The climate hazard group specializes in applying climatology
to problems of food security. We act as intermediaries between institutions
specializing in forecasts, such as the NOAA climate prediction center (
CPC ), and users
of climate forecast and monitoring data, such as the agronomists and economists
evaluating food security for the Famine Early Warning System Network (
www.fews.net ). CHG
has evolved over the last 4 years as part of the ongoing collaboration between
the USGS
international
program and the University of California Santa Barbara Geography department.
The current status of this evolution involves the production of existing CHG
products for consumption via this website, and the development of new approaches
to rainfall prediction and evaluation.
MRF Standard Precipitation Index
The standard precipitation index is a statistical translation of rainfall
values, rescaled to a range and distribution of values similar to those
of a normal distribution with a mean of zero and a standard deviation of
1. This transformation makes it easy to identify values with typical
rainfall (they'll be close to zero). Exceptionally wet or dry areas
will have values greater than +/- 2.0.
The medium range forecast (MRF) SPI map is a comparison of the projected
rain for the following 7 days with the same 7 days in the Collaborative
Historical African Rainfall Model (CHARM) historical database. The
MRF forecast rainfall fields were obtained from the Climate Prediction Center
(
CPC ). An
SPI value greater than zero indicates that the projected rains are wetter
than usual for those 7 days, while a negative SPI values indicates a drier-than-normal
accumulation for the upcoming 7 days. A larger absolute value (i.e.
greater positive value, or less negative value) indicates a more extreme event
for the upcoming 7 days. For caveats about the appropriate use of this
data, please click
here.
RFE Standard Precipitation Index
The standardized precipitation index is a statistical translation of rainfall
values, rescaled to a range and distribution of values similar to those
of a normal distribution with a mean of zero and a standard deviation of
1. This transformation makes it easy to identify values with typical rainfall
(they'll be close to zero). Exceptionally wet or dry areas will have values
greater than +/- 2.0.
The RFE rainfall is modeled rainfall based on a variety of atmospheric
and surface parameters and is produced every 10 days. The MRF SPI compares
the 10-day accumulation with a theoretical distribution derived from the
Collaborative Historical African Rainfall Model (CHARM) historical database.
The RFE dekadal data is downloaded from the Africa Data Dissemination Service
(
ADDS). Aggregating
multiple dekadal files allows for comparisons of current data at a longer
time frame. The number of dekads to be aggregated is variable and can be done
up to multi-year scales. Regardless of accumulation period, the normalizing
of the SPI means that even for longer time periods an SPI value greater than
zero indicates that the modeled rains from the RFE are wetter than usual for
those 10 days (when compared to the CHARM for the identical 10 days), while
negative SPI values indicate drier than normal conditions for the 10 days.
A larger absolute value (i.e. greater positive value, or less negative value)
indicates a more extreme event over the 10 day period.
Forecast Interpretation Tool
Use of the theoretical distribution to represent the likelihood of rainfall
accumulations allows for a link to be made between rainfall accumulations
and the probability of that accumulation being realized. This information
is potentially important to a variety of individuals and agencies.
This information can currently be supplied based only on the historical data.
With this new method forecast probabilities can be used to provide predictive
assessments based on current environmental factors affecting precipitation.
Integrating a forecast map with the historical precipitation amounts could
provide a link between historical probabilities and potential rainfall accumulations
for an upcoming growing season.
The proposed method for integrating historical rainfall distributions
and probability forecasts uses four major steps in arriving at new distribution
parameters reflecting the rainfall probability given by the forecast information.
First the tercile breaks are established based on the historical probability
distribution function (pdf) calculated at each pixel. Second, values
are randomly drawn from each tercile in the proportions described by the
forecast map based on the original distribution function. The new values
are then used to calculate a new shape and scale parameter given the proportionately
drawn values. By performing the second and third step several times,
it is possible to create a distribution of gamma distribution parameters
that can be used to describe the new rainfall distribution function.
This new pdf represents a synthesis of historical data and forecast probabilities,
and can be used to describe the probability of given rainfall values in a
variety of ways including likely rainfall accumulation ranges, or the likelihood
of receiving less than a particular critical accumulation.
Publications
