laicsiifes/IntegraCAR-LULC-500-2012-2015

IntegraCAR-LULC-500 (2012–2015): Historical High-Resolution Optical Benchmark for Land Use and Land Cover Segmentation

Dataset Summary

IntegraCAR-LULC-500 (2012–2015) is a high-resolution remote sensing dataset designed for historical Land-Use and Land-Cover (LULC) semantic segmentation, temporal domain adaptation, and bi-temporal land change detection in the context of the Brazilian Rural Environmental Registry (Cadastro Ambiental Rural - CAR).

Developed by researchers at the Instituto Federal do Espírito Santo (IFES) in collaboration with the Instituto de Defesa Agropecuária e Florestal do Espírito Santo (IDAF), this dataset serves as the historical bi-temporal counterpart to the official SIBGRAPI 2026 dataset:

• Current Era Benchmark (2019–2020): laicsiifes/IntegraCAR-LULC-500
• Historical Era Benchmark (2012–2015): laicsiifes/IntegraCAR-LULC-500-2012-2015 (This repository)

This collection comprises 1,000 paired $2048 \times 2048$ pixel image tiles (500 optical airborne orthophotos (Vexcel UltraCam-Lp sensor) + 500 ground-truth segmentation masks) at $0.5\text{ m/pixel}$ spatial resolution, covering the exact same geographic footprints and rural property centroids as the 2019–2020 benchmark across the 10 official micro-regions of Espírito Santo, Brazil.

---

Key Research & Training Use Cases

1. Historical Land-Cover Segmentation: Train, fine-tune, and evaluate deep convolutional and Vision Transformer architectures on historical sub-meter optical aerial imagery.
2. Temporal Domain Adaptation & Generalization: Assess model robustness when transferring models across a decade (evaluating a 2019–2020 trained model on 2012–2015 imagery and vice versa), exploring spectral shifts, sensor calibration differences, and illumination variances.
3. Pixel-Level Bi-Temporal Change Detection (2012–2015 vs. 2019–2020): Because tile footprints are 100% geographically aligned with laicsiifes/IntegraCAR-LULC-500, this dataset enables direct pair-wise change detection to quantify deforestation, forest restoration, agricultural expansion, and urban/infrastructure encroachment over a 7- to 8-year baseline.
4. Historical CAR Environmental Compliance Auditing: Provides automated benchmarking for rural property environmental regularization under the Brazilian Forest Code (Código Florestal, Law 12.651/2012), validating whether Permanent Preservation Areas (APPs) and Legal Reserves (RL) existed prior to regulatory cutoff thresholds.

---

Data Acquisition & Geoprocessing Pipeline

Data acquisition is based on official spatial data infrastructure maintained by GeoBases (https://geobases.es.gov.br/), the Instituto Jones dos Santos Neves (IJSN), and the Instituto Estadual de Meio Ambiente e Recursos Hídricos (IEMA):

1. Airborne Optical Survey (2012–2015): Unlike the 2019–2020 dataset (which utilized orbital imagery from the South Korean KOMPSAT-3 and KOMPSAT-3A satellites), the historical 2012–2015 imagery originates from a statewide airborne photogrammetric survey (Levantamento Aerofotogramétrico do Espírito Santo) contracted by the IEMA (Instituto Estadual de Meio Ambiente e Recursos Hídricos) and executed by Hiparc Geotecnologia.
   • Sensor / Camera: Microsoft Vexcel UltraCam-Lp high-precision large-format digital metric aerial camera.
   • Inertial & Positioning System: NovAtel SPAN-CPT IMU with dual-frequency DGPS L1/L2 RTK and Omnistar differential correction.
   • Native Resolution: 0.25 m (25 cm) Ground Sample Distance (GSD) panchromatic and 0.55 m multispectral (RGBI), orthorectified across the entire state of Espírito Santo via the official layer geonode:iema_ortofotomosaico_es_025m_2012-2015.
   • Standardized Resolution: Resampled to 0.5 m/pixel (EPSG:4326) in 2048 x 2048 RGB tiles to ensure 1:1 geographic and spatial parity with the 2019–2020 KOMPSAT imagery.
   • (Note: The split names retain satellite_train, satellite_val, satellite_test strictly to maintain seamless programmatic and DataLoader schema parity with the 2019–2020 benchmark).
2. Reference Ground Truth (2012 LULC Mapping): Official high-precision vector mapping (Mapeamento de Uso e Ocupação do Solo do Espírito Santo - 2012, IEMA/SEAMA/GeoBases), photo-interpreted by specialist cartographers and remote sensing geographers.
3. Tile Extraction & Geographic Parity: Each coordinate query extracts a $2048 \times 2048$ pixel window centered on the CAR rural property ($1,024\text{ m} \times 1,024\text{ m}$ on the ground). Polygons were rasterized with exact grid snapping to the optical tiles.

Geographically Stratified Splits

The dataset inherits the balanced 10-microregion stratification of the IntegraCAR project (50 samples per micro-region):

• satellite_train / mask_train: 300 tiles (30 per micro-region)
• satellite_val / mask_val: 100 tiles (10 per micro-region)
• satellite_test / mask_test: 100 tiles (10 per micro-region)

---

Semantic Class Taxonomy

Ground truth masks follow the harmonized 5-class CAR-oriented taxonomy established in the IntegraCAR project:

Class ID	Class Name	Description	GeoBases / IEMA 2012 Source Classes	Portuguese Term
0	Vegetation Areas	Native Atlantic Rainforest, secondary forests, riparian woods, mangroves, marshes, restinga, and high-altitude grasslands.	Floresta Nativa, Mangue, Restinga, Vegetação de Altitude	Áreas de Vegetação
1	Agropastoral Areas	Pasturelands, cultivated crops (coffee, papaya, eucalyptus plantations, sugarcane, banana), and actively managed farmland.	Pastagem, Café, Silvicultura, Agricultura	Áreas Agropastoris
2	Infrastructure	Buildings, rural houses, paved highways, unpaved rural roads, mining quarries, and constructed surfaces.	Edificações, Vias Pavimentadas/Não Pavimentadas, Mineração	Infraestrutura
3	Water Bodies	Perennial and intermittent rivers, drainage channels, lakes, lagoons, reservoirs, and aquaculture ponds.	Rios, Lagos, Lagoas, Represas	Corpos d'Água
4	Macega	Unmanaged fallow land, brushlands, dense scrub, and coarse tall herbaceous vegetation in early transitional stages.	Vegetação Secundária em Estágio Inicial / Pousio / Macega	Macega

---

Dataset Structure & Features

The dataset is structured on Hugging Face as a DatasetDict containing 6 splits:

DatasetDict({
    mask_train: Dataset({ features: ['image', 'filename', 'latitude', 'longitude', 'municipio', 'microestad'], num_rows: 300 }),
    mask_val: Dataset({ features: ['image', 'filename', 'latitude', 'longitude', 'municipio', 'microestad'], num_rows: 100 }),
    mask_test: Dataset({ features: ['image', 'filename', 'latitude', 'longitude', 'municipio', 'microestad'], num_rows: 100 }),
    satellite_train: Dataset({ features: ['image', 'filename', 'latitude', 'longitude', 'municipio', 'microestad'], num_rows: 300 }),
    satellite_val: Dataset({ features: ['image', 'filename', 'latitude', 'longitude', 'municipio', 'microestad'], num_rows: 100 }),
    satellite_test: Dataset({ features: ['image', 'filename', 'latitude', 'longitude', 'municipio', 'microestad'], num_rows: 100 })
})

Feature Definitions

• image (PIL.Image.Image): High-resolution tile ($2048 \times 2048$ pixels, RGB for satellite orthophoto; single-channel 8-bit integer mask for segmentation).
• filename (str): Unique identifier matching the official CAR property code (e.g., ES-3200102-AE8D329D041843D28D7CFE7941120F48.tif).
• latitude (float): SIRGAS 2000 UTM Zone 24S northing coordinate (Y) of the tile center.
• longitude (float): SIRGAS 2000 UTM Zone 24S easting coordinate (X) of the tile center.
• municipio (str): Municipality name in Espírito Santo (e.g., Afonso Cláudio, Linhares).
• microestad (str): Official SEDES planning micro-region (e.g., Sudoeste Serrana, Rio Doce).

---

Getting Started & Python Usage

1. Basic Loading (Single Historical Epoch)

from datasets import load_dataset
import numpy as np

# Load the historical 2012-2015 benchmark
ds_hist = load_dataset("laicsiifes/IntegraCAR-LULC-500-2012-2015")

sat_tile = ds_hist["satellite_train"][0]["image"]
mask_tile = ds_hist["mask_train"][0]["image"]

print("Satellite tile shape:", np.array(sat_tile).shape)
print("Mask unique labels:", np.unique(np.array(mask_tile)))

2. Bi-Temporal Joint Loading (2012–2015 vs. 2019–2020)

Because both datasets share the identical filename conventions and split splits, you can load them side-by-side for change detection:

from datasets import load_dataset
import numpy as np

# Load both temporal eras
ds_past = load_dataset("laicsiifes/IntegraCAR-LULC-500-2012-2015")
ds_present = load_dataset("laicsiifes/IntegraCAR-LULC-500")

# Access matching tiles by filename
sample_past = ds_past["satellite_test"][0]
sample_present = ds_present["satellite_test"][0]

assert sample_past["filename"] == sample_present["filename"], "Coordinate alignment mismatch!"

img_2012 = np.array(sample_past["image"].convert("RGB"))
img_2020 = np.array(sample_present["image"].convert("RGB"))

print(f"Property: {sample_past['filename']}")
print(f"Location: {sample_past['municipio']} ({sample_past['microestad']})")
print(f"Comparing 2012-2015 vs 2019-2020 imagery for change analysis.")

---

Related Repositories

• IntegraCAR-LULC-500 (2019–2020): laicsiifes/IntegraCAR-LULC-500
• IntegraCAR-LULC-10K (2019–2020): laicsiifes/IntegraCAR-LULC-10K
• IntegraCAR-LULC-10K (2012–2015): laicsiifes/IntegraCAR-LULC-10K-2012-2015

---

Citation Information

If you utilize IntegraCAR-LULC-500 (2012–2015) in your research, please cite the foundational SIBGRAPI 2026 paper:

@inproceedings{integracar_lulc_2026,
  title     = {IntegraCAR-LULC-10K: A High-Resolution Optical Satellite Dataset for LULC Segmentation in the Brazilian Rural Environmental Registry},
  author    = {Albertino, Calebe and Lima, Gabriel M. B. and Oliveira, Vin{\'{i}}cius R. and Ribeiro, Arthur R. V. and Oliveira, Eliza K. S. and Souza, Eduardo H. P. and Dalvi, Otavio G. and Komati, Karin S. and Andrade, Jefferson O. and Boldt, Francisco A. and Paix{\~{a}}o, Thiago M.},
  booktitle = {Proceedings of the 39th SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI)},
  year      = {2026}
}

---

Acknowledgments

This research is conducted under the IntegraCAR Project by:

• Instituto Federal do Espírito Santo (IFES) – Serra, ES, Brazil.
• Instituto de Defesa Agropecuária e Florestal do Espírito Santo (IDAF) – Vitória, ES, Brazil.
• FAPES (Grant No. 1048/2025, Project DI 016/2025 - IntegraCAR: Integração do Cadastro Ambiental Rural no Estado do Espírito Santo; Grants 1023/2022, 055/2026, 374/2026).
• CNPq (DT-2 Grant 302726/2023-3 and Grant 407742/2022-0).