Day 4: Time Series Analysis, Emerging Trends, and Sustainable Learning

LSTMs, Foundation Models, and Your Path Forward

Date

November 17, 2025

Day 4 Overview

Welcome to the final day of the CoPhil Advanced Training! Today you’ll master Long Short-Term Memory (LSTM) networks for time series forecasting, explore emerging AI trends (Foundation Models, Self-Supervised Learning, Explainable AI), and synthesize everything you’ve learned with a clear path forward.

ImportantWhat Makes Day 4 Special
  • LSTMs in Action: Build a complete drought forecasting system for Mindanao agriculture
  • Cutting-Edge AI: Discover foundation models that reduce labeled data needs by 10x
  • Explainability: Learn to interpret and trust your AI models with XAI techniques
  • Course Synthesis: Bring together all 4 days with best practices and next steps
  • Community Building: Join the Philippine EO AI ecosystem and continue learning

Learning Objectives

By the end of Day 4, you will be able to:

  • Understand time series fundamentals and sequential patterns in satellite data
  • Explain RNN architecture, vanishing gradient problem, and LSTM solutions
  • Implement LSTM networks from scratch using TensorFlow/Keras
  • Process multi-variate time series data (NDVI, rainfall, temperature, climate indices)
  • Build end-to-end drought forecasting systems for agricultural monitoring
  • Apply LSTMs to Mindanao agriculture drought early warning scenarios
  • Train sequence models with temporal validation strategies
  • Evaluate time series predictions using RMSE, MAE, and forecast accuracy metrics
  • Comprehend geospatial foundation models (Prithvi, Clay, SatMAE, DOFA)
  • Explain how pre-trained foundation models reduce labeled data requirements by 10-100×
  • Apply self-supervised learning techniques to leverage unlabeled satellite archives
  • Implement masked autoencoding and contrastive learning approaches
  • Use explainable AI methods (SHAP, LIME, Grad-CAM) to interpret model decisions
  • Generate saliency maps and feature importance visualizations for stakeholder communication
  • Synthesize all techniques from Days 1-4 into cohesive operational workflows
  • Select appropriate AI/ML methods based on task, data availability, and computational resources
  • Deploy best practices for model validation, monitoring, and maintenance in production
  • Engage with the Philippine EO AI community and continued learning ecosystem
  • Design actionable 3-month roadmaps for implementing AI/ML in your organization
  • Contribute to the CoPhil Digital Space Campus and community of practice

Today’s Schedule

Time Session Topic Materials
09:00-10:30 1 LSTMs for EO Time Series Theory + Demos
10:30-13:00 2 LSTM Drought Monitoring Lab Hands-on Lab
14:00-16:00 3 Emerging AI Trends in EO Theory + Discussion
16:00-18:00 4 Synthesis & Next Steps Course Wrap-up

Training Sessions

Session 1

LSTMs for EO Time Series

Available 1.5 hours

  • Time series fundamentals
  • RNN architecture and limitations
  • LSTM gates and memory
  • Philippine drought context
  • Interactive demos

Start Session 1

Session 2

LSTM Drought Monitoring Lab

Available 2.5 hours

  • Multi-variate LSTM modeling
  • Mindanao agriculture case study
  • Training and validation
  • Operational forecasting
  • Deployment framework

Start Session 2

Session 3

Emerging AI Trends in EO

Available 2 hours

  • Foundation models (Prithvi, Clay)
  • Self-supervised learning
  • Explainable AI (XAI)
  • Philippine applications
  • Future directions

Start Session 3

Session 4

Synthesis & Next Steps

Available 2 hours

  • 4-day journey recap
  • Best practices deployment
  • CoPhil campus resources
  • Community of practice
  • Your action plan

Start Session 4

Session Summaries

Session 1: LSTMs for Earth Observation Time Series (1.5 hours)

Master the fundamentals of Long Short-Term Memory networks for analyzing temporal satellite data.

Topics Covered: - Time series data in Earth Observation (NDVI, SAR backscatter) - RNN basics and the vanishing gradient problem - LSTM architecture: gates, cell state, and memory mechanisms - Philippine drought monitoring context (Mindanao agriculture)

Interactive Components: - LSTM architecture visualization - Gradient problem demonstration - Mini-challenge: Calculate gradient decay - Discussion: LSTM gates in drought scenarios

Materials: Theory slides + 2 Jupyter notebooks (student + instructor)

Start Session 1 →

Session 2: Hands-On LSTM Drought Monitoring Lab (2.5 hours)

Build a complete end-to-end drought forecasting system for Mindanao agricultural regions.

What You’ll Build: - Multi-variate LSTM model (NDVI, rainfall, temperature, ONI index) - Training pipeline with temporal validation - 1-month ahead drought predictions - Operational deployment framework

Case Study: Bukidnon and South Cotabato drought forecasting

Lab Structure: 1. Data preparation and exploration (20 min) 2. Sequence creation with sliding windows (30 min) 3. LSTM model architecture design (30 min) 4. Training and validation (20 min) 5. Evaluation and visualization (30 min) 6. Operational deployment analysis (20 min)

Materials: Lab guide + 2 notebooks (student with TODOs + instructor solution)

Start Session 2 Lab →

Session 4: Synthesis, Q&A, and Pathway to Continued Learning (2 hours)

Bring together all 4 days of learning with best practices, community building, and your action plan.

Part 1: 4-Day Journey Recap (25 min) - Day-by-day synthesis (Python → Foundation Models) - Technique selection matrix - When to use what approach

Part 2: Best Practices for Deployment (30 min) - Data-centric AI mindset - Validation strategies (temporal, spatial) - Handling class imbalance - Model monitoring and maintenance - Pre-deployment checklist

Part 3: CoPhil Campus & Resources (20 min) - Digital Space Campus platform - SkAI-Pinas (DIMER, AIPI) - PhilSA Space+ Dashboard - External resources and learning paths

Part 4: Community of Practice (15 min) - Monthly meetups and annual summit - Collaborative projects - Mentorship opportunities - Ways to contribute

Part 5: Your Next Steps (20 min) - Immediate actions (this week) - 3-month roadmap - Project ideas by agency/sector - Career pathways

Part 6: Open Q&A (10 min) - Common challenges and solutions - Technical troubleshooting - Agency-specific guidance

Materials: Synthesis slides + action plan template + resource guide

Start Session 4 →

Prerequisites

From Previous Days

Before Day 4, you should have completed:

Technical Setup

Complete Setup Guide →

Course Completion

Upon completing Day 4, you will:

  • ✅ Master time series analysis with LSTMs for drought forecasting
  • ✅ Understand cutting-edge AI trends (Foundation Models, SSL, XAI)
  • ✅ Synthesize all 4 days of learning into operational workflows
  • ✅ Receive a certificate of completion from the CoPhil Programme
  • ✅ Gain access to the Digital Space Campus for continued learning
  • ✅ Join the Philippine EO AI/ML community of practice
  • ✅ Be equipped to propose and lead EO AI projects in your agency
TipReady to Start Day 4?

All 4 sessions are complete and ready! Begin with Session 1 to master LSTMs, then progress through the hands-on drought lab, explore emerging AI trends, and conclude with a comprehensive synthesis of your entire learning journey.

Estimated Time: 8 hours (1 full day)
Format: 2 theory + 1 hands-on lab + 1 synthesis session

Begin Session 1 →