• 5-day repeat (three satellites)

Sentinel-3 (Ocean/Land)

• Ocean and land monitoring
• Sea surface temperature
• Ocean color, vegetation

Sentinel-5P (Atmosphere)

• Air quality monitoring
• Atmospheric composition

Timing: 4 minutes

Key Points: - Six Sentinel missions operational (1, 2, 3, 5P, 6) with 4, 7-12 planned - 2025 Update: Sentinel-1C launched December 2024, restoring 6-day global repeat - 2025 Update: Sentinel-2C operational January 2025, creating 3-satellite constellation with 5-day repeat - Today we focus on Sentinel-1 and Sentinel-2 - most relevant for Philippine DRR/CCA/NRM - Sentinel-3 important for marine/coastal applications - Sentinel-5P monitors air pollution (useful for Manila air quality)

Transition: “These missions generate petabytes of free data. Let’s see how this data is used…”

Copernicus Applications

Emergency Management

• Flood mapping
• Fire detection
• Disaster response

Climate & Environment

• Deforestation monitoring
• Agricultural monitoring
• Water quality assessment

Copernicus supports applications in emergency management, health care, agriculture, and environment, with crucial impact on society, climate, and economy.

Sentinel-1: SAR Mission

Sentinel-1 Overview

Mission Configuration (2025)

• Sensor: C-band Synthetic Aperture Radar
• Satellites: 1A, 1C (operational 2025)
• Orbit: Polar sun-synchronous
• All-weather, day/night capability

Key Advantage: Penetrates clouds and works at night

Sentinel-1 is a radar imaging mission with C-band SAR enabling all-weather, day-and-night observations unaffected by clouds - crucial in tropical regions like the Philippines.

Sentinel-1 Technical Specifications

Parameter	Value
Sensor Type	C-band SAR (5.405 GHz)
Revisit Time	6-12 days (constellation)
Swath Width	250 km (IW mode)
Spatial Resolution	5m × 20m (IW mode)
Polarization	VV + VH or HH + HV
Orbit	693 km altitude

Sentinel-1 operates in Interferometric Wide (IW) mode with ~5m by 20m spatial resolution. With two satellites, the revisit cycle is 6 days globally. Note: Sentinel-1B became inoperative in 2022; Sentinel-1C launched in 2024 to restore 6-day revisit.

SAR: How It Works

1. Satellite sends microwave pulses to Earth
2. Signal reflects from surface
3. Sensor measures backscatter intensity
4. Different surfaces = different backscatter

SAR satellites send microwave signals and measure the backscatter. Water appears dark (low backscatter), urban areas appear bright (strong backscatter), and vegetation shows moderate backscatter.

Sentinel-1 Polarization

What is Polarization?

• Orientation of radar wave
• VV: Vertical send/receive
• VH: Vertical send/Horizontal receive
• HH: Horizontal send/receive

Applications

• VV: Good for water/flood mapping
• VH: Sensitive to volume scattering (vegetation)
• VV/VH Ratio: Discriminates surface types

Sentinel-1 IW mode over land typically provides VV and VH polarizations. Different polarizations are sensitive to different surface characteristics.

Backscatter Characteristics by Target

Surface Type	Backscatter	Appearance	Reason
Water (smooth)	Very Low	Dark/Black	Specular reflection
Urban/Buildings	Very High	Bright White	Corner reflectors
Forest/Vegetation	Medium-High	Gray	Volume scattering
Agricultural Fields	Medium	Light Gray	Surface roughness
Bare Soil (dry)	Low-Medium	Dark Gray	Smooth surface
Bare Soil (wet)	Medium	Medium Gray	Increased dielectric

Key Insight: Water appears dark, structures appear bright - the basis for flood mapping!

Understanding backscatter is crucial for interpreting SAR imagery. Smooth surfaces (water) reflect signals away = dark. Rough surfaces and structures reflect signals back = bright.

Sentinel-1 Imaging Modes

Interferometric Wide Swath (IW) - 250 km swath - 5m × 20m resolution - Default over land - Philippine standard mode

Extra Wide Swath (EW) - 400 km swath - 20m × 40m resolution - Maritime/polar regions

Strip Map (SM) - 80 km swath - 5m × 5m resolution - Emergency response - High detail needed

Wave (WV) - Ocean waves - Not used for land

Interferometric Wide (IW) mode is used over land and provides the best balance of resolution and coverage. All Philippine acquisitions use IW mode.

Sentinel-1 Data Products

Level-1 GRD (Ground Range Detected)

• Multi-looked (reduced speckle)
• Projected to ground range
• Most commonly used
• Faster to process
• Smaller file size
• Applications:
  • Change detection
  • Classification
  • Flood mapping
  • Ship detection

Level-1 SLC (Single Look Complex)

• Preserves phase information
• Complex-valued pixels
• Required for InSAR
• Larger files
• Applications:
  • Ground deformation
  • Interferometry
  • Coherence analysis
  • Subsidence monitoring

For most applications including flood mapping and land cover, GRD products are sufficient. SLC products are needed for advanced interferometric applications like volcano monitoring.

GRD vs SLC - Which to Choose?

Factor	GRD	SLC
Use Case	Most applications	Interferometry only
Processing	Ready to use	Complex processing
File Size	~1 GB	~4 GB
Speckle	Reduced	Full speckle
Phase	Not preserved	Preserved
Typical User	Most analysts	Advanced specialists

For this training and most Philippine applications: Use GRD

Unless you specifically need interferometry (earthquake/volcano deformation), always use GRD products. They’re easier to work with and sufficient for 90% of applications.

Sentinel-1 Pre-Processing “Under the Hood”

What Happens Before You See SAR Data?

For this training, we use pre-processed Sentinel-1 GRD data. Here’s what happens “under the hood”:

S1 Processing Pipeline:

1. GRD Download → Raw ground-range detected amplitude
2. Radiometric Calibration → Convert to backscatter coefficient (σ⁰)
3. Terrain Correction (RTC) → Remove topographic distortions using DEM
4. Speckle Filtering → Reduce SAR noise (Lee, Refined Lee, or Gamma-MAP filters)
5. Conversion to dB → γ⁰ (gamma-naught) in decibels for visual interpretation
6. Tiling/Clipping → Extract area of interest

For Day 3 flood mapping labs: We provide analysis-ready patches with these steps already applied

P0 IMPROVEMENT APPLIED: Explicit S1 pre-processing assumptions before U-Net flood lab (Day 3).

Key Teaching Points: - Participants need to understand what “analysis-ready” means - RTC removes terrain effects - critical in mountainous Philippines - Gamma-naught (γ⁰) in dB is standard for land applications - Speckle filtering is essential but can blur edges - These steps are computationally intensive - we pre-process for efficiency

Timing: 3 minutes

Transition: “Now let’s look at Sentinel-2 optical data…”

Sentinel-1 Applications

Flood Mapping

Copernicus Sentinel-1 Flood Monitoring

• Water appears dark in SAR
• Works through clouds
• Near real-time monitoring

Deformation Monitoring

Sentinel-1C Interferogram of Northern Chile

• InSAR technique
• Millimeter precision
• Volcano and earthquake monitoring

Sentinel-1’s ability to penetrate clouds makes it invaluable for flood mapping during typhoons. The left image shows actual flood monitoring using Sentinel-1 data. The right image shows an interferogram from Sentinel-1C demonstrating ground deformation detection with millimeter precision in northern Chile.

Philippine Example: Flood Monitoring

November 2020: Typhoon Ulysses

• Extensive flooding in Luzon
• Sentinel-1 detected flood extent
• Rapid mapping capability

Key Benefits

• No cloud interference
• Quick response time
• Used for rapid damage assessment
• Supported emergency response

During Typhoon Ulysses, Sentinel-1 provided crucial flood extent mapping when optical satellites couldn’t see through clouds. This data supported PAGASA and NDRRMC response efforts.

Sentinel-2: Optical Mission

Sentinel-2 Overview

Key Specifications:

• 13 spectral bands (visible, NIR, SWIR)
• 10m to 60m spatial resolution
• 290 km swath width
• 5-day revisit (three satellites operational 2025)
• L1C & L2A processing levels

Philippine Example: Mayon Volcano

Sentinel-2 monitoring of 2018 eruption

• False-color composite highlighting lava flows
• 10m resolution captures detail
• 5-day revisit enables continuous monitoring

This Sentinel-2 false-color image of Mayon Volcano from 2018 shows recent lava flows. With Sentinel-2C operational in 2025, the three-satellite constellation provides 5-day global revisit.

Sentinel-2 Spectral Bands

Band	Name	Wavelength (nm)	Resolution	Purpose
B1	Coastal Aerosol	443	60m	Aerosol correction, water color
B2	Blue	490	10m	Water bodies, atmospheric
B3	Green	560	10m	Vegetation health
B4	Red	665	10m	Vegetation discrimination
B5	Red Edge 1	705	20m	Vegetation stress detection
B6	Red Edge 2	740	20m	Vegetation classification
B7	Red Edge 3	783	20m	Vegetation stress, chlorophyll
B8	NIR	842	10m	Biomass, water bodies
B8A	NIR Narrow	865	20m	Atmospheric correction
B9	Water Vapor	945	60m	Atmospheric correction
B10	SWIR Cirrus	1375	60m	Cirrus cloud detection
B11	SWIR 1	1610	20m	Moisture content, fire
B12	SWIR 2	2190	20m	Moisture, geology, soil

All 13 Sentinel-2 bands: Four 10m bands (B2-B4, B8) for detailed mapping, six 20m bands including the unique Red Edge trio (B5-B7) for vegetation analysis, and three 60m atmospheric bands (B1, B9, B10) for corrections. The Red Edge bands are Sentinel-2’s signature capability.

Red Edge Bands - Sentinel-2’s Special Capability

What is Red Edge?

• Transition zone between red and NIR (700-780nm)
• Three dedicated bands (B5, B6, B7)
• Sensitive to chlorophyll content
• Unique to Sentinel-2 among free satellites

Applications:

• Early vegetation stress detection
• Crop health monitoring
• Forest disease identification
• Pre-harvest yield estimation

Philippine Use Cases:

• Rice crop assessment - Monitor crop health and nitrogen status during panicle initiation in Nueva Ecija rice paddies
• Coconut disease detection - Identify stem bleeding disease and pest infestations through spectral signatures
• Mangrove health monitoring - Track vegetation stress and recovery in Palawan mangrove forests using red edge indices

Red edge bands detect stress weeks before visible bands show changes

Red Edge bands are Sentinel-2’s superpower. They detect vegetation stress weeks before visible bands show changes - crucial for precision agriculture and forest health monitoring.

Sentinel-2 Data Products

Level-1C (L1C)

• Top-of-Atmosphere reflectance
• Radiometrically corrected
• Geometrically refined
• No atmospheric correction
• Use: If you need raw data for custom processing

Level-2A (L2A)

• Bottom-of-Atmosphere (surface) reflectance
• Atmospherically corrected
• Analysis-ready
• Scene Classification Layer included
• Use: For most applications - RECOMMENDED

Always use Level-2A when available - it’s analysis-ready!

L2A data is atmospherically corrected and ready for analysis. Unless you have a specific reason to use L1C, always choose L2A products.

Sentinel-2 Band Combinations

True Color (Natural) - RGB: B4-B3-B2 (Red-Green-Blue) - Looks like a photograph - Good for visual inspection

False Color Infrared - RGB: B8-B4-B3 (NIR-Red-Green) - Vegetation appears RED - Classic for vegetation assessment

SWIR Composite (Agriculture) - RGB: B11-B8-B2 - Highlights crop moisture - Soil moisture visible

SWIR-NIR-Red (Burn/Fire) - RGB: B12-B8-B4 - Active fires appear BRIGHT - Burn scars dark purple

Different band combinations highlight different features. False Color (NIR-Red-Green) is the most common for vegetation mapping - healthy vegetation appears bright red.

Sentinel-1 vs Sentinel-2

Complementary Capabilities

Aspect	Sentinel-1	Sentinel-2
Sensor Type	Radar (SAR)	Optical (MSI)
Weather	All-weather	Cloud-affected
Time	Day & night	Daytime only
Resolution	5m × 20m	10m / 20m / 60m
Revisit	6-12 days	5 days
Bands	Polarizations (2)	Spectral bands (13)
Best For	Water, structure, moisture	Vegetation, land cover, color

S1 Flood Mapping Tip

VV polarization shows dark water (low backscatter from smooth surfaces)

Best practice: Compare pre-event vs post-event delta for reliability

Synergy: Pair S1 (flood extent through clouds) with S2 (vegetation damage when clear) for complete impact assessment

Sentinel-1 and Sentinel-2 are complementary. SAR penetrates clouds and works at night, while optical provides rich spectral information about surface properties.

MICRO-EDIT APPLIED: Added S1 vs S2 flood tip to set up Day 3 logic - VV dark water, pre/post comparison, and S1+S2 synergy.

Synergistic Use Cases

Flood Mapping

• S1: Detect water extent (through clouds)
• S2: Assess damage to vegetation/crops (when clear)
• Combined: Complete flood impact assessment

Forest Monitoring

• S1: Detect structural changes, biomass
• S2: Identify tree species, health
• Combined: Comprehensive forest mapping

Using both missions together provides more complete information. After a typhoon, S1 maps floods under clouds while S2 assesses vegetation damage where it’s clear.

☕ 5-Minute Break

Stretch Break

Stand up • Grab water • Back in 5 minutes

Timing: 5 minutes

Instructor Actions: - Announce break clearly - Mention return time - Leave slides/browser open for questions - Be available for quick questions during break

When Resuming: - Wait 1-2 minutes for everyone to return - Quick recap: “We’ve covered Sentinel-1 SAR and Sentinel-2 optical. Now we’ll explore the Philippine EO ecosystem.”

Data Access Methods

Platform Choices for This Training

Which Platform for Which Task?

Understanding where to work is crucial - don’t try to train deep models on GEE or download 100GB in Colab!

Task	Platform	Why	Limitations
Data Prep & Exploration	Google Earth Engine	Petabyte catalog, no download, cloud composites	Export limit 32 MB (tile large areas), no deep learning training
ML Training (RF, shallow)	GEE or Colab	RF works in GEE; small data in Colab	GEE memory limits; Colab free tier quotas
Deep Learning (CNN, U-Net)	Local GPU / Colab Pro	Requires PyTorch/TensorFlow	Colab free = limited GPU time; large models need local resources
Large-Scale Processing	CoPhil Mirror Site / COARE	400TB local data, HPC resources	Requires account; learning curve for APIs
Quick Viz & Download	Copernicus Browser	Interactive, fast previews	Manual selection; bulk downloads tedious

Quotas/Pitfalls to know: - GEE: Memory errors with large computations (tile exports!) - Colab Free: GPU disconnects after inactivity; limited sessions/day - CoPhil/Digital Space Campus: Hosts this training’s materials + local data access

P1 IMPROVEMENT APPLIED: Explicit platform choices table prevents common mistakes.

Key Teaching Points: - GEE is NOT for training U-Nets or downloading GBs - Colab free tier has GPU quotas - manage expectations - CoPhil infrastructure enables local processing - Choose the right tool for the right job

Timing: 3 minutes

Transition: “Let’s look at these platforms in detail…”

Copernicus Data Space Ecosystem (2025)

New Platform (2023+)

• Replaced SciHub
• Modern interface
• API access
• Free registration

Features

• Search by location/date
• Preview before download
• Direct download
• Bulk processing

URL: https://dataspace.copernicus.eu

The Copernicus Data Space Ecosystem is the new portal replacing the legacy SciHub. All Sentinel data are free and open.

SentiBoard Dashboard (October 2025)

• Real-time mission status
• Data availability insights
• Acquisition plans
• Quality metrics
• Interactive dashboard

SentiBoard is a new feature (October 2025) providing real-time insights into Copernicus Sentinel missions and data availability.

Google Earth Engine

Planetary-Scale Platform

• Petabyte-scale data catalog
• All Sentinel-1 & Sentinel-2 data
• Cloud-based processing
• Free for research/education
• No download needed!

We’ll use GEE extensively in this training

URL: https://earthengine.google.com

Google Earth Engine provides ready access to all Sentinel data with cloud-based processing. We’ll use GEE extensively for data access and preprocessing.

Alternative Data Sources

Alaska Satellite Facility (ASF)

• Sentinel-1 specialist
• User-friendly interface
• Preprocessing tools
• https://asf.alaska.edu

AWS Open Data

• Sentinel-2 on AWS
• Cloud-optimized
• Pay for compute only
• Programmatic access

Now Available: CoPhil Mirror Site in the Philippines! (Operational 2024-2025)

MICRO-EDIT APPLIED: Mirror Site is NOW operational (not “coming soon”).

Multiple platforms provide Sentinel data access. The CoPhil Mirror Site provides local data access in the Philippines for faster downloads - critical for reducing international bandwidth bottlenecks.

Key Point: 400TB capacity, focused on PH scenes, free for government/academia/researchers

Part 2: Philippine EO Ecosystem

Overview of Philippine EO Landscape

• PhilSA: Space data and operations
• NAMRIA: National mapping and geospatial data
• DOST-ASTI: AI and remote sensing R&D
• PAGASA: Climate and weather data

Beyond European satellites, the Philippines has its own geospatial data platforms and agencies that provide crucial complementary data and support.

Philippine Space Agency (PhilSA)

PhilSA: National Space Agency

Established: August 2019

Mandate:

• Central civilian space agency
• Promote space data use
• Build national capacity
• Support DRR, CCA, NRM
• Co-chair CoPhil programme

Website: https://philsa.gov.ph

PhilSA is the central civilian space agency established in 2019. As co-chair of CoPhil, PhilSA is leading national efforts to leverage Copernicus data.

SIYASAT Data Portal

Purpose (2025)

• Secure data archive operational
• Visualization system
• Data distribution
• Maritime & terrestrial monitoring

Data Types

• NovaSAR-1 radar imagery
• AIS ship tracking data
• Processed products
• Analysis-ready data

Timing: 3 minutes

Key Points: - SIYASAT (Secure Interactive Yield-Assessment & SAR Analytics Tools) is PhilSA’s operational data portal - 2025 Status: Fully operational with NovaSAR-1 data access - Provides secure archive for Philippine government agencies - Critical for maritime security (illegal fishing, territorial monitoring) - Registration required - priority for government agencies

Transition: “Beyond PhilSA, DOST-ASTI is leading AI innovation for EO applications…”

Space+ Data Dashboard

• User-friendly web portal
• Browse satellite imagery
• Visualization tools
• Download datasets
• No programming required
• Open to government, researchers, public

The Space+ Data Dashboard democratizes access to satellite data, making it available to local governments, researchers, and even students without requiring programming skills.

PhilSA 2025 Initiatives

Space Business Innovation Challenge

• Empowers Filipino innovators
• Free satellite data access
• Build solutions for local needs
• Earth observation focus
• Weather & environmental data

Training Programs

• Downstream data utilization
• Practical applications
• Capacity building nationwide
• Partnership with DOST

PhilSA’s 2025 initiatives focus on empowering innovators and building capacity across the Philippines to use satellite data effectively.

COARE Infrastructure

Computing and Archiving Research Environment

• High-performance computing
• Data archiving capabilities
• Science cloud facilities
• Supports data-intensive research
• Enables AI/ML workflows

COARE provides the computational infrastructure needed for processing large satellite datasets and running AI/ML models at scale.

NAMRIA

NAMRIA: National Mapping Authority

National Mapping and Resource Information Authority

Role:

• Official mapping agency
• Authoritative geospatial data
• Topographic maps
• Hazard maps
• Land cover datasets

Website: https://www.geoportal.gov.ph

NAMRIA is the national mapping agency responsible for topographic maps, hydrographic data, and authoritative geospatial datasets.

NAMRIA Geoportal

One-Stop Shop for Philippine Geospatial Data

• National basemaps (1:50,000 scale)
• Administrative boundaries
• Topographic maps
• Thematic maps
• Downloadable shapefiles and rasters

The NAMRIA Geoportal is the go-to source for official Philippine spatial data including boundaries, topographic maps, and land cover data.

Land Cover Mapping Project

Latest: 2020 National Land Cover

Classes:

• Forest types
• Agriculture
• Built-up areas
• Water bodies
• Wetlands
• Barren land

Data Formats:

• Shapefile
• GeoTIFF
• CSV
• GeoJSON
• KML
• PNG

Portal: https://land-cover-mapping-project-namria.hub.arcgis.com

NAMRIA’s land cover datasets are valuable for validation and comparison with satellite-derived classifications. The 2020 national land cover map is available in multiple formats.

HazardHunterPH Portal

Comprehensive Hazard Assessment Platform

Hazard Types:

• Earthquake-induced hazards
• Active fault lines
• Tsunami susceptibility
• Liquefaction zones
• Landslide hazards

Applications:

• Disaster risk assessment
• Land use planning
• Infrastructure siting
• Emergency preparedness

URL: https://hazardhunter.georisk.gov.ph/map

HazardHunterPH provides critical hazard data for DRR planning. These maps can be overlaid with satellite data for vulnerability assessments.

How NAMRIA Complements Sentinel Data

Sentinel Imagery Provides:

• Current conditions
• Frequent updates
• Large-area coverage
• Multi-temporal analysis

NAMRIA Data Provides:

• Ground truth for validation
• Training labels for ML models
• Historical baselines
• Official classifications
• Hazard context

MICRO-EDIT: Using NAMRIA/Space+ as Label Sources

For Day 2 Palawan RF lab: 1. Download NAMRIA land cover shapefile (authoritative classes) 2. Overlay on Sentinel-2 imagery 3. Extract training points per class (forest, agriculture, water, etc.) 4. Train Random Forest classifier 5. Validate predictions against NAMRIA hold-out samples

Space+ Dashboard also provides admin boundaries and infrastructure layers for context in all visualizations.

Example: Use Sentinel-2 to map current land cover, validate against NAMRIA’s official 2020 map, detect changes

MICRO-EDIT APPLIED: Explicit link between NAMRIA/Space+ as label sources & validation layers, not just context.

NAMRIA data provides essential ground truth and context for satellite-based analysis. Official boundaries and hazard maps enhance satellite data interpretation.

Show one example overlay: NAMRIA training polygons (land cover classes) overlaid on S2 RGB composite.

DOST-ASTI AI Initiatives

DOST-ASTI Overview

Advanced Science and Technology Institute

• Lead agency for EO and AI R&D
• Remote sensing expertise
• Machine learning development
• National AI infrastructure
• P2.6 billion investment until 2028

Website: https://asti.dost.gov.ph

DOST-ASTI leads several projects at the intersection of remote sensing, AI, and big data, with significant national investment of P2.6 billion through 2028.

DATOS: Remote Sensing Help Desk

• Remote Sensing and Data Science Help Desk
• Rapid analytics during disasters
• Flood mapping from satellite imagery
• Damage assessment
• Crop mapping (rice, sugarcane)
• Road network detection
• Supports emergency response agencies

During past typhoons, DATOS used satellite images to produce flood maps and give them to disaster response agencies within hours. They also worked on crop mapping and infrastructure detection using AI.

SkAI-Pinas Programme

Philippine Sky AI Program (2021-2028)

• Flagship AI R&D programme
• Part of P2.6B DOST AI investment
• Democratize AI across Philippines
• Remote sensing & big data focus

Impact (2025)

• 300+ institutions supported
• Universities & colleges
• SMEs & research teams
• Local government units

Timing: 3 minutes

Key Points: - 2025 Update: SkAI-Pinas is part of DOST’s P2.6 billion AI investment (2024-2028) - Focuses on democratizing AI for EO applications - Provides infrastructure, training, and support - Successfully engaged 300+ institutions nationwide - Critical for building national AI capacity

Examples: - Universities using SkAI for flood mapping research - LGUs leveraging AI for disaster preparedness - SMEs building EO-based agricultural solutions

Transition: “SkAI provides the platform. DIMER provides the models. PANDA brings it all together…”

DIMER: AI Model Hub

Democratized Intelligent Model Exchange Repository

• Digital “model store”
• Pre-trained AI models
• Ready-to-use
• Filipino-specific challenges

Available Models

• Landslide detection
• Traffic surveys
• Crop monitoring
• Land cover classification
• Flood detection

DIMER lowers barriers to AI by enabling end-users to reuse optimized AI models. If you need a model for land cover or flood detection, check DIMER first before building from scratch.

AIPI: AI Processing Interface

Purpose

• Streamline large-scale remote sensing tasks
• Reduce computational barriers
• Run AI models on ASTI servers
• Process hundreds of images efficiently

Example: Apply an AI model to 100 Sentinel-2 images over entire region without your laptop

AIPI allows users to run large computations on ASTI’s servers without heavy local processing. This is especially valuable when processing many satellite images with AI models.

ALaM: Automated Labeling Machine

Challenge

Creating labeled training data is:

• Time-consuming
• Expensive
• Requires expertise
• Major bottleneck for AI

ALaM Solution

• Automate labeling process
• Crowdsourcing capabilities
• Expert validation
• Build training datasets

Result: Faster creation of high-quality training data for Filipino contexts

Creating labeled data is a big challenge in EO AI. ALaM tries to address this through automation and crowdsourcing, making it easier to build training datasets for Philippine applications.

How DOST-ASTI Tools Work Together

1. ALaM creates training data
2. Train models and share via DIMER
3. Process large datasets with AIPI
4. Deploy for operational use via SkAI-Pinas
5. Support disaster response through DATOS

These initiatives work together to create a complete ecosystem from data labeling through model development to operational deployment.

PAGASA & Other Data Sources

PAGASA: Weather & Climate Data

Philippine Atmospheric, Geophysical and Astronomical Services Administration

Data Types:

• Historical rainfall
• Temperature records
• Typhoon tracks
• Climate forecasts
• Weather observations

Integration: Combine with satellite data for climate analysis

PAGASA provides meteorological and climate data that can be integrated with satellite observations for comprehensive analysis. Rainfall data + flood maps, for example.

Synergy: Satellite + Ground Data

Satellite Data (Sentinel)

• Spatial coverage
• Consistent acquisition
• Multiple variables
• Time series

Ground Data (Philippine agencies)

• Point validation
• Ground truth
• Meteorological context
• Local expertise

Combined = More robust analysis and higher confidence

The Philippines is developing a rich EO ecosystem. The goal of CoPhil is to ensure you can leverage both international (Sentinel) and national data together effectively.

CoPhil Infrastructure

CoPhil Mirror Site

• Philippines-based Copernicus data repository
• Local mirror of Sentinel data
• Focus on Philippine region
• Faster access (no international bandwidth)
• Reliable availability
• Operational by 2025
• Hosted by PhilSA with CloudFerro support

The CoPhil Mirror Site will locally store Sentinel data focused on the Philippines, enabling much faster downloads than from European servers.

Digital Space Campus

Purpose

• Online learning portal
• Training materials repository
• Self-paced learning
• Community of practice
• Knowledge sharing

Content

• Course presentations
• Jupyter notebooks
• Datasets
• Guides & tutorials
• Forum discussions

All materials from this 4-day training will be available on the Digital Space Campus for future reference and for colleagues who couldn’t attend.

Building a Sustainable EO Ecosystem

1. Data Access: Mirror Site + Data Space Ecosystem
2. Processing: COARE + AIPI + Google Earth Engine
3. Models: DIMER repository
4. Training: Digital Space Campus
5. Operations: DATOS + Agency integration
6. Community: SkAI-Pinas network

Result: Complete infrastructure for operational EO AI/ML**

CoPhil is building sustainable infrastructure ensuring you can continue working with Sentinel data and AI/ML tools after this training, without starting from scratch.

Summary & Integration

Key Takeaways

1. Copernicus provides free, high-quality satellite data globally
2. Sentinel-1 (SAR) works day/night, all-weather - essential for tropics
3. Sentinel-2 (optical) provides rich spectral information - 5 day revisit
4. Multiple access methods available (Data Space, GEE, Mirror Site)
5. Philippine agencies provide complementary data and expertise
6. CoPhil infrastructure supports sustainable capacity building

You now understand the full picture from data sources through national infrastructure. This foundation supports all subsequent AI/ML work.

How It All Connects

Satellite data + Philippine ground truth + AI models + processing infrastructure = Operational applications for DRR, CCA, and NRM

Example Workflow: Flood Mapping

1. Acquire Sentinel-1 SAR data (GEE or Mirror Site)
2. Validate with PAGASA rainfall data
3. Process using AI model from DIMER
4. Scale processing via AIPI
5. Combine with NAMRIA hazard maps
6. Deliver to NDRRMC via DATOS

This is what integrated EO capacity looks like!

A real-world workflow leverages multiple data sources, AI tools, and institutional partnerships. This is exactly what CoPhil aims to enable.

What’s Next in Day 1?

Session 2 (Next)

• AI/ML fundamentals
• Supervised vs unsupervised learning
• Neural networks basics
• Data-centric AI
• → 5-min Concept Check (3 questions)

Sessions 3 & 4

• Hands-on Python (GeoPandas, Rasterio)
  • Pre-run pip installs to save time
• Google Earth Engine tutorial
  • Start from ready script, modify filters only
• Access real Sentinel data
• SCL cloud/shadow masking (not QA60)

Now that you understand where data comes from, we’ll learn how AI/ML can extract information from these data, then get hands-on with actual satellite imagery.

P1 IMPROVEMENTS NOTED: - Concept check after Session 2 - Pre-run installations in Session 3 - Ready GEE script in Session 4 (modify filters live, don’t type from scratch)

Session Summary

What We Covered:

✅ Copernicus Programme & free/open data policy
✅ Sentinel-1 (SAR) - all-weather monitoring
✅ Sentinel-2 (Optical) - 13 bands, 10m resolution
✅ 2025 Updates: 1C & 2C operational
✅ Philippine EO infrastructure (PhilSA, NAMRIA, DOST-ASTI)
✅ Data access platforms (SIYASAT, Geoportal, SkAI-Pinas)

Timing: 2 minutes

Key Takeaways: - You now understand WHERE EO data comes from - You know WHAT satellites provide what data - You know HOW to access data (both EU and PH platforms) - Ready for Session 2: Learning WHAT to do with this data using AI/ML

Q&A

Copernicus & Sentinel

• Mission specifications?
• Data products & formats?
• Access methods?
• Processing levels?

Philippine EO Ecosystem

• Agency roles & mandates?
• Data access procedures?
• Integration with Copernicus?
• P2.6B AI investment details?

Timing: 3-5 minutes for Q&A

Facilitation Tips: - Encourage questions from different agency participants - Share knowledge across government, academic, private sectors - If technical question about access, note for Session 4 (GEE hands-on) - Keep responses brief - detailed hands-on coming in Sessions 3-4

Common Questions & Answers: - “How do I sign up for GEE?” → Will cover in Session 4 - “Which satellite should I use?” → Depends on application (Day 2 topic) - “Is SIYASAT public?” → Registration required, priority for gov agencies - “Can I combine Sentinel-1 and -2?” → Yes! Synergistic use cases discussed today

Next: Session 2

Core Concepts of AI/ML for Earth Observation

Coming up after break:

• What is AI/ML and why for EO?
• The EO AI/ML workflow
• Supervised vs unsupervised learning
• Introduction to deep learning
• Data-centric AI approaches

See you in Session 2! 🚀

Thank You!

Contact & Resources

European Platforms:
Copernicus Data Space: https://dataspace.copernicus.eu
CoPhil Programme: https://www.cophil.eu

Philippine Platforms:
PhilSA: https://philsa.gov.ph
NAMRIA Geoportal: https://www.geoportal.gov.ph
DOST-ASTI: https://asti.dost.gov.ph

Session 1 Complete!

Total timing: ~120 minutes including breaks and Q&A

Before Session 2: - Short bio break (5-10 minutes) - Check that all participants are ready - Ensure presentation materials for Session 2 are loaded

15-minute break before Session 2