Skip to main content
Back to challenges
Challenge

Agricultural Crop Damage Assessment with Sentinel-2 NDVI Anomalies and TorchGeo

Prolonged droughts and altered monsoon patterns caused by El Niño lead to widespread crop stress and yield loss across major agricultural regions. Agricultural parametric insurers require satellite computer vision pipelines capable of comparing post-event Multispectral Earth Observation imagery against historical Normalized Difference Vegetation Index (NDVI) baselines to automate loss validation. In this challenge, you will construct a satellite imagery computer vision pipeline using TorchGeo and PyTorch. Your model will ingest 12-band Sentinel-2 GeoTIFF rasters, compute spectral NDVI anomaly maps, and perform semantic segmentation of stressed crop fields. Your pipeline will output field-level damage ratios, estimate lost tonnage, and generate verified spatial overlays for claims adjusting.

Machine LearningHosted by Vera
Sign in free to start
Challenge brief

What you are building

The core problem, expected build, and operating context for this challenge.

Build a satellite image segmentation model using TorchGeo to detect agricultural crop stress from Sentinel-2 multispectral NDVI anomalies.

Delivery guide

How work is evaluated

Evaluation

Evaluates semantic segmentation mIoU and field-level damage ratio estimation against ground truth raster masks.

Datasets

Challenge datasets

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

Evaluation rubric

How submissions are scored

These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.

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

raster_resolution_match

Input and output spatial masks maintain identical pixel dimension resolution

Binary check

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

Dimension 2

segmentation_miou

Mean Intersection over Union score on damaged crop mask • target: 0.82 • 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

What you should walk away with

  • Build a PyTorch spatial segmentation pipeline using TorchGeo datasets and samplers.

  • Process 12-band multispectral Sentinel-2 satellite imagery.

  • Compute Normalized Difference Vegetation Index (NDVI) temporal anomaly metrics.

  • Segment stressed vs healthy agricultural field polygons automatically.

Resources and assets

Reference links and supporting material

Dataset notes

Synthetic multispectral GeoTIFF raster patches representing Sentinel-2 imagery of agricultural zones before and after El Niño drought events.

Challenge setup

Evaluates semantic segmentation mIoU and field-level damage ratio estimation against ground truth raster masks.

Challenge input

JSON with field_id, raster_tiff_path, historical_baseline_ndvi

TorchGeo

Specialized PyTorch extension for satellite raster processing.

Evaluated output

JSON with damaged_area_pct, mean_ndvi_drop, crop_stress_category

Evaluation checks
  • Input and output spatial masks maintain identical pixel dimension resolution
  • Mean Intersection over Union score on damaged crop mask • target: 0.82 • range: 0-1
Configured evaluation evidence
  • Benchmark: AgriCropBench
  • Segmentation Miou target: 0.82
  • 1 public reference case
Configured runtime
  • Python execution harness
  • Python sandbox (unavailable on Versalist)

Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.

View technical recipe

Configured tools

Action Space
  • Zed · Required
  • Amazon Bedrock · Optional
  • Claude 3.5 Haiku · Optional

Evaluation contract

  • raster_resolution_match · Weight 1
  • segmentation_miou · Weight 1

Metrics

  • segmentation_miou · Target 0.82

Benchmarks

  • AgriCropBench
  • materials

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON with field_id, raster_tiff_path, historical_baseline_ndvi

    Output: JSON with damaged_area_pct, mean_ndvi_drop, crop_stress_category

    Execution harness: Python

Evaluation tools

  • TorchGeo · Geospatial Vision Framework

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Agricultural Crop Damage Assessment with Sentinel-2 NDVI Anomalies and TorchGeo