Career Profile
I am a geospatial data scientist and researcher specializing in the spatiotemporal analysis of landscape dynamics, carbon modeling, and GeoAI. My work focuses on building end-to-end machine learning and deep learning pipelines to monitor forest ecosystems, map permafrost wetlands, and model forest fire dynamics and pyrogenic emissions at regional scales.
Core Expertise: GeoAI, Deep Learning for Computer Vision (Semantic Segmentation & Regression), Remote Sensing, Time Series Analytics, Carbon Accounting, and Climate-Disturbance Interactions.
Technical Stack: Languages & ML Frameworks: Python, PyTorch, PyTorch Lightning, Scikit-Learn, XGBoost, Hugging Face Transformers, TorchGeo. Spatial Data Science: GDAL, Rasterio, GeoPandas, Rioxarray, Dask, QGIS, ArcGIS. Photogrammetry: Agisoft Metashape, OpenDroneMap (ODM).
🌍 Open to global relocation opportunities and innovative research or industry collaborations in GeoAI, climate tech, and remote sensing.
Education
Grade: 4.875
- My PhD thesis will be about Geoinformation modeling and geoecological assessment of the spatiotemporal dynamics of forest and swamp geosystems in Western Siberia using remote sensing data and machine learning methods
- Currently I’m on my final year and going to defend my PhD thesis soon
Grade: 4.5
- Master thesis on vegetation modelling of Western Siberian forests using deep learning methods
Grade: 5.0
- Bachelor thesis about fire dynamics in forest tundras of Western Siberia
- Awarded with a winner diploma of Ya-Profi ecology olympiad in 2018 and 2020
- All-Russian Olympiad Grant beneficiary
- Scholarship from Rosneft
Experiences
- Long-Term Boreal Forest Dynamics Modeling (1984–2024)
Designed and trained SegFormer deep learning models to map and analyze 41-year dynamics of 5 key forest ecosystem attributes (dominant tree species, site productivity, forest type, growing stock, stand age) across the Khanty-Mansi Autonomous Okrug (534,800 km²).
Quantified major ecological shifts, mapping a sharp decline in lichen-dominated highland forests and a significant post-disturbance transition from primary pine to secondary broadleaved stands.
- Wetland Type Mapping (Numto Nature Park)
Developed and compared 15 machine learning models (4 classical ML, 11 deep learning) to classify ecologically sensitive peatlands and sporadic permafrost zone mires.
- Remote Sensing Processor
Optimized data and ML workflows by developing Remote Sensing Processor, an open-source geospatial ML Python library used to accelerate and automate raw satellite and terrain data preprocessing, patch generation, and deep learning pipeline workflows for regional-scale modeling.
- Spatiotemporal Wildfire Dynamics in the Forest Tundra (1985–2018)
Reconstructed a 33-year history of wildfire dynamics in northwestern Siberia using Normalized Burn Ratio (NBR) derived from Landsat imagery.
Analyzed climate-fire interactions, proving a strong correlation between burned area and weather conditions.
- Precision Agriculture Project Managed soil sampling field campaign across multiple agricultural plots.
Processed high-resolution UAV multispectral imagery using Agisoft Metashape to construct orthomosaics and calculate localized vegetation indices (NDVI) for crop health monitoring.
- Corporate Carbon Footprint & Pyrogenic Emission Modeling
Evaluated the economic and ecological viability of corporate carbon footprint reduction strategies for mining industry via targeted forest fire prevention programs across the Krasnoyarsk and Irkutsk regions.
Quantified burn severity and structural ecosystem damage across comparative control zones, evaluating areas with active firefighting infrastructure against baseline, unmanaged landscapes using satellite-derived spectral indices (NBR and dNBR).
- Post-Agricultural Forest Recovery & Afforestation Analysis
Assessed the health and development conditions of actively planted forest stands on abandoned agricultural lands for the Altai Afforestation Project.
Provided critical spatial data to verify the integrity and growth trajectory of the afforestation project, supporting the rigorous calculation of regional carbon sequestration offsets.
- Urban life quality assessment
Engineered and calculated 14 distinct ecological and environmental variables across Russia’s 120 largest metropolitan areas to drive comprehensive urban life quality assessments.
Estimated the shares of protected areas, green and industrial zones in the total area of cities, the density of the street and road network, the change in the area of the green zones, the built-up area and the water level in reservoirs over time, the density of buildings per capita and the gap between the rate of development of the territory and the rate of population change in cities using different spatial datasets.
Certifications
Pet projects
is my pet project, an open-source Python library for for geospatial ML and data processing.
-
End-to-End Geospatial ML Framework: RSP is designed to streamline the entire geospatial machine learning lifecycle – from raw imagery preprocessing to deep learning model training and spatial inference map generation.
-
Multi-Sensor Data Pipelines: RSP includes automated preprocessing pipelines for Landsat and Sentinel-2 satellite data, implementing cloud masking, atmospheric correction, pansharpening, and Sentinel-2 20/60m band upscaling (super-resolution).
-
Advanced Raster & DEM Analytics: RSP can generate seamless multi-band mosaics, compute hundreds of spectral indices via Awesome Spectral Indices, and calculate terrain attributes (aspect, slope, curvature, hillshade).
-
Geospatial Dataset Optimization: RSP includes a robust tile-generation engine that rasterizes vector maps, partitions large-scale rasters into synchronized patches, and structures them into an optimized ML-ready custom format (RSPDS) using Hugging Face Datasets and XBatcher.
-
Flexible Hybrid ML/DL Engine: RSP provides unified, high-level API wrapper natively supporting both classical machine learning (e.g. SVM, Random Forest, Gradient Boosting) and advanced Deep Learning architectures (e.g. U-Net, DeepLab, SegFormer) for semantic segmentation and regression.
-
Explainability & Validation Tooling: RSP provides explainable AI capabilities by adapting SHAP to compute pixel-level band importance in geospatial models, alongside custom TorchMetrics evaluation pipelines.
-
Distribution: RSP is available across PyPI and Conda and uses GitHub Actions CI/CD pipelines for automated linting and unit testing.
RSP also have a GUI - the ML4QGIS plugin.