Aquaculture pond detection and change analysis

Illegal shrimp and fish ponds cause substantial mangrove loss along Southeast Asian coasts. Indonesia and Vietnam have lost substantial mangrove cover over the past 2 decades. However, regulators relied on occasional field surveys. Therefore, years could pass between new pond construction and enforcement.

Aquaculture ponds can resemble rivers, tidal flats, and estuaries in optical satellite images. Moreover, tides change whether the same place appears as water or land. Tropical clouds also obscure single images. Therefore, reliable detection needs changes over time. Comparing image stacks helps separate stable natural water from newly built ponds.

I built a pipeline that compares Sentinel-2 images across dates in 3 stages. First, the water indices NDWI, MNDWI, and AWEI locate possible water bodies from reflected light. Shape filters then separate rectangular human-made ponds from natural water. Finally, differences over time identify ponds built within a chosen monitoring period.

I validated the pipeline on Sumatra and Kalimantan’s coasts in Indonesia. Moreover, I tested it in Vietnam’s Mekong Delta and Ca Mau. Their different tides and wavelength responses tested transfer across regions.

The outputs link pond growth with mangrove loss. Therefore, they provide spatial evidence for policy reports investigating possible illegal coastal conversion.