Illuminating… de-illuminating

This morning I went to the library in the hope of finding a book on fieldwork planning. But most book with this title talk about how primary/secondary teachers organise field trip. However, to my delight I found a book on vegetation mapping (Alexander and Millington, 2000), which is quite relevant and informative. I regret that I did not find this book earlier.

The book talked about Countryside Survey and Land Cover Mapping, both involve collecting data from multi-spectral images and field survey. and are similar to my project. There are also chapters on moorland vegetation classification. It cited an article on blanket bog classification in International Journal of Remote Sensing, Vol.8, and I will check this out tomorrow.

However, in the afternoon after I browsed the webpage of CS2000 and LCM2000, I realised that though they provide a resolution up to 25m, they do not contain species information but land cover such as ``bog'' or ``improved grass''. Besides, the price of LCM2000 is very high (500P per 1km^2 Grid cell for academic use).

After I cross-checked the ESA vegetation map and my unsupervised classification result, it reveals that my classification has very little consistency with the actual vegetation distributioni on the ESA map. Using NIR band the vegetation can be easily identified but most of them are not moss as supposed. Thus I think visible bands should also be considered when doing classification. Luckily the ESA map indicates bog and bare peat land cover, thus I can use it in my field survey. See if I can find more Sphagnum patches this time.

I'll go on contact LCM2000 see if the map is useful anyway. The LCM1990 and LCM2000 dataset may serve as a base for model validation... if I can get my hand on them!