Summary of Problem Statement Explainer Session - 1 | Bharatiya Antariksh Hackathon 2025

Summary of the Video: "Problem Statement Explainer Session - 1 | Bharatiya Antariksh Hackathon 2025"

This video is a detailed explainer session covering multiple problem statements for the Bharatiya Antariksh Hackathon 2025. Various mentors present the challenges, objectives, datasets, methodologies, tools, and evaluation criteria for each problem statement, providing participants with clear guidance on how to approach and solve them. The session also includes logistical information about participation, submission, and the hackathon platform.

Main Ideas, Concepts, and Lessons Conveyed

1. Problem Statement 1: Forest Fire Prediction and Spread Simulation

• Objective:
  • Predict forest fire risk zones for the next day using machine learning/deep learning.
  • Simulate the spread of fire over time (1 to 12 hours) from high-risk zones.
• Importance:
  • Forest fires cause biodiversity loss, carbon emissions, and economic damage.
  • Current detection systems are reactive, not predictive.
• Methodology:
  • Use geospatial datasets (burnt area, temperature, precipitation, wind, humidity, elevation, land use, etc.).
  • Generate probability maps classifying zones into high, moderate, low, or no risk.
  • Simulate fire spread using Cellular Automata (CA) or machine learning models.
• Tools & Data:
  • Satellite data (thermal anomalies from 2017 onwards).
  • Python libraries: scikit-learn, TensorFlow, QGIS, matplotlib, R.
  • Google Earth Engine for data access.

2. Problem Statement 2: AI-Based Helpbot for Information Retrieval from MOSC Website

• Objective:
  • Develop an AI chatbot that extracts and understands static/dynamic information from the MOSC web portal (satellite data center).
  • The chatbot should answer user queries conversationally, mining data across documents, PDFs, webpages, images, and metadata.
• Key Features:
  • Context-aware query answering.
  • Modularity to extend to other problems.
• Methodology:
  • Scrape structured and unstructured data from the MOSC portal.
  • Use NLP and machine learning techniques (Retrieval Augmented Generation, LLMs).
  • Train chatbot on data to answer queries about satellite products, data formats, extents, etc.
• Tools & Technologies:
  • NLP libraries: spaCy, NLTK, Dialogflow, Rasa, Node.js.
  • LLMs and retrieval-augmented generation frameworks.
• Evaluation Metrics:
  • Intent recognition accuracy.
  • Entity recognition accuracy.
  • Response completeness and consistency.

3. Problem Statement 3: Monitoring Air Pollution from Space

• Objective:
  • Estimate particulate matter (PM2.5, PM10) concentrations over India using satellite radiance data integrated with ground measurements and reanalysis data.
• Challenges:
  • Satellite measures columnar aerosol effects; surface concentration estimation requires vertical distribution and boundary layer info.
• Methodology:
  • Use visible radiance data from INSAT 3D satellites.
  • Integrate surface PM measurements from CPCB and atmospheric boundary layer data from reanalysis datasets (NARR2, MOS).
  • Train AI/ML models (e.g., random forest) to estimate surface PM concentrations.
• Data & Tools:
  • Satellite radiance data, ground-based PM data, reanalysis atmospheric data.
  • Cloud masking for clear sky pixel selection.
  • AI/ML modeling frameworks.
• Expected Outcome:
  • Spatial maps of PM2.5 and PM10 concentrations over India at ~1° x 1° resolution.

4. Problem Statement 4: Chain-of-Thought Based LLM System for Complex Spatial Analysis

• Objective:
  • Develop an LLM-based reasoning framework to automate complex geospatial workflows (e.g., flood vulnerability mapping) from natural language queries.
• Key Features:
  • Chain-of-thought reasoning linking multi-step geoprocessing tasks.
  • Integration of heterogeneous data sources and APIs.
  • User interface with chatbot interaction and map-based visualization.
• Methodology:
  • Build agents for data gathering, reasoning, geoprocessing, and visualization.
  • Use retrieval-augmented generation (RAG) to fetch relevant data and documentation.
  • Example: generating flood hazard maps by collecting and analyzing multiple datasets.
• Tools & Technologies:
  • Open-source geospatial tools: QGIS, GDAL APIs, Google Earth Engine.
  • Open LLM models (e.g., LLaMA, Mistral).
  • Visualization and chatbot frameworks.
• Evaluation:
  • Robustness tested by applying models

Category

Educational

Video