← Back to Overview
Interactive Visualizations & Products

Product 1

Spectral similarity cross-matching of 62,009 candidate points with 701 field reference polygons (~1 ha; 608 classified + 93 unlabeled) via 64-dimensional L2-renormalized cosine similarity. Access static deliverables, open scripts, data tables, and explore pasture condition categories.

Documentation, Code & Open Data

Static Deliverables & Repository

Integrated access to technical reports, presentation decks, Google Earth Engine & Python routines, and full processed datasets available for direct download on GitHub.

📄
Technical Document

Technical Report (Product 1)

Official progress report "Product 1 — Detailed Sampling Plan and Sampling Improvement Strategies" (August 17, 2026), detailing sample expansion, L2-renormalized cosine similarity, SOM Bayesian filtering, CBERS-4A visual inspection, and field campaign strategy.

⬇ Download Report (PDF) 569 KB
📊
Presentation Slides

Presentation Deck

Executive and technical slide deck: "Scaling Ground Truth" — summarizing findings, temporal evolution of typologies (2019–2025), comparative accuracy charts, and field validation statistics.

⬇ Download Slides (PDF) 11 MB
🛰️
Cloud Assets

Google Earth Engine Assets

Cloud-processed raster image collections and feature datasets available in LAPIG/WRI Earth Engine repositories:

MapBiomas 10m Stable Pasture Map (Col. 4 | 2020–2025) projects/ee-vieiramesquita-mapbiomas/docs/assets/mapbiomas_brasil_10m_pasture_col4_stable_2020_2025
Pasture Condition (Sentinel-2 / CVP 2025) projects/ee-gpw-app/docs/assets/lapig_cerrado_s2_pasture_cvp_2025
Stratified Samples (50k) with Collection 11 Classes (2025 & Mode 2020–2025) projects/ee-gpw-app/docs/assets/lapig_pontos_estratificados_cvp_2025_50k_with_MapBiomasC11_fixed
💻
Source Code

Scripts

Open algorithms in Google Earth Engine Code Editor and Python/DuckDB routines for end-to-end reproduction:

Scripts (Google Earth Engine)

Stable Pasture Export (2020–2025) MapBiomas 10m (Collection 4)
Open in GEE ↗
Annual EVI Mean for Pastures (Sentinel-2) Monthly medians with 15-day moving window
Open in GEE ↗
Stratified Random Sampling (50k points) Based on Pasture Vigor Condition (CVP 2025)
Open in GEE ↗
Monthly Spectral Series Extraction (Sentinel-2) 2019–2025 · 50k Samples
Open in GEE ↗
Alpha Earth Embeddings Series Extraction 2019–2025 · 50k Samples (64-dimensional vectors)
Open in GEE ↗

Local Processing Scripts (Python & DuckDB on GitHub)

gerar_top3_pivotada_50k.py Dot-product computation, Top-3 ranking, Md3 metric, and Parquet/CSV/JSON exports (50k Series · 350k total)
View on GitHub ↗
gerar_top3_pivotada_12k.py 12k-Series pipeline: MapBiomas union (Col11 + 85k), coordinate correction, dot product, and Md3 exports (88k total)
View on GitHub ↗
exportar_referencia_embrapa.py Consolidation of 701 ground-truth points and static typology distribution export
View on GitHub ↗
📦
Open Data

Generated Datasets

Complete processed datasets (combining 50,000 persistent Sentinel-2 samples and 12,009 MapBiomas samples in 2025, totaling 62,009 candidate points across the series, plus 701 field reference areas) available for direct download in Parquet, CSV, and JSON formats on GitHub:

Dataset Series / Year Size / Sample Count Downloads (GitHub)
Md3 Pivoted Table 2025 50k Series · 2025 50,000 points · 16 attributes CSV JSON Parquet
Md3 Pivoted Table 2024 50k Series · 2024 50,000 points CSV JSON Parquet
Md3 Pivoted Table 2023 50k Series · 2023 50,000 points CSV JSON Parquet
Md3 Pivoted Table 2022 50k Series · 2022 50,000 points CSV JSON Parquet
Md3 Pivoted Table 2021 50k Series · 2021 50,000 points CSV JSON Parquet
Md3 Pivoted Table 2020 50k Series · 2020 50,000 points CSV JSON Parquet
Md3 Pivoted Table 2019 50k Series · 2019 50,000 points CSV JSON Parquet
Md3 Pivoted Table 2025 12k Series · 2025 12,009 points (5,453 Col 4 S2 + 6,556 Landsat) CSV JSON Parquet
Md3 Pivoted Table 2024 12k Series · 2024 12,034 points CSV JSON Parquet
Md3 Pivoted Table 2023 12k Series · 2023 12,080 points CSV JSON Parquet
Md3 Pivoted Table 2022 12k Series · 2022 12,149 points CSV JSON Parquet
Md3 Pivoted Table 2021 12k Series · 2021 12,281 points CSV JSON Parquet
Md3 Pivoted Table 2020 12k Series · 2020 12,386 points CSV JSON Parquet
Md3 Pivoted Table 2019 12k Series · 2019 12,483 points CSV JSON Parquet
Field Reference Base (701 Points) Ground Truth Reference 701 points · 7 typologies JSON Parquet
Pasture Condition Categories

Pasture Condition Classification Scheme

701 in situ field reference areas (~1 ha polygons) surveyed across the Cerrado biome (608 classified samples and 93 unclassified units retained for unsupervised steps). Dictionary Reconciliation & Integrity Rule: An initial 5-class proposal is being reconciled with the 7 implemented categories to construct a single versioned dictionary; category distributions remain provisional until this formal reconciliation is completed.

Productive Pasture (Pasto Produtivo)
High forage grass coverage, vigorous density, and absence of severe degradation signs.
Pasture with Weeds (Pasto com Ervas)
Pasture with moderate to severe infestation of herbaceous and ruderal invasive plants.
Pasture with Shrubs (Pasto com Lenhosas)
Significant presence of shrub and woody canopy layers across the pasture support area.
Intermediate Pasture (Intermediário)
Intermediate vegetative vigor stage, with mixed coverage of forage and bare soil.
Biological Degradation (Deg. Biológica)
Prominent bare soil, visible erosion signs, and severe loss of carrying capacity.
Natural Regeneration (Reg. Natural)
Areas undergoing natural vegetative recovery and succession of native Cerrado physiognomies.
Miscellaneous (Miscelânea)
Atypical, transitional, or mixed compositions that do not fit into standard categories.
Interactive Visualizations

Spatial Distribution & Md3 Classification

Cross-matching of MapBiomas sample points with 701 field reference points via 64-dimensional dot product. Explore samples on the map, filter by year, pasture vigor (CVP), and similarity thresholds, or draw a rectangle to analyze custom geographic areas.

50,000
MapBiomas Samples (50k-Series · Year 2025)
100% of points visible
Filters Default
Loads high-resolution sample points for the selected year.
Minimum Cutoff: Md3 ≥ 0.00
MapBiomas Sample Distribution (50k)
Dynamic
Field Reference (701 Points)
Ground Truth

Points show the spatial distribution of geolocated MapBiomas samples across the Cerrado. Use the bounding box tool or vigor/Md3 filters to explore custom subsets.

Samples in Active Area / Filter (0 displayed)

Displaying up to the first 100 sample records sorted by cosine similarity.
Sample ID Target Coords (Lat, Lon) CVP Vigor Reference Match (FID) Assigned Typology Top-1 Similarity Mean Md3 Reference Coords (Lat, Lon)
Neural Networks & Time Series Curation

SOM Analysis (Self-Organizing Maps)

Methodology based on 12×12 Kohonen Self-Organizing Maps (144 neurons) for organizing multivariate Sentinel-2 time series, evaluating spectral-temporal neighborhood coherence, and filtering candidate pseudo-labels via Bayesian smoothing.

🧠

12×12 SOM Topological Curation (144 Neurons)

Component 06 — Under Development

• Input Dimensionality: Standardized monthly indices (NDVI, EVI, NDII, NDWI, PSRI, CAI) structured as 12×J for annual cycles or 72×J for the full 6-year multi-year series (2020–2025).
• Bayesian Smoothing: Dirichlet-Multinomial smoothing over neuron neighborhood compositions (with Normal-Normal conjugate models evaluated as an alternative) to triage candidates into: Keep, Forward for CBERS-4A Inspection, or Reject.
• Preparation & Quality Controls: Feature z-score normalization, missing observation handling, monitoring of quantization and topographic errors, seed stability testing, and strict segregation of reference evidence from candidate pseudo-label frequencies.