Experimental 3.5: separate environmental and land-tenure tab; public SICAR status, IBAMA and ICMBio embargo intersections, ICMBio notices, SEMAS/PA LDI and associated administrative proceedings, and exact-number judicial metadata through CNJ/DataJud. Public ITERPA layers for glebas, settlements, SIGEF/SNCI and land applications cover Para only. Parallel read-only requests use a five-minute consultation budget and preserve incomplete-source statuses. Automatic HTML/JSON/CSV reports and PDF export. This screening cannot certify ownership, legal/environmental clearance or the absence of all public proceedings.
Experimental 3.4: public INPE Brazil Data Cube Landsat Collection 2 Level-2 scenes integrated with Planetary Computer in manual catalog, batch commands and automatic PRODES acquisition. Same-product mirrors are merged and used when a band download fails. Public-source outages are reported. Historical/year coverage is not guaranteed.
Experimental 3.3: consistent calibrated natural-color display, sampled in-parcel QA ranking before spectral download, same-year candidate fallback for download failures or missing pixels, no cloud-limit relaxation and explicit missing-year reporting. Original spectral data and sensor resolution are preserved. Structured temporal/thematic/spatial review argumentation and an offline accuracy evaluator are included; no eighty-percent thematic accuracy claim is validated.
Experimental 3.2: one-click command execution, immediate main-thread QGIS raster import or archive-only destination, exact requested scene-cloud constraints, annual missing-scene reporting, import quality checks and PRODES command routing. Acquisition and QGIS integration tested with synthetic imagery; live acquisition remains to be validated.
Experimental 3.1: classic imagery catalog restored as main entry, task planning in Portuguese with annual or per-date batch downloads, native-pixel display by default, in-parcel missing-data inspection, SLC-off hidden by default, lazy NBR acquisition for selected investigation scenes and technical evidence matrices. Commands are a restricted parser, not a Codex connection or general AI chat.
3.0 experimental: autonomous bounded investigation, per-polygon QA ranking, seasonal intermediate-date searches, hypotheses, versioned official guidance, hash-chained audit, local multimodal Ollama assistance (requires a locally installed vision model), consistent reflectance contrast, UTM focus maps with context inset and grid north, 300-DPI output and automatic evidence package export. Visual hypotheses do not override measured conclusions. No model weights are included and thematic accuracy is not validated.
Previous experimental 2.3: two-vector-input workflow with automatic official PRODES source lookup, anonymous Landsat acquisition, historical search from 1985 by default through two years after each record, calibrated same-pixel temporal comparisons, exploratory NBR/dNBR disturbance and recovery screening, readable recommendations and per-polygon before/after PRODES RGB maps embedded in PDF and HTML. Reports distinguish observation intervals from exact first-clearance dates and never prove fire causation or legal compliance. Technical consolidated-area input remains a supplied classification. QGIS 3.34.4 end-to-end integration verified with synthetic imagery and real vectors; live native-network acquisition and thematic accuracy still require independent validation.
Anonymous Landsat Collection 2 Level-2 and Sentinel-2 Level-2A search by polygon area, scene cloud cover below 20 percent, georeferenced preview, selectable RGB bands, NDVI and a local offline archive. QGIS 3.22-3.99 with GDAL Python and NumPy is required; standard QGIS installations generally include these dependencies. New imagery requires internet and Microsoft Planetary Computer or INPE Brazil Data Cube availability. Experimental release; QGIS 3.34.4 integration verified, other installations remain to be tested.
Desenvolvido em colaboração por Kayky Pessoa de Araujo e OpenAI Codex. Imagens Landsat: USGS/NASA; Sentinel-2: ESA/Copernicus. Catálogo e hospedagem: Microsoft Planetary Computer e INPE/Brazil Data Cube. Pesquisa cenas históricas, permite composições RGB distintas da mesma cena e calcula NDVI com as bandas NIR e vermelha. Permite conferir o limite do imóvel sobre a imagem antes de importar e manter um acervo local por área. Internet é necessária para novos downloads e atualizações; GDAL e NumPy são usados no processamento.