3.3.2 Processing MODIS Data
In order to process MODIS data, there are several remote sensing technique was applied to get the imagery become useable. Vegetation indices were used to
enhance the features of the objects of interest in the study. As the blue band is sensitive to atmospheric conditions, it is used to adjust the reflectance in the red
band as a function of the reflectance in EVI. Cloud covers become a problem where a research using MODIS data
conducted in tropical area like Indonesia. The best way to remove cloud cover and get a clear image data is to make composite of imagery with different dates for the
same location. In this research, the daily MODIS EVI data were composited for each 16 days and resulting of 69 EVI image temporal data ranging from January
2008 to December 2010 with interval of 16 days. The next technique to do is to change the 16-day of EVI MODIS image projection from sinusoidal projection
into geographic projection in order to get the common projection that can be overlaid with other data, and the results were used to do next process. All of the
69 EVI MODIS satellite data ranging from January 2008 to December 2010 then were stacked to create an image with 69 bands that will make it able to be
analyzed as multi temporal image of 16 days interval of EVI data. Next of the process is to subset to 69 bands EVI image with administrative boundary of
Karawang, Subang, and Indramayu Regency and resulting multi temporal EVI image of research area. Once the image was stacked and subsets in to research
area, then the image need to be classified to identify the land cover of the research area. In this research the classification process were done in two steps, first is
using unsupervised classification method then continued with supervised classification based on temporal behavior of land cover.
The iterative self-organized unsupervised clustering algorithm ISODATA of the ERDAS imagines software was used to derive spectral classes from 69 EVI
image data layers. The ISODATA procedure is iterative in that it repeatedly performs an entire classification outputting a thematic raster layer and
recalculates statistics. Self-organizing refers to the way in which it locates clusters with minimum user input. The ISODATA clustering method uses spectral
distance, as in the sequential method, but iteratively classifies the pixels, redefines
the criteria for each class, and classifies again, so that the spectral distance patterns in the data gradually emerge ERDAS, 1997.
It starts from arbitrary cluster means. In each successive clustering, the means of clusters are shifted. A cluster is a group of pixels classes with similar
spectral characteristics. The ISODATA utility repeats the clustering of pixels in the image, until either a maximum number of set iterations has been performed
50, or a maximum coverage threshold is reached set to 1.0. Performing an unsupervised classification is simpler than a supervised classification, because the
cluster signatures are automatically generated by the ISODATA algorithm. The user must predetermine the number of iterations and number of resulting clusters
classes. In this research, separate ISODATA runs were carried out to define 5 to 50 classes with interval of 5.
In each run the desired number of classes is produced by the ISODATA clustering Algorithm. The divergence statistical measure of distance ERDAS,
2003 between defined cluster signatures by run was used to compare the various runs. The best run with a clear distinguished peak in the divergence separability
was selected for further study. After the ISODATA clustering was performed, the image will have result
on many types of land cover on research location. The result from ISODATA still gives a coarse classification and still has too many classes; this is why supervised
classification still needs to be performed to enhance the result. The supervised classification was done by identifying phenology similarity between classes.
Classes that have similar phenology pattern then were combined into one group and so on until each of the groups shows different phenology pattern.
3.3.3 Rice Calendar Regionalization
Rice calendar data is used to identify the growing stages of rice in Karawang, Subang, and Indramayu Regency. The combination between rice
calendar data and Landuse map of West Java Province will provide the information of rice growing season in each area of west java or in other word we
could regionalized growing season of rice in research area.
In this research, growing season of rice will be identified by the MODIS EVI value of rice field that shows the green leaf of rice plantations. Temporal
MODIS EVI data that show the greenest index of rice plantation will give the picture of rice plantation age over the research area and the multi temporal pattern
of MODIS EVI then were analyzed to see the shifting EVI value in research area to determine the rice growing season start and end in several location of research
area.
3.4 Regression Analysis
The multiple linear regression analysis was used to generate the regression equations. Assuming that variables and crop area statistics were independent of
each other and that crop area is a linear combination of multiple predictors, the multiple linear regressions of the districts were carried out using the selected
predictors by the stepwise regression. Multiple linear regressions are used, when y is considered a function of
more than one independent x variables. Stepwise regression removes and adds variables to the regression model for the purpose of identifying a useful subset of
the predictors. The main approaches are:
- Forward selection, which involves starting with no variables in the model,
trying out the variables one by one and including them if they are statistically significant.
- Backward elimination, which involves starting with all candidate variables
and testing them one by one for statistical significance, deleting any that are not significant.
The statistical and spatial data pre-processing, provided a data set in tabular format of the EVI classes. The classes were as the predictors with the consolidated
crop area statistic as a response variable. The multiple stepwise regression analysis was used to select variables that were significant distributors of crop area.
The predictors were subjected to a no constant stepwise linear regression in order to eliminate variables that were not significant to the regression. The elimination
was carried out iteratively first entering the variable that explains the most variance in the data, until no more variables could be eliminated. Selected
variables that yielded a negative coefficient value though significant were eliminated since they seemed to suggest that negative crop area is existent. The
selected variables were then used to run the multiple linear regressions for the different crop areas. Only significant predictors of crop area were included in the
multiple linear regression models. The results of the multiple linear regressions were coefficients that had to be spatially distributed in the form of value maps.