sonthuybacha / remote-sensing-aquaculture

Remote sensing: tracking aquaculture in South Vietnam through Google Earth Engine

Remote sensing: Tracking aquaculture in South Vietnam through Google Earth Engine

### Description This is a small project using [Google Earth Engine](https://developers.google.com/earth-engine/), a geospatial processing platform that combines a catalog of 30+ years of satellite images and Google's cloud-based computing power.

The goal of this project is to track the spread of shrimp farms in South Vietnam by analyzing and classifying satellite images from 1988 to 2011.

Region

The area chosen for this study is a rectangle in the province of Ca Mau.

Image processing

I used images from the USGS Landsat 5 TM Raw Scenes (Orthorectified). For each 2-year period:

• The collection of raw images is filtered by area and dates
• The simpleComposite algorithm obtains the images with the lowest cloud scores
• Clipping is used to reduce the image to our specific predefined rectangle

Tasseled Cap conversion

Each image is converted to tasseled cap. The resulting image has 6 bands: brightness, greenness, wetness, fourth, fifth and sixth.

Classification

The image can be decomposed into three major types of land cover: water, vegetation and aquaculture. Samples of each type were selected according to the following map:

For each image, the samples were used to train the classifier. Subsequently a prediction was made for the entire image. In regard to the classification algorithm, I found satisfactory results with a CART model. A confusion matrix and a summary of different accuracy metrics are printed to the console.

Area measurement

Once each pixel in the image is classified into one of the three types (water, vegetation and aquaculture), I measured the area (in square meters) covered by each class.

Results

The following example (which corresponds to the period 2004-2005) summarizes the procedure: (1) the original image (in false color) after processing, (2) the image converted to tasseled cap, and (3) the classified image. We can appreciate that the model does a fairly decent job at identifying the different types of land cover in the image.

The line chart (with the series of the area covered by each type) shows a clear upward trend in the area used for aquaculture in South Vietnam, peaking around 2006.

Next steps

I would have liked to explore a larger area, increase the number of land cover types, and experiment with other sample areas (with the aim of increasing the accuracy of the classifier in cases where the satellite image is not clear).

References

• Assessment of land-cover changes related to shrimp farming in two districts of northern Vietnam using multitemporal Landsat data by Pham Thi Thanh Hien, Martin Béland, Ferdinand Bonn, Kalifa Goïta & Jean-Marie Dubois I did not replicate their approach but reading it was a good starting point.
• MSAN631-03 Geographic Information Systems - Course Notes by David S. Saah, PhD
• Google Earth Engine Docs