Seasonal prediction of lightning activity in North Western Venezuela

Atmospheric Research 172–173 (2016) 147–162
Contents lists available at ScienceDirect
Atmospheric Research
journal homepage: www.elsevier.com/locate/atmosres
Seasonal prediction of lightning activity in North Western Venezuela:
Large-scale versus local drivers
Á.G. Muñoz a,b,⁎, J. Díaz-Lobatón a, X. Chourio a, M.J. Stock a
a
b
Centro de Modelado Científico (CMC), Universidad del Zulia, Maracaibo 4004, Venezuela
International Research Institute for Climate and Society (IRI), Earth Institute, Columbia University, NY, USA
a r t i c l e
i n f o
Article history:
Received 23 July 2015
Received in revised form 8 December 2015
Accepted 23 December 2015
Available online 8 January 2016
Keywords:
Venezuela
Lightning variability
Catatumbo Lightning
Atmospheric electricity
Nocturnal Low Level Jet
a b s t r a c t
The Lake Maracaibo Basin in North Western Venezuela has the highest annual lightning rate of any place in the
world (~200 fl km−2 yr−1), whose electrical discharges occasionally impact human and animal lives (e.g., cattle)
and frequently affect economic activities like oil and natural gas exploitation. Lightning activity is so common in
this region that it has a proper name: Catatumbo Lightning (plural). Although short-term lightning forecasts are
now common in different parts of the world, to the best of the authors' knowledge, seasonal prediction of lightning activity is still non-existent. This research discusses the relative role of both large-scale and local climate
drivers as modulators of lightning activity in the region, and presents a formal predictability study at seasonal
scale.
Analysis of the Catatumbo Lightning Regional Mode, defined in terms of the second Empirical Orthogonal Function
of monthly Lightning Imaging Sensor (LIS-TRMM) and Optical Transient Detector (OTD) satellite data for North
Western South America, permits the identification of potential predictors at seasonal scale via a Canonical Correlation Analysis. Lightning activity in North Western Venezuela responds to well defined sea-surface temperature
patterns (e.g., El Niño-Southern Oscillation, Atlantic Meridional Mode) and changes in the low-level meridional
wind field that are associated with the Inter-Tropical Convergence Zone migrations, the Caribbean Low Level Jet
and tropical cyclone activity, but it is also linked to local drivers like convection triggered by the topographic configuration and the effect of the Maracaibo Basin Nocturnal Low Level Jet. The analysis indicates that at seasonal
scale the relative contribution of the large-scale drivers is more important than the local (basin-wide) ones,
due to the synoptic control imposed by the former. Furthermore, meridional CAPE transport at 925 mb is identified as the best potential predictor for lightning activity in the Lake Maracaibo Basin.
It is found that the predictive skill is slightly higher for the minimum lightning season (Jan–Feb) than for the
maximum one (Sep–Oct), but that in general the skill is high enough to be useful for decision-making processes
related to human safety, oil and natural gas exploitation, energy and food security.
© 2016 Elsevier B.V. All rights reserved.
1. Introduction
Characterizing lightning activity in different geographical regions is
of great importance both for research and forecast applications
(Barnes and Newton, 1982; Court and Griffiths, 1982). There is strong
evidence pointing to a relationship between flash rate and other thunderstorm parameters, such as precipitation rate (Lee, 1990; Goodman,
1990; Baker et al., 1995). Moreover, there is a growing interest in studying the modulation of lightning distribution and frequency due to interannual phenomena, like El Niño-Southern Oscillation (ENSO) regional
teleconnections (Goodman et al., 2000; Hamid et al., 2001; Chronis
et al., 2008), or even the effect of climate change on lightning activity
⁎ Corresponding author at: International Research Institute for Climate and Society
(IRI), Earth Institute, Columbia University, NY, USA.
E-mail address: agmunoz@iri.columbia.edu (Á.G. Muñoz).
http://dx.doi.org/10.1016/j.atmosres.2015.12.018
0169-8095/© 2016 Elsevier B.V. All rights reserved.
(Goodman and Christian, 1993). One of the main issues impacting this
kind of research has been the availability of long time series of data.
The acquisition of atmospheric electricity data is difficult due to the
nature of the events. Prior research was hampered by the absence of
measurements that accurately quantify the frequency and distribution
of lightning activity in the planet (Christian et al., 2003). However, the
use of space-based sensors allows the attainment of data in a more effective way than ground-based sensors and World Meteorological Organization (WMO) thunder day statistics determined by local observers
(World Meteorological Organization (WMO), 1953). Furthermore, satellite images represent an ideal platform for investigating lightning
over large regions, and have now been available for several years. In particular, the spaceborne optical sensors Optical Transient Detector (OTD)
on the MicroLab-1 satellite, and the Lightning Imaging Sensor (LIS) of
the Tropical Rainfall Measuring Mission (TRMM) satellite now offer
lightning density data for over seventeen years (Cecil et al., 2014). For
technical details on the acquisition, characteristics, instrument
148
Á.G. Muñoz et al. / Atmospheric Research 172–173 (2016) 147–162
Caribbean Sea
I
II
Colombia
Lake
Maracaibo
III
IV
as
rin
Ba
Venezuela
Fig. 1. Lake Maracaibo Basin, in North Western Venezuela. Roman numerals identify each quadrant.
efficiency and merging process of the LIS–OTD dataset, and the associated spatial and temporal lightning density distribution see Christian et al.
(2003), Albrecht et al. (2009), Cecil et al. (2014) and references therein.
The Catatumbo region, located in the southwestern quadrant of the
Lake Maracaibo Basin (LMB) in North Western Venezuela (Fig. 1), is
the location with the highest density rate of lightning for the entire
planet (Albrecht et al., 2009; Muñoz and Díaz-Lobatón, 2011), with
flash density rates surpassing the 200 fl km− 2 yr−1. Albrecht et al.
(2009) and Bürgesser et al. (2012) have presented a characterization
of this lightning phenomenon using LIS–OTD and the WWLLN (Virts
et al., 2013) data, respectively. In this region, lightning activity has
such an extraordinary frequency that it is commonly grouped as a single
phenomenon, usually known as “Catatumbo Lightning” (plural) or
“Maracaibo's Lighthouse”1. Comparing lightning mortality rate data,
Muñoz et al. (2015a) estimated that it is about three times more probable for a person to be struck by a lightning discharge in Catatumbo
than in the continental United States. Besides the obvious implications
for human safety, lightning discharges also impact important economic
activities in the region: killing or injuring cattle in one of the most productive regions of Venezuela in terms of meat and dairy, and delaying or
interrupting oil and natural gas exploitation, in a country that holds the
world's largest proven oil reserves (~20% of global reserves). Therefore,
it is imperative to be able to forecast lightning activity in the LMB.
Short-term (i.e., from one week up to a few days in advance) lightning forecasts are becoming more and more common in several regions
of the planet (Burrows et al., 2005; McCaul et al., 2009; Shafer and
Fuelberg, 2008; Lynn et al., 2012; Zepka et al., 2014). Although
the short-term methodology and model validation that the Centro
de Modelado Científico (CMC, or Center for Scientific Modeling,
Venezuela) is developing for the LMB will be discussed elsewhere, for
the sake of completeness it is worthwhile to mention that CMC is following the International Research Institute for Climate and Society's
(IRI) Ready–Set–Go! approach (Hellmuth et al., 2011): in order to
provide useful lightning activity forecasts well in advance, a skillful
probabilistic seasonal prediction indicating the expected flash density
rate for the next three-months should be available to the decisionmakers at least one or two months prior (the Ready stage), followed
by shorter-term forecasts once the three month period has arrived (the
Set stage); this ensures that the decision-makers are continuously
aware of the expected conditions, and, if needed, can prepare communities, farmers and the gas/oil industry to take action (the Go! stage).
This research addresses the predictability study required for the
Ready stage of the approach. The datasets and methodologies are introduced in Sections 2 and 3. The description of the LMB and generalities
about the Catatumbo Lightning are presented in Section 4. The identification of potential predictors and the predictive skill for the minimum
(Jan–Feb, JF) and maximum (Sep–Oct, SO) flash density rate seasons
are discussed in Sections 5 and 6, respectively. Finally, the last section
deals with the discussion and concluding remarks.
2. Datasets
This section describes the lightning, atmospheric fields and seasurface temperature datasets used in the research. Unless otherwise
indicated, the period 1996–2013 was used for all datasets.
2.1. Lightning data
1
This phenomenon seems to be famous in Venezuela for being the “first planetary
ozone layer regenerator”. The authors have not found any direct evidence or serious study
indicating the validity of such a claim. Tropospheric ozone is actually poisonous, and the
minimum transit time for ozone (~6 months) to the stratospheric layers is much longer
than its typical lifetime (~22 days).
Two lightning datasets are used in the present study, provided by
NASA's Global Hydrology Resource Center (GHRC, http://ghrc.nsstc.
nasa.gov). The first dataset is the LIS–OTD merged 2.5° Low Resolution
Time Series (LRTS), at monthly resolution. The second dataset is the
Á.G. Muñoz et al. / Atmospheric Research 172–173 (2016) 147–162
LIS Granule Science Dataset (see (Goodman et al., 2007) and http://ghrc.
nsstc.nasa.gov); it has 4–7 km resolution and was only used in this work
to compute monthly lightning climatologies (1998–2013) for each one
of the quadrants in the LMB, as well as minor analyses and comparisons
with the LRTS dataset.
Details on the OTD and LIS-TRMM lightning sensors, data and quality
control procedures are provided by several authors (Christian et al.,
2000; Cecil et al., 2014; Boccippio et al., 2002; Boccippio et al., 2000;
Goodman and Boeck, 2007). Differences in the magnitudes between
the two datasets are due to their different spatial resolutions and the
post-processing method used in each case.
2.2. Atmospheric fields
This research explores a subset of the physical variables studied by
Díaz-Lobatón (2012) as potential predictors for lightning activity in
the LMB. Sections 5 and 6 further elaborate on the selection of these
variables.
The components of the wind field (u , v ,w) and Convective Available
Potential Energy (CAPE) were used to identify potential predictors;
other variables like specific humidity and temperature were also considered in the analysis to provide a better understanding of the physical
processes. The fields were extracted from the 20th Century Reanalysis
(20CR, e.g., (Compo and Coauthors, 2011)); a high resolution model
output produced with the Weather and Research Forecast (WRF)
model: the North Western South America Retrospective Simulation
(NOSA30k, for details see Muñoz and Recalde (2010) and Muñoz and
Díaz-Lobatón (2011)); and the operational Climate Forecast System
Version 2 (CFSv2) forecasts (Saha and Coauthors, 2014). All these
datasets are freely available via the IRI's Data Library (http://iridl.ldeo.
columbia.edu).
The 20CR is a comprehensive global atmospheric circulation dataset
at 6-hourly temporal and 2° spatial resolutions. It was produced assimilating only surface pressure reports and using observed monthly seasurface temperature and sea-ice distributions as boundary conditions.
NOSA30k is a free-run, 6-hourly and 30 km resolution, simulation,
spanning 13 yrs (1996–2008). It was produced as part of the Latin
American Observatory partnership (Muñoz et al., 2012; Muñoz et al.,
2010). Its precipitation field has been recently evaluated by Ochoa
et al. (2013).
CFSv2 dataset is available at monthly temporal and 0.937° spatial
resolution. It improves nearly all aspects of the data assimilation and
forecast model components of the system (Saha and Coauthors, 2014),
increases the length of skillful Madden–Julian Oscillations forecasts
from 6 to 17 days, and significantly improves global SST forecasts over
its predecessor. In this study, forecasts for JF were initialized in December, while SO forecasts were initialized in August. A 24-member ensemble mean was used in both cases.
2.3. Sea-surface temperature
The study uses the monthly extended reconstructed sea-surface
temperature (SST) dataset (ERSSTv3b, for details see Smith et al.
(2008)) extracted from the U.S. National Oceanic and Atmospheric Administration/National Climatic Data Center (NOAA/NCDC) archives.
This SST dataset combines readings from ships and buoys on a 2° × 2°
grid and does not include satellite data. In the predictability experiments, December SST was used as a potential predictor for lightning
in JF, and August SST for SO.
3. Methodology
All anomalies were computed with respect to the long-term mean of
the corresponding field. A total of 216 months were used to analyze if
there was a statistically significant long-term (18 yrs) linear trend in
the LMB's mean lightning density rate time series, using an F-test of
149
the linear model versus a constant model. Upon finding a significant
trend, the initial time series was detrended for further analyses. Due
to the low resolution of the LTRS dataset, the LIS granule data was
used to compute the monthly climatology for the four quadrants (or
zones) of the LMB (Fig. 1). Unless otherwise indicated, the LTRS dataset
was used for the remainder of the analysis and predictability study
itself.
Due to its high resolution, the NOSA30k dataset was used to analyze
the diurnal cycle of several atmospheric variables inside the LMB. For
analysis of the basin as a whole, the 20CR (which has a larger time period) was used instead.
Multivariate statistical methods were used in the present study. An
Empirical Orthogonal Function (EOF) analysis was performed on the entire time series of lightning density rate and on seasonal averages of the
same variable for Jan–Feb and Sep–Oct, from 1996 to 2013. The principal components (PC), or timeseries associated with the spatial patterns
obtained (EOF), were correlated using Pearson's coefficient.
The statistical models were built using Principal Component Regression (PCR) and Canonical Correlation Analysis (CCA). These are well
known multivariate regression methods that have been applied to seasonal climate forecasting for some time (for a recent example, see
Recalde-Coronel et al. (2014)). In PCR, each predictand is regressed
using a linear combination of the predictor's EOF. CCA is a generalization
that calculates linear combinations of EOF of both a set of predictors and
predictands, identifying pairs of combinations (i.e., canonical variates or
modes) such that their correlations are maximized. Thus, the canonical
modes describe the preferred coupled spatial patterns relating predictors and predictands, and are presumed to be physically meaningful
(that is not warranted a priori). In this study, CCA is conducted using
IRI's Climate Predictability Tool (CPT; available online from the IRI at
http://iri.columbia.edu/our-expertise/climate/tools/cpt/). CPT provides
information that diagnoses the underlying coupled patterns, and also
cross-validated forecast skill metrics that permit the assessment of the
associated potential predictability. Once the best model has been identified, it is possible to forecast both within the historical training period
(hindcasts) and subsequently for future seasons.
To avoid artificial skill, CPT verifies the goodness, or skill, of the
resulting predictions using cross-validation (Barnston and van den
Dool, 1993). Due to the short number of years for analysis of lightning
data, here a cross-validation window of one year is used,2 meaning
that one year from the time series is held out, predicted and later verified, as a simulated independent case outside of the training sample
(e.g. Barnston and van den Dool, 1993; Mason and Stephenson, 2008).
This process is repeated such that each year in the dataset is forecasted
with the climatological data redefined each time a new year is withheld,
and so that after processing all years the mean values of the skill metrics
are provided. In this study the following metrics were used: Kendall's τ,
Spearman correlation coefficient, Hit Skill Score and Relative Operating
Characteristics (ROC) (Mason and Stephenson, 2008). Forecasts were
computed in terms of the probabilities associated with three categories:
below normal, normal and above normal lightning density rates. Since
lightning density rates in the region do not exhibit a Gaussian distribution, they were always transformed before building any model using
empirical cumulative density functions and quantile renormalization.
4. Generalities and lightning variability
The region of interest is located in the northwestern part of
Venezuela (see Fig. 1), between 7.5N–11.5N and 73W–70W. It is
dominated by two important geographic phenomena. The first one is
the Cordillera de los Andes, which splits into two branches: the Perijá
mountains, which head northward and constitute a natural border
2
In some experiments, like the PCR ones, 3 and 5 years were also used for the crossvalidation window. The results show stable results in terms of the number of modes identified for those models.
150
Á.G. Muñoz et al. / Atmospheric Research 172–173 (2016) 147–162
between Colombia and Venezuela, and the Venezuelan Andes mountain
range, which heads towards the Northeast before disappearing in front
of the Caribbean Sea. With mean altitudes between 1500 m and 3700 m,
both the Perijá Cordillera and the Venezuelan Cordillera de los Andes
surround the second feature: the Lake Maracaibo, the largest lake in
South America with an area of almost 13,000 km2.
The northern region of the LMB is dominated by a xerophytic climate, characterized by poor soils and extreme barrenness. Nevertheless,
the basin is characterized by a tropical moist climate, influenced by the
presence of the lake. Annual precipitation shows a clear meridional gradient from the south of Lake Maracaibo (3500 mm) to its northern limit
(125 mm). The annual precipitation cycle is bimodal, with a minimum
in January and primary and secondary maxima in September–October–November (SON) and AMJ seasons (see, for example Velásquez
(2000) and Pulwarty et al. (1998)).
Lightning activity is significantly higher in the southwestern
quadrant of the LMB (Zone III in Fig. 1). The high resolution LIS granule
data confirms the location of the lightning hotspot in Catatumbo,
close to the mouth of the Catatumbo River, as reported by Albrecht
et al. (2009) and Bürgesser et al. (2012). A second hotspot, with
lower mean density rates, is also present in quadrant III close the
Venezuelan–Colombian border (approximately at 9N, 73W).
absolute maximum in September, and a mean maximum density rate
of ~200 fl km−2 yr−1, approximately twice as much as the SO seasonal
mean for the other quadrants. Zones I and II, located closer to the Caribbean Sea, exhibit similar behavior and monthly flash density rates, with
a bimodal cycle and primary and secondary maxima in September and
May, respectively. On the other hand, quadrant IV, eastwards of Zone
III, presents three maxima, namely (in descending order) October, August and April.
In order to better understand the regional context of the lightning
activity observed through the annual cycle in the LMB (Fig. 2), mean
monthly flash density rate anomalies were computed using the LRTS
dataset (Figs. 3 to 5). While January and February exhibit minimum
lightning activity in North Western South America (Fig. 3), a positive
anomaly cluster begins to strengthen in Colombia in April–May between 5N and 9N, and around 75W. It monotonically strengthens in
the following months, quickly shifting to 7N–10N and migrating eastward (Fig. 4). In September (Fig. 4) a new cluster appears in the Amazon
Basin. After the LMB's maximum in September, the monthly flash
density rate decreases rapidly to its minimum value again around
December.
4.1. Decadal trend
The daily maximum of lightning activity tends to happen between
1800 and 0400 local solar times (LST), respectively (Albrecht et al.,
2009; Bürgesser et al., 2012). The first lightning hotspot (close to the
Catatumbo River mouth) is more active between 2400 and 0400 LST,
while for the second hotspot tends to start and end slightly earlier
(2000 to 0200 LST) in the day.
The slightly less than two-decades-long period (18-year) available is
not sufficient to properly assess if there is a long-term trend in the LMB
lightning time series that is related to climate change. Nonetheless, a
statistically significant (p b 0.01) positive linear trend was found in
this study; this deserves future attention and will be analyzed
elsewhere.
4.2. Annual cycle
5. Potential predictors
Understanding if the Catatumbo Lightning are related to purely local
climate drivers or to regional/global-scale forcings may give a hint about
which are the best potential predictors. A priori, in this study the distinction between scales is considered to be an artificial one, as the
( fl km -2 yr -1)
Zone III presents a distinctive annual cycle (Fig. 2) compared to the
other quadrants of LMB; it shows a unimodal distribution with an
4.3. Diurnal cycle
Fig. 2. Monthly lightning density rate (LIS granule dataset), in [fl km−2 yr−1] for Zones I (blue), II (green), III (purple) and IV (yellow). See Fig. 1. (For interpretation of the references to
color in this figure legend, the reader is referred to the online version of this chapter.)
Á.G. Muñoz et al. / Atmospheric Research 172–173 (2016) 147–162
( fl km -2 yr -1)
Fig. 3. Monthly lightning density rate anomaly (LTRS dataset) for North Western South America, in [fl km−2 yr−1]. Months: JFMA.
(fl km -2 yr -1)
Fig. 4. Same as in Fig. 3 but for MJJA.
151
152
Á.G. Muñoz et al. / Atmospheric Research 172–173 (2016) 147–162
(fl km -2 yr -1)
Fig. 5. Same as in Fig. 3 but for SOND.
LMB is not a closed system. However, if “local” potential predictors are
those defined only within the geographical domain of the basin, then
the distinction may be plausible.
It is unlikely that local potential predictors alone are capable of
explaining the LMB lightning variability, but instead that there are contributions from both local and regional/global climate drivers. Thus, key
questions to ask are: (a) what is a minimal set of potential predictors?,
and (b) does a particular “scale” have a preponderant influence on LMB
lightning variability? For example, given that ENSO is the main climate
driver at seasonal-to-interannual timescales, is it possible that it also
controls most part of the lightning activity in the region? Or is a local
driver responsible of the majority of the explained variance?
Falcón et al. (2000) suggested that the Catatumbo Lightning are
caused by methane concentration coming from the swamps present in
the neighborhood of the Catatumbo River mouth. Their pyroelectric
model is based in symmetry properties of the methane molecule:
under temperature changes (as the ones related to the atmospheric
lapse rate) the methane tends to self-polarize and if a key concentration
is achieved in the top layer of the cloud (methane is lighter than water
vapor) a self-maintained electrical discharge would drive lightning activity every night. This model fails to explain the daily and annual lightning cycles in the LMB, as it predicts higher activity in the dry season
because evaporation will help to increase methane concentrations,
and less lightning in the wet ones as the methane will be washed-out
by rainfall (Bürgesser et al., 2012). Furthermore, the model predicts
lightning only from sunset to sunrise, as the solar radiation photodissociates the CH4 molecules, again contradicting the lightning observations (see Section 4.3).
After analyzing observed and model data, and along with several expeditions to Catatumbo, Muñoz and Díaz-Lobatón (2011) and DíazLobatón (2012) pointed out that lightning activity in the LMB is
influenced by several climate drivers that modulate convection and
local wind and shear magnitudes, namely, the seasonal variation of
the trade winds and the Caribbean Low Level Jet (CLLJ) (Amador,
2008) during boreal autumn and winter, the Inter-Tropical Convergence Zone (ITCZ) migrations, surface heating due to incoming shortwave radiation, upward motion triggered by topography and the presence of the lake, traveling waves and tropical cyclones (see for example
Durán-Quesada et al. (2010), Whyte et al. (2008) and references therein). In particular, Díaz-Lobatón (2012) analyzed the transformations between potential and kinetic energies in the LMB, and showed that the
seasonal behavior of local CAPE and low level meridional velocities in
the basin have statistically significant correlations with lightning density rate; these variables were therefore proposed as potential predictors.
In order to understand the seasonal impact of LMB's low level wind
field in lightning activity, the next subsection discusses first its role in
the diurnal lightning cycle.
5.1. Maracaibo Basin Nocturnal Low Level Jet
The behavior of the low level meridional wind field is related to what
will be referred to as the Maracaibo Basin Nocturnal Low Level Jet (MBNLLJ). This is a characteristic feature of the diurnal cycle in the LMB
that modulates convection via moisture advection and thermodynamic
instability, and that is closely linked to the CLLJ, thus connecting the
diurnal and seasonal (annual) cycles.
Muñoz and Díaz-Lobatón (2012) analyzed the diurnal cycle of the
MB-NLLJ through high resolution (4 km) WRF simulations and its impact in observed local surface pressure (bottom panel of Fig. 6), finding
that the southwestward oriented winds coming from the Venezuelan
Gulf begin to increase in speed around noon, maximizing between
1700 and 1800 LST (absolute daily minimum of surface pressure).
Á.G. Muñoz et al. / Atmospheric Research 172–173 (2016) 147–162
153
Fig. 6. Typical sub-daily variability of temperature (top panel), relative humidity (central panel) and surface pressure observed in Maracaibo city. Visualized data corresponds to February
14–16, 2011 (10 min time resolution). The horizontal lines indicate mean values for each variable.
Modified from Muñoz and Díaz-Lobatón (2012).
After sunset, the LLJ decreases speed, disappearing around 0400 (secondary minimum of daily surface pressure) when cold downslope
winds coming to the basin from the mountains are at their maximum.
The northward component of the anomalous wind field achieves its
maximum around 0900 (not shown), as does the daily surface pressure
and then changes sign around noon, when the cycle starts again. The
overall magnitude of the meridional wind in the LMB is negative (note
that Fig. 7 is showing anomalies), its magnitude changing depending
on the location and time of the day. The general characteristics of the
cycle are well reproduced by NOSA30k (Fig. 7), and the wind velocities
at surface level are in good agreement with observations. NOSA30k
seems to be adequate to analyze the mechanisms generating the LLJ.
Inertial oscillations are considered to be a major contributing factor
for the formation of low level jets; they are related to a nocturnal
decoupling of surface flows due to different day-night horizontal turbulent stresses, and an imbalance between the Coriolis and the pressure
gradient forces, producing super-geostrophic winds (Blackadar's mechanism, see Wiel et al. (2010) and references therein). Nonetheless, as
expected for these latitudes, the MB-NLLJ cannot be explained purely
in terms of inertial oscillations; for example, the predicted internal period, a function of the inverse of the Coriolis parameter, is around 72.7 h,
or around 3 times that observed. Instead, the most important contributing factors are related to thermal forcing in the diurnal oscillation of the
planetary boundary layer winds above sloping terrain (Holton's mechanism (Holton, 1967)). Preliminary results show that the MB-NLLJ is well
described by the recent Du-Rotunno model (Du and Rotunno, 2014),
which considers both the Blackadar and Holton mechanisms. The
details of this analysis are outside the scope of the present work and
will be discussed in a companion paper.
The daily maxima of lightning activity in the LMB coincide with the
NLLJ variability between 1800 and 0400 LST. Analysis of these phenomena confirms a direct relationship between negative meridional wind
anomalies and lightning flash density rate, leading to the identification
of a potential predictor, with the following physical explanation. During
the afternoon, the LLJ transports moisture from the Caribbean and Lake
Maracaibo to the southwestern part of the basin. Around 1630 the meridional winds are so intense that they are capable of crossing the Andes
and Perijá cordilleras, producing orographic precipitation and advecting
part of the moisture out of the LMB. During the afternoon, the temperature decreases sharply, rapidly increasing the relative humidity and
providing suitable conditions for convection (Fig. 6). After sunset (approximately 1800), thermal forcing induces downslope colder flows
from the mountains to the warmer lake, gradually decreasing the negative meridional wind anomaly in the basin (see differences between the
two upper panels in Fig. 7). These processes increase instability in the
basin, especially in Zone III, and convection occurs. Clouds develop
with heights on the same scale as the surrounding mountains, with
lightning density being proportional to approximately the fifth power
of the cloud height (e.g., Price and Rind, 1992; Wong et al., 2013).
Around 0400 LST (see Fig. 7, upper right panel) the mean meridional
wind anomaly in the LMB changes sign, the orographic convection decreases as the base of the clouds moves away from the mountains, and
154
Á.G. Muñoz et al. / Atmospheric Research 172–173 (2016) 147–162
1930
0130
0730
1330
Fig. 7. Meridional-vertical section of meridional wind anomalies (NOSA30k) along 71.5W, at 1930 LST (00Z, upper left panel), 0130 (06Z, upper right panel), 0730 (12Z, lower left) and
1330 (18Z). Topography is shown in black. The Catatumbo region is located around 9.4N.
vertical wind shear breaks the clouds. For an animation illustrating
some of these processes, see Suplemental Material.
The role of the MB-LLJ suggests that meridional wind anomalies at
925 mb could be a potential predictor for lightning activity. The analysis
also reveals that the intensity of the LLJ and its moisture advection have
a seasonal dependence, which will be discussed in the next subsection.
5.2. Seasonal-to-interannual role of low level jets
A simple way to analyze how much of the Catatumbo Lightning variability responds to a regional signal (or vice-versa) is to compare the
standardized basin-wide mean flash density rate with the principal
components of the same variable for a wider domain in North Western
South America (22N–1N, 82W–58W). The principal component analysis considering all months in the time series reveals that the second
EOF, explaining 19% of the total variance, is associated with a regional
lightning pattern of variability located between the northern border of
Colombia and Venezuela, and thus it was selected for further analysis
(the first and third EOFs are discussed in the next paragraph). The standardized Pearson coefficient between these two time series indicates a
significantly high correlation (0.66, p b 0.01, see Fig. 8). Due to their similarity, the second principal component was denominated Catatumbo
Lightning Regional Mode (CLRM). The identification of this mode is also
important for the predictability study discussed in the next section, as
it is key to recognize the physical identity of the modes being used by
the CCA approach (the actual LMB mean of flash density rate can be
used directly in the PCR methodology).
Considering only the first three EOFs of lightning density rate in
North Western South America, it was found that the CLRM corresponds
to the second and third EOFs in the JF and SO seasons, with around 17%
of the total explained variance in both cases. The other two patterns, associated with lightning activity around the El Chocó/Darién Gap (between Colombia and Panama) and the Amazon Basin, respectively,
explain 50% and 11% for JF, and 32% and 27% for SO. The analysis also revealed that the CLRM is the only pattern showing a trend (cf. observed
decadal trend discussed in Section 4.1). A PCR using only these three
EOFs indicates that around 78% (for both JF and SO) of the LMB lightning
activity can be explained by those patterns. This suggests that the
Catatumbo Lightning have a strong dependence on regional climate
drivers.
It was mentioned before that several climate drivers have an impact in the LMB lightning activity. After analyzing the link between
lightning and LLJ diurnal cycles, two logical assumptions to make
are that (a) the meridional wind field is also a good predictor at
seasonal-to-interannual timescale, and (b) the behavior of this field
in the LMB is modified by the mentioned seasonal climate drivers.
Á.G. Muñoz et al. / Atmospheric Research 172–173 (2016) 147–162
155
-4
-2
0
2
4
Catatumbo Lightnings Regional Mode
LMB mean
Jan
1996
Jan
1998
Jan
2000
Jan
2002
Jan
2004
Jan
2006
Jan
2008
Jan
2010
Jan
2012
Jan
2014
Time
Fig. 8. Variability of the mean LMB lightning density rate anomaly (red) and Catatumbo Lightning regional mode (blue). Time series have been detrended and standardized (units in
standard deviations). (For interpretation of the references to color in this figure legend, the reader is referred to the online version of this chapter.)
4
This hypothesis simplifies the potential predictability study as it reduces the number of candidates to explore, but also brings a physically sound argument to the analysis: the lightning activity likely
depends on multiple factors at the same time, which must have an
impact on the meridional wind field and, in consequence, should
provide a higher predictive skill than considering each one of the climate drivers independently.
As indicated by Díaz-Lobatón (2012), and as in the case of the diurnal cycle, the basin-wide meridional wind anomaly at 925 mb is significantly anti-correlated (p b 0.001) to LMB mean lightning density rate
for the entire period under study (Fig. 9). The same statement is true
for the CLRM. Nonetheless, as convection requires moisture availability,
unstable atmospheric parcels and an initial lifting force, the meridional
wind field by itself does not necessarily account for the instability condition. A simple way to combine instability with the advective effect of
the LLJ dynamics in the LMB is to consider the CAPE transport by the
meridional wind field. Compared to the meridional wind velocity, this
variable has a better synchronization (Fig. 9) with the LMB lightning
activity and the CLRM, and thus was selected as potential predictor for
the study.
The availability and transport of moisture are critical factors that determine the strength of convection, via the moisture flux divergence
term. The main moisture source in the region is the Caribbean Sea
(the second is the Lake Maracaibo itself), while the main seasonal advection mechanisms are related to the CLLJ, the ITCZ and, especially in certain years, tropical cyclones. However, the meridional wind field can
also play a role against lightning activity proliferation: a sufficiently
strong wind circulation may inhibit convection, because it transports
moisture out of the region, or due to a strong vertical wind shear. This
is partially why the DJF season displays a minimum of lightning activity
-4
-3
-2
-1
0
1
2
3
Flash density rate
v
vCAPE
Jan
1996
Jan
1997
Jan
1998
Jan
1999
Jan
2000
Jan
2001
Jan
2002
Jan
2003
Jan
2004
Jan
2005
Jan
2006
Jan
2007
Jan
2008
Time
Fig. 9. Time series of the flash density rate (red, LIS-OTD), meridional wind anomaly (blue) and meridional CAPE advection anomaly (in yellow, the last two from the 20CR). The units are
standard deviations. (For interpretation of the references to color in this figure legend, the reader is referred to the online version of this chapter.)
156
Á.G. Muñoz et al. / Atmospheric Research 172–173 (2016) 147–162
in North Western South America (see Fig. 3); it corresponds to one of
the two seasonal enhancements of the CLLJ, the other being the JJA season (Whyte et al., 2008; Amador, 2008).
At regional level, the ITCZ's northward migration is associated
with the positive anomaly that appears around March in southern
Colombia (Fig. 3). The centroid of the lightning anomaly pattern
sketched in Figs. 3 and 4 does not tend to go further North because
from June onwards the CLLJ, located along latitude 15 N, intensifies
again (Whyte et al., 2008) and keeps convective activity confined to a
corridor between 7N and 12N (Fig. 4). The CLLJ shifts slowly westward
in the following few months, pushing the positive anomaly pattern eastward. In September the winds related to the CLLJ have changed from a
typical July (maximum) value of 14 m s−1 at or below 925 mb, to barely
8 m s−1. This creates suitable conditions for enhanced convection and
lightning activity for that month on both (Venezuelan and Colombian)
sides of the Cordillera de los Andes (Fig. 5). Soon after, the ITCZ starts
migrating southward, and consequently the convection decreases
significantly.
At a local level, as mentioned before, September exhibits the maximum value of lightning activity in the LMB. This is remarkable for
Zone III, where intricate topographic features, moisture convergence
and orographic updrafts enhance local convection. Zones I and II are,
on the contrary, more exposed to the influence of the Caribbean
winds, and consequently do not show as high levels as the other
quadrants (Fig. 2). Indeed, their proximity to the Caribbean Sea and
more exposure to the intensified low level winds of the CLLJ makes
the convective activity in Zones I and II to decrease in July, and that is
the reason for the bimodal annual behavior of both lightning and rainfall. Zone IV shows different behavior because it involves simultaneously the southeastern part of the LMB, the tallest section of the Venezuelan
Andes cordillera and even a section related with the Venezuelan Llanos
(savannah), which extends eastward of the Cordillera de los Andes.
Strong winds in May and September are typical for this region
(these winds are called “Barinas”, which also gives the name to the
Venezuelan province located there), and are related to katabatic
winds (Pulwarty et al., 1998). It is suggested here that these winds
could be the cause of the minimal values evidenced for Zone IV, as
shown in Fig. 2.
This analysis confirms that there is a seasonal modulation of lightning activity by several climate drivers, which impact the meridional
velocity field in the LMB, the moisture availability and therefore
convective activity. The most important seasonal driver is the CLLJ,
which interacts with the MB-NLLJ (these LLJs are decoupled in boreal
winter, and strongly coupled in boreal autumn), modulating lightning
activity and, more generally, convection when it happens. The ITCZ
and tropical cyclones (see for example year 2005 in Fig. 9, a historical record of tropical cyclone activity in the Caribbean) also tend to modify
the location and intensity of the CLLJ, and ultimately the effect of the
MB-NLLJ on the Catatumbo Lightning. The choice of meridional wind
anomalies, or meridional CAPE transport, as a potential predictor that
includes the contribution of several climate drivers is thus physically
justified.
6. Predictive skill
This section discusses the predictability analysis of lightning activity
in the LMB. Reanalysis and perfect-prognosis simulations were used in
the previous sections to diagnose relationships between variables, but
since there is interest in evaluating the predictive skill for operational
forecasts, observed SST fields and actual CFSv2 forecasts are used as
the source for potential predictors.
The final candidate predictor identified in the previous section is
CAPE meridional transport (vCAPE) at 925 mb; meridional winds at
925 mb provided similar but slightly worse results than vCAPE, and
thus those are not reported here. SST is the most common potential predictor field in seasonal forecasts (e.g., Mason and Baddour, 2008), and
Table 1
Predictors and predictands used in the experiments. The LMB geographic extension is defined as in Section 4.
Variable
Type
Domain
Role
Denomination
SST
SST
vCAPE
vCAPE
FDR
FDR
FDR
CLRM
Field
Field
Field
Field
Field
Field
Time series
Time series
25N–10S, 154–354
25N–10S, 254–354
16N–6N, 280–292
11.5N–7.5N, 73W–70W
22N–1N, 82W–58W
11.5N–7.5N, 73W–70W
LMB mean
EOF
Predictor
Predictor
Predictor
Predictor
Predictand
Predictand
Predictand
Predictand
PacAtl-Dec
PacAtl-Aug
NWSA
LMB
NWSA
LMB
LMBm
CLRM
therefore it was used as a reference to compare with vCAPE. Four
predictands were used: the Flash Density Rate (FDR) field for North
Western South America (NWSA), FDR for the LMB, its basin-wide
mean, and the CLRM. Table 1 summarizes these and other details.
A large number of experiments were performed in order to identify
the best models and domains for the potential predictors. It is important
to note that the basin-wide mean FDR gave results that were very similar to, but never better than, the ones obtained with the CLRM. This was
expected, as the CLRM could be considered an EOF-filtered version of
the dominating FDR spatial pattern in the LMB, thus capturing better
the relationships with potential predictors. The seven models with the
best performance are reported in Tables 2 and 3 for the seasons under
study. Their names provide information about the variable used as potential predictor, the multivariate regression method and the domain
or mode considered in the calculation.
Most models were able to capture physically meaningful patterns,
appearing consistently in the different final experiments. For example,
SST-PCR-CLRM (a model that regresses the CLRM using December's
SST field from a wide domain involving the Pacific and Atlantic, see
Fig. 10) showed that ENSO (EOF1), the Atlantic Meridional Mode
(Chiang and Vimont, 2004) (EOF2), a Western Atlantic/Caribbean SST
mode in phase with the Eastern Tropical Pacific (EOF3), and a (positive)
trending SST mode in the Caribbean (EOF4), play key roles in the lightning variability for the boreal winter (their combined loadings are
shown in the upper left panel of Fig. 10). These modes appear in several
other models for the same season.
On the other hand, the autumn lightning activity in LMB tends to be
less influenced by the Central Pacific and more by a dipolar SST mode
Table 2
Cross-validation skill metrics for the best models in boreal winter. Predictor–predictandCCA modes are shown for CCA models. The best results are presented in bold.
Model (JF)
Modes
Kendall's τ
Spearman
HSS (%)
ROCb
ROCa
SST-CCA-NWSA
SST-CCA-LMB
SST-PCR-CLRM
vCAPE-CCA-NWSA
vCAPE-CCA-LMB
vCAPE-PCR-NWSA
vCAPE-PCR-CLMR
5–4–3
6–2–2
4
8–2–1
4–2–1
10
10
.168
.399
.359
.340
.405
.556
.569
.50
.45
.39
.39
.55
.48
.69
11.76
25.00
16.67
16.67
25.00
33.33
41.67
.80
.74
.83
.65
.68
.65
.79
.85
.81
.61
.75
.82
.78
.85
Table 3
Cross-validation skill metrics for the best models in boreal autumn. Predictor–predictandCCA modes are shown for CCA models. The best results are presented in bold.
Model (SO)
Modes
Kendall's τ
Spearman
HSS (%)
ROCb
ROCa
SST-CCA-NWSA
SST-CCA-LMB
SST-PCR-CLRM
vCAPE-CCA-NWSA
vCAPE-CCA-LMB
vCAPE-PCR-NWSA
vCAPE-PCR-CLMR
5–7–2
10–2–1
4
9–2–1
4–2–1
3
1
.212
.248
.346
.105
.196
.503
.490
.14
.48
.32
.30
.40
.62
.55
0.0
25.00
0.0
0.0
8.33
41.67
33.33
.71
.79
.71
.57
.68
.81
.74
.54
.78
.57
.67
.72
.71
.71
Á.G. Muñoz et al. / Atmospheric Research 172–173 (2016) 147–162
157
Fig. 10. SST-PCR-CLRM model. Combined loadings for the SST EOFs (left), temporal scores (middle, SST appears in red, CLRM in green), and CLRM's loadings (right), for both JF (top) and SO
(bottom). Canonical correlations are 0.73 and 0.66, respectively. (For interpretation of the references to color in this figure legend, the reader is referred to the online version of this
chapter.)
between the Western Tropical Atlantic and the Eastern Tropical Pacific,
an homogeneous inter-basin SST pattern similar to winter's EOF3, a
trending SST mode in the Western Caribbean (similar to the EOF4
found for winter, although with a more modest trend), and a meridional
dipole between the Tropics and the North Atlantic (again, their combined loadings are shown in Fig. 10, lower left panel).
The analysis of the models using vCAPE as potential predictor indicates that they are consistent with the expected physical behavior of
the involved variables. The dominant vCAPE modes present in the
NWSA domain in most models are related to high convection in El
Chocó/Darién Gap, the CLLJ and several local patterns related to orographic convection, and coastal and inland (valley) winds. In particular,
the experiments confirmed that the CLLJ plays a key role in controlling
the seasonal variation of lightning as it intensity and mean location
changes between winter and autumn. For example, the vCAPE-PCRCLMR model (Fig. 11) clearly shows a northeastern–southwestern pattern in the region of the LMB that is suggested to be related to a “constructive cross-timescale interference” (Muñoz et al., 2015b) between
the CLLJ and the MB-NLLJ in autumn (lower panel of Fig. 11), as
discussed in the previous section. These LLJs are “decoupled” in the JF
season, providing lower than normal moisture availability to the LMB
and low lightning activity.
In this study, the Kendall's τ and Spearman correlation coefficients
are used to give a general idea of the goodness of the forecast, and to
measure how in-phase are the observations and the produced seasonal
forecasts. The Hit Skill Score (HSS) indicates how frequently the
category with the highest forecast probability is verified, and it is defined such that random “hits” are not considered. Discrimination, or
how well a forecast distinguishes between below normal, normal or
above normal categories, is an extremely important attribute, because
it indicates whether any potentially useful information is actually
being provided. Here, ROC curves (diagrams comparing proportion of
hit rates versus false alarms, Figs. 12 and 13) and their areas under the
curve are used as metrics of discrimination (in Tables 2 and 3, ROCa
and ROCb indicate the ROC areas for the above-normal and belownormal categories, respectively).
As hypothesized in Section 5, models using meridional winds and
vCAPE as potential predictors tend to outperform those using SST. The
best models use variables defined over the LMB (i.e., “local” variables).
Naturally, that does not imply that lightning in the LMB is not modulated by regional and global factors (the recent analysis just showed that
they are), but it answers the questions posed in this study previously:
local potential predictors that are sensitive to regional and global climate drivers, as vCAPE at 925 mb or meridional winds, are the best ones.
The ROC curves show that models using vCAPE have good discrimination, especially vCAPE-PCR-CLMR for winter (Fig. 13g) and vCAPEPCR-NWSA for SO (Fig. 13f), but SST models also show good discrimination. As a matter of fact, winter's SST-PCR-CLRM model (Fig. 12e) has
the highest discrimination of any model for the below-normal category, and the SST-CCA-NWSA model (Fig. 12a) has basically the
same ROC areas and even steeper ROC curves than the best winter
model (Fig. 13g).
158
Á.G. Muñoz et al. / Atmospheric Research 172–173 (2016) 147–162
Fig. 11. vCAPE-PCR-CLMR model. Combined loadings for the vCAPE EOFs (left), temporal scores (middle, vCAPE appears in red, CLRM in green), and CLRM's loadings (right), for both JF
(top) and SO (bottom). Canonical correlations are 0.96 and 0.76, respectively. (For interpretation of the references to color in this figure legend, the reader is referred to the online version
of this chapter.)
Overall, the best models are those built with PCR. The winter model
with the best combination of the different scores is vCAPE-PCR-CLMR,
while for autumn it is vCAPE-PCR-NWSA. There is slightly higher potential predictability in JF, which is attributed here to the role of ENSO in the
modulation of atmospheric circulations in the region of study, but this
requires further research. Above-normal ROC areas are in general better
for winter, and since it is the season with less convective activity, this
higher discrimination for above-normal lightning activity is useful for
decision-making in the “dry season”. For SO, the discrimination is in
general better for the below-normal category; the SST-CCA-LMB
model (Fig. 13d) is the only one having high and similar ROC areas for
both below- and above-normal categories.
Although the scores reported here for these models are in general
better than for rainfall amount in the same region (Kendall's \tau ~0.10.3), a few ideas are being explored currently in order to increase the
skill of lightning forecasts even more. It is unclear at the moment if
using higher resolution atmospheric fields as potential predictors may
help. It was shown that NOSA30k is very good to study the diurnal
cycle in the LMB, but a 13-year simulation is too short to formally diagnose potential predictability with this dataset. A new multi-physics set
of numerical simulations including lightning output is being developed
(some short experiments are already publicly available) (Chourio and
Muñoz, 2015) to explore the role of high resolution potential predictors.
Perhaps the most important improvement for this potential forecast
system is to have high resolution predictands.
7. Concluding remarks
The first lightning predictability study at seasonal scale for the Lake
Maracaibo Basin, and perhaps the first one globally, was discussed in
this paper. Lightning predictive skill was quantified using multiple metrics and different PCR and CCA models. Skill tends to be slightly higher in
the JF season than in SO, probably because of the higher predictability of
ENSO and its influence in convective activity in the Tropics. Nonetheless,
both seasons show higher skill than the typical values for rainfall
amount in the same region. This could be related to the fact that lightning density rate is a frequency measure, and it has been shown in
other studies (e.g., Moron et al., 2007) that seasonal frequency tend to
be more predictable than seasonal amount or intensity (at least for rainfall, which is also related to convection in the Tropics).
Most of the lightning activity in the whole LMB is related to a regional mode located between Northeastern Colombia and Northwestern
Venezuela: the Catatumbo Lightning Regional Mode. The Catatumbo
Lightning, located in the southwestern quadrant of the LMB, are thus a
combination of regional-to-global-scale factors and local ones. Besides
the complex topography, the most important local factor is the Maracaibo Basin Nocturnal Low Level Jet, which controls the diurnal cycle of
lightning activity, but its seasonal intensity and moisture transport are
modulated via coupling and decoupling with the Caribbean Low Level
Jet, thus explaining the observed seasonal lightning variability in the
basin. Other key climate drivers are the ITCZ and hurricane activity.
Á.G. Muñoz et al. / Atmospheric Research 172–173 (2016) 147–162
159
Fig. 12. Relative Operating Characteristics for JF (left column) and SO (right column) seasons, showing the results for the SST-CCA-NWSA (top), SST-CCA-LMB (middle) and SST-PCR-CLMR
models. Curves correspond to the above- (blue) and below-normal categories (red). See Tables 2 and 3. (For interpretation of the references to color in this figure legend, the reader is
referred to the online version of this chapter.)
The net contribution of these regional-to-global drivers is to modify the
moisture availability and wind circulation in the basin, thus controlling
convection. In consequence, it was no surprise to find that the meridional CAPE transport at 925 mb in both North Western South America or
in the LMB is in general a better potential predictor than regional or
global-scale SST fields, which indeed have an impact in atmospheric circulation and moisture availability, but do not seem to be enough to
uniquely define the role of the different physical drivers acting on the
region's lightning activity. The low-level meridional CAPE transport
does a better job capturing them.
The predictability study presented here was designed as the first
stage of a Ready–Set–Go approach (Hellmuth et al., 2011) to provide
useful lightning hazard information to decision-makers in the LMB.
Although the analysis was performed for one-month lead time operational lightning forecasts, the same methodology could be used to
explore the potential predictive skill at longer lead times, and for
other seasons. On the other hand, the analysis suggests potential predictors that could be used to explore lightning predictability at subseasonal scales (the Set stage of the approach). This should be explored
in the near future.
160
Á.G. Muñoz et al. / Atmospheric Research 172–173 (2016) 147–162
Fig. 13. Relative Operating Characteristics for JF (left column) and SO (right column) seasons, showing the results for the vCAPE-CCA-NWSA (top), vCAPE-CCA-LMB (second row), vCAPEPCR-NWSA (third row) and vCAPE-PCR-CLMR models. Curves correspond to the above- (blue) and below-normal categories (red). See Tables 2 and 3. (For interpretation of the references
to color in this figure legend, the reader is referred to the online version of this chapter.)
Á.G. Muñoz et al. / Atmospheric Research 172–173 (2016) 147–162
This study has considered only flash density rate as predictand, but
no distinction was made in terms of the type of the electrical discharge
(intra-cloud, IC, versus cloud-ground, CG) being forecasted, which is a
piece of information that is very important for decision-makers in the
LMB. A reliable lightning sensor network must be put to work in the region so a continuous, quality controlled and high resolution lightning
dataset reporting daily proportion of IC/CG discharges is available for
nowcasting and forecasting. Since the IC/CG time series will be too
short to build reliable statistical models, dynamical or hybrid models
could be used once they have been calibrated with observations. The
charge separation and lightning formation mechanisms are understood
well enough (see Virts et al. (2013) and references therein) to provide
forecasts of IC/CG ratios. Indeed, the dynamical structure of a thunderstorm and its microphysics seem to be related to lightning frequency
and type (Goodman et al., 1988), stronger updrafts leading to more frequent IC lightning activity (Price and Rind, 1992), (Williams et al.,
1999), due to more frequent interactions between ice phase hydrometeors within the mixed phase region of the convective cloud (Buechler
et al., 2000). Since strong updrafts are typical in the Catatumbo region
due to orographic forcing and the presence of the MB-NLLJ, all the
necessary physical ingredients seem to be already present, without
any need of additional factors (e.g., methane).
Supplementary data to this article can be found online at http://dx.
doi.org/10.1016/j.atmosres.2015.12.018.
Acknowledgments
The authors acknowledge the use of Global Hydrology Resource Center lightning data (http://ghrc.msfc.nasa.gov). The final manuscript
benefited from several comments from anonymous reviewers. The authors are also grateful to Drs. Simon Mason and Catherine Pomposi for
enriching discussions, to the International Research Institute for Climate
and Society (IRI) Data Library Team, especially Mike Bell, for their help
making available the required datasets, and to Dr. Daniel MartnezTong for useful discussions about LIS Granule Science data, data mining
and post-processing. This work was partially funded by CMC-GEO-0210.
MJS was partially funded by CONDES-LUZ-CC-0015-08.
References
Albrecht, R., Goodman, S., Buechler, D., Chronis, T., 2009. Tropical frequency and distribution of lightning based on 10 years of observations from space by the Lightning Imaging Sensor (LIS). Preprints. Fourth Conf. on Meteorological Applications of Lightning
Data, Phoenix, AZ, Amer. Meteor. Soc, P2.12.
Amador, J.A., 2008. The intra-americas sea low-level jet. Ann. N. Y. Acad. Sci. 1146 (153),
188. http://dx.doi.org/10.1196/annals.1446.012.
Baker, M.B., Christian, H.J., Latham, J., 1995. A computational study of the relationships
linking lightning frequency and other thundercloud parameters. Quart. J. Roy. Met.
Soc. 121, 1525.
Barnes, S.L., Newton, C.W., 1982. Thunderstorms in the synoptic setting, in thunderstorms.
In: Kessler, E. (Ed.), A Social, Scientific, and Technological Documentary, Vol. 2, Thunderstorm Morphology and Dynamics. U.S. Dept. of Commer, Washington, D. C.,
pp. 109–171.
Barnston, A.G., van den Dool, H.M., 1993. A degeneracy in cross-validated skill in regressionbased forecasts. J. Clim. 6, 963–977 (doi:10.1175/1520-0442(1993)006b0963:ADICVS2.0.CO;2).
Boccippio, D.J., Driscoll, K., Koshak, W., Blakeslee, R., Boeck, W., Mach, D., Buechler, D.,
Christian, H.J., Goodman, S.J., 2000. The optical transient detector (OTD): instrument
characteristics and cross-sensor validation. J. Atmos. Oc. Tech 17, 441–458.
Boccippio, D.J., Koshak, W.J., Blakeslee, R.J., 2002. Performance assessment of the optical
transient detector and lightning imaging sensor, I, predicted diurnal variability.
J. Atmos. Ocean. Technol. 19, 1318–1332.
Buechler, D.E., Driscoll, K.T., Goodman, S.J., Christian, H.J., 2000. Lightning activity within a
tornadic thunderstorm observed by the optical transient detector (OTD). Geophys.
Res. Lett. 27 (15), 2253–2256.
Bürgesser, R.E., Nicora, M.G., Ávila, E.E., 2012. Characterization of the lightning activity of
“Relámpago del Catatumbo”. J. Atmos. Solar-Terrestrial Phys. 77, 241–247. http://dx.
doi.org/10.1016/j.jastp.2012.01.013.
Burrows, W.R., Price, C., Wilson, L.J., 2005. Warm season lightning probability prediction
for Canada and the northern United States. weather Forecast. 20, 971–988. http://
dx.doi.org/10.1175/WAF895.1.
161
Cecil, D.J., Buechler, D.E., Blakeslee, R.J., 2014. Gridded lightning climatology from TRMMLIS and OTD: dataset description. Atmos. Res. 135–136, 404–414. http://dx.doi.org/
10.1016/j.atmosres.2012.06.028.
Chiang, J.C.H., Vimont, D.J., 2004. Analagous meridional modes of atmosphere–ocean
variability in the tropical Pacific and tropical Atlantic. J. Climate 17 (21),
4143–4158.
Chourio, X., Muñoz, Á.G., 2015. Multi-member Simulations to Support Lightning Research
in North Western South America. Centro de Modelado Científico de la Universidad del
Zulia Available at http://iridl.ldeo.columbia.edu/expert/home/agmunoz/.Catatumbo/.
NWSA_Lightning/.
Christian, H.J., et al., 2000. Algorithm Theoretical Basis Document (ATBD) for the Lightning Imaging Sensor (LIS).
Christian, H.J., et al., 2003. Global frequency and distribution of lightning as observed from
space by the Optical Transient Detector. J. Geophys. Res. 108 (D1), 4005. http://dx.
doi.org/10.1029/2002JD002347.
Chronis, T.G., Goodman, S.J., Cecil, D., Buechler, D., Robertson, F.J., Pittman, J., Blakeslee,
R.J., 2008. Global lightning activity from the ENSO perspective. Geophys. Res. Lett.
35, L19804. http://dx.doi.org/10.1029/2008GL034321.
Compo, G.P., et al., 2011. The twentieth century reanalysis project. Q. J. R. Meteorol. Soc.
137, 1–28. http://dx.doi.org/10.1002/qj.776.
Court, A., Griffiths, J.F., 1982. Thunderstorm climatology. In: Kessler, E. (Ed.), Thunderstorms: A Social, Scientific, and Technological Documentary, Vol. 2, Thunderstorm Morphology and Dynamics. U.S. Dept. of Commer., Washington, D. C.,
pp. 11–52.
Díaz-Lobatón, J., 2012. Energética de los Relámpagos del Catatumbo. Trabajo Especial de
Grado. Translation: Energetics of the Catatumbo Lightning. BSc Thesis. Dpto. de Física,
Facultad de Ciencias de Universidad del Zulia. Maracaibo 102pp.
Du, Y., Rotunno, R., 2014. A simple analytical model of the nocturnal low-level jet over the
Great Plains of the United States. J. Atmos. Sci. 71, 3674–3683. http://dx.doi.org/10.
1175/JAS-D-14-0060.1.
Durán-Quesada, A., Gimeno, L., Amador, J.A., Nieto, R., 2010. Moisture sources for Central
America: identification of moisture sources using a Lagrangian analysis technique.
J. Geophys. Res. 115, D05103. http://dx.doi.org/10.1029/2009JD012455.
Falcón, M., Pitter, W., Muñoz, Á.G., Barros, T., Viloria, A., Nader, D., 2000. Modelo
Electroatmósferico del Relámpago sobre el Ro Catatumbo. Sci. J. Exp. Faculty of Sc.
(Ciencia) 8 (2), 155–167.
Goodman, S.J., 1990. Predicting thunderstorm evolution using ground-based lightning detection networks. NASA Tech. memo. TM-103521.
D., M., Christian, H., Blakeslee, R., Boccippio, D., Goodman, S., Boeck, W., 2007. Performance assessment of the optical transient detector and lightning imaging sensor.
Part II: clustering algorithm. J. Geophys. Res. 112, D09210. http://dx.doi.org/10.
1029/2006JD007787.
Goodman, S.J., Christian Jr., H.J., 1993. In: Gurney, R.J., al., e. (Eds.), Global Observations of
Lightning, in Atlas of Satellite Observations Related to Global Change. Cambridge
Univ. Press, Cambridge, pp. 191–219.
Goodman, S.J., Buechler, D.E., Wright, P.D., Rust, W.D., 1988. Lightning and precipitation
history of a microburstproducing storm. Geophys. Res. Lett. 15, 1185–1188.
Goodman, S.J., Buechler, D.E., Knupp, K., Driscoll, K.T., McCaul, E.W., 2000. The 1997–1998
El Niño event and related wintertime lightning variations in the southeastern United
States. Geophys. Res. Lett. 27, 541–544.
Goodman, S., Buechler, D., McCaul, E., 2007. Lightning, in Our Changing Planet: The View
From Space. Cambridge Univ. Press, pp. 44–52.
Hamid, E.F., Kawasaki, Z.-I., Mardiana, R., 2001. Impact of the 1997–98 El Niño event on
lightning activity over Indonesia. Geophys. Res. Lett. 28, 147–150.
Hellmuth, M.E., Mason, S.J., Vaughan, C., van Aalst, M.K., Choularton, R., 2011. A better climate for disaster risk management. International Research Institute for Climate and
Society (IRI). Columbia University, New York, USA.
Holton, J.R., 1967. The diurnal boundary layer wind oscillation above sloping terrain.
Tellus 19, 199–205. http://dx.doi.org/10.1111/j.2153–3490.1967.tb01473.x.
Lee, A.C.L., 1990. Bias elimination and scatter in lightning location by the VLF arrival time
difference technique. J. Atmos. Ocean. Technol. 719.
Lynn, B.H., Yair, Y., Price, C., Kelman, G., Clark, A.J., 2012. Predicting cloud-to-ground and
intracloud lightning in weather forecast models. Weather Forecast. 27, 1470–1488.
http://dx.doi.org/10.1175/WAF-D-11-00144.1.
Mason, S.J., Baddour, O., 2008. Statistical modelling. In: Troccoli, A., Harrison, M.,
Anderson, D.L.T., Mason, S.J. (Eds.), Seasonal Climate: Forecasting and Managing
RiskVol. 82 of Nato Science Series. IV: Earth and Environmental Sciences, Springer
Netherlands, Dordrecht.
Mason, S.J., Stephenson, D., 2008. How do we know whether seasonal climate forecasts
are any good? In: Troccoli, A., Harrison, M., Anderson, D.L.T., Mason, S.J. (Eds.), Seasonal Climate: Forecasting and Managing RiskVol. 82 of Nato Science Series. IV:
Earth and Environmental Sciences, Springer Netherlands, Dordrecht, pp. 259–289
McCaul, E.W., Goodman, S.J., LaCasse, K.M., Cecil, D.J., 2009. Forecasting lightning threat
using cloud-resolving model simulations. Weather Forecast. 24, 709–729. http://dx.
doi.org/10.1175/2008WAF2222152.1.
Moron, V., Robertson, A.W., Ward, M.N., Camberlin, P., 2007. Spatial coherence of tropical
rainfall at the regional scale. J. Clim. 20, 5244–5263. http://dx.doi.org/10.1175/
2007JCLI1623.1.
Muñoz, Á.G., Díaz-Lobatón, J.E., 2011. The Catatumbo Lightning: A Review. Mem. XIV Int.
Conf. Atmos. Electr. pp. 1–4.
Muñoz, Á.G., Díaz-Lobatón, J.E., 2012. Relámpagos del Catatumbo y el Flujo energético
Medio en la Cuenca del Lago de Maracaibo. Reporte Público CMC-GEO-01-2012.
Centro de Modelado Científico (CMC). Universidad del Zulia (12 pp).
Muñoz, Á.G., Recalde, C., 2010. North Western South America Retrospective Simulation.
CMC-OLE2-INAMHI. Available at http://iridl.ldeo.columbia.edu/SOURCES/.U_Zulia/.
CMC/.
162
Á.G. Muñoz et al. / Atmospheric Research 172–173 (2016) 147–162
Muñoz, Á.G., López, M.P., Velásquez, R., et al., 2010. An environmental watch system for
the Andean countries: El Observatorio Andino. Bull. Amer. Meteor. Soc. 1645–1652
http://dx.doi.org/10.1175/2010BAMS2958.1.
Muñoz, Á.G., Ruiz, D., Ramírez, P., León, G., Quintana, J., Bonilla, A., Torres, W., Pastén,
M., Sánchez, O., 2012. Risk Management at the Latin American Observatory. Risk
Management — Current Issues and Challenges, N. Banaitiene, Ed., InTech.
pp. 532–556.
Muñoz, Á.G., Núñez, A., Chourio, X., Díaz-Lobatón, J., Márquez, R., Moretto, P., Juárez, M.,
Casanova, V., Quintero, A., Zurita, D., Colmenares, V., Vargas, L., Salcedo, M.L.,
Padrón, R., Contreras, L., Parra, H., Vaughan, C., Smith, D., 2015a. Reporte Final de la
Expedición Catatumbo: Abril 2015. Reporte Público CMC-01-2015. Centro de
Modelado Científico (CMC). Universidad del Zulia (20 pp. DOI: 10.13140/
RG.2.1.1351.0566).
Muñoz, Á.G., Goddard, L., Robertson, A.W., Kushnir, Y., Baethgen, W., 2015b. Cross-time
scale interactions and rainfall extreme events in southeastern South America for
the austral summer. Part I: potential predictors. J. Clim. 28, 7894–7913. http://dx.
doi.org/10.1175/JCLI-D-14-00693.1.
Ochoa, A., Pineda, L., Crespo, P., Willems, P., Cayetano, S., De Cuenca, U., Brussel, V.U.,
Engineering, H., 2013. Evaluation of TRMM 3B42 Precipitation Estimates and WRF
Retrospective Precipitation Simulation Over the Pacific–Andean Region of Ecuador
and Peru. pp. 3179–3193 (doi:10.5194/hess-18-3179-2014).
Price, C., Rind, D., 1992. A simple lightning parameterization for calculating global lightning distributions. J. Geophys. Res. Atmos. 97, 9919–9933. http://dx.doi.org/10.
1029/92JD00719.
Pulwarty, R., Barry, R., Hurst, C., Sellinger, K., Mogollon, L., 1998. Precipitation in the venezuelan andes in the context of regional climate. Meteorog. Atmos. Phys. 67, 217–237.
Recalde-Coronel, G.C., Barnston, A.G., Muñoz, Á.G., 2014. Predictability of December–April
rainfall in coastal and Andean Ecuador. J. Appl. Meteorol. Climatol. 53, 1471–1493.
http://dx.doi.org/10.1175/JAMC-D-13-0133.1.
Saha, S., et al., 2014. The NCEP climate forecast system version 2. J. Clim. 27, 2185–2208.
http://dx.doi.org/10.1175/JCLI-D-12-00823.1.
Shafer, P.E., Fuelberg, H.E., 2008. A perfect prognosis scheme for forecasting warm-season
lightning over Florida. Mon. Weather Rev. 136, 1817–1846. http://dx.doi.org/10.
1175/2007MWR2222.1.
Smith, T.M., Reynolds, R.W., Peterson, T.C., Lawrimore, J., 2008. Improvements to NOAA's
historical merged land–ocean surface temperature analysis (1880–2006). J. Climate
21, 2283–2296. http://dx.doi.org/10.1175/2007JCLI2100.1.
Velásquez, R., 2000. Mecanismos Físicos de Variabilidad Climática y de eventos extremos
en Venezuela (Thesis). Universidad de Costa Rica, San José de Costa Rica (118 pp.).
Virts, K.S., Wallace, J.M., Hutchins, M.L., Holzworth, R.H., 2013. Highlights of a new
ground-based, hourly global lightning climatology. Bull. Am. Meteorol. Soc. 94,
1381–1391. http://dx.doi.org/10.1175/BAMS-D-12-00082.1.
Whyte, F., Taylor, M., Stephenson, T., Campbell, J., 2008. Features of the caribbean low
level jet. Int. J. Climatol. 28, 119–128.
Wiel, B.J.H.V.d., Moene, A.F., Steeneveld, G.J., Baas, P., Bosveld, F.C., Holtslag, A.A.M., 2010.
A conceptual view on inertial oscillations and nocturnal low-level jets. J. Atmos. Sci.
67, 2679–2689.
Williams, E.R., Boldi, B., Matlin, A., Weber, M., Hodanish, S., Sharp, D., Goodman, S.,
Raghavan, R., Buechler, D., 1999. The behavior of total lightning activity in severe
Florida thunderstorms. Atmos. Research 51, 245–265.
Wong, J., Barth, M.C., Noone, D., 2013. Evaluating a lightning parameterization based on
cloud-top height for mesoscale numerical model simulations. Geosci. Model Dev. 6,
429–443. http://dx.doi.org/10.5194/gmd-6-429-2013.
World Meteorological Organization (WMO), 1953. World Distribution of Thunderstorm
Days, Publ. 21, TP 6 and Suppl. (1956), WMO, Geneva, Switzerland.
Zepka, G.S., Pinto, O., Saraiva, A.C.V., 2014. Lightning forecasting in southeastern Brazil
using the WRF model. Atmos. Res. 135–136, 344–362. http://dx.doi.org/10.1016/j.
atmosres.2013.01.008.