Land Classification of Hakuba, Japan

Satellite data form May to June, 2024

Infrared Image

1 / 6

True Colour Image

2 / 6

NDVI Image

3 / 6

Object Based Classification

4 / 6

Pixel Based Classification

5 / 6

Supervised Pixel Classification

6 / 6

The idea behing this project was figuring out how the basic image classifications work. It started by taking the Satellite imagry form the European public acces satellite data. Then I imported the .tif file into ArcGIS Pro. Following that, I altered the color data on the map to get the Infrared Image from the True Colour Image. This is usefull because healthy plants reflect more infrared energy. The brigther the red stops are on the Infrared Map, the healthier the plants are growing.

After creating the Infrared map, I made the NDVI Map. This map lays out the infrared data in a more presentable way. So here, the darker the green is, the healthier the plants are! This has practical implacations for agriculture. If you have an industurial far, you can fly a LiDAR drone overtop of it. When you take the IR data, you can see the sections of the farm that emmit less IR Light. Basically, it shows you the areas of the farm where the crops are struggeling to grow. Knowing that information allows farmers to bolster their crop yeilds.

Moving on to the image classifications, I found a range of success with the different methoeds. I believe the Object Based Classification turned out the best. The process was unsupervised, this means that I did'nt tell the computer what to classify. You can see that it created a lot of classes for the mountain top, as there are very contrasting colours present. Its flaw was the valey where Hakuba is. The computer struggled to classify data in the valley as the colors are quite dark. If you zoom in, you can see that it did classify the valley though. It would have been better if the computer had colour coded the reigons so it would be more visable.

The pixel based classification worked allright in terms of visability. The dark mountain valley is clearly visible in this image. It is still true that the computer struggled to discern objects in the valley as things like the town itself and the lake were classed together. The benifit of this methoed was the ease of viewing the map. its very clear what is where as you know that each pixel is some area in real life. This lets you know what is prominant in the general area inside one pixel in real life.

Lastly, the Supervised Pixel Classification. The Supervised Pixel Classfication allowed me to select the range of reflected light and manually calss the data. Then, the computer searches for all pixels with light data in that range to fill out your classification. This enables me to say that the lake in the bottom of the map is a lake, whereas the unsupervised classification missed that. An issue with this methoed was huma error and not creating enough classes. The bright yellow class is a river but becasue i didnt creat a class for the villege, the computer classed the villege as a river and partly as the mountain. Another issue was the mountain top itself. I had overclassed the mountain top. The overclassing makes it appear as one big blob on my map, which dosn't look nice. If I were to do it again, I would create more classes so the map would be more detailed.

Back to My Work
Back Home