RL-Driven Sustainable
Land-Use Allocation

Optimizing ecosystem service values in the Lake Malawi Basin through deep reinforcement learning

25 kmStudy radius
50×50Grid cells
9Land-use classes
Explore

Background

Research Concepts

This study applies deep reinforcement learning to optimize land-use allocation around Lake Malawi, maximizing ecosystem service values while respecting ecological constraints such as water zone buffering and habitat contiguity.

Sentinel-2 Land Cover

Land Cover · 2024

ESA 10m resolution satellite imagery (2024), downsampled to 10% for land cover classification. Provides 9-class land cover maps derived from the ESA WorldCover product.

View data source

MODIS Evapotranspiration

ET Data · 2024

NASA Moderate Resolution Imaging Spectroradiometer (MODIS) providing evapotranspiration (ET) data at regional scale via Microsoft Planetary Computer.

View data source

ESVD / Costanza et al.

ESV Reference · 2014

Ecosystem Services Value Database from Costanza et al. (2014), "Changes in the global value of ecosystem services." Calibrated to Lake Malawi using Zuze 2013 wetland anchor ($554/ha/yr).

View data source
Experiment

Current ESV Distribution

Low ESV
High ESV
Water
Trees
Flooded
Crops
Built Area
Bare Ground
Snow/Ice
No Data
Rangeland

Optimized Allocation