WMOD00482

• Title: Microphysical Observations Over the Atlas Mountains in Morocco
• Date: 12/23/1988
• Weather Modification - Doc Type: Report

the horizontal and 600 mb in the vertical.
Dynamical and microphysical structures give
the general airflow over the mountains,
estimate the ice particle growth rates and fall
velocities, and predict the chain of events
leading from nucleation to precipitation.
A comparison of model wind and temperature
estimates with aircraft measurements was made
over the target area on 24 February 1988. The
errors were small in the 100-km domain
upwind mountain crest; wind directions and
temperatures were closely reproduced. The
largest errors were downwind of the barrier
crest.

4. STATISTICAL DESIGN AND
EVALUATION

The development and implementation of
scientific evaluation procedures have been
considered to be essential for the success of the
project since its inception. The Government of
Morocco set up a scientific analysis and
evaluation team. with a specific area of work
assigned to each scientist. A USBR scientist
was assigned to collaborate with the Moroccan
scientists in each area of responsibility. Much
of this work has been reported elsewhere
(Baddour and Rasmussen, 1988; Baddour et aL,
1989; EI Majdoub et al., 1989). Therefore, this
section focuses on the statistical evaluation.
Severe drought has forced the Moroccan
program to operational cloud seeding where all
potential precipitation cases are treated when
feasible. This generaIJy limits the statistical
design choices to target only or target control,
with historical data supplying the nontreated
sample. The lack of randomization means that
any P values obtained are only estimates of
P values as defined in classical statistics.
The target area (fig. 1) for the Moroccan
program was selected based on a major water
source to the Oued Oum Er Rbia Basin and
good geographic and topographic characteristics
for cloud seeding. A -control area, the upper
reaches of the eastern Tensift Basin (fig. 1),
was selected because it is upwind of the target
for most of the important winter storms,
contamination by the cloud seeding is unlikely,
and its precipitation is well correlated with that
of the target area.
Available target and control historical
data consisted of daily streamflow. No
precipitation data were available for high
elevation sites within the target area.
A target-control design with streamflow
as the primary response variable was selected
for the Moroccan program. Evaluation is to be
performed with monthly and seasonal time
units. Hydrographs for target and control

steamflow indicated that the primary runoff
period is November through June while the
spring melt from high elevations starts in
March and trails to a base flow in July. Fast
streamflow response to some storms was
indicated suggesting that monthly analyses
might be productive.
A secOIid variable, volume of
precipitation obtained from application of the
Rhea orographic precipitation model (Rhea,
1978), was found useful in predicting target
streamflow. The model is two-dimensional,
steady-state, and multilayer. Input is data from
rawinsonde ascents. Adaptation of the model
to Moroccan terrain and weather conditions is
discussed elsewhere in these proceedings by
EI Majdoub et al.
With model-produced volume
precipitation integrated over the target area for
the November to April period and target and
control streamflow summed over the November
to July period, the variance explained by the
control streamflow and the model output was
86% by least squares multiple regression. The
model-generated volume precipitation for
monthly time periods was not well correlated
with measured flows since much of the
high-elevation precipitation, which occurs
during tbe December to March period, does
not TUn off until April tbrough July.
Consequently, the model does not appear
useful in predicting monthly flows.
Comparisons with a few mid-level
elevation stations did indicate good capability
by the model to predict monthly precipitation.
Correlations for individual stations were
typically 0.7 to 0.85 (EI Majdoub et aL, these
proceedings. )
The LAD (lease-absolute-deviations)
n:gression and MRPP (multiresponse
pl~rmutation procedures), as described by
Mielke and Medina (1982. 1987), were selected
to yield evaluation procedures that are applied
through computer programs. Estimates of
P values (estimates because of the lack of
randomization) are produced. Estimates of
treatment effects are obtained by use of the
well-known double ratio.
Values of the project duration were
obtained according to several probability-of-
detection level settings, given 10 and 15%
simulated increases. The LAD- and MRPP-
based evaluation procedures were employed in
a rerandomization approach where 100 samples
were selected at random from the historical
data and a treatment amount applied to target
quantities. The P values were estimated from
the modified samples and the initial historical
data set then employed to obtain the
probability-of-detection, defined as (number of