Skip to main content
Loading navigation
Back to challenges

Build a Multimodal Flood and Satellite Exposure Analytics Agent with Google ADK

Satellite imagery and multimodal sensor signals are crucial for rapid flood loss verification following extreme storms. Build a multimodal flood assessment agent using Google Agent Development Kit (ADK) and Gemini 1.5 Pro. Achieve 90%+ alignment with expert structural damage classifications on satellite and aerial drone imagery.

Machine LearningHosted by Vera
Sign in free to start

What you are building

Develop a multimodal agent using Google ADK and Gemini models to process post-disaster satellite imagery alongside property risk records.

How work is evaluated

Evaluation

Evaluates multimodal damage classification precision from paired satellite images and property metadata.

Datasets

Public datasets, plus the private datasets you have access to.

How submissions are scored

Dimensions
2 scoring checks
Binary
2 pass or fail dimensions
Ordinal
0 scaled dimensions
Dimension 1

valid_json_schema

Output complies with required multimodal assessment structure.

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Dimension 2

damage_grade_accuracy

Accuracy of structural damage severity classification (NONE, MODERATE, SEVERE) • target: 0.9 • range: 0-1

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Learning goals

  • Implement multimodal agents using Google Agent Development Kit (ADK).

  • Leverage Gemini 1.5 Pro multimodal vision processing capabilities.

  • Combine satellite visual inputs with tabular elevation and hydrological metrics.

  • Validate inundation depth predictions against ground truth survey markers.

Resources

Dataset notes

Dataset of 200 post-flood satellite image pairs with associated elevation and parcel metadata.

Challenge setup

Evaluates multimodal damage classification precision from paired satellite images and property metadata.

Challenge input

Multimodal payload containing image_url and parcel_metadata

Google ADK

Google standard SDK for Gemini multi-modal agent workflows.

Evaluated output

JSON with flood_inundated (bool), estimated_depth_ft, and severity_grade

Evaluation checks
  • Output complies with required multimodal assessment structure.
  • Accuracy of structural damage severity classification (NONE, MODERATE, SEVERE) • target: 0.9 • range: 0-1
Configured evaluation evidence
  • Benchmark: MultimodalGeoBench
  • Damage Grade Accuracy target: 0.9
  • 1 public reference case
Configured runtime
  • Python execution harness
  • Python sandbox (unavailable on Versalist)

Shows the first task, up to three tools, and up to three evidence items.

View technical recipe

Configured tools

Environment
  • Modal · Required
Action Space
  • Google · Optional
  • Google ADK · Optional
Policy Serving
  • Modal · Required

Evaluation contract

  • valid_json_schema · Weight 1
  • damage_grade_accuracy · Weight 1

Metrics

  • damage_grade_accuracy · Target 0.9

Benchmarks

  • MultimodalGeoBench
  • general

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: Multimodal payload containing image_url and parcel_metadata

    Output: JSON with flood_inundated (bool), estimated_depth_ft, and severity_grade

    Execution harness: Python

Evaluation tools

  • Google ADK · Agent Framework

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Build a Multimodal Flood and Satellite Exposure Analytics Agent with Google ADK