Undergraduate Geodetic Engineering student at Universitas Diponegoro, Semarang, Indonesia. I build remote-sensing and geospatial ML pipelines for environmental monitoring SAR optical change detection, InSAR deformation, and land cover classification with an emphasis on validation rigour and honest scoping.
- SAR–optical fusion (Sentinel-1 + Sentinel-2) for change detection and land cover mapping
- InSAR time series for land subsidence and coastal deformation (LiCSBAS)
- Geospatial machine learning (XGBoost, LightGBM, Random Forest) with spatial cross validation
- Google Earth Engine + Python pipelines (rasterio, geopandas, GDAL) and spatial databases (PostGIS)
Sentinel-1 / Sentinel-2 fusion pipeline screening a ~38,800 ha Indonesian oil palm concession for forest loss after the EU Deforestation Regulation cutoff (2020-12-31). The primary detector is unsupervised SAR change point detection radar penetrates the near permanent tropical cloud cover that blinds optical only methods confirmed by Sentinel-2 optical drops. Detected loss is cross validated against three independent satellite references (RADD, Hansen GFC, JRC TMF), reaching ~0.81–0.85 strict-pixel and ~0.87–0.90 patch area precision. A supervised XGBoost model is included as a documented comparison, with its honest non-result reported as part of the contribution. Capability demonstration, not a legal grade compliance product.
Stack: Google Earth Engine, Python, rasterio, XGBoost, Folium
Sentinel-1 SAR flood frequency and LiCSBAS InSAR subsidence mapping for the Semarang–Demak coast (2020–2025), fused into a per-kelurahan coastal-flood vulnerability typology. Built for the IEEE GRSS REACT EO4SDG challenge. SDG 11.5 / 13.1.
Stack: Sentinel-1, LiCSBAS, ESA SNAP, Python, geopandas
End-to-end remote sensing pipeline mapping industrial land cover in the Karawang corridor, West Java, via Sentinel-1 / Sentinel-2 fusion and XGBoost classification. Produces 1,063 industrial polygons with documented human-in-the-loop refinement and an explicit limitations section.
Stack: Python, rasterio, geopandas, XGBoost, ESA SNAP
Drone imagery photogrammetry with GCP-based geodetic validation orthophoto and DSM generation and accuracy assessment using OpenDroneMap / WebODM.
Stack: OpenDroneMap, WebODM, GCP survey, Python
Async REST orchestrator for resource constrained binary tools, demonstrated with COLMAP for Structure-from-Motion reconstruction. Single tenant FastAPI service with subprocess timeout, payload validation, and VRAM bounded preprocessing.
Stack: Python, FastAPI, COLMAP
| Domain | Tools and Methods |
|---|---|
| Remote sensing | Sentinel-1/2, Landsat, ESA SNAP, LiCSBAS, Google Earth Engine |
| Change detection / InSAR | SAR backscatter time-series, change-point detection, InSAR deformation |
| Geospatial ML | XGBoost, LightGBM, Random Forest, scikit-learn, spatial cross-validation |
| Geospatial Python | rasterio, geopandas, GDAL, shapely |
| Databases | PostgreSQL, PostGIS |
| Backend | FastAPI, async subprocess orchestration |
| Geodetic computation | Map projections, least-squares adjustment, satellite geodesy |
- Deep learning for semantic segmentation in remote sensing
- Operational SAR/InSAR monitoring workflows (LiCSBAS, time-series)
- Reproducible geospatial ML pipelines determinism, spatial CV, validation rigour
- Email (academic): devonrama@students.undip.ac.id
- Email (personal): devonrama.w@gmail.com
- Location: Semarang, Indonesia