PhD researcher at the University of Oulu, specializing in GeoAI, remote sensing, and AI-driven hydrological modeling. My research involves creating deep learning frameworks to estimate snow hydrology parameters using Earth observation data from multiple sensors. I'm dedicated to merging multi-sensor imagery to enhance the accuracy of snow cover fraction and snow water equivalent estimates, which are crucial for hydrological modeling and analyzing climate impacts.
Research interests:
- •AI for hydrological modeling
- •Multi-temporal and multi-sensor data fusion
- •Snow water equivalent and snow cover estimation
- •Foundation models for Earth observation
- •Climate-focused computer vision
*Award-Winning Thesis
My PhD research designs and evaluates deep learning architectures for extracting physically meaningful snow hydrology parameters from multi-sensor Earth observation data. The framework integrates SAR and optical satellite imagery through multi-temporal data fusion to improve estimation of snow cover fraction and snow water equivalent.
The contribution lies in bridging remote sensing feature extraction with hydrological modeling requirements, enabling:
Designed and validated a multi-task deep learning architecture that jointly solves three interconnected Nordic snow monitoring problems of current depth estimation, 7-day forecasting, and extreme event detection, using a single shared encoder trained on 10 years of real operational data (SMHI + MESAN).
This project confidently applies deep learning methods to estimate snow hydrology parameters, directly utilizing findings from a published study and imagery data. Our clear goal is to build a robust data processing pipeline, covering every essential step from data acquisition and model selection to training strategies and performance evaluation.
We used DenseFusion for 6D pose estimation to help people with upper-limb disabilities grasp objects. We used RGB, depth, and mask images. We also created our own dataset using Unity and HTC Vive headsets, comprising 3D models, camera parameters, and RGB images for DenseFusion. See More
The objective of this project was to perform semantic segmentation on the images of crop fields, background, and weeds. To achieve this, three different segmentation models were deployed U-Net, Attention U-Net, and DeepLabV3+. See More