Land cover using Landsat-5

Land-cover classification in mountain areas using Landsat-5

Aim of the method/technique	Describe the long term changes in the land cover categories and the NDVI vegetation index in a study area.
Scale – spatial and temporal	Vegetation raster map (30x30m grid), study area level.
Brief description	Recently the use of remote sensing data for monitoring LEDD is increasing due to the availability of satellite images that allows the monitoring and assessment of environmental change. Satellite-derived vegetation index such as the Normalized Difference Vegetation Index (NDVI) is an estimate of vegetation productivity that can be used to relate green vegetation cover and the vegetation productivity with LEDD. In order to quantify vegetation changes from Landsat-5 TM, satellite images from specific year in sequential order from the 80th to nowadays at 10 years intervals are required.
Data requirements	Landsat 5-TM image from the 80th to nowadays at 10 years intervals are required. With this it is possible to create the vegetation raster map (30x30m grid) and NDVI map to do the frequency distribution of the vegetation types for each period, transition matrix, or Markov chains. Relation among biophysical variables and the changes can be estimated by different GLM statistical analyses.   The procedure to obtain a correct land cover classification with the satellite image is: Correction of the images to normalize to be compares the different period image. Ortho-rectification (the root mean square error (RMSE) <0.5 pixels) and radiometric corrections. Random Training (RT) data for the supervised classification: RT was based on a random selection of 800 single pixels for all the study area. The land-cover category to which each selected pixel corresponds was identified using ortho-photographs and field work. Based on the main broad vegetation types that can be identified across the altitudinal gradients of most mountainous regions, four major land-cover categories are considered: forest, scrubland, dense grasslands, and sparse grasslands. Random Forest (RF) method from R (package randomForest, December 2009) is used classify the images. The accuracy of the classifications is tested using samples taken other 800 Random Training data (Congalton 1991). To assess the accuracy of the land-cover classification, we used Cohen´s Kappa Index of Agreement.
Main applications in grazing land regions	For tescription of land cover are useful in assessing the impacts of drivers of LEDD in a grasslands region.
Strengths and weaknesses	Cost-effective efficient tool that allow the monitorization of land degradation on large areas with low effort once vegetation classification has been performed

• < Prev
• Next >