3. Standardized Runoff Index
We define the standardized runoff index (SRI) as the unit
standard normal deviate associated with the percentile of
hydrologic runoff. Calculation of SRI follows the general approach
employed by Mckee et al. (1993) to estimate the SPI.
Like the SPI, the SRI can be calculated for runoff totals accumulated
over different durations (e.g., 1-month, 9-month),
and for different spatial aggregations depending on source
runoff data resolution and desired application. SPIs, for instance,
are calculated by NOAA on a climate division basis
and by state agencies on a county level basis. The procedure
for calculating the SRI includes the following steps:
1. A retrospective time series of runoff at the desired temporal
and spatial level of aggregation is obtained by simulation,
and a probability distribution is fit to the sample
represented by the time series values.
2. The distribution is used to estimate the cumulative
probability of the accumulated runoff value of interest (either
the current value or one from a retrospective date.
3. The cumulative probability is converted to a standard
normal deviate (with zero mean and unit variance), which
can either be calculated from a numerical approximation
to the normal cumulative distribution function (CDF) or
extracted from a table of values for the normal CDF that
is readily available in statistics textbooks or on the World
Wide Web.
McKee et al. (1993) select the Gamma distribution for
fitting monthly precipitation data series, and suggest that
the procedure can be applied to other variables relevant to
drought, e.g., streamflow or reservoir contents. In pursuing
this suggestion for runoff, we note that distributions
other than the Gamma may be more appropriate, depending
on the runoff variable’s retrospective sample characteristics
(especially skew and kurtosis), which vary greatly by geographic
location. Figure 1 shows the example of a 3-month
total simulated runoff during March, April and May from
the years 1955-2005, averaged over the area of the Feather
River basin. The Gamma and two-parameter lognormal
(LN2) distributions for the sample are plotted, and the figure
makes clear that any value of runoff can equally well be
expressed in terms of its percentile (top axis) or the standardized
index (bottom axis). Here the LN2 distribution
provides a better fit overall than the Gamma distribution,
although the Gamma distribution may perform better for
low values of runoff. The 3-parameter lognormal and Generalized
Extreme Value distributions may have even better
general applicability for runoff over widely varying hydroclimatic
regimes. The use of a fitted distribution for runoff
carries the risk that a poor fit is achieved, causing percentiles
and hence SRI values to be misestimated. One alternative
is to avoid the distribution selection by empirically estimating
percentiles, although the risk of misestimation due to
sampling uncertainties remains. Care must be taken to negotiate
the first two steps above so as to minimize errors
in estimating the probability of runoff, particularly in arid
regions where runoff may be intermittent.