275
Precipitation Estimate Using NEXRAD Ground-Based Radar Images
and lower relative difference compared with rain gauge measurements than the Stage
III data. Overall, Stage III overestimated ~20% precipitation (in 2001), whereas
MPE underestimated 7% (in 2004). Yet, Westcott et al. (2008) showed that MPE
underestimated average county-level monthly precipitation over nine states in the
midwestern United States. At a daily temporal scale, MPE overestimated precipitation depth for precipitation events with low values and predicted similar or lower
precipitation depth for precipitation events with larger values (Westcott et al. 2008).
Young and Brunsell (2008) evaluated both Stage III (1998–2002) and MPE (2003–
2004) precipitation products using approximately 1200 rain gauges in the Missouri
River Basin. They found that MPE performed better than Stage III for warm seasons
but worse than Stage III in cold seasons. Despite the overall improvement of MPE
over Stage III, MPE bias reached ~40% in cold seasons and ~20% in warm seasons.
The spatial mismatch between the scale of precipitation estimates by NEXRAD
and rain gauges is worth noting. Precipitation variability at a small spatial scale
reported in previous research (e.g., Krajewski et al. 2003) may lead to inconclusive
and misleading comparisons between the areal NEXRAD estimates and observations from a single rain gauge (Kitchen and Blackall 1992; Habib et al. 2009). The
subgrid variability effect was emphasized by Young et al. (2000). Wang et al. (2008)
evaluated a case using the precipitation from the nine NEXRAD grids surrounding
the rain gauge and suggested to limit NEXRAD and a single rain gauge comparison
under uniform precipitation events when a dense rain gauge network is not available.
Recently, Habib et al. (2009) have highlighted the importance of using a dense rain
gauge network to validate NEXRAD. The comparison of MPE products with pixelaverage gauge precipitation is expected to reduce the effect of single-gauge uncertainty and lead to more accurate evaluation of NEXRAD errors (Habib et al. 2009).
By using a dense rain gauge network in south Louisiana, Habib et al. (2009) found that
the bias between MPE and rain gauge observations is very small over an annual scale;
however, the bias reached ±25% of the total precipitation depth for half of the events
and exceeded 50% for 10% of events in 2004–2006. The large differences between
NEXRAD and rain gauge observations are expected to have significant implication
for the application of NEXRAD data and evaluation of the quality of NEXRAD
precipitation products; therefore, necessary corrections must be made before their
application in earth system modeling (Jayakrishnan et al. 2004). Although NEXRAD
provides precipitation data with much better spatial sampling frequencies, compared
with rain gauges, the estimates from NEXRAD are less accurate.
Efforts were exerted to improve the accuracy of NEXRAD using rain gauge measurements using various numerical schemes of interpolation. Seo (1998) and Seo et
al. (1990, 1999) used cokriging and simple kriging with varying local means (SKlm)
methods to correct NEXRAD precipitation products using rain gauge observations.
Steiner et al. (1999) applied a BA method to correct NEXRAD in Goodwin Creek,
a small research watershed in northern Mississippi. Haberlandt (2007) applied kriging with external drift (KED) and indicator kriging with external drift (IKED) for
the spatial interpolation of hourly precipitation from rain gauges using additional
information from radar, which clearly outperformed the univariate interpolation
methods. Li et al. (2008) developed a linear regression–based kriging method to
calibrate daily NEXRAD precipitation using rain gauge data and applied it in Texas
Precipitation Estimate Using NEXRAD Ground-Based Radar Images
and lower relative difference compared with rain gauge measurements than the Stage
III data. Overall, Stage III overestimated ~20% precipitation (in 2001), whereas
MPE underestimated 7% (in 2004). Yet, Westcott et al. (2008) showed that MPE
underestimated average county-level monthly precipitation over nine states in the
midwestern United States. At a daily temporal scale, MPE overestimated precipitation depth for precipitation events with low values and predicted similar or lower
precipitation depth for precipitation events with larger values (Westcott et al. 2008).
Young and Brunsell (2008) evaluated both Stage III (1998–2002) and MPE (2003–
2004) precipitation products using approximately 1200 rain gauges in the Missouri
River Basin. They found that MPE performed better than Stage III for warm seasons
but worse than Stage III in cold seasons. Despite the overall improvement of MPE
over Stage III, MPE bias reached ~40% in cold seasons and ~20% in warm seasons.
The spatial mismatch between the scale of precipitation estimates by NEXRAD
and rain gauges is worth noting. Precipitation variability at a small spatial scale
reported in previous research (e.g., Krajewski et al. 2003) may lead to inconclusive
and misleading comparisons between the areal NEXRAD estimates and observations from a single rain gauge (Kitchen and Blackall 1992; Habib et al. 2009). The
subgrid variability effect was emphasized by Young et al. (2000). Wang et al. (2008)
evaluated a case using the precipitation from the nine NEXRAD grids surrounding
the rain gauge and suggested to limit NEXRAD and a single rain gauge comparison
under uniform precipitation events when a dense rain gauge network is not available.
Recently, Habib et al. (2009) have highlighted the importance of using a dense rain
gauge network to validate NEXRAD. The comparison of MPE products with pixelaverage gauge precipitation is expected to reduce the effect of single-gauge uncertainty and lead to more accurate evaluation of NEXRAD errors (Habib et al. 2009).
By using a dense rain gauge network in south Louisiana, Habib et al. (2009) found that
the bias between MPE and rain gauge observations is very small over an annual scale;
however, the bias reached ±25% of the total precipitation depth for half of the events
and exceeded 50% for 10% of events in 2004–2006. The large differences between
NEXRAD and rain gauge observations are expected to have significant implication
for the application of NEXRAD data and evaluation of the quality of NEXRAD
precipitation products; therefore, necessary corrections must be made before their
application in earth system modeling (Jayakrishnan et al. 2004). Although NEXRAD
provides precipitation data with much better spatial sampling frequencies, compared
with rain gauges, the estimates from NEXRAD are less accurate.
Efforts were exerted to improve the accuracy of NEXRAD using rain gauge measurements using various numerical schemes of interpolation. Seo (1998) and Seo et
al. (1990, 1999) used cokriging and simple kriging with varying local means (SKlm)
methods to correct NEXRAD precipitation products using rain gauge observations.
Steiner et al. (1999) applied a BA method to correct NEXRAD in Goodwin Creek,
a small research watershed in northern Mississippi. Haberlandt (2007) applied kriging with external drift (KED) and indicator kriging with external drift (IKED) for
the spatial interpolation of hourly precipitation from rain gauges using additional
information from radar, which clearly outperformed the univariate interpolation
methods. Li et al. (2008) developed a linear regression–based kriging method to
calibrate daily NEXRAD precipitation using rain gauge data and applied it in Texas