water

Article
Validation of Satellite Estimates (Tropical Rainfall
Measuring Mission, TRMM) for Rainfall Variability
over the Pacific Slope and Coast of Ecuador
Bolívar Erazo 1,2,3, *, Luc Bourrel 1 , Frédéric Frappart 1,4 , Oscar Chimborazo 5 ID
, David Labat 1 ,
Luis Dominguez-Granda 6 , David Matamoros 6 and Raul Mejia 7
1 UMR 5563 GET, Université de Toulouse—CNRS—IRD—OMP—CNES, 14 Avenue Edouard Belin,
31400 Toulouse, France; [email protected] (L.B.); [email protected] (F.F.);
[email protected] (D.L.)
2 Empresa Pública Metropolitana de Agua Potable y Saneamiento (EPMAPS–Agua de Quito),
Quito 17-03-0330, Ecuador
3 Departamento de Ingeniería Civil y Ambiental, Facultad de Ingeniería Civil y Ambiental, Escuela
Politécnica Nacional, Quito 17-01-2759, Ecuador
4 UMR 5566 LEGOS, Université de Toulouse—CNRS—IRD—OMP—CNES, 14 Avenue Edouard Belin,
31400 Toulouse, France
5 Department of Atmospheric and Environmental Sciences, University at Albany, State University of New
York, Albany, NY 12222, USA; [email protected]
6 Facultad de Ciencias Naturales y Matemáticas, Centro del Agua y Desarrollo Sustentable, Campus Gustavo
Galindo, Escuela Superior Politécnica del Litoral (ESPOL) Km. 30.5 Vía Perimetral, P.O. Box 09-01-5863,
Guayaquil, Ecuador; [email protected] (L.D.-G.); [email protected] (D.M.)
7 Instituto Nacional de Meteorología e Hidrología (INAMHI), Proceso Desconcentrado Cuenca del Guayas,
Guayaquil 090150, Ecuador; [email protected]
* Correspondence: [email protected]

Received: 28 February 2017; Accepted: 5 February 2018; Published: 16 February 2018

Abstract: A dense rain-gauge network within continental Ecuador was used to evaluate the quality
of various products of rainfall data over the Pacific slope and coast of Ecuador (EPSC). A cokriging
interpolation method is applied to the rain-gauge data yielding a gridded product at 5-km resolution
covering the period 1965–2015. This product is compared with the Global Precipitation Climatology
Centre (GPCC) dataset, the Climatic Research Unit–University of East Anglia (CRU) dataset, the
Tropical Rainfall Measuring Mission (TRMM/TMPA 3B43 Version 7) dataset and the ERA-Interim
Reanalysis. The analysis reveals that TRMM data show the most realistic features. The relative bias
index (Rbias) indicates that TRMM data is closer to the observations, mainly over lowlands (mean
Rbias of 7%) but have more limitations in reproducing the rainfall variability over the Andes (mean
Rbias of −28%). The average RMSE and Rbias of 68.7 and −2.8% of TRMM are comparable with
the GPCC (69.8 and 5.7%) and CRU (102.3 and −2.3%) products. This study also focuses on the
rainfall inter-annual variability over the study region which experiences floods that have caused high
economic losses during extreme El Niño events. Finally, our analysis evaluates the ability of TRMM
data to reproduce rainfall events during El Niño years over the study area and the large basins of
Esmeraldas and Guayas rivers. The results show that TRMM estimates report reasonable levels of
heavy rainfall detection (for the extreme 1998 El Niño event) over the EPSC and specifically towards
the center-south of the EPSC (Guayas basin) but present underestimations for the moderate El Niño
of 2002–2003 event and the weak 2009–2010 event. Generally, the rainfall seasonal features, quantity
and long-term climatology patterns are relatively well estimated by TRMM.

Keywords: rainfall variability; El Niño events; in situ data; TRMM; Ecuadorian Pacific slope and
coast; Esmeraldas and Guayas basins

Water 2018, 10, 213; doi:10.3390/w10020213 www.mdpi.com/journal/water

Water 2018, 10, 213 2 of 23

1. Introduction
Spatio-temporal analysis of rainfall is crucial for water-resource management including water
supply, risk management, sustainable agriculture and hydrological infrastructure. These aspects
must be addressed and discussed before promulgating public policies in order to achieve the best
climate-adapted development. Over South America, at continental scale, the rainfall distributions
and related processes, such as moisture sources and transport, atmospheric circulation over oceans
and continents, and the Andes range forcing, are fairly well-documented [1–4]. At regional scale, the
Ecuadorian Pacific slope and coast (EPSC) is an area of particular interest due to its physiographic
features (surface, altitudinal range and the considerable horizontal distance from the coastal border to
the watershed division on the high Andes) because they have a strong impact on the spatial variability
of rainfall. In addition, the El Niño-Southern Oscillation (ENSO) is commonly identified as the main
driver of temporal rainfall variability along the Ecuadorian coastal region and how the influence of
this is different on the Andes [5].
High-rainfall events over the EPSC, generally associated with El Niño events, are responsible for
increases in runoff that cause major floods over Ecuador and Peru. The results of a 35-year simulation
of rivers’ runoff over the Pacific Slope and coast of South America (PSCSA) showed that 15% of the total
PSCSA runoff comes from the EPSC [6] making it one of the main runoff surfaces over the PSCSA. By
comparison, the Peruvian Pacific slope produces 17% of the PSCSA runoff over the area that is six times
the area of the EPSC [6]. This highlights the importance of conducting more detailed climatological
and hydrological studies over the whole EPSC and also in its two largest basins (the Guayas and
Esmeraldas basins) considering different types of ENSO events in terms of strength and seasonality.
The rainfall distribution and anomalous heavy rainfall in the coastal area of Ecuador are known
to be related to the strong positive Sea-Surface Temperature Anomalies (SSTA) in the El Niño 1 + 2
region (N1 + 2) located between 0–10◦ S/80–90◦ W [7]. The spreading of atmospheric instability in
the N1 + 2 Pacific region to the eastern escarpment of the Andes could be a result of the temporary
eastward shift of the Walker circulation [8]. Moreover, over the Andes, the rainfall patterns are driven
by the influence of both the Pacific Ocean and the Amazon basin [8] and the combinations of regional
and local atmospheric processes which interact with the topography [5,7,9–11]. Currently, various
datasets are available to study different aspects of the Ecuadorian climate, such as the spatio-temporal
rainfall patterns over this region. These datasets include the best rainfall estimates from gauge analyses
such as the best-estimate precipitation rate with multiple independent precipitation estimates of the
Tropical Rainfall Measuring Mission (TRMM) sensors and rain-gauge analysis (TRMM 3B43 monthly
Version 7 product) or called TRMM Multi-Satellite Precipitation Analysis (TMPA/3B43) [12] that later
we will name only TRMM.
A few studies have investigated the rainfall patterns over Ecuadorian areas [9,13,14], but
they do not examine in detail the entire EPSC surface. One of the objectives of this study is to
better understand rainfall behavior over the EPSC. A critical step to achieving this goal consists in
identifying the best regionally available dataset (e.g., based on synoptic observations from in situ
networks, model reanalyzes, or derived from remote sensing) to represent the rainfall patterns over
the EPSC. Consequently, this identified dataset will provide a more realistic framework to advance
further hydro-climatic studies. Of course, in situ observations, that pass a quality-control process,
constitute the most valuable source of information for climate studies. However, post-processing of
satellite information contribute to enhancing the products that are only based on in situ observations,
particularly on areas where it is too difficult to install a weather station. Over South America,
TRMM products were used for regional analyses of rainfall variability already tested, e.g., the
Peruvian [15–17] and Central Andes [18], Brazil [19], Andean–Amazon River Basins [20,21] or the
Amazon Basin [22]. Therefore, this study aims to test whether TRMM information represents the
climatological conditions of the EPSC obtained from in situ observations better than the other three
datasets (Global Precipitation Climatology Centre (GPCC), Climatic Research Unit–University of East
Anglia (CRU), and ERA-Interim Reanalysis). To determine which global dataset provides the better
Water 2018, 10, 213 3 of 23

results, a monthly 5-km resolution product was generated from the rain gauge network that covers the
entire EPSC region. This product served as reference for the comparison with the four other datasets.
This study is organized as follows: the details of the study area are presented in Section 2, whilst
Section 3 presents the data. First, the quality-control process applied to the information of all available
rain gauges in Ecuadorian territory that are maintained by the Meteorological and Hydrological
National Institute of Ecuador (INAMHI); second, the process to generate the 5-km gridded rainfall
dataset applying the cokriging method (COK) to the rain-gauge data; and last, a brief overview of the
other rainfall products. Section 4 presents methods for comparing the different products based on
statistical metrics, principal component analysis (PCA), and an analysis of selected El Niño rainfall
events corresponding to different types and amplitudes. Section 5 is dedicated to the summary of the
results. Section 6 presents the discussion. Finally, Section 7 presents the conclusions of this work.

2. Study Area
Ecuador is located in north-western South America, between Colombia and Peru, between
81.03◦ W–75.16◦ W, 1.48◦ N–5.04◦ S. Ecuador extends from the Pacific coast in the west to the Amazon
plain in the east. Following a north–south direction, the Andes range crosses the entire Ecuadorian
territory. Along this section, the Andes are divided in two main chains, the western and eastern ranges.
These two quasi-parallel lines form an inter-Andean zone characterized by several valleys where
many human settlements are found, including the capital of Ecuador, Quito. The highest watershed
altitude divides the territory into two large drainage surfaces, with main flow directions towards the
Pacific Ocean and the Amazon basin respectively (Figure 1). Over the western Andes slopes, rainfall is
produced from moist air coming from the Pacific Ocean, whilst over the western Andes the moisture
comes from the Amazon basin and the Atlantic Ocean. The eastern side, through the trade winds,
generally receive more moist air than the western slope [5,23]. In addition, the inter-Andean valleys
are influenced by both the oceanic and continental air masses [5], the prevailing easterly moisture flow
extends across the mountains depending on the speed of trade winds, especially in the south (around
3◦ S), where the mountain chain is generally lower [10].
Our study area, the EPSC, is delimited to the west by the Pacific Ocean and to the east by the
Andes watershed division. From west to east, the EPSC can be divided into the coastal region, the
low-altitude coastal cordillera (extending from 1◦ N to 2◦ S, with a maximum altitude of 860 m.a.s.l.),
an inland low valley, the western flanks of the Andes, the western high Andes ranges until they reach
the tropical glaciers, and finally an inter-Andean region in the north (Figure 1). The EPSC covers an
area of ~116,436 km2 and represents about 47% of Ecuadorian territory with a total wide range of
altitudes varying from 0 to 5870 m.a.s.l. from the coastal border to the higher Andes summits. Due
to the complex topography of the study area, 74 basins are delimited according to level five of the
Pfafstetter methodology [24]. The Esmeraldas and Guayas basins are the largest of the EPSC, covering
19,680 km2 and 32,300 km2 respectively, together representing 44.6% of the EPSC surface.
The singular rainfall distribution of the EPSC is related to the two relevant mountains chains. The
coastal border is characterized by low rainfall (<600 mm/year); the rainfall amount increases over the
low coastal cordillera; and eastwardly, between this chain and the start of the Andes foothills, rainfall
amounts reach the maximum of the region (>2000 mm/year). Then, to the east, rainfall decreases
with altitude towards the high Andes (~400 to 1200 mm) [25]. Over the entire region, large rainfall
variability is associated with the influence of the Pacific Ocean warming during extreme El Niño
events [7], which induce extensive floods that can become devastating during the extreme El Niño
years [26] over the lowlands.
Water 2018, 10, 213 4 of 23
Water 2018, 10, x FOR PEER REVIEW 4 of 23

1. Topographical
Figure 1.
Figure Topographicalmap mapofofcontinental
continentalEcuador
Ecuador obtained
obtained using thethe
using digital elevation
digital model
elevation of the
model of
the Shuttle Radar Topographic Mission (SRTM) at 1 km of spatial resolution. The Pacific slopecoast
Shuttle Radar Topographic Mission (SRTM) at 1 km of spatial resolution. The Pacific slope and and
(study(study
coast area) is delineated
area) with awith
is delineated red aline
redand
linethe
andtwo
thelargest basinsbasins
two largest (Esmeraldas and Guayas)
(Esmeraldas with
and Guayas)
orange
with and purple
orange lines lines
and purple respectively. The other
respectively. hydrographic
The other basinsbasins
hydrographic are delineated with black
are delineated lines
with black
according
lines to level
according to5level
of the5 Pfafstetter methodology.
of the Pfafstetter There are
methodology. in total
There are74
inbasins forbasins
total 74 the Pacific slope
for the and
Pacific
coast and three basins for the Amazon slope.
slope and coast and three basins for the Amazon slope.

3. Data

3.1. In
3.1. In Situ
Situ Rainfall
Rainfall Data
Data
The in
The in situ
situ observations
observations areare composed
composed ofof the
the monthly
monthly rainfall
rainfall records
records from
from 325
325 selected
selected
meteorological stations (262 gauges on the Pacific slope and coast, and 63 on the Amazon slope)slope)
meteorological stations (262 gauges on the Pacific slope and coast, and 63 on the Amazon with
with at least
at least 10 years
10 years of data
of data overover the 1965–2015
the 1965–2015 period
period (Figure
(Figure 2). The
2). The meteorological
meteorological stations
stations network
network is
is managed by INAMHI. The method applied to select the stations with valid long-term
managed by INAMHI. The method applied to select the stations with valid long-term records is records is
described in Section 3.1.1.
described in Section 3.1.1.
Water 2018, 10, 213 5 of 23
Water 2018, 10, x FOR PEER REVIEW 5 of 23

Figure
Figure 2. The2. rain
The gauges
rain gauges
of theofPacific
the Pacific slopecoast
slope and and (EPSC)
coast (EPSC) network
network (the
(the 262 262 stations
stations are
are represented
represented using circles and the colors represent the rainfall record length of the stations for the
using circles and the colors represent the rainfall record length of the stations for the 1965–2015 period).
1965–2015 period). The rain gauges of the Amazon slope (63 stations) are represented using triangles.
The rain gauges of the Amazon slope (63 stations) are represented using triangles. The study area
The study area (EPSC) is delineated using a black and bold line. The Esmeraldas and Guayas basins
(EPSC) is delineated using a black and bold line. The Esmeraldas and Guayas basins are delineated
are delineated using orange and purple lines, respectively. The topography is represented using the
usingdigital
orange and purple lines, respectively. The topography is represented using the digital elevation
elevation model (SRTM at 1 km of spatial resolution).
model (SRTM at 1 km of spatial resolution).
3.1.1. Data Homogenization and Validation
3.1.1. Data Homogenization and Validation
The raw data from the rain gauge network operated by INAMHI (370 gauges) were first
quality-checked
The raw data with fromthe themethodology
rain gaugeapplied
network in [27,28].
operatedIn the
byfollowing
INAMHI analysis a valid station
(370 gauges) were first
record should contain at least 10 years of observations and pass
quality-checked with the methodology applied in [27,28]. In the following analysis a valid the quality-assessment andstation
regionalization processes using the Regional Vector Method (RVM) [29]. The RVM assumes that for
record should contain at least 10 years of observations and pass the quality-assessment and
the same rainfall regime in a climatic zone, the total annual rainfall presents a
regionalization processes using the Regional Vector Method (RVM) [29]. The RVM assumes that for
pseudo-proportionality (little random variation) associated with the rain distribution in the zone.
the same
Basedrainfall
on thisregime
method,in atheclimatic zone,
coherence of the
thetotal
gaugeannual
data rainfall
was checkedpresents a pseudo-proportionality
grouping the gauges by
(littlewatersheds
random variation) associated with the rain distribution
and altitudinal ranges. Then, they were regrouped iteratively in the zone. Based on thistheir
to check method,
the coherence
homogeneity.of The
the main
gaugestatistical
data was checked
criteria groupingstations
for regrouping the gauges by watersheds
are based on thresholdsand altitudinal
applied to
ranges.the Then, they
standard were regrouped
deviation iteratively
of the differences to check
between annualtheir homogeneity.
pluviometrical The main
indices statistical
of stations and thecriteria
regional vector
for regrouping indices;
stations areandbasedto the
on correlation
thresholdscoefficient
applied tobetween the regional
the standard vectorofand
deviation theannual
differences
betweenpluviometric values of stations
annual pluviometrical [27,28].
indices ofDuring
stationsthis
andprocess, an exhaustive
the regional geographical
vector indices; and tosupervision
the correlation
was conducted using a background isohyet map with all data under review.
coefficient between the regional vector and annual pluviometric values of stations [27,28]. This allowed excludingDuring
stations with doubtful data and those that did not correspond to any group and did not represent
this process, an exhaustive geographical supervision was conducted using a background isohyet map
real climate zones. After this process, 45 gauges were excluded.
with all data under review. This allowed excluding stations with doubtful data and those that did
Belonging to the EPSC, 262 stations were grouped and selected; however, due to the
not correspond to anyover
low-gauge density group the and
Amazondid region,
not represent real climate
only 63 stations were zones. After
taken into this process,
account 45 gauges
for this region.
were excluded.
Belonging to the EPSC, 262 stations were grouped and selected; however, due to the low-gauge
density over the Amazon region, only 63 stations were taken into account for this region. The
information from these stations was used to perform an interpolation of the rainfall for all the
Ecuadorian continental territory.
Water 2018, 10, 213 6 of 23

3.1.2. Rainfall Data Interpolation

Mountainous or sparsely populated regions often lack stations, so the meteorological information
that represents these places does not exist or is limited. To solve these problems, spatial interpolation
methods are convenient approaches. These techniques create continuous data over the region with
missing information from sampled point values adjacent to a determined location.
The 325 validated gauges were interpolated using the cokriging method (COK) [30] for the
whole Ecuadorian territory but the results are only applied to the EPSC. The Digital Elevation Model
(DEM) of the Shuttle Radar Topographic Mission (SRTM) at 1-km resolution, provided by the National
Aeronautics and Space Administration–National Geospatial-Intelligence Agency (NASA–NGA) and
available at [31] was used as the external covariable for COK. Considering the gauge network density
and the covariable resolution, the results of COK were obtained at a spatial resolution of 5 km. The R
language and the libraries raster [32], gstat [33], were mainly used to perform the COK on the rainfall
records. The library automap [34] was used to get the best fit for the variogram model. The best values
to fit the variogram model were selected by testing the following four models: exponential, spherical,
Gaussian and Matern–Stein. Considering the whole period, the best-fit indices were obtained with the
exponential model, and therefore it was chosen.
The COK method was selected to interpolate the rainfall data because it allows avoiding the
instability caused by highly redundant secondary data [35]. Improved results over the Andes have
already been shown using this method [36] as well as over complex terrain in general [37]. However,
despite the advantages of COK, the main weakness of the Kriging method is the tendency to produce
maximum rainfall values over the summits rather than on the slopes [38]. Despite this limitation,
the COK interpolation produces better results compared to those obtained in a previous study using
simple Kriging and Cressman [39]. The gridded rainfall data obtained with COK represents adequately
the rainfall variation by altitude. This better representation is thanks to the DEM data used by the
COK method, which allows representing coherent rainfall changes by altitude, especially over the
slope and the Andes.
In order to show a brief summary of the EPSC features and the results obtained with the validated
gauges, Figure 3a shows the topography of the study area using SRTM (1 km of spatial resolution)
and Figure 3b presents a three-dimensional view of the relief. The average annual rainfall map for the
1965–2015 period obtained using the COK spatial rainfall interpolation is presented in Figure 3c. The
monthly mean rainfall variability averaged over the whole region is presented in Figure 3d.
The rainfall spatial distribution (Figure 3c) shows the lowest rainfall region (<750 mm/year)
located on the central coastal border between latitudes 1◦ S and 2.5◦ S. This is associated with the
limited displacement of the Inter-Tropical Convergence Zone (ITCZ) to the south in normal conditions,
which is truncated by the cold water of the south-west Pacific inhibiting the development of convection
processes [40]. Towards to the east, the rainfall increases over the low-altitude coastal cordillera (~750
to 1500 mm/year) and reaches the highest rainfall (~1500 to 3500 mm/year) on an inter-valley between
the low coastal cordillera and the Andes (1◦ N to 1◦ S and 79.5◦ W to 78.5◦ W). Finally, on the eastern
side of the EPSC, over the Andes range, the rainfall decreases (~750 to 1750 mm/year).
Water 2018, 10, 213 7 of 23
Water 2018, 10, x FOR PEER REVIEW 7 of 23

(mm)
3500

3000

0 0
2500

2000

Latitude
Latitude

2S 2S
1500

1000

4S 4S
Altitude (m.a.s.l)
500

0
81 W 80 W 79 W 78 W 81 W 80 W 79 W 78 W

Longitude Longitude

(a) (c)

(b) (d)
Figure 3. (a) Pacific slope and coast (EPSC) topography; and (b) south–north perspective
Figure 3. (a) Pacific slope and coast (EPSC) topography; and (b) south–north perspective
3D-topography view showing the low-altitude costal cordillera (west) and the Andes (east); (c)
3D-topography view showing the low-altitude costal cordillera (west) and the Andes (east);
multiyear average rainfall distribution from 325 rain gauges (262 stations from the EPSC) for the
(c) multiyear
1965–2015average rainfallwith
period obtained distribution from
the cokriging 325 rain method
interpolation gauges (COK)
(262 stations from
at 5 km grid the EPSC)
resolution in for
the 1965–2015 period
mm/year; and obtained
(d) the rainfallwith the cokriging
averaged interpolation
over the whole method (COK)
region (mm/month) for theat1965–2015
5 km grid resolution
period.
in mm/year; and (d) the rainfall averaged over the whole region (mm/month) for the 1965–2015 period.
3.2. Rainfall Products
3.2. Rainfall
TheProducts
gridded in situ rainfall obtained with COK interpolation as described in Section 3.1.2 was
used to evaluate four commonly used gridded monthly rainfall datasets. They include two global
The gridded in situ rainfall obtained with COK interpolation as described in Section 3.1.2 was used
gridded gauge-analysis products (CRU and GPCC), a reanalysis product (ERA-Interim) and the
to evaluate four commonly used gridded monthly rainfall datasets. They include two global gridded
satellite-based estimate product from the TRMM monthly rainfall estimates with 3B34 algorithm,
gauge-analysis products
version 7. The (CRU
details of and GPCC),
the products usedaare
reanalysis
shown inproduct
Table 1. (ERA-Interim) and the satellite-based
estimate product from the TRMM monthly rainfall estimates with 3B34 algorithm, version 7. The
details of the products used are shown in Table 1.
Water 2018, 10, 213 8 of 23

Table 1. Description of the regional gridded interpolated rain-gauge product and four external rainfall
datasets used in this study.

Product and Spatial Time Period Using Period

Source and Reference
Version Resolution Available (Common Period)
Ecuadorian rain gauges
Meteorological and
Interpolated Hydrological National
5 km (0.04◦ ) 1965–2015 1970–2015
Rainfall (Cokring) Institute of Ecuador
(INAMHI)
Global rain-gauge dataset
Global Precipitation
Climatology Centre
GPCC, v7 0.5◦ 1901–2013 (GPCC), National 1970–2013
Meteorological Service
of Germany [41]
Climate Research Unit
CRU, v3.22 0.5◦ 1901–2013 (CRU), University of 1970–2013
East Anglia [42]
Reanalysis
European Centre for
ERA-Interim
Medium-Range
(Synoptic monthly 0.125◦ 1979–present 1979–2015
Weather Forecasts
mean)
(ECMWF) [43]
Satellite
National Aeronautics
and Space
Tropical Rainfall
Administration
Measuring Mission 0.25◦ 1998–2015 1998–2015
(NASA) Goddard
(TRMM) 3B43 V7
Space Flight Centre
[44]

4. Methods

4.1. Evaluation of the Rainfall Products

The interpolated in situ observations obtained with COK were considered as the reference data to
assess the quality of the four rainfall products described in Section 3.2. Three standard comparison
metrics were used: the Pearson correlations coefficient (Crr), the Root Mean Squared Error (RMSE)
and the Relative bias index (Rbias), the latter expressed in percentage (Equations (1)–(3)):
! !
n
1 P(λ, ϕ, t) − P(λ, ϕ, t) x (λ, ϕ, t) − x (λ, ϕ, t)
n − 1 t =1 ∑
∑
Crr = (1)
σp σx
s
n
1
RMSE =
n ∑ [( P(λ, ϕ, t) − x (λ, ϕ, t)]2 (2)
t =1

∑nt=1 ( P(λ, ϕ, t) − x (λ, ϕ, t))

Rbias = × 100 (3)
∑nt=1 P(λ, ϕ, t)
where n is the number of months; P is the monthly observed interpolated rainfall of the grid located at
the coordinates (λ, ϕ) in the month t; P and σP , represent respectively the mean and standard deviation
of P; x, corresponds to the monthly series rainfall of the compared product; x and σx represent,
respectively, the mean and standard deviation of x.
Water 2018, 10, 213 9 of 23

The comparisons were performed over the longest time period of common availability by
resampling the in situ observations at the same original resolution of the four tested products (see
Table 1): for ERA-Interim at 0.125◦ , for TRMM at 0.25◦ and for the rainfall products GPCC, CRU at 0.5◦ .

4.2. Evaluation of the Tropical Rainfall Measuring Mission (TRMM) Rainfall Product
Principal Component Analysis (PCA) (or the Empirical Orthogonal Functions (EOF))
technique [45,46] is a commonly used method to characterize the spatio-temporal variability of physical
fields in climate-related studies. It was applied to the centered and deseasonnalized anomalies of
the interpolated monthly gridded in situ observations obtained in Section 3.1.2, over the 1965–2015
period. Given that, TRMM shows the best comparison metrics in terms of the criteria presented in
4.1 among the four tested products; we performed a more detailed evaluation of the TRMM data
over the overlapping period with the gridded in situ observations (1998–2015). In particular, the
PCA was also applied on the centered and deseasonalized anomalies to analyze the spatio-temporal
variability of rainfall estimates from TRMM and compare them to those obtained using the gridded in
situ observations.
Furthermore, the TRMM estimates were also evaluated during El Niño events over the entire
EPSC area and over the two largest river basins (Esmeraldas and Guayas basins). The results of the
comparison will be presented for each study region and each El Niño event as the percentage of Rbias
of the TRMM rainfall estimates and the total rainfall episodes.
The duration of El Niño rainfall events were selected according to the consecutive positive values
of the Southern Oscillation Index (SOI) [47]; El Niño intensity was ranked according to the NOAA
Oceanic Niño Index (ONI) [48]; sea-surface temperature anomaly (SSTA) as weak (0.5 ◦ C to 0.9 ◦ C
SSTA), moderate (1 ◦ C to 1.4 ◦ C SSTA) and extreme (≥2 ◦ C SSTA). Finally, the events’ ranking was
fine-tuned according to the observed rainfall quantity in the study area. The events intensity for
the EPSC was also adjusted regarding the SOI values because the ONI only considers the mean of
extended reconstructed sea-surface temperature anomalies (ERSST.v4) [49] in the El Niño 3.4 region
(5◦ N–5◦ S, 120◦ –170◦ W). According to these criteria, and for the overlapping data period, the El Niño
events selected were: El Niño 1998 (January 1998–September 1998) ranked as extreme, the El Niño of
2002–2003 (January 2002–September 2003), 2007–2008 (September 2007–September 2008) ranked as
moderates and El Niño of 2009–2010 (October 2009–September 2010) ranked as weak. It should be
noted that only a part of El Niño 1997–1998 (from January 1998) could be analyzed due to the data
availability of TRMM.

5. Results

5.1. Rainfall Products Comparison

The first step consists of determining the rainfall product that provides the better rainfall estimates
compared with in situ observations. The results of the comparison between in situ observations
interpolated with COK and the four rainfall products described in Section 3.2 are presented in Table 2
and Figure 4.
Water 2018, 10, 213 10 of 23

Table 2. Results of the comparison between the four rainfall products and in situ gridded (interpolated
with COK) observations. The monthly time-series comparison is shown, along with the gridded
resolution of the comparison, the mean annual precipitation with the mean bias difference of each
product with the interpolated observations. Finally, the results of the three comparison metrics:
correlation, RMSE and relative bias. The comparison was made in reference to the common temporal
periods and the original grid resolutions of the four rainfall products compared (see Table 1).
Water 2018, 10, x FOR PEER REVIEW 10 of 23

Mean Annual
The two rainfall estimates that provide the best results are GPCC and TRMM products with Crr
Precipitation Mean Mean of Relative
higher than 0.8Compared
and Rbias lower than
Resolution 6%. The
(mm/year)advantage
major and ofMean
TRMM over GPCC
RMSE withis itsBias
higher
Index (Rbias)
Rainfall Source Times Correlation
spatial resolution (0.25° against 0.5°) and the
(Degrees) availability
Difference withof more frequently updated
Obs data.(%) TRMM
with Respect to
Series with Obs
Observations
underestimates the total rainfall (−12.5%) with a Rbias, mean −2.8%, 0.82 andObservations
in % Crr and RMSE of(mm/year) 68.7,
respectively (Table 2). TRMM exhibits the (in Parenthesis)
best score in the lowlands, with high correlation, low
RMSE and lowCommon
Interpolated Rbias compared to the other products (Figure 4). Nevertheless, the performance of
observations
TRMM is quite lowperiods 0.008
over the high-altitude 1537.1
regions, as already mentioned in several studies [16,53–57]
(COK) respectively
(Figure 4a,c). This can be explained by the fact that the rainfall data from TRMM are derived through
ERA-Interim Total 4117.4
an inverse approach from the brightness
1979–2015 0.125 temperature at the cloud0.69
top [58]. The 288.2
TRMM processing −177.1
precipitation (+167.9%)
scheme of microwave and infrared (IR) data has to cope with the highly heterogeneous terrain with
1345.5
TRMM varying
3B43 V7brightness temperatures0.25
1998–2014 which affects the rainfall estimates [18]. Generally,
0.82 68.7 the microwave −2.8
(−12.5%)
signal as seen from space is strongly dependent on surface type [17], and is affected by the presence
1331.8
of topography
GPCC V7 that is especially 0.5
1970–2013 the case over the Andes [59]. The 0.83
Andes cause strong
69.8 scattering of5.7
(−13.4%)
the electromagnetic waves emitted by the Precipitation Radar (PR). This is a large source of error for
1468.2
CRUrainfall
V3.22 estimation [38,44] and also
1970–2013 0.5 may severely affect the infrared retrieved estimation
0.66 102.3 of rainfall
−2.3
(−4.5%)
[54].

ERA−Interim GPCC CRU

TRMM 3B43 1.0 1.0
1.0 1.0

0.8 0.8 0.8 0.8

0 0
0 0

0.6 0.6 0.6 0.6

−2 −2 −2 −2
0.4 0.4 0.4 0.4

−4 0.2 −4 0.2 −4 0.2 −4 0.2

0.0 0.0
−81 −80 −79 −78 −81 −80 −79 −78 0.0 0.0
−81 −80 −79 −78 −81 −80 −79 −78

(a)
GPCC (mm) CRU (mm)
ERA−Interim (mm) TRMM 3B43 (mm) 900 900
900 900
800 800 800
800
700 700 700 700
0 0 0 0
600 600 600 600
500 500 500 500
−2 −2 400 −2 −2
400 400 400
300 300 300 300
−4 200 −4 200 200 200
−4 −4
100 100
100 100
0 0
−81 −80 −79 −78 0 0
−81 −80 −79 −78
−81 −80 −79 −78 −81 −80 −79 −78

(b)
(%) GPCC (%) CRU (%)
ERA−Interim TRMM 3B43 (%)
200 200
200 200
100 100 100
100
0 0 0 0 0 0
0 0

−100 −100 −100 −100

−2 −200 −2 −200 −2 −200 −2 −200

−300 −300 −300 −300

−4 −400 −4 −400 −4 −400 −4 −400
−500 −500
−500 −500
−81 −80 −79 −78 −81 −80 −79 −78
−81 −80 −79 −78 −81 −80 −79 −78

(c)

Figure 4. Maps of (a) linear correlations; (b) root mean squared error (RMSE) (in mm) and; (c)
Maps bias
Figure 4.relative of (a)
(in linear correlations;
%) between (b)
interpolated in root mean squared
situ observations error
with COK (RMSE)
and the four(in mm)rainfall
studied and; (c) relative
bias (in %) between
datasets interpolated
(ERA-Interim, TRMMin situV7,
3B43 observations
GPCC V7, CRUwith COK and the four studied rainfall datasets
V3.22).
(ERA-Interim, TRMM 3B43 V7, GPCC V7, CRU V3.22).
Water 2018, 10, 213 11 of 23

An analysis of variance (ANOVA) was performed on the four global rainfall products and the in
situ one, all resampled at 0.5◦ of spatial resolution (corresponding to 10,572 samples in the EPSC from
1998 to 2013 at monthly time-scale) in order to determine if the differences between the datasets are
significant. The different datasets do not perfectly follow the normal distribution, but they have almost
all similar variances (20,388, 14,849, 18,646, 24,670 mm2 for the observations, TRMM, GPCC and CRU,
respectively), except ERA-Interim (95,579 mm2 ). Results of the ANOVA give a F-statistic of 2839 and a
p-value close to 0. As the ERA-Interim does not perfectly fit the assumption of the ANOVA and also as
the results of the comparison with the in situ data show important differences with the in situ dataset,
an ANOVA on the in situ product, TRMM, GPCC and CRU was then performed. A F-statistic of 33.85
and a p-value of 7.6 × 10−22 were obtained. A final ANOVA on the in situ data, TRMM and GPCC
was performed as CRU provides quite different results from the other datasets. A F-statistic of 6.78
and p-value of 0.0011 were obtained. In all the cases, the null hypothesis was rejected, showing that
the results presented in Table 2 and Figure 4 are statistically significant.
Considering the global products only based on in situ observations, better results are obtained
using the GPCC product than the CRU (correlation 0.83 versus 0.66 and RMSE of 69 versus 102 mm).
This is likely to be due to the use of a larger number of rain gauges in GPCC than in CRU [42,50]. ERA
Interim reanalysis monthly mean data show a large rainfall overestimation over the central region of
the EPSC (Guayas basin). This was already observed in other studies in the same latitude range as
Ecuador [51] and over the Peruvian Andes [52].
The two rainfall estimates that provide the best results are GPCC and TRMM products with Crr
higher than 0.8 and Rbias lower than 6%. The major advantage of TRMM over GPCC is its higher
spatial resolution (0.25◦ against 0.5◦ ) and the availability of more frequently updated data. TRMM
underestimates the total rainfall (−12.5%) with a Rbias, mean Crr and RMSE of −2.8%, 0.82 and
68.7, respectively (Table 2). TRMM exhibits the best score in the lowlands, with high correlation, low
RMSE and low Rbias compared to the other products (Figure 4). Nevertheless, the performance of
TRMM is quite low over the high-altitude regions, as already mentioned in several studies [16,53–57]
(Figure 4a,c). This can be explained by the fact that the rainfall data from TRMM are derived through
an inverse approach from the brightness temperature at the cloud top [58]. The TRMM processing
scheme of microwave and infrared (IR) data has to cope with the highly heterogeneous terrain with
varying brightness temperatures which affects the rainfall estimates [18]. Generally, the microwave
signal as seen from space is strongly dependent on surface type [17], and is affected by the presence
of topography that is especially the case over the Andes [59]. The Andes cause strong scattering of
the electromagnetic waves emitted by the Precipitation Radar (PR). This is a large source of error for
rainfall estimation [38,44] and also may severely affect the infrared retrieved estimation of rainfall [54].

5.2. Identification of Spatio-Temporal Rainfall Variability with TRMM

As a second step, as showed in 5.1, the TRMM product was chosen among the other rainfall
products because of its evaluation scores, which were some of the best, and also because of its higher
spatial resolution (0.25◦ × 0.25◦ ).
PCA was applied on deseasonalized anomalies to both the TRMM product and the gridded
interpolated observations averaged at 0.25◦ of spatial resolution over 1998–2015. According to North’s
rule of thumb [60], the first five PCA modes are significant, but only the first three PCA modes will
be discussed because the higher-order modes explain a variance lower than 5% and are difficult to
interpret. Figure 5 shows these three first modes, which represent 76% (53, 13 and 10%, respectively)
of the explained total variance for the gridded observations and 86% (69, 10, and 7%, respectively) for
TRMM. The spatial structures of both rainfall products are very similar, especially the first spatial and
temporal component, which accounts for the highest explained variance. The explained variances
obtained are lower for the observations resampled at the 0.25◦ × 0.25◦ resolution than for TRMM.
This difference is because the PCA method is less able to represent (as explained variance) the more
detailed spatial distribution of observed resampled grid rainfall than the TRMM estimates.
Water 2018, 10, 213 12 of 23
Water 2018, 10, x FOR PEER REVIEW 12 of 23

Obs. rainfall EOF 1, Var: 53% TRMM rainfall EOF 1, Var: 69%
(mm) (mm)
500 500
400 400
0 0
300 300 Crr = 0.91
RMSE=55 mm
−2 −2
200 200

−4 100 −4 100
0 0
−81 −79 −81 −79
1st Temporal Component (PC1)
1.0 Observations
TRMM 3B43
0.5 Crr = 0.95
RMSE=0.07 mm
0.0

−0.5
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
(a)
Obs. rainfall EOF 2, Var: 13% TRMM rainfall EOF 2, Var: 10%
(mm) (mm)
400 400
300 300
0 0
200 200
Crr = 0.91
−2 100 −2 100 RMSE=45 mm
0 0
−4 −100 −4 −100
−200 −200
−81 −79 −81 −79
2nd Temporal Component (PC2)
1.0 Observations
TRMM 3B43
0.5
Crr = 0.57
0.0 RMSE=0.27 mm
−0.5

−1.0
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016

(b)
Obs. rainfall EOF 3, Var: 10% TRMM rainfall EOF 3, Var: 7%
(mm) (mm)
100 100
50 50
0 0
0 0
−50 Crr = 0.81
−2 −50 −2 RMSE=34 mm
−100 −100
−4 −4 −150
−150
−200 −200
−81 −79 −81 −79
3rdTemporal Component (PC3)
1.0 Observations
TRMM 3B43
0.5
Crr = 0.82
0.0 RMSE=0.19 mm
−0.5

−1.0
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016

(c)

Figure 5. Comparison of spatial and temporal components of the three first principal component
Figure 5. Comparison of spatial and temporal components of the first (a), second (b) and third (c) first
analysis (PCA) modes obtained on the deseasonalised rainfall anomalies for the gridded
Principal Component Analysis (PCA) modes obtained on the deseasonalised rainfall anomalies for the
observations and TRMM estimates. The comparison was performed at the spatial resolution of the
gridded
TRMM observations
3B43 product and TRMM
(0.25°) and estimates. The comparison
over the common was performed
period of availability at the
of the two spatial
datasets resolution
(1998–
of the TRMM 3B43 V7 product (0.25°) and over the common period of availability of the two datasets
2015).
(1998–2015).
5.3. Identification of Spatio-Temporal Rainfall Variability during El Niño-Southern Oscillation (ENSO)
Events
Very with TRMM
similar spatial and temporal patterns, with Crr = 0.91 and 0.95 and RSME = 55 mm and
0.07 respectively,
As a third were found
and final step,for theregions
three first PCA
weremode
studied(Figure 5a). For
for extreme, the spatial
moderate component,
and weak El Niño small
events selected in Section 4.2 over the whole EPSC and its two largest basins
differences can be observed in terms of amplitude. TRMM represents more rainfall variability: (Esmeraldas and lower
Guayas). Time series and maps of average rainfall (1998–2015 period), as well as
over the Andes foothills and higher over the lowlands. The temporal component is dominated by temporal events
the and cumulative
signature of therainfall
ENSO maps over especially
events, each study theareaextreme
and for each
eventElof
Niño
1998event
andare thepresented
moderate in event
Figure 6.
of 2002–2003. For the second PCA mode, showing a meridional dipole, the spatial patterns are
very similar between TRMM and the re-gridded deseasonalized anomalies’ observations (Crr = 0.91,
RMSE = 45 mm), although the amplitudes of the dipole centers are smaller for TRMM, especially over
the north of the EPSC (ITCZ migration region) and on the coastal center and south border (Humboldt
Water 2018, 10, 213 13 of 23

Current influence). The temporal components are quite different (Crr = 0.57 and RMSE = 0.27). In
particular, TRMM exhibits a stronger variability with respect to the observations (Figure 5b). Similar
spatial and temporal patterns with, respectively, Crr = 0.81 and 0.82 and RSME = 34 mm and 0.19 were
found for the third PCA mode. However, a clear east–west gradient is present in the in situ gridded
rainfall. It does not appear so clearly for TRMM (Figure 5c).

5.3. Identification of Spatio-Temporal Rainfall Variability during El Niño-Southern Oscillation (ENSO) Events
with TRMM
As a third and final step, three regions were studied for extreme, moderate and weak El Niño
events selected in Section 4.2 over the whole EPSC and its two largest basins (Esmeraldas and Guayas).
Time series and maps of average rainfall (1998–2015 period), as well as temporal events and cumulative
rainfall maps over each study area and for each El Niño event are presented in Figure 6.

5.3.1. Pacific Slope and Coast

For the EPSC, the monthly averaged time variations of the observed rainfall were well estimated
in the TRMM product. They present a low Rbias of −6.9% with respect to the observations. As shown
in Table 3, the 1998 El Niño event was the strongest rainfall event of the entire period 1998–2015 over
the EPSC with 75.1% above the pluri-annual monthly average rainfall. Next was the moderate El
Niño event of 2007–2008, with a cumulative rainfall event of 9.3% above the monthly average. The
moderate and weak events of 2002–2003 and 2009–2010, with an average rainfall of −3.5% and −0.8%,
respectively, were the closest to the monthly average of the EPSC. The best represented event by TRMM
was the moderate 2002–2003 El Niño, and the worst was the weak 2009–2010 El Niño (Figure 6a).
The spatial and temporal results presented in Figure 6a show that the 1998 El Niño, for the
9 months considered, produced a maximum cumulative rainfall event between 3500 and 4000 mm
over most of the lowlands and the high rainfall was observed towards the south until 2.5◦ S. During
2002–2003 El Niño, a maximum cumulative rainfall event of 6000 mm in 21 months was located in
the northern region (~1◦ N to 0.5◦ S). For the 2007–2008 El Niño, the maximum cumulative rainfall
event of 3500 mm in 13 months was located over the north of the region (~1◦ N to 0.5◦ S). During the
2009–2010 El Niño, with a duration of 12 months, the maximum cumulative rainfall event of 3500 mm
was located in the north (~1◦ N to 1◦ S). For all of these four events, high rainfall was delimited at the
west and east by the low coastal cordillera and the Andes range, respectively (Figure 6a).

5.3.2. Esmeraldas Basin

In the Esmeraldas basin, the monthly average rainfall was generally underestimated by TRMM,
with an average Rbias of −16.4%. The rainfall during El Niño events of 1998, 2002–2003, 2007–2008
and 2009–2010 exceeded by 43.2%, 3.4%, 8.1% and 5.9% the pluri-annual monthly average rainfall
observations. The extreme 1998 El Niño was the major rainfall event. The Rbias comparison of rainfall
observations and TRMM estimates of El Niño events of 1998, 2002–2003, 2007–2008 and 2009–2010 are
presented in Table 3.
The rainfall during the extreme El Niño of 1998 and the weak 2009–2010 El Niño are, respectively,
the best and worst estimated using TRMM. The rainfall during the weak 2009–2010 El Niño was the
most underestimated by TRMM. Regarding the spatial rainfall distribution of the events (Figure 6b),
the maximum accumulated rainfall was observed in the central and southern regions of this basin.
The upstream basin (in the Andes) was, in all cases, the region with least rainfall and lower variability.
This spatial distribution is in accordance with the zonal rainfall distribution showed by the multiyear
average rainfall map and the first spatial PCA mode.
Water 2018, 10, 213 14 of 23

Table 3. Percentages of cumulated observed rainfall with respect to the monthly average (period 1998–2015) for each El Niño event (1998, 2002–2003, 2007–2008
and 2009–2010) and for each study region (Ecuadorian Pacific Slope and Coast, Esmeraldas and Guayas basins) and the corresponding Rbias (in %) of the TRMM
rainfall estimates.

Ecuadorian Pacific Slope and Coast Esmeraldas Basin Guayas Basin

Time Period Observed Rainfall TRMM Estimation Observed Rainfall TRMM Estimation Observed Rainfall TRMM Estimation
Event with Respect of Observed Rainfall Event with Respect of Observed Rainfall Event with Respect of Observed Rainfall
to the Average (%) (%) to the Average (%) to the Average (%)
Available period:
−6.9 −16.4 −8.5
Jan 1998–Dec 2015
ENSO: Jan 1998–Sep 1998 75.1 7.7 43.2 6.3 100.7 8.5
ENSO: Jan 2002–Sep 2003 −3.5 −2.4 3.4 −12 −21.3 5.3
ENSO: Sep 2007–Sep 2008 9.3 −11 8.1 −13.1 0.8 −14.2
ENSO: Oct 2009–Sep 2010 −0.8 −17.1 5.9 −23.7 −7.5 −18.9
Water 2018, 10, 213 15 of 23
Water 2018, 10, x FOR PEER REVIEW 15 of 23

600
Mean rainfall observations
Mean rainfall TRMM

500
El Niño 1998
El Niño 2002−2003

mean rainfall (mm)

El Niño 2007−2008

400
El Niño 2009−2010

300
200
100
0

Dec
May
Mar
Dec

Nov
Sep
May

Jan
Mar

Feb
Nov
Sep

Aug
Apr

Oct
Jan

Jun
Feb

Aug
Apr

Oct
Jun

Jul
Jul
El Niño Jan 1998−Sep 1998
(m m )
4000

500
Mean rainfall Observations
3500 El Niño 1998 Observations

mean rainfall (mm)

El Niño 1998 TRMM

400
0 3000

200 300
2500

2S
2000 100

1500
1000
0

4S

Dec
May
Dec

Nov
Mar

Sep
May

Jan
Mar

Nov
Sep

Feb

Aug
Apr

Oct
Jan

Jun
Feb

Aug
Apr

Oct
Jun

Jul
Jul
500
0
81 W 80 W 79 W 78 W

El Niño Jan 2002−Sep 2003

(m m )
6000
500

Mean rainfall Observations

El Niño 2002-2003 Observations
5000
100 200 300 400

El Niño 2002-2003 TRMM

mean rainfall (mm)

0
4000

2 S 3000

2000
4 S
1000
0

Dec
May
Mar
Dec

Nov
Sep
May

Jan
Feb
Mar

Nov
Sep

Aug

Oct
Jan

Apr

Jun
Feb

Aug

Oct
Apr

Jun

Jul
Jul

0
81 W 80 W 79 W 78 W

El Niño Sep 2007−Sep 2008

(mm)
3500
500

Mean rainfall Observations

El Niño 2007-2008 Observations
3000
100 200 300 400

El Niño 2007-2008 TRMM

mean rainfall (mm)

0
2500

2000
2S
1500

1000
4S
0

500
May

Dec
Mar

Sep
May

Dec

Nov
Jan
Mar

Sep

Feb

Aug
Nov

Apr

Oct
Jan

Jun
Feb

Aug
Apr

Jun

Oct

Jul
Jul

0
81 W 80 W 79 W 78 W

El Niño Oct 2009−Sep 2010

(mm)
3500
500

Mean rainfall Observations

El Niño 2009-2010 Observations
3000
400

El Niño 2009-2010 TRMM

mean rainfall (mm)

0
2500
300

2000
200

2 S
1500
100

1000
4 S
500
0

May

Dec
Mar

Sep
Dec

Nov
May

Jan
Mar

Feb
Sep

Aug
Nov

Oct
Apr
Jan

Jun
Feb

Aug

Oct
Apr

Jun

Jul
Jul

0
81 W 80 W 79 W 78 W

(a)

Figure 6. Cont.
Water 2018, 10, 213 16 of 23
Water 2018, 10, x FOR PEER REVIEW 16 of 23

0 100 200 300 400 500 600 700

Mean rainfall Observations
Mean rainfall TRMM
El Niño 1998
El Niño 2002−2003

mean rainfall(mm)
El Niño 2007−2008
El Niño 2009−2010

Dec
May
Mar

Sep
Dec

Nov
May

Jan
Mar

Sep

Nov

Feb

Aug
Apr

Oct
Jan

Jun
Feb

Aug
Apr

Oct
Jun

Jul
Jul
El Niño Jan 1998− Sep 1998

100 200 300 400 500

(mm)
Mean rainfall Observations
4000
El Niño 1998 Observations
3500

mean rainfall (mm)

El Niño 1998 TRMM
0.5 N
3000
2500
0
2000
0.5 S 1500
1000
0

Dec
May
Mar

Sep
Dec

Nov
May

Jan
Mar

Sep

Feb

Aug
Nov

Apr
Jan

Jun

Oct
Feb

Aug
Apr

Oct
Jun

Jul
Jul
79.5 W 79 W 78.5 W

El Niño Jan 2002− Sep 2003

500

(mm) Mean rainfall Observations

El Niño 2002-2003 Observations
6000
100 200 300 400
mean rainfall (mm)

El Niño 2002-2003 TRMM

0.5 N 5000

4000
0
3000
0.5 S
2000
0

1000
Dec
May
Dec

Mar

Sep

Nov
May

Jan
Mar

Sep

Nov

Feb

Aug
Apr

Oct
Jan

Jun
Feb

Aug
Apr

Oct
Jun

79.5 W 79 W 78.5 W
Jul
Jul

El Niño Sep 2007− Sep 2008

500

(m m ) Mean rainfall Observations

El Niño 2007-2008 Observations
3000
400

El Niño 2007-2008 TRMM

mean rainfall (mm)

0.5 N
300

2500
200

0
2000
100

0.5 S 1500
0

1000
Dec
May
Mar

Sep
Dec

Nov
May

Jan
Mar

Sep

Feb

Aug
Nov

Apr
Jan

Jun

Oct
Feb

Aug
Apr

Jun

Oct

Jul
Jul

79.5 W 79 W 78.5 W

El Niño Oct 2009− Sep 2010

500

Mean rainfall Observations

(m m )
El Niño 2009-2010 Observations
200 300 400
mean rainfall (mm)

El Niño 2009-2010 TRMM

3000
0.5 N
2500

0 2000
100

1500
0.5 S

1000
0

Dec
May
Mar
Dec

Sep

Nov
May

Jan
Mar

Sep

Feb

Aug
Nov

Apr

Oct
Jan

Jun
Feb

Aug
Apr

Jun

Oct

Jul
Jul

79.5 W 79 W 78.5 W

(b)

Figure 6. Cont.
Water 2018, 10, 213 17 of 23
Water 2018, 10, x FOR PEER REVIEW 17 of 23

100 200 300 400 500 600 700

Mean rainfall Observations
Mean rainfall TRMM
El Niño 1998

mean rainfall (mm)

El Niño 2002−2003
El Niño 2007−2008
El Niño 2009−2010

Dec
May
Mar

Sep
Dec

Nov
May

Jan
Mar

Sep

Feb

Aug
Nov

Apr
Jan

Jun

Oct
Feb

Aug
Apr

Jun

Oct

Jul
Jul
El Niño Jan 1998− Sep 1998

0 100 200 300 400 500 600 700

(m m ) Mean rainfall Observations
4000 El Niño 1998 Observations
El Niño 1998 TRMM

mean rainfall (mm)

0.5 S
3500
3000
1 S
2500
1.5 S 2000
1500
2 S
1000
500

Dec
May
Mar

Sep
Dec

Nov
May

Jan
Mar

Sep

Feb

Aug
Nov

Apr
Jan

Jun

Oct
Feb

Aug
Apr

Jun

Oct

Jul
Jul
80 W 79.5 W 79 W

El Niño Jan 2002− Sep 2003

0 100 200 300 400 500 600 700

(m m ) Mean rainfall Observations

4500 El Niño 2002-2003 Observations
0.5 S
4000 El Niño 2002-2003 TRMM
mean rainfall (mm)

1 S 3500
3000
1.5 S
2500
2000
2 S
1500
1000

Dec
May
Mar

Sep
Dec

Nov
May

Jan
Mar

Sep

Feb

Aug
Nov

Apr
Jan

Jun

Oct
Feb

Aug
Apr

Jun

Oct

Jul
Jul

80 W 79.5 W 79 W

El Niño Sep 2007− Sep 2008

(m m )
100 200 300 400 500 600 700

3000 Mean rainfall Observations

El Niño 2007-2008 Observations
0.5 S
El Niño 2007-2008 TRMM
mean rainfall (mm)

2500
1S
2000
1.5 S
1500
2S
1000
0

Dec
May
Mar

Sep
Dec

Nov
May

80 W 79.5 W 79 W
Jan
Mar

Sep

Feb

Aug
Nov

Apr
Jan

Jun

Oct
Feb

Aug
Apr

Jun

Oct

Jul
Jul

El Niño Oct 2009− Sep 2010

0 100 200 300 400 500 600 700

(m m ) Mean rainfall Observations

3000 El Niño 2009-2010 Observations
0.5 S El Niño 2009-2010 TRMM
mean rainfall (mm)

2500
1 S

2000
1.5 S
1500
2 S
1000
Dec
May
Mar

Sep
Dec

Nov
May

Jan
Mar

Sep

Feb

Aug
Nov

Apr
Jan

Jun

Oct
Feb

Aug
Apr

Jun

Oct

Jul
Jul

80 W 79.5 W 79 W

(c)
Figure 6. Rainfall comparison of the gridded observations and TRMM estimates at 0.25° of spatial
Figure 6. Rainfall comparison of the gridded observations and TRMM estimates at 0.25◦ of spatial
resolution, during the El Niño events of 1998, 2002–2003, 2007–2008 and 2009–2010. The top image
resolution, during the El Niño events of 1998, 2002–2003, 2007–2008 and 2009–2010. The top image
corresponds to the map of the observed pluri-annual monthly average rainfall of the period 1998–
corresponds
2015 andto the
the map of the
monthly observed of
comparison pluri-annual
four El Niñomonthly
events average
with the rainfall of the period
mean rainfall 1998–2015
observed and
and the
estimated by TRMM for (a) the Pacific slope and coast, for (b) the Esmeraldas basin, and for (c) the by
monthly comparison of four El Niño events with the mean rainfall observed and estimated
TRMM for (a)basin.
Guayas the Pacific slope
The other mapsand coast,
are for (b)cumulative
the spatial the Esmeraldas
rainfall basin, and for
observations and(c)the
the Guayas basin.
comparisons
The other maps are the spatial cumulative rainfall observations and the comparisons
of the temporal average series by El Niño event, between observations and TRMM made for of thethe
temporal
(a)
Pacific slope and coast and for (b) the Esmeraldas basin and (c) Guayas basin.
average series by El Niño event, between observations and TRMM made for the (a) Pacific slope and
coast and for (b) the Esmeraldas basin and (c) Guayas basin.
Water 2018, 10, 213 18 of 23

5.3.3. Guayas Basin

In the Guayas Basin, the monthly average rainfall was also underestimated by TRMM, with an
average Rbias of −8.5% (lower than in the Esmeraldas basin). El Niño events of 1998, 2002–2003,
2007–2008 and 2009–2010 presented 100.7%, −21.3%, 0.8% and −7.5% more/less rainfall with reference
to the average of the basin (Table 3). The extreme 1998 El Niño presented the largest impact on the
rainfall variability compared to the EPSC and Esmeraldas basin. The moderate El Niño of 2002–2003
was the best estimated, and the worst was the weak El Niño of 2009–2010.
The distribution of average rainfall (Figure 6c) indicates that it was mostly concentrated in the
northern basin region (normal seasonal ITCZ influence). The largest rainfall amounts occurred in the
center and along the north–south border of the basin during the 1998 El Niño event. The total rainfall
(4000 mm in 9 months) was distributed over most of the lowlands in the region delimited by the Andes
and the low-altitude coastal cordillera. For all events, less rainfall was found in the Andean part of
the basin. This spatial distribution is in accordance with the multiyear average rainfall distribution
and with the zonal and meridional rainfall-variability distribution (N–S increase and W–E decrease)
showed by the spatial PCA modes (EOF1 and EOF2) of Figure 5.

6. Discussion
Our work presents a detailed rainfall distribution for the EPSC, which shows a significant
correlation with orographic features. The high amount of climatology rainfall is concentrated in
the north at the western windward side of the Andes and in the low coastal cordillera due to the
intense low-level convergence when the ITCZ is placed on the north of the equator (almost in line
with the oceanic ITCZ) in austral winter [4]. The mountain slopes exposed perpendicularly to frequent
winds that transport moisture [11] can produce this highest amount of rainfall. This could also be
supported by a larger cloud frequency observed in the north (~0) than the south (~4 S) [61]. The
spatial rainfall distribution over the EPSC is clearly delimited by its two mountain chains, which
act as weather divisions, mainly the Andes, as the major borderline between Pacific and Amazonian
climatic influences. These two chains have permanent interactions with tropospheric flow, which is
more remarkable during the rainy season due the ITCZ seasonal migration and the interannual ENSO
influence periodicity (ranging from 2 to 7 years [62]). The particular case of the coastal border, where
the rainfall amount is minimum, can be related to the influence of the SE Pacific anticyclone and the
cold water upwelling of the Humboldt Current in austral winter [61]. As for the temporary rainfall
distribution over the EPSC, the first rainfall seasons starts in November–December when the ITCZ
begins its southern displacement, then a second marked season, due to the direct influence of the ITCZ
on convective processes, starts in Jan–May reaching a maximum in March. A third season with lower
(minimum) rainfall occurs during Jul–Sep due to the northward shift of the ITCZ, during the northern
hemisphere summer, and the intensified Walker circulation that produces advective low cloud [63].
The largest interannual variability within the EPSC region is mostly produced by the ENSO
conditions and influenced by the seasonal meridional migration of the ITCZ. This relationship
is supported by the fact that the ITCZ migration is delayed (favored) during warm (cool) ENSO
phases [64] because the ITCZ generally migrates toward a differentially warming hemisphere [65].
The spatial component of the second EOF mode is consistent with the higher cloud frequency during
the ITCZ meridional migration over the EPSC and, therefore, closely related to the ENSO events
represented by the first EOF mode. The spatial component of the first EOF specifically reveals the
zonal rainfall variability influence of El Niño events, which is highest over the lowlands, specifically
higher over the center south (Guayas basin), low over the Andes slope, and very low over the Andes.
This was clear, for example, during the extreme 1998 El Niño, with a high rainfall variability impact
towards the center-south according to the spatial rainfall variability presented by the first EOF mode
and the event rainfall accumulation over the Esmeraldas and Guayas basins. The higher rainfall
variability for the Guayas basin (center-south region) than for the Esmeraldas basin (north region) can
be accounted for by evidence of historic strong and extreme El Niño events, which clearly separate
Water 2018, 10, 213 19 of 23

the moist northern Ecuadorian coast, under the normal influence of the ITCZ, from the south coast of
Ecuador, which is the driest region and sensitive to ENSO events [66]. It should be noted that the ITCZ
shift during warm ENSO episodes reduces rainfall by about 100 mm/year along the northern edge of
the normal ITCZ over the eastern Pacific [67], mostly in December–February (DJF) and March–May
(MAM). It is equally important to mention that the western and central Pacific ITCZ shifts southward
by about 2◦ S on typical ENSO conditions, and by about 5◦ S during strong El Niño events (such as in
1983 and 1998) [65] with the longitudinal ITCZ structure modified by ENSO’s zonal rearrangement of
convection [67].
Although the monthly global datasets as GPCC and CRU obtained by interpolating global gauges’
observations allow for fairly good rainfall data for the study region, TRMM 3B34 V7 is the better source
among all the datasets considered in this study. TRMM showed good agreement with gauge data
compared with GPCC and CRU, and it showed to be superior to the global atmospheric reanalysis of
ERA-Interim. Nevertheless, TRMM presents some overestimations over lowlands (mean Rbias of 7%)
and has more underestimations over the Andes (mean Rbias of −28%) when compared with in situ
gauges. For the El Niño rainfall events, TRMM presents mostly underestimations for the considered
El Niño events. This could be explained because the TRMM dataset is the result of the combination
of multiple independent precipitation estimates from the TRMM microwave imager (TMI), visible
and infrared scanner (VIRS), rain gauge data and the precipitation radar (PR). PR underestimates
rainfall rate for extremely intense convective rainfall [68], especially for extreme precipitating systems
that contain significant mixed phase and/or frozen hydrometeors [69], as on the Andes. There is
also the limitation of the VIRS data that provide information of cloud-top height, which do not
correlate well enough with ground precipitation [70]. Different cloud types may have similar cloud-top
temperatures and are associated with different amounts of rainfall at the ground [71]; for higher
convective cloud there are normally underestimations compared to low-level short convection [72].
Finally, the TMI also missed the light and heavy rainfall because of its small scale (swath width of
758.5 km) [73] and/or type of rainfall according to its nature as, for example, the warm rain (derived
from non ice-phase processes in clouds) [74]. As shown by [61], over Ecuadorian territory the average
cloud-top height increases from west to east during the wet season (December–May), which means
W–E rainfall cloud-top height increases; thus, this results in important underestimation over the Andes
against a reasonably small overestimation over lowlands. It could also suggest that during the lower
rainfall season (July–September), as shown in [75], TRMM overestimations over the dry areas could be
attributed to sub-cloud evaporation.

7. Conclusions
Comparison of the gridded observations with the commonly used rainfall datasets from GPCC,
CRU, ERA-Interim reanalysis, and the satellite estimates from TRMM 3B43, showed that the
satellite-based rainfall product provides the more reliable estimates. Overall, considering the 1998–2015
period, there is a good agreement between observations and TRMM with an average lowest RMSE
of 68.7 mm/year and Rbias of −2.8% for the entire EPSC. We can note that, for the lowlands, the
Rbias obtained (7%) are closer (small overestimation) to the observations than for the Andes (−28%)
(underestimation). These results can be related to the uncertainties associated with the TRMM 3B43
algorithm and the errors from the different sensors onboard the satellite (TMI, PR and VIRS) which
are responsible for underestimations of the rainfall during the wet season (December–May) when
top-cloud heights increase from W–E of the EPSC over the Andes slopes and inter-Andean basin.
Very similar spatial and temporal patterns were found, especially for the first mode (Crr = 0.91
and 0.95 and RSME = 55 mm and 0.07 mm), when applying the PCA to deseasonalized anomalies of
rainfall from TRMM 3B43 and in situ gridded observations over the EPSC between 1998 and 2015. For
the spatial component, some differences can be observed in terms of rainfall variability amplitude and
structures form over the Andes foothills (lower for TRMM) and over the lowlands (higher for TRMM).
The first temporal component is dominated by the signature of the ENSO events, especially the extreme
Water 2018, 10, 213 20 of 23

event of 1998. The first PCA spatial mode clearly shows the location of heavy rainfall impact of El
Niño events and their zonal rainfall variability influence, which is highest over the lowland and lower
towards the Andes.
The TRMM 3B43 product showed a generally good capability for providing realistic rainfall
estimates during extreme El Niño 1998 (mean Rbias of +7.7%), and moderate El Niño of 2002–2003
(mean Rbias of −2.4%) over the EPSC. Nevertheless, rainfall for the El Niño 2007–2008 and 2009–2010
events were underestimated by TRMM (mean Rbias of −11% and −17.1%) over the EPSC and more
notably underestimated for the 2009–2010 event for the Esmeraldas (−23.7%) than the Guayas basin
(−18.9%). General good agreement was also found over the Esmeraldas basin for the extreme
El Niño 1998 (mean Rbias 6.3%) and over the Guayas basins for the extreme 1998 and moderate
2002–2003 El Niño events (mean Rbias of +8.5%, +5.3%) in spite of small overestimations. All these
results confirm that TRMM 3B43 V7 reports reasonable levels of heavy rainfall detection over the
EPSC and specifically towards the center-south of the EPSC (Guayas basin) but presents a general
underestimation for the moderate and weak El Niño events. Over the whole EPSC, the seasonal
features and quantity are relatively well estimated by TRMM and the long-term climatology patterns
are well represented. The present study validates the use of remotely sensed rainfall data in regions
with sparse rain-gauge stations and high rainfall variability, taking into account the potentialities and
limitations of satellite estimates.

Acknowledgments: This work was supported by Ecuadorian National Secretary for Higher Education, Science,
Technology and Innovation (SENESCYT). Authors would like to thank INAMHI (Meteorological and Hydrological
National Institute of Ecuador) for providing the complete hydrometeorological raw dataset. We thank three
anonymous reviewers for their constructive comments on our manuscript.
Author Contributions: Bolívar Erazo, Luc Bourrel and Frédéric Frappart conceived, designed the study and
processed the data. David Labat, Oscar Chimborazo, Luis Dominguez-Granda, David Matamoros and Raul Mejia
were involved in the analysis of the results. All authors contributed to the writing of the manuscript.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Zhou, J.; Lau, K.M. Principal modes of interannual and decadal variability of summer rainfall over South
America. Int. J. Climatol. 2001, 21, 1623–1644. [CrossRef]
2. Haylock, M.R.; Peterson, T.C.; Alves, L.M.; Ambrizzi, T.; Anunciação, Y.M.T.; Baez, J.; Grimm, A.M.;
Karoly, D.; Marengo, J.A.; Marino, M.B.; et al. Trends in Total and Extreme South American Rainfall in
1960–2000 and Links with. J. Clim. 2006, 19, 1490–1513. [CrossRef]
3. Garreaud, R.D. The Andes climate and weather. Adv. Geosci. 2009, 7, 1–9. [CrossRef]
4. Garreaud, R.D.; Vuille, M.; Compagnucci, R.; Marengo, J. Present-day South American climate. Palaeogeogr.
Palaeoclimatol. Palaeoecol. 2009, 281, 180–195. [CrossRef]
5. Vuille, M.; Bradley, R.S.; Keimig, F. Climate Variability in the Andes of Ecuador and Its Relation to Tropical
Pacific and Atlantic Sea Surface Temperature Anomalies. J. Clim. 2000, 13, 2520–2535. [CrossRef]
6. Mernild, S.H.; Liston, G.E.; Hiemstra, C.; Beckerman, A.P.; Yde, J.C.; Mcphee, J. The Andes Cordillera.
Part IV: Spatio-temporal freshwater run-off distribution to adjacent seas (1979-2014). Int. J. Climatol. 2016.
[CrossRef]
7. Bendix, A.; Bendix, J. Heavy rainfall episodes in Ecuador during El Niño events and associated regional
atmospheric circulation and SST patterns. Adv. Geosci. 2006, 6, 43–49. [CrossRef]
8. Bendix, J.; Trache, K.; Palacios, E.; Rollenbeck, R.; Goettlicher, D.; Nauss, T.; Bendix, A. El Niño meets La
Niña—Anomalous rainfall patterns in the “traditional” El Niño region of southern Ecuador. Erdkunde 2011,
65, 151–167. [CrossRef]
9. Bendix, J.; Lauer, W. Die Niederschlagsjahreszeiten in Ecuador und ihre klimadynamische Interpretation
(Rainy Seasons in Ecuador and Their Climate-Dynamic Interpretation). Erdkunde 1992, 46, 118–134.
[CrossRef]
Water 2018, 10, 213 21 of 23

10. Emck, P. A climatology of south Ecuador-with special focus on the major Andean ridge as Atlantic-Pacific
climate divide. Ph.D. Thesis, Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), Erlangen, Germany,
July 2007.
11. Rollenbeck, R.; Bendix, J. Rainfall distribution in the Andes of southern Ecuador derived from blending
weather radar data and meteorological field observations. Atmos. Res. 2011, 99, 277–289. [CrossRef]
12. Bolvin, D.T.; Huffman, G.J. Transition of 3B42/3B43 research product from monthly to climatological
calibration/adjustment. In NASA Precipitation Measurement Missions Document; NASA: Washington, DC,
USA, 2015; pp. 1–11.
13. Rossel, F.; Mejía, R.; Ontaneda, G.; Pombosa, R.; Roura, J.; Calvez, R.; Cadier, E. Régionalisation de l’influence
du El Niño sur les précipitations de l’Equateur. Bulletin d'Institut Français D'études Andines 1998, 27, 643–654,
ISSN:0303-7495.
14. Morán-Tejeda, E.; Bazo, J.; López-Moreno, J.I.; Aguilar, E.; Azorín-Molina, C.; Sanchez-Lorenzo, A.;
Martínez, R.; Nieto, J.J.; Mejía, R.; Martín-Hernández, N.; et al. Climate trends and variability in Ecuador
(1966–2011). Int. J. Climatol. 2015. [CrossRef]
15. Lavado, W.; Labat, D.; Guyot, J.L.; Ronchail, J.; Ordoñez, J. TRMM rainfall data estimation over the Peruvian
Amazon-Andes basin and its assimilation into a monthly water balance model. In New Approaches to
Hydrological Prediction in Data-Sparse Regions; International Association of Hydrological Sciences: London,
UK, 2009.
16. Condom, T.; Rau, P.; Espinoza, J.C. Correction of TRMM 3B43 monthly precipitation data over the
mountainous areas of Peru during the period 1998-2007. Hydrol. Process. 2011, 25, 1924–1933. [CrossRef]
17. Mantas, V.M.; Liu, Z.; Caro, C.; Pereira, A.J.S.C. Validation of TRMM multi-satellite precipitation analysis
(TMPA) products in the Peruvian Andes. Atmos. Res. 2014. [CrossRef]
18. Scheel, M.L.M.; Rohrer, M.; Huggel, C.; Villar, D.S.; Silvestre, E.; Huffman, G.J. Evaluation of TRMM
Multi-satellite Precipitation Analysis (TMPA ) performance in the Central Andes region and its dependency
on spatial and temporal resolution. Hydrol. Earth Syst. Sci. 2011, 2649–2663. [CrossRef]
19. Franchito, S.H.; Rao, V.B.; Vasques, A.C.; Santo, C.M.E.; Conforte, J.C. Validation of TRMM precipitation
radar monthly rainfall estimates over Brazil. J. Geophys. Res. Atmos. 2009, 114. [CrossRef]
20. Negrón Juárez, R.I.; Li, W.; Fu, R.; Fernandes, K.; de Oliveira Cardoso, A. Comparison of Precipitation
Datasets over the Tropical South American and African Continents. J. Hydrometeorol. 2009, 10, 289–299.
[CrossRef]
21. Zulkafli, Z.; Buytaert, W.; Onof, C.; Manz, B.; Tarnavsky, E.; Lavado, W.; Guyot, J.-L. A Comparative
Performance Analysis of TRMM 3B42 (TMPA) Versions 6 and 7 for Hydrological Applications over
Andean–Amazon River Basins. J. Hydrometeorol. 2014, 15, 581–592. [CrossRef]
22. Frappart, F.; Ramillien, G.; Ronchail, J. Changes in terrestrial water storage versus rainfall and discharges in
the Amazon basin. Int. J. Climatol. 2013, 33, 3029–3046. [CrossRef]
23. Bendix, J.; Rollenbeck, R.; Reudenbach, C. Diurnal patterns of rainfall in a tropical Andean valley of southern
Ecuador as seen by a vertically pointing K-band Doppler radar. Int. J. Climatol. 2006, 26, 829–846. [CrossRef]
24. Pfafstetter, O. Classification of hydrographic basins: Coding methodology. Unpublished work. 1989.
25. Bourrel, L.; Melo, P.; Vera, A.; Pombosa, R.; Guyot, J.L. Study of the erosion risks of the Ecuadorian Pacific
coast under the influence of ENSO phenomenom: Case of the Esmeraldas and Guayas basins. In Proceedings
of the International Conference on the Status and Future of the World’s Large Rivers, Vienna, Austria, 11–14
April 2011; pp. 11–14.
26. Frappart, F.; Bourrel, L.; Brodu, N.; Riofrío Salazar, X.; Baup, F.; Darrozes, J.; Pombosa, R. Monitoring of the
Spatio-Temporal Dynamics of the Floods in the Guayas Watershed (Ecuadorian Pacific Coast) Using Global
Monitoring ENVISAT ASAR Images and Rainfall Data. Water 2017, 9, 12. [CrossRef]
27. Bourrel, L.; Rau, P.; Dewitte, B.; Labat, D.; Lavado, W.; Coutaud, A.; Vera, A.; Alvarado, A.; Ordoñez, J.
Low-frequency modulation and trend of the relationship between ENSO and precipitation along the northern
to centre Peruvian Pacific coast. Hydrol. Process. 2014, 29, 1252–1266. [CrossRef]
28. Rau, P.; Bourrel, L.; Labat, D.; Melo, P.; Dewitte, B.; Frappart, F.; Lavado, W.; Felipe, O. Regionalization of
rainfall over the Peruvian Pacific slope and coast. Int. J. Climatol. 2016. [CrossRef]
29. Brunet Moret, Y. Homogénéisation des précipitations. ORSTOM 1979, XVI, 147–170.
30. Cressie, N.; Johannesson, G. Fixed rank kriging for very large spatial data sets. J. R. Stat. Soc. Ser. B Stat.
Methodol. 2008, 70, 209–226. [CrossRef]
Water 2018, 10, 213 22 of 23

31. Jarvis, A.; Reuter, H.I.; Nelson, A.; Guevara, E.; others Hole-filled SRTM for the globe Version 4. Available
CGIAR-CSI SRTM 90m Database Httpsrtm Csi Cgiar Org. 2008. Available online: http://www.cgiar-csi.
org/data/srtm-90m-digital-elevation-database-v4-1 (accessed on 15 May 2017).
32. Hijmans, R.J.; van Etten, J. Raster: Geographic data analysis and modeling. R Package Version 2014, 2, 15.
33. Pebesma, E. Spacetime:: Spatio-Temporal Data in R. J. Stat. Softw. 2012, 51, 1–30. [CrossRef]
34. Hiemstra, P.; Hiemstra, M.P. Package “automap”. Compare 2013, 105, 10.
35. Schuurmans, J.M.; Bierkens, M.F.P.; Pebesma, E.J.; Uijlenhoet, R. Automatic Prediction of High-Resolution
Daily Rainfall Fields for Multiple Extents: The Potential of Operational Radar. J. Hydrometeorol. 2007, 8,
1204–1224. [CrossRef]
36. Buytaert, W.; Celleri, R.; Willems, P.; De Bièvre, B.; Wyseure, G. Spatial and temporal rainfall variability in
mountainous areas: A case study from the south Ecuadorian Andes. J. Hydrol. 2006, 329, 413–421. [CrossRef]
37. Diodato, N. The influence of topographic co-variables on the spatial variability of precipitation over small
regions of complex terrain. Int. J. Climatol. 2005, 25, 351–363. [CrossRef]
38. Mourre, L.; Condom, T.; Junquas, C.; Lebel, T.; Sicart, J.E.; Figueroa, R.; Cochachin, A. Spatio-temporal
assessment of WRF, TRMM and in situ precipitation data in a tropical mountain environment (Cordillera
Blanca, Peru). Hydrol. Earth Syst. Sci. 2016, 20, 125–141. [CrossRef]
39. Cressman, G.P. An Operational Objective Analysis System. Mon. Weather Rev. 1959, 87, 367–374. [CrossRef]
40. Mitchell, T.P.; Wallace, J.M. The annual cycle in equatorial convection and sea surface temperature. J. Clim.
1992, 5, 1140–1156. [CrossRef]
41. Schneider, U.; Ziese, M.; Meyer-Christoffer, A.; Finger, P.; Rustemeier, E.; Becker, A. The new portfolio of
global precipitation data products of the Global Precipitation Climatology Centre suitable to assess and
quantify the global water cycle and resources. Proc. Int. Assoc. Hydrol. Sci. 2016, 374, 29–34. [CrossRef]
42. Harris, I.; Jones, P.D.; Osborn, T.J.; Lister, D.H. Updated high-resolution grids of monthly climatic
observations–the CRU TS3. 10 Dataset. Int. J. Climatol. 2014, 34, 623–642. [CrossRef]
43. Dee, D.P.; Uppala, S.M.; Simmons, A.J.; Berrisford, P.; Poli, P.; Kobayashi, S.; Andrae, U.; Balmaseda, M.A.;
Balsamo, G.; Bauer, P.; et al. The ERA-Interim reanalysis: Configuration and performance of the data
assimilation system. Q. J. R. Meteorol. Soc. 2011, 137, 553–597. [CrossRef]
44. Huffman, G.J.; Bolvin, D.T. TRMM and Other Data Precipitation Data Set Documentation; TRMM 3B423B43
Doc; NASA: Greenbelt, MD, USA, 2015; pp. 1–44. [CrossRef]
45. Jolliffe, I.T.; Cadima, J. Principal component analysis: A review and recent developments. Phil. Trans. R. Soc.
A 2016, 374. [CrossRef] [PubMed]
46. Preisendorfer, R.W.; Mobley, C.D. Principal Component Analysis in Meteorology and Oceanography; Elsevier:
Amsterdam, The Netherlands, 1988; Volume 425.
47. Southern Oscillation Index (SOI) | Teleconnections | National Centers for Environmental Information
(NCEI). Available online: https://www.ncdc.noaa.gov/teleconnections/enso/indicators/soi/ (accessed on
27 July 2017).
48. Null, J. El Nino and La Nina Years and Intensities Based on Oceanic Nino index (ONI); NOAA: Silver Spring, MD,
USA, 2015.
49. Huang, B.; Banzon, V.F.; Freeman, E.; Lawrimore, J.; Liu, W.; Peterson, T.C.; Smith, T.M.; Thorne, P.W.;
Woodruff, S.D.; Zhang, H.-M. Extended reconstructed sea surface temperature version 4 (ERSST. v4). Part I:
Upgrades and intercomparisons. J. Clim. 2015, 28, 911–930. [CrossRef]
50. Schneider, U.; Becker, A.; Finger, P.; Meyer-Christoffer, A.; Ziese, M.; Rudolf, B. GPCC’s new land surface
precipitation climatology based on quality-controlled in situ data and its role in quantifying the global water
cycle. Theor. Appl. Climatol. 2014, 115, 15–40. [CrossRef]
51. Maidment, R.I.; Grimes, D.I.F.; Allan, R.P.; Greatrex, H.; Rojas, O.; Leo, O. Evaluation of satellite-based and
model re-analysis rainfall estimates for Uganda. Meteorol. Appl. 2013, 20, 308–317. [CrossRef]
52. Norina, A.; Vegas, F.; Lavado, C. Zappa water resources and climate change impact modelling on a daily
time scale in the Peruvian Andes. Hydrol. Sci. J. 2014, 59, 2043–2059. [CrossRef]
53. Ballari, D.; Castro, E.; Campozano, L. Validation of Satellite Precipitation (TRMM 3B43) in Ecuadorian
Coastal Plains, Andean Highlands and Amazonian Rainforest. ISPRS-Int. Arch. Photogramm. Remote Sens.
Spat. Inf. Sci. 2016, 305–311.
54. Satgé, F.; Bonnet, M.-P.; Gosset, M.; Molina, J.; Lima, W.H.Y.; Zolá, R.P.; Timouk, F.; Garnier, J. Assessment of
satellite rainfall products over the Andean plateau. Atmos. Res. 2016, 167, 1–14. [CrossRef]
Water 2018, 10, 213 23 of 23

55. Manz, B.; Buytaert, W.; Zulkafli, Z.; Lavado, W.; Willems, B.; Robles, L.A.; Rodríguez-Sánchez, J.-P.
High-resolution satellite-gauge merged precipitation climatologies of the Tropical Andes. J. Geophys. Res.
Atmos. 2016. [CrossRef]
56. Dinku, T.; Connor, S.J.; Ceccato, P. Comparison of CMORPH and TRMM-3B42 over mountainous regions of
Africa and South America. In Satellite Rainfall Applications for Surface Hydrology; Springer: Berlin, Germany,
2010; pp. 193–204.
57. Espinoza, J.C.; Chavez, S.; Ronchail, J.; Junquas, C.; Takahashi, K.; Lavado, W. Rainfall hotspots over the
southern tropical Andes: Spatial distribution, rainfall intensity, and relations with large-scale atmospheric
circulation. Water Resour. Res. 2015, 51, 3459–3475. [CrossRef]
58. Levizzani, V.; Amorati, R.; Meneguzzo, F. A Review of Satellite-Based Rainfall Estimation Methods. In
MUSIC–MUltiple-Sensor Precipitation Measurements, Integration, Calibration and Flood Forecasting; European
Commission: Brussels, Belgium, 2002; Volume 70.
59. Mätzler, C.; Standley, A. Technical note: Relief effects for passive microwave remote sensing. Int. J. Remote
Sens. 2000, 21, 2403–2412. [CrossRef]
60. North, G.R.; Bell, T.L.; Cahalan, R.F.; Moeng, F.J. Sampling errors in the estimation of empirical orthogonal
functions. Mon. Weather Rev. 1982, 110, 699–706. [CrossRef]
61. Bendix, J.; Rollenbeck, R.; Göttlicher, D.; Cermak, J. Cloud occurrence and cloud properties in Ecuador. Clim.
Res. 2006, 30, 133–147. [CrossRef]
62. Diaz, H.F.; Markgraf, V. El Niño and the Southern Oscillation: Multiscale Variability and Global and Regional
Impacts; Cambridge University Press: Cambridge, UK, 2000.
63. Campozano, L.; Célleri, R.; Trachte, K.; Bendix, J.; Samaniego, E. Rainfall and Cloud Dynamics in the Andes:
A Southern Ecuador Case Study. Adv. Meteorol. 2016, 2016. [CrossRef]
64. Philander, S.G.H.; El Niño, L.N. The Southern Oscillation; Academic Press: San Diego, CA, USA, 1990;
pp. 904–905.
65. Schneider, T.; Bischoff, T.; Haug, G.H. Migrations and dynamics of the intertropical convergence zone. Nature
2014, 513, 45–53. [CrossRef] [PubMed]
66. Arteaga, K.; Tutasi, P.; Jiménez, R. Climatic variability related to El Niño in Ecuador? A historical background.
Adv. Geosci. 2006, 6, 237–241. [CrossRef]
67. Dai, A.; Wigley, T.M.L. Global patterns of ENSO-induced precipitation. Geophys. Res. Lett. 2000, 27,
1283–1286. [CrossRef]
68. Iguchi, T.; Kozu, T.; Meneghini, R.; Awaka, J.; Okamoto, K. Rain-profiling algorithm for the TRMM
precipitation radar. J. Appl. Meteorol. 2000, 39, 2038–2052. [CrossRef]
69. Rasmussen, K.L.; Choi, S.L.; Zuluaga, M.D.; Houze, R.A. TRMM precipitation bias in extreme storms in
South America. Geophys. Res. Lett. 2013, 40, 3457–3461. [CrossRef]
70. Nastos, P.T.; Kapsomenakis, J.; Philandras, K.M. Evaluation of the TRMM 3B43 gridded precipitation
estimates over Greece. Atmos. Res. 2016, 169, 497–514. [CrossRef]
71. Sorooshian, S.; Hsu, K.-L.; Gao, X.; Gupta, H.V.; Imam, B.; Braithwaite, D. Evaluation of PERSIANN system
satellite–based estimates of tropical rainfall. Bull. Am. Meteorol. Soc. 2000, 81, 2035–2046. [CrossRef]
72. Islam, M.N.; Uyeda, H. Use of TRMM in determining the climatic characteristics of rainfall over Bangladesh.
Remote Sens. Environ. 2007, 108, 264–276. [CrossRef]
73. Stano, G.; Krishnamurti, T.N.; Kumar, T.V.; Chakraborty, A. Hydrometeor structure of a composite monsoon
depression using the TRMM radar. Tellus A 2002, 54, 370–381. [CrossRef]
74. Lau, K.M.; Wu, H.T. Warm rain processes over tropical oceans and climate implications. Geophys. Res. Lett.
2003, 30. [CrossRef]
75. Dinku, T.; Ceccato, P.; Connor, S.J. Challenges of satellite rainfall estimation over mountainous and arid
parts of east Africa. Int. J. Remote Sens. 2011, 32, 5965–5979. [CrossRef]

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).