NDVI data were taken from multi-temporal Terra MODIS VIs Product. The data were
obtained from the NASA Warehouse Inventory Search Tool (WIST) (https://wist.echo.
nasa.gov/api/). The dataset was from the year 2001 to 2009. The product name is the
MOD13Q1 VI which is a 16-day composite with resolution 250 m. The MODIS VIs
product takes two tiles to cover the region; therefore, image mosaic technique was applied
to generate a single image for each 16-day composite. The digital image preprocessing
steps such as radiometric and geometric correction were applied to process raw data. For
radiometric correction, the MODIS images were registered from the optical Terra/MODIS
sensor and cloud contamination was a problem for data acquisition during the rainy season.
The radiometric correction was treated to mitigate this effect resulting in data missing for
some dates especially in the rainy season from June to July. 16-day composite of NDVI
data was averaged into monthly data to calculate VCI. The format of MODIS images is
distributed as Hierarchical Data Format (HDF) with 10 9 10 arc degree-tiles and projected
in the sinusoidal projection. The geometric correction was applied to resample the raw data
into UTM WGS84 datum.
VCI was developed to identify the degree of greenness of each pixel of NDVI values in
relation to the average greenness at the same location overtime. The calculation of VCI is
described in Eq. (1), where NDVImin and NDVImax refer to the absolute minimum and
maximum NDVI of the study period.
VCI ¼ 100 ðNDVI NDVIminÞ=ðNDVImax NDVIminÞ ð1Þ
This study applied the MOD13Q1 VI product because the advantages in the ability to
monitor vegetation condition (Huete et al. 2002). This product was appropriate for
vegetation study at the regional scale in northeast Thailand in terms of spatial and temporal
resolution. The monthly VCI value corresponding to the sample sites was calculated from
the mean of selected 3 9 3 pixel (resolution 250 m per pixel) of each site covering area
about 0.5 km2
. From this step, the monthly VCI at all test sites from year 2001 to 2009 was
processed to analyze drought impact on vegetation at the different land cover type.
Fig. 3 Location of meteorological stations in Northeast Thailand
Nat Hazards
123
2.4 Association rules for drought event
The relationship between SPI and VCI was identified by association rules to find the
occurrence of drought from 2001 to 2009. Association rules are used to identify the
relationship between data items and characterized by two parameters: support and confi-
dence (Dhanya and Nagesh Kumar 2009). Association rule is the pattern of the form
X ? Y where X is the rule antecedent and Y is its consequent and X Y ¼ 0. Support {X}
is a statistical significant measurement and defined by the ratio of the count of the events in
X divided by the total number of the events. The confidence of the rule is defined as the
ratio of support fX [ Yg, which is equal to the percentage of events that occurs in both
X and Y, divided by the support {X}. For the rule X ? Y, the support of antecedent is
called the rule coverage and measures how often it should occur in the database, while the
confidence measures the strength of the rule (Shekhar and Chawla 2003). To assess the
descriptive interest of a rule, Lift is used to measure how far from independence is
X ? Y. The values of Lift range from 0 to ??. X and Y are independent if the Lift’s value
closes to 1, and the rule is not interesting (Azevedo and Jorge 2007).
Monthly SPI at the different timescales and VCI values were converted into discrete
representation in which SPI values were aggregated from seven categories into five
categories. The VCI values were also transformed into five clusters in order to be comparable
and identify drought patterns related to SPI values. The discretization criteria were
based on Sharma (2006) as shown in Table 1.
The sliding window width for both dataset was 1 month. Drought pattern was detected
from the target episodes for extremely dry and dry categories for SPI, and very low and low
for VCI. The minimum support was determined to measure the minimum frequency of
drought episodes in the database. This study searched for drought episodes which occurred
at least 10 % of the time. Association rules were generated using apriori algorithm (Lai
et al. 2006), and the best rules were selected based on the criteria above
NDVI data were taken from multi-temporal Terra MODIS VIs Product. The data wereobtained from the NASA Warehouse Inventory Search Tool (WIST) (https://wist.echo.nasa.gov/api/). The dataset was from the year 2001 to 2009. The product name is theMOD13Q1 VI which is a 16-day composite with resolution 250 m. The MODIS VIsproduct takes two tiles to cover the region; therefore, image mosaic technique was appliedto generate a single image for each 16-day composite. The digital image preprocessingsteps such as radiometric and geometric correction were applied to process raw data. Forradiometric correction, the MODIS images were registered from the optical Terra/MODISsensor and cloud contamination was a problem for data acquisition during the rainy season.The radiometric correction was treated to mitigate this effect resulting in data missing forsome dates especially in the rainy season from June to July. 16-day composite of NDVIdata was averaged into monthly data to calculate VCI. The format of MODIS images isdistributed as Hierarchical Data Format (HDF) with 10 9 10 arc degree-tiles and projectedin the sinusoidal projection. The geometric correction was applied to resample the raw datainto UTM WGS84 datum.VCI was developed to identify the degree of greenness of each pixel of NDVI values inrelation to the average greenness at the same location overtime. The calculation of VCI isdescribed in Eq. (1), where NDVImin and NDVImax refer to the absolute minimum andmaximum NDVI of the study period.VCI ¼ 100 ðNDVI NDVIminÞ=ðNDVImax NDVIminÞ ð1ÞThis study applied the MOD13Q1 VI product because the advantages in the ability tomonitor vegetation condition (Huete et al. 2002). This product was appropriate forvegetation study at the regional scale in northeast Thailand in terms of spatial and temporalresolution. The monthly VCI value corresponding to the sample sites was calculated fromthe mean of selected 3 9 3 pixel (resolution 250 m per pixel) of each site covering areaabout 0.5 km2. From this step, the monthly VCI at all test sites from year 2001 to 2009 wasprocessed to analyze drought impact on vegetation at the different land cover type.Fig. 3 Location of meteorological stations in Northeast ThailandNat Hazards1232.4 Association rules for drought eventThe relationship between SPI and VCI was identified by association rules to find theoccurrence of drought from 2001 to 2009. Association rules are used to identify therelationship between data items and characterized by two parameters: support and confi-dence (Dhanya and Nagesh Kumar 2009). Association rule is the pattern of the formX ? Y where X is the rule antecedent and Y is its consequent and X Y ¼ 0. Support {X}is a statistical significant measurement and defined by the ratio of the count of the events inX divided by the total number of the events. The confidence of the rule is defined as theratio of support fX [ Yg, which is equal to the percentage of events that occurs in bothX and Y, divided by the support {X}. For the rule X ? Y, the support of antecedent iscalled the rule coverage and measures how often it should occur in the database, while theconfidence measures the strength of the rule (Shekhar and Chawla 2003). To assess thedescriptive interest of a rule, Lift is used to measure how far from independence isX ? Y. The values of Lift range from 0 to ??. X and Y are independent if the Lift’s valuecloses to 1, and the rule is not interesting (Azevedo and Jorge 2007).Monthly SPI at the different timescales and VCI values were converted into discreterepresentation in which SPI values were aggregated from seven categories into fivecategories. The VCI values were also transformed into five clusters in order to be comparableand identify drought patterns related to SPI values. The discretization criteria werebased on Sharma (2006) as shown in Table 1.The sliding window width for both dataset was 1 month. Drought pattern was detectedfrom the target episodes for extremely dry and dry categories for SPI, and very low and lowfor VCI. The minimum support was determined to measure the minimum frequency ofdrought episodes in the database. This study searched for drought episodes which occurredat least 10 % of the time. Association rules were generated using apriori algorithm (Laiet al. 2006), and the best rules were selected based on the criteria above
การแปล กรุณารอสักครู่..
