Experience

Satellite Ground Segment Architect — Payload Data Processing — AI Data Processing — 16 satellite missions across Phase 0–E, combining system engineering with hands-on L0–L2 payload data processing, laboratory and on-orbit Cal/Val, and ECSS-compliant software delivery.

:red_circle: Note: This profile is prepared using publicly available information from the literature and reflects concepts I have learned during my professional experience. It does not include any proprietary or confidential information.

Professional summary

Operational satellite ground segment system engineer with ESA Harmony / EOF-EOS (PDGS) experience in ECSS-compliant environments. Contributed across 16 satellite missions covering the full Phase 0–E lifecycle — requirements decomposition (URD/MRD/SRD), MBSE-based architecture (UML/SysML), interface control, and IV&V. Designed and developed L0–L2 payload data processing chains with system-level sizing and trade-offs (throughput, latency, compute capacity), and implemented containerized pipelines using Docker and Kubernetes. Deep mastery of the radiometric chain, proven by building the exact inverse of the Sentinel-2 MSI L0→L1B chain in the open-source Sentinel-2 MSI Synthetic Raw Data Generator project.

Career timeline

Competency depth

The sections below expand on four core competency areas referenced in the timeline above.

System Engineering

Ground-segment and payload-data system engineering across Phase 0–E in ECSS-compliant environments.

• Requirements & architecture — URD/MRD/SRD decomposition; MBSE-based design (UML/SysML); interface control documents and Master ICD maintenance.
• Review & IV&V — Ground Segment Review Board (ESA Harmony); HEEPS simulator reviews; cross-verification testing for SAR and multispectral processors (L1/L2/L3); change control and traceability.
• Sizing & trade-offs — downlink volumes, throughput, latency, compute capacity; C++ vs Python processor implementation trade-offs; GPU/memory planning for AI pipelines.
• Standards & transition — CCSDS, PUS; EOPF transition (Zarr adoption, Sentinel-1 ingestion); PDR/CDR/QR lifecycle participation.
• Container orchestration — Docker and Kubernetes for ground-segment deployment patterns.

Related open-source work: IPF ecosystem · sar-processor

Cal/Val

Laboratory and on-orbit calibration for pushbroom optical instruments — hyperspectral, multispectral, and SWIR.

Ground-segment Cal/Val — Pushbroom hyperspectral spectrometer (L0–L1b)

I perform laboratory calibration of optical hyperspectral spectrometers for spaceborne applications, ensuring the instrument meets its spectral, radiometric, and geometric fidelity requirements. The campaign systematically characterizes detector-related effects, spectral response, radiometric accuracy, and geometric alignment, validating that the instrument will perform to mission specification on orbit.

Calibration objectives

• Spectral — determine the Instrument Spectral Response Function (ISRF) vs. wavelength and pixel position; verify ISRF FWHM; validate spectral oversampling and sampling interval; minimize smile and keystone; confirm spectral linearity and ISRF stability under thermal/mechanical perturbation; determine wavelength calibration accuracy.
• Radiometric — establish a radiometric reference with high multiplicative accuracy over multiple radiance levels; verify optical transmission and detector-response consistency; ensure radiometric and zero-level offset stability; validate detector linearity up to saturation and stray-light contributions.
• Geometric — characterize focal length, aperture, slit geometry and alignment; confirm spatial sampling distance at orbital altitude; correct spatial smile/keystone below the design threshold.

Methodology & steps

• Pre-calibration alignment — mount on a vibration-isolated optical table; verify optical components with a collimated reference beam; record slit orientation and grating alignment. Operate the detector in a dark, thermally controlled environment and characterize dark current and offset for reserved dark pixels.
• Spectral — use a monochromator to supply narrow spectral lines across the spectrometer’s bandwidth; measure ISRF profiles per pixel and extract FWHM. Shift the illumination spot across the FoV to record spectral/spatial shifts (smile/keystone). Simulate thermal and mechanical stresses to evaluate ISRF variation.
• Radiometric — illuminate the entrance slit with a calibrated integrating sphere; derive per-pixel gain and offset. Assess SNR, dark current and readout noise; vary integration time and radiance to confirm linearity. Simulate high-contrast scenes to evaluate stray-light suppression.
• Geometric — map the Point Spread Function with a collimated beam and pinhole mask; measure FoV against design; quantify smile/keystone with distinct spectral lines across the FoV.

Data products & validation — spectral (ISRF maps, wavelength alignment), radiometric (gain, offset, linearity) and geometric (distortion maps, smile/keystone) calibration files; uncertainty estimation against reference standards and environmental stability; validation by applying calibration files to a test dataset and verifying requirement compliance.

On-orbit radiometric, spatial & geometric calibration — NUC / MTF / TOA

I command captures of pseudo-invariant sites (Mauritania Desert, Dome-C, Antarctic) for flatfield images at different TDI stages and exposure times (following the USGS Test Sites Catalog), and night passes over the Atlantic (no clouds, no light) as darkfield images.

Non-Uniformity Correction (NUC) — mean each column of flatfield/darkfield (flatfield_desired, darkfield_desired), then:

Apply NUC and flatfielding simultaneously with a lab-derived dark_offset, and store gain/offset in a Calibration Key Data (CKD) container:

Bad-pixel correction — global pixel variance as threshold; replace flagged pixels with neighbour average. Denoising — Butterworth low-pass (parameters by trial).

Image restoration / MTF — internal-clock timing offset can introduce up to ~7 km positional deviation, so I use runways and bridges as MTF targets when dedicated targets (e.g. Baotou) aren’t available. Build the Edge Spread Function from a high-contrast edge → differentiate to the Line Spread Function → normalized Fourier transform → MTF; derive PSF from the MTF (or simulate a PSF model when noisy); normalize the PSF kernel and convolve, or use Wiener deconvolution.

Band registration — convert to 8-bit + CLAHE dummy bands; SIFT keypoints/descriptors vs. a reference band; FLANN matcher; homography; warp bands to the reference.

Georeferencing — Sentinel-2 bands as image-to-image reference; estimate scene coordinates from TLE; download Sentinel-2 from Earth Engine; feature-match & warp; copy corner coordinates, CRS and transform.

TOA conversion — convert NUC Digital Numbers to radiance with per-band radiometric gain/offset:

Atmospheric correction — Py6S; familiar with MODTRAN and LibRadTran; used FLAASH (ENVI) on Landsat-8 OLI; due to MODTRAN licensing, follow the Landsat 8–9 Cal/Val ADD (p. 776). Mathematical modeling & SNR — optics+sensor integrated modeling to compute total photons collected as a function of attitude (roll, pitch, yaw).

Mission examples: CONNECTA T2.1/T3.x CubeSats · GHG monitoring mission (AIRMO) · hyperspectral spectrometer Cal/Val matrix (QA4EO-aligned).

Remote Sensing & AI/ML

Earth-observation science and machine-learning pipelines across operational and research missions.

• ESA / USGS / NASA products: Sentinel-1 (SAR), Sentinel-2 A/B, Sentinel-5P, Landsat 7/8/9, ASTER, MRO CTX & HiRISE.
• Applications: metallic-mineral exploration · fault-line detection · natural-disaster analysis · NDVI/NDWI/NBR · image segmentation · surface-deformation detection · lineament extraction.
• Tooling (Qt Designer): image-processing pipeline tools; image database search & download utilities.
• ASTER: VNIR/SWIR/TIR channels for mineral & mineral-group identification (mining applications).
• Deep learning (Python, OpenCV, Keras, PyTorch): object detection via color segmentation, template matching, corner/edge/contour detection; feature matching; watershed; CNNs for real-time digit classification and object detection; pyramid representation, sliding window, non-maximum suppression, region proposals; R-CNN, YOLO and SSD; Spiking Neural Networks (SNNs) integrated into EO target-detection chains.
• Research: Turkish Lunar Rover Mission — DEM extraction and ML prototypes for landing-site analysis; planetary geology (Thaumasia Planum, Mars); BSc thesis lineament extraction (Landsat-8 PCA).

Related open-source work: eo-data-embedding

Software Data Processing Development

End-to-end payload data ground-segment software from raw instrument packets to science-ready products.

Multispectral high-resolution camera — preprocessing pipeline (L0–L2)

Responsible for developing the calibration, validation and preprocessing ground-segment software from Level-0 (raw) to Level-2 (science-ready) for pushbroom multispectral optical instruments.

• Level 0 — decoding; missing-package check and flag generation.
• Level 1 — Non-Uniformity Correction (gain/offset key data); dark-current correction; denoising (filters / designed digital signal filters); radiometric conversion from lab or on-orbit key data; MTF compensation / PSF deconvolution; band co-registration (sensor acquisition model or keypoint extraction & matching); georeferencing (GDAL/rasterio; central-pixel metadata coordinate, or — when GNSS is missing — Sentinel-2 reference via TLE + Google Earth Engine, then image-to-image keypoint matching); geolocation-accuracy (CE95); orthorectification.
• Level 2 — atmospheric correction (Py6S).
• Additional — pansharpening (Simple Brovey, Gram-Schmidt, ESRI); image-quality report (PSNR, RMSE, SSIM, MSE, GIQE, CE95, radiometric accuracy) generated to PDF.

Operator-facing Qt GUI and CLI/API; Docker-containerized execution; versioned YAML/JSON configuration; automated PDF Data Quality Reports.

Related open-source work: msi-processor · Sentinel-2 MSI Synthetic Raw Data Generator · IPF ecosystem

Skills