Can a land cover map derived through hyper-temporal ndvi images be improved using ndwi information?

Land cover can be successfully mapped through-the use of hyper-temporal Normalized Difference
Vegetation Index (NDVI) imagery. Normalized Difference Water Index (NDWI) is a different indicator,
not correlated to NDVI, which responds to differences in vegetation water content. As different land
covers have different temporal behaviour of vegetation water content, theoretically NDWI can provide
input to land cover mapping.
This research investigates if additional features of land cover can be mapped after the main variability in
land cover is captured through the use of hyper-temporal NDVI imagery. The main data used in this
research are the 10-day Maximum Value Composite of SPOT VEGETATION NDVI and NDWI
acquired from April 1998 to May 2010. The study area is in Andalucia, Southern Spain.
Three NDVI units were selected for study in this research. They represent one area dominated by
agriculture (NDVI unit 19), one area dominated by forest and semi natural vegetation mixed with
transitional woodland shrub and herbaceous (NDVI unit 45), and one area dominated by forest and semi
natural vegetation (NDVI unit 57). The information in the NDWI images are reduced through an
unsupervised classification procedure called Iterative Self-Organizing Data Analysis (ISODATA) in
ERDAS Imagine software. The differences in NDWI profiles between classes proved consistent across
the years. These differences can only be caused by true differences between the ecosystems they represent.
The process followed by extracting four parameters that characterized the shape of the NDWI-profiles by
class. The four derived NDWI parameters are maximum, minimum, length of cycle and amplitude.
The land cover related data collected were percentage cover of tree, shrub, herb and grass, litter, stone and
soil. Using orthophotos (2004), field data were converted into one square kilometre pixel information. In
turn, that information was correlated with the four NDWI parameters.
NDWI parameter explains 32% variability of tree characteristics in NDVI unit 19, 49% shrubs
characteristics in NDVI unit 45, and 25% shrubs characteristics in NDVI unit 57 at 99% confidence level.
It is evidence that among the four parameters, amplitude is the parameter that give a better relationship
with cover characteristics.
The results of this research show that linear regression models had overall low goodness-of-fit between
NDWI parameters and cover characteristics. It indicates that the chosen cover related data do not
represent the anticipated cover related variability that would explain the differences in derived-NDWI
profiles.