Developed by: Solved - Soluções em Geoinformação
This repository provides the steps to detect aquaculture areas using Landsat Top of Atmosphere (TOA) mosaics.
The detection process focuses on identifying aquaculture areas using Landsat TOA mosaics. The process involves generating annual cloud-free mosaics using Google Earth Engine (GEE) and applying a U-Net deep learning model for segmentation.
For more information about the methodology, please consult the Coastal Zone Algorithm Theoretical Basis Document.
One must have a Google Earth Engine Account (Get Started), be able to create a GEE repository in the code editor and upload the modules in it.
Some sort of GPU capability is also required for the training process.
Landsat TOA Mosaics: Use USGS Landsat Collection 2 Tier 1 TOA imagery. Generate annual cloud-free mosaics from January 1st to December 31st from 1985-2024. Apply a median filter to remove clouds and shadows.
- Script: 1-mosaic-generation.js
Vizualise the training and validation regions, along with the supervised layer available publicly
- Script: 2-train-test-dataset.js
Training Samples: Select training samples based on aquaculture and non-aquaculture categories. No differentiation between coastal and continental aquaculture is made during the segmentation.
Every prediction is a binary set of pixel values. 0 - "non-aquaculture", 1 - "aquaculture"
Semantic Segmentation
Model: Use a U-Net neural network to perform semantic segmentation on local servers.
| PARAMETERS | VALUES |
|---|---|
| Neural network | U-Net |
| Tile-Size | 256 x 256 px |
| Samples | 65400(Train), 16400 (Validation) |
| Attributes | green, red, nir, swir1, NDVI, MNDWI |
| Output | 2 (aquaculture and Not-aquaculture) |
Table 2 - CNN attributes and segmentation parameters. In total, six (6) distinct attributes were used.
- Script: 3-Jupyter Notebook
Gap-fill: Replace no-data values using the nearest available valid class. Temporal Filter: Apply a 3-year moving window to correct temporal inconsistencies.
- Script: 4-gap-fill-temporal-filter.js
| RULE | INPUT (YEAR) | OUTPUT |
|---|---|---|
| - | T1 / T2 / T3 | T1 / T2 / T3 |
| GR | Aqua / N-Aqua / Aqua | Aqua / Aqua / Aqua |
| GR | N-Aqua / Aqua / N-Aqua | N-Aqua / N-Aqua / N-Aqua |
Spatial Filter: Use GEE's connectedPixelCount to remove isolated pixels, ensuring a minimum mapping unit of ~1 ha.
- Script: 5-spatial-filter.js
Frequency Filter: Remove classes with less than 10% temporal persistence.
- Script: 6-frequency-filter.js
| CLASS | REFERENCES |
|---|---|
| AQUACULTURE / SALT-CULTURE | MapBiomas Collection 9, Atlas Dos Remanescentes Florestais da Mata Atlântica (SOS Mata Atlântica, 2020), Barbier and Cox, 2003; Guimarães et al., 2010; Prates, Gonçalves and Rosa, 2010, Queiroz et al., 2013; Tenório et al., 2015; Thomas et al., 2017, Diniz et al., 2021, São José et al., 2022, plus visual inspection. |
SÃO JOSÉ, F. F. DE et al. Mapeamento de viveiros escavados para aquicultura no Brasil por sensoriamento remoto. Embrapa Territorial. Documentos, 144, 2022.
GUIMARÃES, A. S. et al. Impact of aquaculture on mangrove areas in the northern Pernambuco Coast (Brazil) using remote sensing and geographic information system. Aquaculture Research, v. 41, n. 6, p. 828–838, 13 maio 2010.
QUEIROZ, L. et al. Shrimp aquaculture in the federal state of Ceará, 1970–2012: Trends after mangrove forest privatization in Brazil. [s.l: s.n.]. v. 73
TENÓRIO, G. S. et al. Mangrove shrimp farm mapping and productivity on the Brazilian Amazon coast: Environmental and economic reasons for coastal conservation. Ocean & Coastal Management, v. 104, p. 65–77, 2015.
USGS. LANDSAT COLLECTION 1 LEVEL 1 PRODUCT DEFINITION. [s.l.] Earth Resources Observation and Science (EROS) Center, 2017.
XU, H. Modification of normalised difference water index (NDWI) to enhance open water features in remotely sensed imagery. International Journal of Remote Sensing, v. 27, n. 14, p. 3025–3033, 20 jul. 2006.