Deep Learning based multisource data fusion for biomass estimation
Required skills:
1) Programming skills in python.
2) Willingness to learn deep learning frameworks (TensorFlow, PyTorch)
3) Willingness to learn the manipulation of geographical data in Python (GDAL, geopandas, shapely, rasterio, etc)
Estimating aboveground biomass is essential for generating vegetation inventories and understanding carbon cycles. In this sense, it constitutes an important tool for supporting policies aimed on mitigating climate changes. It can be assumed that the spaceborne LiDAR sensor from the GEDI mission provides high accurate estimations of biomass. However, GEDI is a sampling mission resulting in data gaps. Several SAR sensors, e.g. TanDEM-X, Sentinel-1 and NISAR missions, combined with optical sensors and the sampled GEDI data can provide synergistic data with complementary information. The synergistic combination of the different sensors can provide gapless (i.e., wall-to-wall) aboveground biomass with high accuracy. This MSc topic focus on developing a Deep Learning method for estimating forests biomass based on sampled and wall-to-wall data from multiple sensors.
The student will initially revise the literature on biomass estimation and on multimodal deep learning models. The main focus will be on developing an end-to-end trainable multimodal/multi-path deep neural network composed of multiple paths or networks, optimized by a combined loss function, and able to combine LiDAR, SAR and optical imagery for estimating aboveground biomass maps, especially for forest areas.
[1] Qi, W., Saarela, S., Armston, J., Ståhl, G., & Dubayah, R. (2019). Forest biomass estimation over three distinct forest types using TanDEM-X InSAR data and simulated GEDI lidar data. Remote Sensing of Environment, 232, 111283. <https://www.sciencedirect.com/science/article/abs/pii/S0034425719303025?via%3Dihub>
[2] Saarela, S.; Holm, S.; Healey, S.P.; Andersen, H.-E.; Petersson, H.; Prentius, W.; Patterson, P.L.; Næsset, E.; Gregoire, T.G.; Ståhl, G. Generalized Hierarchical Model-Based Estimation for Aboveground Biomass Assessment Using GEDI and Landsat Data. Remote Sens. 2018, 10, 1832. <https://www.mdpi.com/2072-4292/10/11/1832>
[3] Silva, C. A., Duncanson, L., Hancock, S., Neuenschwander, A., Thomas, N., Hofton, M., ... & Dubayah, R. (2021). Fusing simulated GEDI, ICESat-2 and NISAR data for regional aboveground biomass mapping. Remote Sensing of Environment, 253, 112234. <https://www.sciencedirect.com/science/article/pii/S0034425720306076>