each county in the study area was masked out, and based on the total area of corn CDL
pixels contained within the NDVI pixel, the percentage of corn cover was calculated
for each NDVI pixel. For the purpose of this case study, only NDVI pixels with over
95% corn cover were identified as pure corn pixels.
16.4.4.2 Normal NDVI
Normal NDVI refers to a single NDVI curve representing the annual crop production
within a county. These representative NDVI time series are derived by analyzing
identified crop NDVI data within the same county. Based on the pure pixels
identified in the previous section, daily NDVI for all these pure pixels was acquired,
and the normal NDVI value for each day was then calculated by taking the median of
all pure pixel NDVI values for the same day. For example, the normal NDVI value
for day 5 of the normal corn NDVI time series was calculated by taking the median
value of all available day 5 NDVI values from the selected corn pure pixels within
the county. Hence, for each year and each county, there will a single NDVI curve for
each crop representing that crop’s normal NDVI.
16.4.4.3 NDVI Smoothing
As described in the previous section, before utilizing the NDVI for any further
analysis, one of the most important tasks is to remove any data noises within the
time series. It is extremely important to smooth these time series in order to achieve
impartial and accurate results. Daily NDVI, compared to composite NDVI, tends
to have more impurities in the data and requires a more intense smoothing
approach. In this case study, two levels of statistical filtering approach, Best
Index Slope Extraction (BISE) with Savitzky-Golay (SavGol), were carried out
to smooth the daily NDVI. Even though there are other various statistical filtering
methods to remove impurities from the NDVI time series such as running average,
asymmetric Gaussian function, and double-Sigmoidal function, however, the
two-step combination filtering approach using BISE as the first-level filter and
SavGol as the second-level filter provides the best NDVI smoothing results
(Rahman et al. 2016; Shrestha et al. 2017).
BISE is based on a threshold approach that utilizes a fixed moving window to
traverse through the NDVI time series and only accepts the value if it fulfills the
given threshold (Viovy et al. 1992). Similarly, SavGol is based on the weighted
moving average filtering method where the weight is provided as a polynomial of a
certain degree (Chen et al. 2004). In the first level of BISE filtering for this case
study, the moving window of 40 days sliding period was selected as the threshold,
and for the second level of SavGol filtering, the second degree of polynomial
function was used. Besides smoothing the NDVI, the growing season for each
16 Flood Monitoring and Crop Damage Assessment
339
pixels contained within the NDVI pixel, the percentage of corn cover was calculated
for each NDVI pixel. For the purpose of this case study, only NDVI pixels with over
95% corn cover were identified as pure corn pixels.
16.4.4.2 Normal NDVI
Normal NDVI refers to a single NDVI curve representing the annual crop production
within a county. These representative NDVI time series are derived by analyzing
identified crop NDVI data within the same county. Based on the pure pixels
identified in the previous section, daily NDVI for all these pure pixels was acquired,
and the normal NDVI value for each day was then calculated by taking the median of
all pure pixel NDVI values for the same day. For example, the normal NDVI value
for day 5 of the normal corn NDVI time series was calculated by taking the median
value of all available day 5 NDVI values from the selected corn pure pixels within
the county. Hence, for each year and each county, there will a single NDVI curve for
each crop representing that crop’s normal NDVI.
16.4.4.3 NDVI Smoothing
As described in the previous section, before utilizing the NDVI for any further
analysis, one of the most important tasks is to remove any data noises within the
time series. It is extremely important to smooth these time series in order to achieve
impartial and accurate results. Daily NDVI, compared to composite NDVI, tends
to have more impurities in the data and requires a more intense smoothing
approach. In this case study, two levels of statistical filtering approach, Best
Index Slope Extraction (BISE) with Savitzky-Golay (SavGol), were carried out
to smooth the daily NDVI. Even though there are other various statistical filtering
methods to remove impurities from the NDVI time series such as running average,
asymmetric Gaussian function, and double-Sigmoidal function, however, the
two-step combination filtering approach using BISE as the first-level filter and
SavGol as the second-level filter provides the best NDVI smoothing results
(Rahman et al. 2016; Shrestha et al. 2017).
BISE is based on a threshold approach that utilizes a fixed moving window to
traverse through the NDVI time series and only accepts the value if it fulfills the
given threshold (Viovy et al. 1992). Similarly, SavGol is based on the weighted
moving average filtering method where the weight is provided as a polynomial of a
certain degree (Chen et al. 2004). In the first level of BISE filtering for this case
study, the moving window of 40 days sliding period was selected as the threshold,
and for the second level of SavGol filtering, the second degree of polynomial
function was used. Besides smoothing the NDVI, the growing season for each
16 Flood Monitoring and Crop Damage Assessment
339