161
Space Analog’s Searching
to Improve Deterministic
Forecasting Using Analog
Ensemble Method Over Morocco
Badreddine Alaoui, Driss Bari and Yamna Ghabbar
Abstract
The forecasting of surface weather parameters is a key
stone to every weather and climate application, and it is
mainly conducted by deterministic approaches, where
several uncertainty sources are faced. These limita-
tions enhanced the reflection toward using ensemble
forecasting methods, which provides helpful tools for
decision-making. In this work, we aim to develop a low
computational cost ensemble forecasting approach based
on the analog ensemble method (AnEn), to predict six
surface parameters (T2m, WS10m, WD10m, RH2m,
SURFP, and MSLP) at 10 airports of Morocco for 24
forecast hours. For this goal, we use hourly observations
and forecasts from the operational mesoscale numerical
model AROME covering the 4-year period (2016–2019).
The latter was split into training (2016–2018) and test-
ing (2019) periods. In AnEn, the selection of analogs
commonly considers only historical data for each grid
point in the study domain closest to the observation site.
Herein, we propose two novelties: firstly, a new weight-
ing strategy for predictors where we use three machine
learning algorithms (linear regression, XGBoost, and
random forest) to assign predictors’ weights. Secondly,
since AnEn requires larger training dataset to enhance
chances to find best analogs, we extend the search space
by integrating neighboring grid points. Thus, the analog
detection is based here on 16 nearest grid points. As a
result of the combination of these two techniques, it is
found that the machine learning weighting strategies
proved an improvement of performances in bias and root
mean square error for different lead times and locations.
Since the new space neighboring strategy maximizes
the chances to find the best analogs, clear improvements
were perceived for most airports. However, perfor-
mances remain geographically dependent. In some air-
ports, where topography is heterogeneous, applying this
new analog searching strategy might lead to some wors-
ening since weather conditions can vary at a hectometric
scale.
Keywords
Analog ensemble · Machine learning · Space
neighborhood · AROME · Space analog search
1 Introduction
Weather surface parameters play a key role in all weather
and climate applications. For most world weather centers,
predicting those parameters is based on determinism which
raises several limitations and uncertainty sources (Eckel &
Mass, 2005; Orrell et al., 2001; Paimazumder & Mölders,
2009). Ensemble forecasting provides a reliable alterna-
tive since it covers all the potential prediction scenarios
(Leith, 1974), but compared to determinism, the latter
requires high computational capacities. Analog ensemble
method is one of the most effective and low-cost ensemble
methods. This method has shown high success in several
applications (Alessandrini et al., 2018; Davò et al., 2016;
Delle Monache et al. 2013). The fundamental idea behind
is to look for all the similar weather situations to the cur-
rent one in a set of past forecasts, analogs, provided by
the same numerical weather prediction (NWP) model. For
a given location, whenever the most analogous past situa-
tions are defined, their associated observations construct our
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024
H. Chenchouni et al. (eds.), Recent Advancements from Aquifers to Skies in Hydrogeology, Geoecology, and Atmospheric Sciences,
Advances in Science, Technology & Innovation, https://doi.org/10.1007/978-3-031-47079-0_36
B. Alaoui (*) · D. Bari
Direction Générale de la Météorologie, CNRM, Casablanca,
Maroc
e-mail: alaoui.badreddine.abe@gmail.com
B. Alaoui · Y. Ghabbar
Hassania School of Public Works, LaGeS/MoNum, Casablanca,
Maroc