Data & methods - Land System Sustainability Lab | Tohoku University

Earth observation

We work primarily with open satellite archives, chosen to match the scale and time span of the question:

Source	Typical use
Landsat (5/7/8/9)	Long-term land-cover change, land-surface temperature
Sentinel-1 / Sentinel-2	Recent high-resolution mapping, cloud-resilient monitoring
MODIS / VIIRS	Daily-to-annual regional and global time series
Global land-cover and forest-change products	Baselines and cross-validation
Satellite atmospheric products	Air-quality and pollution exposure

Geospatial data science

Analyses are built on cloud and desktop geospatial stacks: Google Earth Engine for large-area time-series processing, GDAL and Python for local processing pipelines, and ArcGIS Pro, ENVI, and QGIS for mapping, image analysis, and cartography.

Statistics & modelling

Spatial statistics — autocorrelation, spatial regression, hot-spot and change detection.
Driver attribution — linking observed land-system change to socio-economic and biophysical factors.
Land-system modelling — scenario simulation, including agent-based and cellular approaches to land-use change.
Deep learning — convolutional and transformer models for classification, segmentation, and prediction on Earth-observation data, implemented in PyTorch.

Reproducibility

We aim to keep analyses scripted end-to-end, version-controlled, and documented so that results can be reproduced and extended. Code and derived datasets are released where licensing and agreements allow — see Data & code.