Download as pdf or txt
Download as pdf or txt
You are on page 1of 19

Atmospheric Research 233 (2020) 104708

Contents lists available at ScienceDirect

Atmospheric Research
journal homepage: www.elsevier.com/locate/atmosres

Influence of PBL parameterization schemes in WRF_ARW model on short - T


range precipitation's forecasts in the complex orography of Peruvian Central
Andes

Aldo S. Moya-Álvareza, , René Estevana, Shailendra Kumara, Jose L. Flores Rojasa, Joel J. Ticsea,
Daniel Martínez-Castroa,b, Yamina Silvaa
a
Instituto Geofísico del Perú, Lima, Peru
b
Instituto de Meteorología de Cuba, La Habana, Cuba

A R T I C LE I N FO A B S T R A C T

Keywords: The study evaluated the sensitivity of the precipitation forecast in the central Andes of Peru of Weather Research
Central Andes and Forecasting (WRF) model to change the planetary boundary layer (PBL) schemes. In that region is located
WRF model the Mantaro basin, which is one of the most important in the region. Here, the rainfall is very important to the
PBL schemes agriculture and to the reserves of drinking water. The simulations were carried out with ten PBL schemes for
Mantaro basin
19 days in January, February, and March, between 2009 and 2012. Based on the statistical analysis (model vs.
Verification
observation), the more efficient schemes were determined and analyses of the vertical profiles of some variables
are shown. As a result, the schemes that most helped the model in rainfall forecasting were MYNN3 (general and
north sector of the basin), Bou-Lac (central sector) and Bretherton-Park (southern sector). The model generally
overestimated rainfall in the northern basin, underestimated in the center, and in the south some schemes
overestimated and others underestimated. In addition, it was concluded that the boundary layer is more stable in
the model than in the observations. The schemes that generated the most rainfall were those that generated a
more unstable boundary layer with weaker wind speeds, at least with easterly winds. Another conclusion is that
the height of the boundary layer for rainy days in the region at 18 UTC oscillates around 1000 m and that,
generally, the wind's velocity changes very little or decreases within the boundary layer and increases above it.

1. Introduction The Mantaro river basin, where Huancayo city is located, is one of the
most important in the entire region. Here, the rainfall in warm period is
In the present work evaluates the influence of Planetary Boundary very important to the agriculture because they guarantee the water
Layer schemes (PBL) in Weather Research and Forecasting model reserves necessary for this activity (Martínez et al., 2006), so studying
(WRF) (Skamarock et al., 2008) on short-range precipitation forecast in the ability of meteorological models to forecast precipitation is very
the Peruvian Central Andes, which has very complex topography. The important.
central Andes of Peru constitute a region with complex orography, in The first version of WRF developed began in the second half of the
which elevations can exceed 5000 m in some areas (Fig. 1, D2). For this 1990s as a collaboration between the National Center for Atmospheric
reason, the work of atmospheric models becomes very complex, mainly Research (NCAR), the National Oceanic and Atmospheric
due to the low horizontal representation of grid points and the defining Administration (NOAA), the Air Force Time Agency (AFWA), the Naval
role of mesoscale processes on the formation of climate in the region. Research Laboratory, the University of Oklahoma and the Federal
Satellite-based climatology of the intense rainfall events showed that Aviation Administration (FAA), all of the United States of America. The
topography of the Andes mountain affects the geographical locations of WRF model is widely used for this proposes in the world. In the present
rainfall events over the South American continent (Kumar et al., work the experiments were carried out with the model version 3.7 of
2019a), and their interaction with different directional surface flow the model, developed in 2015. In this cases, source codes were taken
could affect the rainfall intensity and hydrometeors size and their from the WRF user page (http://www2.mmm.ucar.edu/wrf/users/
concentration in the central part of the Andes (Kumar et al., 2019b). download/get_sources.html). Several studies have used WRF to


Corresponding author.
E-mail address: amoya@igp.gob.pe (A.S. Moya-Álvarez).

https://doi.org/10.1016/j.atmosres.2019.104708
Received 1 July 2019; Received in revised form 9 September 2019; Accepted 14 October 2019
Available online 23 October 2019
0169-8095/ © 2019 Elsevier B.V. All rights reserved.
A.S. Moya-Álvarez, et al. Atmospheric Research 233 (2020) 104708

Fig. 1. Model domains used in the simulations. Terrain elevation is indicated with color shades in meters above sea level. In both domains, thick contour indicates the
limits of the Mantaro Basin.

conduct studies in different mountainous regions, including parts of the several physical schemes and model ability to predict extreme rainfall
Andes (Mercader et al., 2010; Wardah et al., 2011; Joon-Bum and in the central Andes of Peru, respectively. Recently, Moya-Álvarez et al.
Sangil, 2017; Junquas et al., 2017; Zeyaeyan et al., 2017; Chawla et al., (2019), evaluated the sensitivity of the model to the resolutions of the
2018; Moya-Álvarez et al., 2018b, 2019). domains in the forecast of rainfall in the region. However, for this re-
The boundary layer plays an important role in precipitation-forming gion, there are no studies aimed at evaluating the sensitivity of the WRF
processes, as these depend, among other things, on the moisture content model to the different PBL schemes. Due to this, and to the importance
and air flows at the lower levels of the troposphere. For example, for the of precipitations from the social and economic point of view, the pre-
occurrence of rainfall or storms, there must be high humidity, a suffi- sent work acquires great importance for the prediction of rainfall in the
ciently negative lift index, which implies high instability (Rasmussen region.
and Blanchard, 1998; Schneider and Dean, 2008). These conditions, In the work, research is carried out to evaluate how the WRF model
depend largely on the processes in the boundary layer, so for a good simulates rainfall in the central Peruvian Andes through the statistics of
prediction of the phenomenon, the model should achieve a good de- ten PBL schemes and 19 case studies. For validation, observation data
scription of the processes in that layer. Thus, precipitation is very from 19 stations in the region are used. Additionally, two case studies
sensitive to PBL schemes (Shin and Hong, 2011). In this sense, Chawla were used corresponding to two rainy days for which atmospheric
et al. (2018) showed that the selection of MP scheme influences the soundings are available on the station “Huayao”.
spatial pattern of rainfall, while the choice of PBL and cumulus para-
meterizations, influence the magnitude of rainfall in the model simu-
lations. Therefore, the selection of an appropriate scheme is very im- 2. Data and methodology
portant in meteorological modeling.
Previous studies have characterized the sensitivity of simulated 2.1. Model configuration
boundary layer parameters with different parametrization schemes in
the WRF model. Shahzad et al., 2016 studied the impact of PBL schemes The simulations were produced for two domains (Fig. 1) with one-
on high-resolution rainfall simulation over the selected region of Hi- way nesting, whose characteristics are specified in Table 1. The model
malaya, Karakorum and Hindukush. Recent studies (Xu and Zhao, was initialized at 00 UTC (Universal Time Coordinated) with 36 h of
2000; Lynn et al., 2001; Wisse and de Arellano, 2004; Jankov et al., forecast horizon for all cases (00 UTC is approximately between 14 and
2005; Lin-Jing and Wey-Lun, 2016) have shown that the evolution of 16 h before the expected diurnal convection). The initial and boundary
precipitating systems and specially, convective systems, is very sensi- conditions were taken from the Global Operational Final Analyses
tive to the PBL parameterization and, in particular, can take advantages (FNL) of the National Center of Environmental Prediction (NCEP)
of the application of higher-order turbulence closure schemes. Lin-Jing (available online at https://rda.ucar.edu/datasets/ds083.2/), which
and Wey-Lun (2016) also showed that the choice of PBL scheme has a has a temporal resolution of 6 h and a horizontal resolution of 1° x 1°.
significant effect on precipitation and Gbode et al. (2018) investigated
the sensitivity of different physics schemes in the WRF model during a Table 1
West African monsoon regime. Main characteristics of the domains and initial and boundary conditions.
In the study zone, there is a clear climatic seasonality, with a dry Characteristics Domain 1 (D1) Domain 2 (D2)
period, extended from May to September and a rainy period, extended
Central point Lat: 10° S Lat: 12.25819° S
from October to April, in which most of the rainfall occurs in the
Lon: 75° W Lon: 74.8356° W
country, including the Centrals Andes (Silva et al., 2008; Aceituno, Horizontal step 18 km 6 km
1989). In a recent paper (Junquas et al., 2017), the influence of oro- Dimensions (XYZ) 115 × 140 × 50 115 × 142 × 50
graphy on the diurnal cycle of rainfall in the Central Andes has been Vertical levels 50 50
investigated using WRF, and Moya-Álvarez et al. (2018b, 2019) eval- Time step 90 s 36 s
Initial and border conditions FNL (1°) Simulation of domain 1
uated the sensitivity of the WRF rainfall forecast to the change of

2
A.S. Moya-Álvarez, et al. Atmospheric Research 233 (2020) 104708

Table 2
PBL and SFC layers schemes used in simulations.
Set PBL scheme Short name Closure type SFC layer PBLH defitition

1 Asymmetrical Convective Model version 2 ACM2 1.0 non-local MM5 similarity Rib calculated above neutral buoyancy level
2 Mellor–Yamada–Janjic MYJ 1.5 local ETA similarity TKE-prescribed threshold
3 Yonsei University YSU 1.0 non-local MM5 similarity Rib calculated from sfc
4 Bougeault–Lacarrère scheme BouLac 1.5 local MM5 similarity TKE-prescribed threshold
5 Mellor–Yamada–Nakanishi–Niino level 2.5 MYNN25 1.5 local MYNN TKE-prescribed threshold
6 Mellor–Yamada–Nakanishi–Niino level 3.0 MYNN3 2.0 local MYNN TKE-prescribed threshold
7 Quasi-Normal Scale Elimination QNSE 1.5 local QNSE TKE-prescribed threshold
8 Grenier– Bretherton–McCaa GBM 1.5 local MM5 similarity TKE-prescribed threshold
9 University of Washington BP 1.5 local MM5 similarity Rib threshold
10 Shin-Hong PBL scheme SH Scale-aware scheme MM5 similarity TKE-prescribed threshold

Fig. 2. Precipitation, registered in each day of the cases selected for the study.

Table 3
Stations used to model verification.
N° Name Lon Lat Altitude (m) Rainy days

1 Cerro de Pasco North −76.300 −10.693 4260 17


2 Marcacomapocha North −76.325 −11.405 4413 18
3 Yantac North −76.400 −11.334 4600 17
4 Ricran North −75.525 −11.542 3500 19
5 Huayao Center −75.339 −12.034 3308 19
6 Ingenio Center −75.288 −11.881 3450 16
7 Laive Center −75.400 −12.252 3990 17
8 Comas Center −75.100 −11.748 3300 19
9 Jauja Center −75.479 −11.783 3322 19
10 Santa Ana Center −75.221 −12.004 3295 19
11 Jarpa Center −75.400 −12.125 3726 19
12 Viques Center −75.234 −12.163 3186 18
13 Lircay South −74.729 −12.983 3150 19
14 Acobamba South −74.559 −12.864 3236 17
15 La Quinua South −74.135 −13.034 3260 18
16 Pichalca South −75.085 −12.406 3570 17
17 Pampas South −74.866 −12.393 3260 19
18 Huancalpi South −75.237 −12.542 3800 19
19 Huancavelica South −75.000 −12.780 3676 19

The model's topography was constructed with the topographic data (Iacono et al., 2008), was used as a radiation model, and for convection,
set of the digital elevation model of the Shuttle Radar Topography Grell – Freitas scheme (Grell and Freitas, 2014). The soil model was the
Mission (available online at SRTM; https://dds.cr.usgs.gov/srtm/ Unified Noah Land Surface Model (Tewari et al., 2004). For the PBL, ten
version2_1/), which has a resolution of 90 m (Rodríguez et al., 2006; schemes were tested Table 2:
Farr et al., 2007). The parameterization schemes were selected based on
the sensitivity study upon previous work for this region by Moya-
2.1.1. Asymmetrical convective model version 2 (ACM2)
Álvarez et al. (2018a). The Morrison scheme was used for microphysics
(Morrison et al., 2009); The RRTMG (Rapid Radiative Transfer Model This is a first order, (hybrid closure) scheme and features non-local
upward mixing and local downward mixing (Pleim, 2007). Pleim

3
A.S. Moya-Álvarez, et al. Atmospheric Research 233 (2020) 104708

can better represent the PBL in regimes of higher static stability com-
pared to nonlocal schemes in similar regimes (Shin and Hong, 2011).

2.1.5. Mellor–Yamada–Nakanishi–Niino level 2.5 scheme (MYNN25) is a


1.5 closure order (Nakanishi and Niino, 2006)
Local closure scheme and predicts sub-grid TKE terms. Expressions
of stability and mixing length are based on the results of large eddy
simulations and the expressions of mixing length are more applicable to
a variety of static stability regimes.

2.1.6. Mellor–Yamada–Nakanishi–Niino level 3.0 scheme (MYNN3)


Similar to MYNN25, but with second closure order. Reasonably
depicts statically stable boundary layer simulations supporting radia-
tion fog development (Nakanishi and Niino, 2006).

2.1.7. Quasi-normal scale elimination (QNSE) scheme


This is a 1.5 closure order. Local closure scheme with TKE predic-
tion (Sukoriansky et al., 2005). Provides realistic depiction of potential
temperature profiles, PBL height, and kinematic profiles based on ob-
servational data and corresponding large eddy simulations (Kosovic and
Curry 2000).

2.1.8. Grenier–Bretherton–McCaa (GBM)


A 1.5-order local closure to depict a PBL influenced by stratocu-
Fig. 3. Geographical distribution of the stations used in the investigation.
Within the brown circle, the stations corresponding to the northern sector of the
mulus clouds; vertical fluxes of TKE are enhanced for better comparison
basin, within the red circle, those corresponding to the central zone, and within to large eddy simulations. Reductions to the stratocumulus cloud deck
the blue circle, those located in the southern sector. (For interpretation of the owing to vertical mixing are found to be well handled for vertical grid
references to color in this figure legend, the reader is referred to the web ver- spacing at or smaller than 15 mb; this could be relevant for depicting
sion of this article.) the impact of stratocumulus clouds on buoyancy preceding potentially
severe convection (Grenier and Bretherton, 2001).
(2007) validates the use of the ACM2 scheme owing to its support of
PBL heights similar to those based on afternoon wind profiler data from 2.1.9. The university of washington scheme (BP) is a 1.5 closure order
radar. (Bretherton and Park, 2009)
Local TKE closure scheme from the Community Earth System Model
(CESM). This is an attempts to improve upon the GBM; changes from
2.1.2. Mellor–Yamada–Janjic scheme (MYJ)
the GBM include accounting for relatively longer time steps and diag-
This is a 1.5 closure order prognostic TKE (turbulent kinetic energy)
nosis of TKE instead of forecasting.
scheme with local vertical mixing (Janjić, 2002). Improves upon Mel-
lor–Yamada 1.5-order local scheme (Mellor and Yamada 1974, 1982).
2.1.10. Non-local Shin-Hong PBL scheme (SH) is a scale-aware scheme
(Shin and Hong, 2015)
2.1.3. Yonsei University scheme (YSU) Include scale dependency for vertical transport in convective PBL.
YSU is a widely first-order and non-local PBL scheme with an ex-
plicit entrainment layer (Hong et al., 2006). Accurately simulates the 2.2. Study period and model verification
vertical mixing in buoyancy-driven PBLs with shallower mixing in
strong-wind regimes. The statistical experiments were conducted for 19 dates between
2009 and 2012 years (Fig. 2), selected by taking into account that
2.1.4. Bougeault–Lacarrère scheme (BouLac) sufficient rainfall has to occur in order to allow the performance of the
This is a 1.5 closure order multi-layer. Local closure scheme and has model to be evaluated with the PBL schemes considered. Note in
a TKE prediction option (Bougeault and Lacarrere, 1989). The scheme Table 3 that the selected meteorological stations recorded rainfall in

Fig. 4. Average geopotential fields in 700 hPa, climatological (a), 19 cases studied (b) and anomalies of (b) respect to (a).

4
A.S. Moya-Álvarez, et al. Atmospheric Research 233 (2020) 104708

Fig. 5. Average geopotential fields in 200 hPa, climatological (a), 19 cases studied (b).

Fig. 6. Average relative humidity fields in the layer 700–500 hPa, a, climatological, b, 19 cases studied, and c, anomalies of “b” respect to “a”.

most cases, and even some recorded rainfall in all cases. The cases
studied were selected in the months of January, February and March,
according to the marked climatic seasonality in Mantaro basin, since
the three months are in the rainy period of the year, where 92% of the
rainfall recorded in the study region occurs (Silva et al., 2008). For this
reason, no cases were selected in the cold period of the year.
The verification of the results is carried out by using a point ver-
ification technique for the forecast of accumulated rainfall in 24 h. The
last 24 h of the 36 simulation hours were taken for verification (7 to 7 h,
local time, taking the first 12 h as spin up). With that objective, model
output was interpolated using the Cressman method (Cressman, 1959)
to the station's points coordinates and compared with the “in situ”
measured precipitation data, taken from the official station's network of
the Meteorological Service (SENAMHI). The coordinates of the stations
considered for the study are shown in Table 3, and the geographical
distribution of these can be seen in Fig. 3 (the Mantaro basin is marked
in dark line). The atmospheric sounding data used for the two case
studies included in the investigation were taken from two launches
made at the station “Huayao”, February 25 and March 7, 2019.
The markedly seasonal regime of precipitations in the Mantaro
basin is associated with markedly favorable synoptic conditions in this
Fig. 7. Spatial distribution of rainfall, average of the 19 cases studied.
season in relation to the annual climatology. In Fig. 4, which shows the
behavior of the geopotential field in 700 hPa, both climatological (a)
and for the 19 cases studied (b), a weakening over Peru of the influence
of the South Atlantic anticyclone dorsal can be observed in relation to
the annual climatology. This can be verified in the field of geopotential

5
A.S. Moya-Álvarez, et al. Atmospheric Research 233 (2020) 104708

Table 4
Model statistics (mm and %) for the ten PBL schemes, average of all cases and stations.
Statistic ACM2 BL BP GBM MYJ M25 M30 QNSE SH YSU

(mm)
BIAS −1.62 0.72 −2.15 0.59 −0.13 −1.72 −1.86 −1.42 −0.85 −0.92
MAE 9.67 9.78 9.56 11.25 9.64 9.51 9.25 9.68 9.39 9.37
RMSE 12.19 12.34 12.17 15.08 12.23 11.88 11.53 12.12 11.90 11.85
(%)
BIAS −12.7 5.7 −16.8 4.6 −1.0 −13.5 −14.5 −11.1 −6.6 −7.2
MAE 78.3 79.2 78.2 92.6 77.9 75.9 74.4 78.1 75.9 75.3
RMSE 99.9 100.5 100.2 125.8 99.5 95.0 92.6 98.5 95.6 95.6

Fig. 8. Box and whiskers diagrams of the rainfall produced by the model with each PBL scheme and the precipitation observed.

Table 5
Statistical significance of the differences in precipitation forecast among all PBL schemes evaluated.
Schemes ACM2 BL BP GBM MYJ M25 M30 QNSE SH YSU

ACM2 X 0.000 0.238 0.002 0.008 0.830 0.622 0.693 0.113 0.160
BouLac X X 0.000 0.825 0.000 0.000 0.000 0.000 0.000 0.000
BP X X X 0.000 0.000 0.249 0.435 0.052 0.000 0.001
GBM X X X X 0.224 0.000 0.000 0.000 0.010 0.007
MYJ X X X X X 0.000 0.000 0.000 0.021 0.014
MYNN25 X X X X X X 0.544 0.308 0.002 0.003
MYNN3 X X X X X X X 0.151 0.001 0.001
QNSE X X X X X X X X 0.058 0.089
SH X X X X X X X X X 0.648
YSU X X X X X X X X X X

anomalies (c), which are negative over the whole country, with a significantly over Peru and shows positive anomalies in relation to the
maximum fall in the Peruvian border with Brazil and Bolivia. This si- climatology. This can be seen in (c), which shows anomalies > 17% in
tuation facilitates the circulation of local winds over the region, such as central and southern Peru.
the circulation of valley and mountains, as well as convective devel- Junquas et al. (2017), studied the influence of the interaction of
opment. synoptic mechanisms with orography in southeastern Peru on the
On the other hand, in the 200 hPa it is clearly observed that the diurnal cycle of rainfall, and demonstrated that this interaction is the
average of the 19 cases studied reflects a clear anticyclonic circulation “key step to understand how the atmospheric circulation influences the
centered in Bolivia (Fig. 5b), known in the region as the “high of Bo- patterns of precipitation variability on longer time-scales”. Among
livia” (Lenters and Cook, 1997), which favors rainfall processes over the other things, they found “diurnal thermally driven circulation at local
region due to the strong divergence generated in the upper troposphere scale, including upslope (downslope) wind and moisture transport
over much of Peru. Climatologically (Fig. 5a) this circulation has its during the day (night)”.
center displaced to the northeast of Brazil. Now, Fig. 7 shows that the highest rainfall in the 19 cases studied
Finally, Fig. 6 shows the fields of average relative humidity in layer occurs towards the center and south of the basin, which is consistent
700–500 hPa, climatology (a) and for the 19 cases studied (b), where it with the highest positive relative humidity anomalies observed in
can be seen that during these rain events the relative humidity increases Fig. 6c. This means that, even though the northern sector of the basin

6
A.S. Moya-Álvarez, et al. Atmospheric Research 233 (2020) 104708

Fig. 9. Height (above ground level) of the PBL of the ten schemes, average of all cases and stations at 18 UTC.

Fig. 10. Model vertical profiles of equivalent potential temperature, predicted with the ten PBL schemes, average of all cases and stations at 12 and 18 UTC.

Table 6
Vertical gradient of θe(K) of layer 2–1000 m for the then PBL schemes, average of all cases and stations at 12 and 18 UTC.
Period ACM2 BL BP GBM MYJ M25 M30 QNSE SH YSU

12 UTC 0.52 0.52 0.54 0.50 0.57 0.55 0.56 0.57 0.54 0.54
18 UTC −0.03 −0.48 −0.19 −0.34 −0.46 −0.33 −0.28 −0.18 −0.24 −0.25

7
A.S. Moya-Álvarez, et al. Atmospheric Research 233 (2020) 104708

Table 7
Vertical gradient of θe(K) of layer 10–200 m for the ten PBL schemes, average of all cases stations at 18 UTC.
Period ACM2 BL BP GBM MYJ M25 M30 QNSE SH YSU

18 UTC −0.09 −0.52 −0.11 −0.33 −0.25 −0.21 −0.14 −0.13 −0.17 −0.18

Fig. 11. Vertical profiles of wind speed a) and vertical velocity b), predicted with each PBL scheme used in the research, average of all cases and stations at 18 UTC.

Table 8 3. Results
Bias, wind speed, wind direction (average of layer 10–2000) and gradient of θe
(average of layer 10–1000 m) for each PBL scheme, average of all cases and 3.1. Ability of WRF model to precipitation's forecast with different PBL
stations. Schemes in Peruvian Andes
Schemes BIAS (mm) Gradient θe(K) Wind speed Wind direction
(m/s) (degrees) Table 4 Bias, RMSE and MAE for the 10 PBL schemes used in the
verification. To obtain the statistics shown on the table, all the cases
GBM 0.59 −0.48 0.74 135
and all the stations used in the work were considered. As can be seen,
BL 0.72 −0.46 0.75 118
MYJ −0.13 −0.34 0.83 108 almost all the schemes underestimate the rainfall over the region (with
M25 −1.72 −0.33 0.83 97 an average of approximately −7%, according to the bias shown in
YSU −0.92 −0.28 0.92 110 Table 4). In this sense, the lowest bias is shown by the MYJ scheme
SH −0.85 −0.25 0.91 109
(−0.13 mm, −1.0%). Here, BouLac and GBM overestimated rainfall,
M30 −1.86 −0.24 0.78 94
QNSE −1.42 −0.19 1.06 106 the latter to a lesser extent. However, the smallest errors were obtained
BP −2.15 −0.18 1.08 85 with the MYNN3 scheme, which showed results slightly better than
ACM2 −1.62 −0.03 1.22 74 MYNN25, SH and YSU, but notably better than the rest of the schemes.
MYNN3 is similar to MYNN25, but with second closure order. In this
case, it is striking that the rest of the schemes are of first or 1.5 order of
has the highest elevations, the circulation of humid air from the closure. From now on, in some tables and figures instead of BouLac,
Amazon to the southern portion of the basin plays a fundamental role in MYNN25 and MYNN3; BL, MYNN25 and MYNN3 will be used, in order
the formation of rainfall. It can be concluded then that the orographic to optimize the space in the tables columns or in the figures description.
effect plays an important role, but there will be more rainfall in the A broader view of the results can be seen in the Fig. 8, showing the
areas of moister air, although the elevations are lower. box and whiskers diagrams of the rainfall produced by the model with
each PBL scheme and the precipitation observed. In the diagrams, “x”
represents the average value and the horizontal line of each box is the
median. The vertical limits of the box represent the quartiles, the upper
whisker, the maximum value and the lower whisker, the minimum
value. The black horizontal line is the average observed precipitation.

8
A.S. Moya-Álvarez, et al. Atmospheric Research 233 (2020) 104708

Fig. 12. Vertical profiles of the “u” and “v” wind components, average of all cases and stations, predicted with each PBL scheme used in the investigation.

In this graphic, the outliers were eliminated. For this reason, we consider that the MYNN3, YSU and SH schemes
In the figure, it can be seen that the MYJ scheme presents the closest are the ones with the best results. A negative feature of the MYNN3
average to that observed, in coincidence with the smallest bias shown in scheme is its problem in generating extreme rainfall. Thus, it improves
Table 4. Also the BL and GBM schemes show average values relatively the quality of the rainfall forecast, but extreme precipitation events can
close to that observed. However, in the cases of MYJ and GBM the be skipped.
variability of the model is less than that of the observed precipitation, Taking into account that the results of the rainfall forecast quality
which shows the inability of the model to generate extreme rainfall indicators with the different PBL schemes are relatively close to the
values with these schemes. In the case of BL, the variability is very high, naked eye, the t-test was applied to determine the statistical sig-
which implies that the model produces too large rainfall values, which nificance of the differences for a 95% confidence interval, whose results
do not actually occur. Note that the maximum value recorded with this are shown in Table 5. The table shows that in most cases the difference
scheme is much higher than the observation and that produced by the between the results of the schemes are statistically significant. Among
other schemes. The average values produced by the rest of the schemes the schemes that showed the smallest difference between them can be
underestimate the observed precipitation, as shown in the bias values of mentioned ACM2 and QNSE, ACM2 and MYNN3, YSU and SH and
Table 4. In this case, the SH and YSU schemes are the ones whose MYNN25 and MYNN3.
variability is closer to that of the observed precipitation. In an attempt to understand why some models produced more
In the case of the MYNN3 scheme, it can be observed that it un- precipitation than others, we constructed the vertical profiles of some
derestimates the rain to a greater extent than other schemes; however, meteorological variables, important to understand the structure of the
Table 4 shows that this configuration produced the lowest forecast er- planetary boundary layer in the area in the days used in the study. First,
rors (MAE and RMSE). The above can be interpreted as that MYNN3 is a the height of the boundary layer of the model was obtained, which is
good scheme from the point of view of the precipitation forecast, even shown in Fig. 9. As can be seen, most of the schemes showed similar
though it presents relatively high negative bias. It is important here to heights, with the exception of the BP scheme, which showed a height
understand that a model can have bias near zero and large errors in the well below the rest, and the QNSE scheme, with a height of almost
forecast. From the point of view of the precipitation forecast, the most 2 km. The rest of the schemes obtained heights around 1 km. The
important indicators in this case would be the MAE and the RMSE. The average of all the schemes was 1025 m. As it has been said, there is no
fact of having a well-defined bias also adds the possibility of applying a survey data of the boundary layer for the dates studied, and conse-
bias correction procedure with a greater chance of success than with quently, there is no data of observed height of said layer, however, the
schemes that have bias close to zero. average height obtained in this investigation is consistent with that

9
A.S. Moya-Álvarez, et al. Atmospheric Research 233 (2020) 104708

Fig. 13. Behavior of the bias, vertical gradient of θe (average of layer 10–1000 m) and wind speed, wind direction (average of layer 10–2000), predicted with each
scheme (average of all cases and stations), ordered on the x axis from the most unstable to the least unstable PBL.

10
A.S. Moya-Álvarez, et al. Atmospheric Research 233 (2020) 104708

Fig. 14. wind circulation close to 500 and 2000 m generated by the QNSE scheme at 18 UTC. In color, wind speed.

obtained, the profiles of the equivalent potential temperature θe of each


boundary layer scheme were constructed for the first 2 km of height
above the surface. Fig. 10 shows the profiles for 12 and 18 UTC,
averaged between all cases and stations considered. As can be seen, the
results for the 12 UTC reflect a stable boundary layer, as expected,
without large differences between the schemes. Therefore, from now
on, most of the analysis will be done only for the 18 UTC profiles. The
vertical profiles at 18 UTC in general show an unstable boundary layer,
some schemes more than others. In this sense, the scheme that showed
the most instability was the BL scheme, which at the same time was one
of the two schemes with which the most precipitation model produced.
Note that only with the GBM scheme the model produced more pre-
cipitation than BL.
From these profiles, has been calculated the vertical gradient of
equivalent potential temperature, average of the first 1000 m of height.
Precisely the three schemes that built a more unstable boundary layer,
were the schemes with which the model produced more rainfall. They
even overdid it (Table 6).
It is interesting that for 0–1000 m, the MYJ scheme generated a
boundary layer more unstable than GBM, however, in Table 7 it is
observed that in the first 200 m, the PBL was more unstable when the
GBM scheme was used. This speaks about the importance of the tem-
perature and humidity vertical distribution in the lower boundary
layer.
Fig. 11 shows the vertical profiles of wind speed a) and vertical
velocity b) for each PBL scheme used in the research. In this case, the
most interesting thing is that the two schemes that produced more
Fig. 15. Vertical profiles of wind speed, observed on February 25 (blue) and
precipitation (GBM and BL) visually showed the lowest wind speed in
March 7 (red), 2019. (For interpretation of the references to color in this figure the first 1000 m in relation to the rest of the schemes, which corre-
legend, the reader is referred to the web version of this article.) sponds to that they were two of the schemes that generated the most
unstable boundary layer. On the other hand, the BL and MYJ schemes
showed the limit layer with ascending vertical movements superior to
obtained in an observation dated February 25, 2019, which resulted in
the rest of the schemes, mainly above 800 m. The exception in this case
1034 m above ground level.
was the GBM scheme, which showed upward vertical movements of
Taking into account the height values of the boundary layer

11
A.S. Moya-Álvarez, et al. Atmospheric Research 233 (2020) 104708

Fig. 16. Cross section of the relief's height in the vicinity of the Mantaro Valley, centered on latitude −12.034° (a) and vertical profiles of the Froude number for
February 25 and March 7, 2019 (b). On the vertical axis, the height above sea level.

Fig. 17. Behavior of the statistics, calculated for the northern, central and southern sectors of the basin.

Table 9
Model statistics (mm) of the ten PBL schemes, average of all cases and stations of north, center and south.
Statistics ACM2 BL BP GBM MYJ M25 M30 QNSE SH YSU

North BIAS 1.8 4.7 3.0 6.3 4.2 1.3 0.6 2.9 2.5 0.8
MAE 8.5 9.0 9.4 12.0 8.0 7.4 7.0 8.8 8.0 8.0
RMSE 12.0 11.9 12.5 18.0 10.8 9.4 8.8 11.6 10.1 10.0
Center BIAS −4.0 −2.2 −5.0 −2.3 −2.7 −4.6 −4.2 −4.5 −3.7 −0.9
MAE 10.4 9.7 10.1 11.5 9.9 10.0 9.9 9.9 10.0 10.4
RMSE 12.3 11.9 12.3 14.4 12.2 12.3 12.2 11.9 12.2 13.1
South BIAS −0.8 1.9 −1.7 0.7 0.4 −0.1 −0.5 −0.2 0.6 −0.2
MAE 9.5 10.3 9.0 10.6 10.2 10.1 9.8 10.0 9.4 8.3
RMSE 12.1 13.0 11.9 14.2 13.1 12.8 12.3 12.7 12.4 10.9

12
A.S. Moya-Álvarez, et al. Atmospheric Research 233 (2020) 104708

Table 10 least unstable. Here it can be seen more clearly that the bias is more
Bias, wind speed, wind direction (average of layer 10–2000) and gradient of θe positive in that the boundary layer is more unstable, the wind speed is
(average of layer 10–1000 m) for each PBL scheme, average of all cases and weaker and its direction is more southeastern. The correlation coeffi-
stations at the north sector of the basin. cients obtained of the bias with each variable are −0.80 for θe, −0.61
Schemes BIAS Gradient θe(K) Wind speed Wind direction for the wind speed and 0.84 wind direction respectively.
(mm) (m/s) (degrees) In Fig. 11 a) it could be observed that a particular characteristic of
the profiles of wind speed in the region is that it remains with little
BL 4.71 −0.51 0.83 273
MYJ 4.21 −0.48 1.05 253
change in its intensity during the first 350 m, but then it begins to de-
GBM 6.30 −0.35 0.77 252 crease until around the 1000 m, when it starts to increase again. The
M25 1.33 −0.30 0.74 346 above can be related to the orography of the terrain, which does not
M30 0.65 −0.30 0.67 350 allow the increase of wind with height, which generally takes place in
YSU 0.82 −0.29 0.77 330
the boundary layer on flat or low rough terrain. However, there is an-
SH 2.46 −0.28 0.75 326
BP 2.96 −0.26 0.91 343 other element that can give rise to that particularity. Fig. 14 shows the
QNSE 2.89 −0.12 1.08 22 wind circulation at 516 and 2055 m generated by the QNSE scheme at
ACM2 1.85 −0.02 0.92 195 18 UTC. Here it can be seen that at 516 m there is an interaction of the
flow of the east with the west coming from the ocean, the latter as a
result of the circulation of breeze, resulting in a weakening of the
Table 11 easterly winds, however, already at the height of the 2055 m the flow of
Bias, wind speed, wind direction (average of layer 10–2000) and gradient of θe
the west disappears and consequently the flow of the east returns to
(average of layer 10–1000 m) for each PBL scheme, average of all cases and
increase remarkably to this to that level.
stations at the central sector of the basin.
In Peru, atmospheric soundings are not carried out frequently,
Schemes BIAS (mm) Gradient θe(K) Wind speed Wind direction therefore, there is no data on the vertical profiles of the variables for the
(m/s) (degrees)
selected study period. However, two cases corresponding to two laun-
BL −2.18 −0.63 0.64 71 ches made at the “Huayao” station were analyzed. Two observations are
MYJ −2.65 −0.57 0.61 65 not enough to perform a characterization or to evaluate the behavior of
GBM −2.27 −0.51 0.81 88 a model in a given region, but they can be useful as a reference in this
M25 −4.61 −0.50 0.62 80
regard. In that sense, Fig. 15 shows two vertical profiles of wind cor-
M30 −4.21 −0.44 0.61 85
YSU −0.88 −0.41 0.81 76 responding to the soundings made on February 25 and March 7, 2019.
SH −3.66 −0.40 0.81 76 In the figure it can be seen that indeed the wind speed changes little
BP −5.01 −0.38 0.79 71 with the height in the first meters, and sometimes even decreases in-
QNSE −4.52 −0.34 0.92 67 distinctly in either case. From the height marked 960 m, it is observed
ACM2 −3.97 −0.20 0.95 77
that the wind definitely increases. Here, also is interesting that both
profiles are very similar to each other from about 350 m. Notice that the
Table 12 trend curves (second order polynomial) are very similar in both cases.
Bias, wind speed, wind direction (average of layer 10–2000) and gradient of θe The above means that the wind profiles built by the model can corre-
(average of layer 10–1000 m) for each PBL scheme, average of all cases and spond to the usual behavior of this variable in the region.
stations at the southern sector of the basin. Taking into account the vertical wind profiles observed at the
Huayao station on February 25 and March 7, 2019, it was decided to
Schemes BIAS (mm) Gradient θe(K) Wind speed Wind direction
(m/s) (degrees) perform a stability assessment based on the Froude number (Fr). The
Froude number is a way of characterizing the movement of an air parcel
BL 1.88 −0.73 1.76 56 in the face of wind shear and orography and expresses the ratio between
MYJ 0.38 −0.60 1.53 58
kinetic and potential energy according to the following formula:
GBM 0.72 −0.55 1.46 57
M25 −0.06 −0.49 1.33 60 V
M30 −0.48 −0.44 1.19 75 Fr =
(N ∗ hmax )
YSU −0.19 −0.30 1.73 57
SH 0.59 −0.06 1.86 51 where V is the flow velocity, N is the frequency of Brunt- Väisäla and
BP −1.69 0.32 1.55 62
hmax is the height of the barrier. The Brunt-Väisälä Frequency re-
QNSE −0.24 0.41 1.75 57
ACM2 −0.80 0.44 1.49 60 presents the maximum vertical oscillation frequency of propagation of
the internal gravity waves and expresses the stability of a parcel of
atmospheric fluid when it is forced by the wind to move over the
lesser magnitude. orographic accidents.
Table 8 shows wind speed, wind direction (average of layer If the Froude number equals or slightly exceeds 1, there is a prob-
10–2000) and gradient of θe (average of layer 2–1000 m) for each PBL ability that mountain waves will occur, if instead it is much < 1, the air
scheme, ordered from the most unstable to the least unstable schemes. flow is not strong enough to pass to the other side of the top of the
In the table it can be seen visually, that, in general, the most unstable mountain and is blocked: In the case of values much > 1, the air flow
schemes showed vertical profiles of wind with speeds that are weaker goes up the slope of the mountain and down the other side, without
than those that are less unstable. At the same time, wind direction tends significant oscillations (Gossard and Hooke, 1975; Balnes, 1987; Nappo,
to be more east southeast or southeast in the more unstable boundary 2002). Other results related to the Froude number can be found in the
layer, while in the less unstable boundary layer, wind tends to be more works of Smith, 1979; Jiang, 2003 and Miglietta and Buzzi, 2004. In
east northeast or near to east, in both cases, with some exceptions. South America, we can find the recent work of Vásquez and Falcón
Observe in Fig. 12, that with all the schemes the wind's meridional (2015), who calculated the vertical profiles of the Froude number for
component tends to be positive at a greater height, but it is more so several case studies of mountain waves in the planetary boundary layer
with the GBM, BL and MYJ schemes. in the Venezuelan Andean region.
The above can be better seen in Fig. 13, which shows the behavior In our work, the vertical profile of Fr was constructed for a west
of the variables in the Table 6 for each PBL scheme in graphic form. In wind, predominant in the two cases observed. The difference between
this case, the schemes have been ordered from the most unstable to the 4456 m (highest point east of the Mantaro Valley) and the height of the

13
A.S. Moya-Álvarez, et al. Atmospheric Research 233 (2020) 104708

Fig. 18. Behavior of the bias and the vertical gradient of θe (average of layer 10–1000 m), predicted with each scheme at the north sector of the basin, ordered from
the most unstable to the least unstable PBL.

first wind observation over the station (3320 m) was taken as the height were obtained in the southern sector.
of the barrier (Fig. 16a). Also in Table 9 it can be seen that in each region the schemes that
In Fig. 16b it can be seen that the Froude number increases obtained the best results were not the same, although the MYNN3
smoothly with height, but from approximately 3939 m increases in both scheme worked better than the rest in the northern sector with accep-
cases, mainly in the case of February 25, reaching values greater than table results in the center and south sectors in relation to the rest of the
one (1.42). Vásquez and Falcón (2015), obtained values of Fr at the schemes. The best statistics in each case are marked in red.
height of up to 1.9, but for higher wind speeds. This result implies that Similar to Tables 9, 10–12 show the vertical average of wind speed,
in the case of March 7, the air flow would not be able to exceed the top wind direction and gradient of θe for each PBL scheme at the north,
of the mountain, while in the case of February 25, the air flow that center and south sectors, ordered in each case from the schemes that
moves over 3939 m could overcome the barrier of the mountain, but generated the most unstable boundary layer to the least unstable ones.
not the air flow that moves below 3939 m. It is very likely that in In Table 10 it can be visually determined that the most unstable
conditions of the valley, where the Huayao station is framed, usually schemes have higher positive biases; however, the relationship of the
only the air flow near the top of the mountains can exceed it, mainly bias with the wind is not well defined. Thus, the correlation coefficient
due to the fact that in this area the winds are weak in relation to open between bias and θe was 0.52, while for wind in general they were very
regions. The above may imply that the orographic contribution inside low. Consequently, in Fig. 18 only the bias and θe curves are shown.
the valley to the formation of precipitation is not significant, perhaps in From the above, we concluded that in this sector of the basin atmo-
relation to the air flows that reach the mountains that surround the spheric stability is a fundamental factor in the formation of rainfall.
valley both in one sector and in another. In the central Andes of Peru, Table 11 shows that in the central sector of the basin all the schemes
no work has been carried out directly related to the detection of underestimated the precipitation, and in this case the schemes that
mountain waves, so in the future some research could be carried out in generated a more unstable boundary layer were those that under-
this regard, taking into account that this topic is not a direct object of estimated to a lesser extent because, as in the northern sector, they were
this work. those that generated more rainfall. In this sector, coefficients of corre-
In a more detailed analysis of the effectiveness of PBL schemes in lation were found of −0.58 for the gradient of θe and − 0.28 for the
precipitation forecasting it was possible to verify that the behavior of wind speed. Thus, in the center of the basin, atmospheric stability plays
the model is not the same in all sectors of the basin. Fig. 17, which a fundamental role in the formation of precipitation, which was also
shows the statistics calculated for the northern, central and southern generated to a greater extent with lower wind speeds (view Fig. 19).
sectors of the basin, reflects that the overestimations of the model are In the southern sector, unlike the northern and central sectors of the
mainly concentrated in the northern sector, while in the center and in basin, the results showed a low correspondence between the vertical
the south the model general underestimates the accumulated rainfall. In gradient of θe and the bias (−0.02), however, the correlation coeffi-
Table 7 it can be observed that, in general, the greater errors were also cients with speed and direction of wind, although they were not high,
obtained for the northern sector of the basin, while the lowest ones were better than in other sectors of the basin, in this case, −0.38

14
A.S. Moya-Álvarez, et al. Atmospheric Research 233 (2020) 104708

Fig. 19. Behavior of the bias, vertical gradient of θe (average of layer 10–1000 m) and wind speed (average of layer 10–2000 m), predicted with each scheme at the
central sector of the basin, ordered from the most unstable to the least unstable PBL.

and − 0.36 for wind speed and direction, respectively. From the above which predicted 2985 m. In Fig. 9 we observed that QNSE also predicts
we inferred that in the south of the basin the precipitations can have a the boundary layer much higher than the rest of the schemes, so it can
mainly advective character, that is to say, the precipitations form in an be deduced that this scheme greatly overestimates this parameter in the
area out of the basin and later they are transferred towards the basin. In region.
this case, figures, similar to Figs. 18 and 19 are not shown, as they are In Fig. 20 it is observed that the model is not able of reproducing the
not very informative. atmospheric instability shown in the vertical profile of observed
As it could be observed, the model in general tends to underestimate equivalent potential temperature, which is a determining factor in the
the precipitation over the study region, mainly in the central sector of formation of clouds and precipitations, in fact, we saw that in the north
the basin. So far, we have analyzed the possible causes of WRF pro- and central sectors of the basin, the vertical gradient of this variable
ducing more precipitation with some schemes than with others, how- showed the best correlation coefficients with the model's bias and it was
ever, it is interesting to know why this underestimation of rainfall oc- shown that the model produces more precipitation when the boundary
curs in the region. In this sense, we have analyzed the vertical profiles layer is more unstable.
of the equivalent potential temperature (Fig. 20) corresponding to the For these days the Table 14, shows that the model underestimated
two cases with atmospheric sounding in the “Huayao” station, which as the rainfall recorded with all the PBL schemes, although the forecasts
we know, is located precisely in the central sector of the basin. In this were closer to what was observed in the case of February 25, when the
case, as can be seen in Table 13, a boundary layer height of 1034 m was water vapor mixing ratio of the model was closer to that observed
observed at 18 UTC, while the schemes predicted a height between 900 (Fig. 21). The model for February 25 underestimated precipitation by
and 1700 m approximately, with the exception of the QNSE scheme, 34.7%, average among all schemes. The schemes that most approached

15
A.S. Moya-Álvarez, et al. Atmospheric Research 233 (2020) 104708

Fig. 20. Vertical profiles of equivalent potential temperature, predicted with the ten PBL schemes and observation, corresponding to February 25 (a) and march 07
(b), 2019 at 18 UTC.

Table 13
Height of PBL observed and predicted by the model with each PBL scheme in each case.
Date OBS ACM2 BL BP GBM MYJ M25 M30 QNSE SH YSU

02/25/2019 1034 1593 1717 1587 1246 914 1294 948 2985 1459 1478

Table 14
Precipitation observed and predicted by the model with each PBL scheme in each case.
Date OBS ACM2 BL BP GBM MYJ M25 M30 QNSE SH YSU

02/25/2019 9.8 7.5 7.8 8.8 9.0 4.5 3.6 5.6 5.7 5.9 5.6
03/07/2019 11.2 0.5 1.8 0.1 1.0 1.9 2.2 1.9 1.7 1.2 1.4

the observed precipitation were GBM, BP, BL and ACM2. rainy years in the studied region. The results of the simulations were
From the above, it follows that the absolute humidity predicted by compared with the rainfall registered in 19 stations in the region, all
the model is not sufficient to generate the precipitation that takes place, located in the vicinity of the Mantaro river basin, one of the largest and
however, there is another element that can be decisive in the fact that most important in the area. The bias, RMSE and MAE statistics were
WRF underestimates the rainfall in the region. Fig. 22, which shows the calculated and the schemes that help the model to forecast rainfall in
vertical profiles of predicted and observed temperature, reflects that the the region were determined. As part of the results, analyses of the
model predicts values of temperatures lower than those observed, vertical profiles of different meteorological variables are shown, which
which indicates that the warming of the low layer of the troposphere, help to explain why some schemes produce more rainfall than others.
simulated, is not enough, and consequently, the boundary layer be- Taking into account that there is no atmospheric sounding for the
comes more stable in simulations than reality. period studied and with the aim of taking them as a reference, simu-
lations were also carried out for two days of the year 2019 in which
4. Discussion and conclusions radiosonde data are available, which allowed us to delve into the
possible causes of the underestimations of the model in the precipita-
The study evaluated the sensitivity of the precipitation forecast in tion forecast, detected generally in the results. These two cases also
the central Andes of Peru of WRF model to change of the planetary offered an important reference on the vertical behavior of the wind.
boundary layer schemes, based on the important role played by the Thus, the model obtained the best results in the rainfall forecasting
processes in this part of the troposphere in the formation of clouds and when the following schemes were used: local MYNN3 (general and
precipitation. In the research, 10 planetary boundary layer schemes north of basin), local BL (central sector of the basin) and non-local YSU
were tested, of 12 available in version 3.7 of the model. The simulations (southern sector of the basin). In this case, the first was better than the
were carried out for 19 selected days of the months of January, rest in general and in the north of the basin, while it showed acceptable
February and March, between the years 2009 and 2012, which were results in the central and in the southern sectors.

16
A.S. Moya-Álvarez, et al. Atmospheric Research 233 (2020) 104708

Fig. 21. Vertical profiles of water vapor mixing ratio predicted with the ten PBL schemes and observation, corresponding to February 25 (a) and march 07 (b), 2019
at 18 UTC.

Fig. 22. Vertical profiles of temperature, predicted with the ten PBL schemes and observation, corresponding to February 25 (a) and march 07 (b), 2019 at 18 UTC.

The overall model overestimated precipitation in the northern Álvarez et al. (2018a) studied the precipitation forecast with WRF on
basin, while it underestimated significantly in the center. In the south, the Peruvian central Andes and obtained biases remarkably positive
some schemes overestimate and others underestimated. The best total however, Moya-Álvarez et al. (2018b) showed that the positive biases
results were obtained for this sector. Junquas et al. (2017) and Moya- are obtained when the precipitations are weak and that for

17
A.S. Moya-Álvarez, et al. Atmospheric Research 233 (2020) 104708

precipitations above of certain thresholds, the model underestimate. possible factors.


The difference between both results is a consequence of the fact that if By way of summary, the use of the MYNN3 scheme is suggested for
we evaluate the precipitation forecast of the model taking relatively the operational work of the WRF model in the region, provided that the
long periods without discriminating from the accumulated rainfall, the entire basin is to be covered. For domains that cover only a certain
model tends to overestimate, mainly due to the fact that there are more sector of the same, BL or YSU schemes may be used, depending on the
days without rain or with little rain than days with heavy rainfall; and work sector.
model in days of little rain, tends to overestimate.
It can also be concluded that one of the possible physics causes of Funding
the underestimation of the model is related to the inability of the model
to heat the lower layer of the troposphere to the values shown in the This research was funded by FONDECYT, CONCYTEC, Peru (grants
observations. Consequently, the boundary layer is more stable in the 010-2017-FONDECYT).
model than in reality. The study showed that the schemes that gener-
ated the most rainfall were those that generated a more unstable Declaration of Competing Interest
boundary layer with weaker wind speeds, at least with easterly winds,
statistically predominant in the sample taken. The presence of pre- None.
dominantly easterly winds in the selected rainy days, coincides with the
statements of other authors, who have concluded that in the Peruvian Acknowledgments
central Andes, the most intense rains occur when the moist air, coming
from The Amazon, flows over the region and converges in the afternoon Present study comes under the project “MAGNET-IGP:
with the mountain-valley circulation, being responsible for the occur- Strengthening the research line in physics and microphysics of the at-
rence of convective summer storms in this region (Garreaud, 1999; mosphere (Agreement N° 010-2017-FONDECYT)”. We would like to
Vuille, 1999; Vuille et al., 2000; Garreaud et al., 2003; Vuille and thank the FONDECYT, CONCYTEC, Peru, for financial support and
Keimig, 2004). Inter-American Institute for Cooperation on Agriculture (IICA) for ad-
One of the more important factors that characterizes the PBL ministrative support. This work was done using computational re-
schemes is the way they conceive vertical mixing. In this sense, schemes sources, HPC-Linux Cluster, from Laboratorio de Dinámica de Fluidos
with local closure only consider the closest layers to determine the Geofísicos Computacionales at Instituto Geofísico del Perú (grants 101-
conditions at a given level, while those that work with non-local closure 2014-FONDECYT, SPIRALES2012 IRD-IGP, Manglares IGP-IDRC,
also consider the more distant layers. In this work, YSU scheme (non- PP068 program. We would also like to thank NCEP for FNL analysis
local) in general did not provided the best results, but Table 4 shows data and SENAMHI for observational precipitation data.
that the MAE and RMSE obtained in the precipitation forecast with this
scheme were among the lowest 4. In a preliminary study, Moya-Álvarez References
et al. (2018a), had shown (with other dates) that in this region, the
model generates fewer errors in rainfall forecasting (MAE, RMSE) with Aceituno, P., 1989. On the functioning of the southern oscillation in the South American,
the YSU scheme than with local scheme MYJ (with the same config- sector. Part II. Upper-air circulation. J. Clim. (4), 341–355. http://journals.ametsoc.
org/doi/pdf/10.1175/15200442%281989%29002%3C0341%3AOTFOTS%3E2.0.
uration of soil, cumulus, radiation and microphysics used in this re- CO%3B2.
search), which is also confirmed in this new investigation. In that case, Balnes, P.G., 1987. Upstream blocking and airflow over mountain. Annu. Rev. Fluid
only YSU and MYJ schemes were compared. However, Lin-Jing and Mech. 19, 75–97. https://people.eng.unimelb.edu.au/pbaines/baines_x1987a.pdf.
Bougeault, P., Lacarrere, P., 1989. Parameterization of orography-induced turbulence in a
Wey-Lun (2016) for a large area covering the West Pacific Ocean, mesobeta–scale model. Mon. Weather Rev. https://doi.org/10.1175/1520-
mainland China, and the East Indian Ocean, showed that the simula- 0493(1989)117<1872:POOITI>2.0.CO;2.
tions with MYJ reproduces better distribution of rain belts, while YSU Bretherton, C.S., Park, S.A., 2009. Newmoist turbulence parameterization in the com-
munity atmosphere model. J. Clim. 22, 3422–3448. https://doi.org/10.1175/
strongly overestimates the precipitation intensity. Zeyaeyan et al. 2008JCLI2556.1.
(2017), determined that for the northwest of Iran, the scheme MYJ gave Chawla, I., Osuri, K., Mujumdar, P., Niyogi, D., 2018. Assessment of the Weather Research
better results in the precipitation forecast that the schemes YSU and SH. and Forecasting (WRF) model for simulation of extreme rainfall events in the upper
Ganga Basin. Hydrol. Earth Syst. Sci. 22, 1095–1117. https://doi.org/10.5194/hess-
Sultan et al. (2017) found too, for Himalaya, that YSU (non-local)
22-1095-2018.
scheme increases rainfall significantly as compared to MYNN25 (local), Cressman, G.P., 1959. An operational objective analysis system. Mon. Weather Rev. 87,
and ACM2 schemes. In that case, MYNN25 with MYNN surface layer 367–374. https://doi.org/10.1175/1520-0493(1959)087<0367:AOOAS>2.0.CO;2.
schemes produced more realistic storm structure and rainfall than other Farr, T.G., Rosen, P.A., Caro, E., Crippen, R., Riley, D., Hensley, S., Kobrick, M., Paller, M.,
Rodríguez, E., Ladislav, R., Seal, L., Shaffer, S., Shimada, J., Umland, J., Werner, M.,
schemes. Note that in the present investigation for the Central Andes, Oskin, M., Burkman, D., Alsdorf, D., 2007. The shuttle radar topography mission.
the scheme with the best results in general was the MYNN3 scheme, Rev. Geophys. 45. https://doi.org/10.1029/2005RG000183.
which is similar to MYNN25, but with second order of closure. In this Garreaud, R., 1999. Multi-scale analysis of the summertime precipitation over the central
andes. Mon. Weather Rev. 127, 901–921. http://www.dgf.uchile.cl/rene/PUBS/
case, the simulations were made also with MYNN schemes for surface Altiplano_multi_MWR.pdf.
layer. In this investigation, in the two cases in which the vertical pro- Garreaud, R., Vuille, M., Clement, A.C., 2003. The climate of the Altiplano: observed
files of equivalent potential temperature, temperature and water vapor current conditions and mechanisms of past changes. Palaeogeogr. Palaeoclimatol.
Palaeoecol. 194, 5–22. http://www.atmos.albany.edu/facstaff/mathias/pubs/
mixing ratio were compared with observation, the water vapor mixture Garreaud_et_al_2003.pdf.
ratio values of MYNN3 were the closest to the observed values than Gbode, I.E., Dudhia, J., Ogunjob, K.O., Ajayi, V.O., 2018. Sensitivity of different physics
others schemes. Another endorsement in favor of the MYNN schemes schemes in the WRF model during a West African monsoon regime. Theor. Appl.
Climatol. 136, 133–751. https://doi.org/10.1007/s00704-018-2538-x.
was shown in the works of Gbode et al. (2018), which concluded that
Gossard, E.E., Hooke, W.H., 1975. Waves in the Atmosphere. Academic Press, San Diego,
the Grell-Freitas cumulus scheme (also used in this research) works pp. 456. http://www.scirp.org/(S(lz5mqp453edsnp55rrgjct55))/reference/
better than its predecessors, especially when it was combined with the ReferencesPapers.aspx?ReferenceID=1256805.
Grell, G.A., Freitas, S.R., 2014. A scale and aerosol aware stochastic convective para-
PBL MYNN25 scheme.
meterization for weather and air quality modeling. Atmos. Chem. Phys. 14,
However, Ray and Pattnaik (2019) determined that for India, the 5233–5250. https://doi.org/10.5194/acp-14-5233-2014.
MRF (non-local) and ACM2 (hybrid) schemes showed better statistical Grenier, H., Bretherton, C.S., 2001. A moist PBL parameterization for large-scale models
indicators in rainfall forecasting than the local BL and MYNN25 and its application to subtropical cloud-topped marine boundary layers. Mon.
Weather Rev. 129, 357–377. https://doi.org/10.1175/1520-0493(2001)
schemes. The above indicates that in some cases better results can be 129<0357:AMPPFL>2.0.CO;2.
obtained in the rainfall forecast using local schemes and in others, non- Hong, S.-Y., Noh, Y., Dudhia, J., 2006. A new vertical diffusion package with an explicit
local schemes, it may depend on the region studied, among others treatment of entrainment processes. Mon. Weather Rev. https://doi.org/10.1175/

18
A.S. Moya-Álvarez, et al. Atmospheric Research 233 (2020) 104708

MWR3199.1. Nakanishi, M., Niino, H., 2006. An improved Mellor–Yamada level-3 model: its numerical
Iacono, M.J., Delamere, J.S., Mlawer, E.J., Shephard, M.W., Clough, S.A., Collins, W.D., stability and application to a regional prediction of advection fog. Bound.-Layer
2008. Radiative forcing by long–lived greenhouse gases: calculations with the AER Meteorol. 119, 397–407. https://doi.org/10.1007/s10546-005-9030-8.
radiative transfer models. J. Geophys. Res. 113, D13103. http://citeseerx.ist.psu. Nappo, C.J., 2002. An introduction to atmospheric gravity waves, Int. Geophys. Ser., 8.
edu/viewdoc/download?doi=10.1.1.370.4357&rep=rep1&type=pdf. San Diego, California. https://www.elsevier.com/books/an-introduction-to-
Janjić, Z.I., 2002. Nonsingular Implementation of the Mellor–Yamada Level 2.5 Scheme atmospheric-gravity-waves/nappo/978-0-12-385223-6.
in the NCEP Meso Model. NCEP Office Note 437, pp. 61. http://citeseerx.ist.psu.edu/ Pleim, J.E., 2007. A combined local and nonlocal closure model for the atmospheric
viewdoc/download?doi=10.1.1.459.5434&rep=rep1&type=pdf. boundary layer. part I: model description and testing. J. Appl. Meteorol. Climatol. 46,
Jankov, I., Gallus, W.A., Segal, M., Koch, S.E., 2005. The impact of different WRF model 1383–1395. https://doi.org/10.1175/JAM2539.1.
physical parameterizations and their interactions on warm season MCS rainfall. Rasmussen, E.N., Blanchard, D.O., 1998. A baseline climatology of sounding-derived
Weather Forecast. 20 (6). https://doi.org/10.1175/WAF888.1. supercell and tornado forecast parameters. Wea. Forecast 13, 1148–1164. https://
Jiang, Q., 2003. Moist dynamics and orographic precipitation. Tellus 55A, 301–316. doi.org/10.1175/1520-0434(1998)013,1148:ABCOSD.2.0.CO;2.
https://doi.org/10.3402/tellusa.v55i4.14577. Ray, D., Pattnaik, S., 2019. Evaluation of WRF planetary boundary layer parameterization
Joon-Bum, J., Sangil, K., 2017. Sensitivity study on high-resolution WRF precipitation schemes for simulation of monsoon depressions over India. Meteorog. Atmos. Phys.
forecast for a heavy rainfall event. Atmosphere 8, 96–112. https://doi.org/10.3390/ https://doi.org/10.1007/s00703-019-0656-3.
atmos8060096. Rodríguez, E., Morris, S.C., Belz, J.E., 2006. A Global Assessment of the SRTM
Junquas, C., Takahashi, K., Condom, T., Espinosa, J.-C., Chavez, S., Sicart, J.-E., Lebel, T., Performance. Photogramm. Eng. Remote. Sens. (3), 249–260. https://doi.org/10.
2017. Understanding the influence of orography on the precipitation diurnal cycle 14358/PERS.72.3.249.
and the associated atmospheric processes in the Central Andes. Clim. Dyn. 1–23. Schneider, R.S., Dean, A.R., 2008. A comprehensive 5-year severe storm environment
https://link.springer.com/article/10.1007/s00382-017-3858-8. climatology for the continental United States. In: Preprints, 24th Conf. on Severe
Kosovic, B., Curry, J., 2000. A large eddy simulation study of a quasi-steady, stably Local Storms. 16A. Amer. Meteor. Soc, Savannah GA, pp. 4. http://ams.confex.com/
stratified atmospheric boundary layer. J. Atmos. Sci. 57. https://doi.org/10.1175/ ams/pdfpapers/141748.pdf.
1520-0469(2000)057<1052:ALESSO>2.0.CO;2. Shahzad, Sultan, Lin, Hui, Sajjad, H., Amir, M., Atif, I., Ijaz, A., Zaheer, B.A., Iftikhar, A.,
Kumar, et al., Kumar, S., Silva, Y., Moya-Álvarez, A.S., Martínez-Castro, D., 2019a. 2016. Impact of pbl schemes on high-resolution rainfall simulation over the selected
Seasonal and regional differences in extreme rainfall events and their contribution to region of Himalaya, Karakorum and Hindukush. Arab. J. Earth Sci. 4, 81–94.a.
the world's precipitation: GPM observations. Adv. Meteorol. 2019. https://doi.org/ Shin, H.H., Hong, S.-Y., 2011. Intercomparison of planetary boundary-layer para-
10.1155/2019/4631609. meterizations in the WRF model for a single day from CASES-99. Bound.-Layer
Kumar, S., Silva-Vidal, Y., Moya-Álvarez, A.S., Martínez-Castro, D., 2019b. Effect of the Meteor. 139, 261–281. https://doi.org/10.1007/s10546-010-9583-z.
surface wind flow and topography on precipitating cloud systems over the Andes and Shin, H.H., Hong, S.-Y., 2015. Representation of the subgrid-scale turbulent transport in
associated Amazon basin: GPM observations. Atmos. Res. 225, 193–208. convective boundary layers at gray-zone resolutions. Mon. Weather Rev. 143,
Lenters, J.D., Cook, K.H., 1997. On the origin of the bolivian high and related circulation 250–271. https://doi.org/10.1175/MWR-D-14-00116.1.
features of the south American climate. J. Atmos. Sci. 54, 656–677. https://doi.org/ Silva, Y., Takahashi, K., Chávez, R., 2008. Dry and wet rainy seasons in the Mantaro river
10.1175/1520-0469(1997)054<0656:OTOOTB>2.0.CO;2. basin (Central Peruvian Andes). Adv. Geosci. 14, 261–264. https://hal-insu.archives-
Lin-Jing, Q., Wey-Lun, Q., 2016. Intercomparison of different physics schemes in the WRF ouvertes.fr/hal-00297108.
model over the Asian summer monsoon region. Atmos. Ocean. Sci. Lett. 9, 169–177. Skamarock, W., Klemp, J., Dudhia, J.A., 2008. Description of the Advanced Research
https://doi.org/10.1080/16742834.2016.1158618. WRF Version 3, NCAR Technical Note, NCAR/TN–468+STR, National Center for
Lynn, B.H., Stauffer, D.R., Wetzel, P.J., Tao, W.K., Alpert, P., Perlin, N., Baker, R.D., Atmospheric Research (NCAR), Mesoscale and Microscale Meteorology Division.
Munoz, R., Boone, A., Jia, Y.Q., 2001. Improved simulation of Florida summer con- Boulder, Colorado, USA.
vection using the PLACE land model and a 1.5-order turbulence parameterization Smith, R.B., 1979. The Influence of Mountains on the Atmosphere, Advances in
coupled to the Penn State-NCAR mesoscale model. Mon. Weather Rev. 129, Geophisics, 21. Academic Press, pp. 87–230. http://adsabs.harvard.edu/full/
1441–1461. https://serials.unibo.it/cgi-ser/start/it/spogli/df-s.tcl?prog_art= 1981ESASP.165...37S.
8234696&language=ITALIANO&view=articoli. Sukoriansky, S., Galperin, B., Perov, V., 2005. Application of a new spectral theory of
Martínez, A.G., Núñez, E., Silva, Y., 2006. Vulnerability and adaptation to climate change stably stratified turbulence to the atmospheric boundary layer over sea ice. Bound.-
in the Peruvian Central Andes. In: Results of a Pilot Study, in: International Layer Meteorol. 117, 231–257. https://doi.org/10.1007/s10546-004-6848-4.
Conference on Southern Hemisphere Meteorology and Oceanography (ICSHMO), 8, Tewari, M., Chen, F., Wang, W., Dudhia, J., LeMone, M.A., Mitchell, K., Ek, M., Gayno, G.,
Foz Do Iguac¸U. Roceedings, S˜ao Jos'e dos Campos, INPE, CD-ROM ISBN Wegiel, J., Cuenca, R.H., 2004. Implementation and verification of the unified NOAH
85–17–00023-4, 2006, pp. 297–305. http://www.met.igp.gob.pe/publicaciones/ land surface model in the WRF model. In: 20th Conference on Weather Analysis and
2000_2007/V&AMantaro.pdf. Forecasting/16th Conference on Numerical Weather Prediction, pp. 11–15. https://
Mellor, G.L., Yamada, T., 1974. A hierarchy of turbulence closure models for planetary ams.confex.com/ams/84Annual/techprogram/paper_69061.htm.
boundary layers. J. Atmos. Sci. 13, 1791–1806. https://doi.org/10.1175/1520- Vásquez, H., Falcón, E., 2015. Ondas de montaña en la capa límite planetaria de la región
0469(1974)031%3C1791:AHOTCM%3E2.0.CO%3B2. andina venezolana. Rev. Bras. Metorologia 30, 241–253. https://doi.org/10.1590/
Mellor, G.L., Yamada, T., 1982. Development of a turbulent closure model for geophysical 0102-778620130626.
fluid problems. Rev. Geoph. Space Phys. 20, 851–875. http://www.fap.if.usp.br/ Vuille, M., 1999. Atmospheric circulation over the Bolivian Altiplano during dry and wet
~hbarbosa/uploads/Teaching/Modclim2010a/MellorYamada1982.pdf. periods and extreme phases of the Southern Oscillation. Int. J. Climatol. 19,
Mercader, J., Codina, B., Sairouni, A., Cunillera, J., 2010. Resultados del modelo 1579–1600. http://www.atmos.albany.edu/facstaff/mathias/pubs/Vuille_1999.pdf.
meteorológico WRF-ARW sobre Cataluña, utilizando diferentes parametrizaciones de Vuille, M., Keimig, F., 2004. Interannual variability of summertime convective cloudiness
la convección y la microfísica de nubes. Tethys 7, 77–89. https://doi.org/10.3369/ and precipitation in the Central Andes derived from ISCCP-B3 data. J. Clim. 17,
tethys.2010.7.07. 3334–3348. http://www.atmos.albany.edu/facstaff/mathias/pubs/Vuille_Keimig_
Miglietta, M.M., Buzzi, A., 2004. A numerical study of moist stratified flow regimes over 2004.pdf.
isolated topography. Q. J. R. Meteorol. Soc. 130, 1749–1770. https://doi.org/10. Vuille, M., Bradley, R.S., Keimig, F., 2000. Interannual climate variability in the Central
1256/qj.02.225. Andes and its relation to tropical Pacific and Atlantic forcing. J. Geophys. Res. 105
Morrison, H., Thompson, G., Tatarskii, V., 2009. Impact of cloud microphysics on the (12), 447–460. https://agupubs.onlinelibrary.wiley.com/doi/pdf/10.1029/
development of trailing stratiform precipitation in a simulated squall line: 2000JD900134.
Comparison of one- and two-moment schemes. Mon. Weather Rev. 137, 991–1007. Wardah, T., Kamil, A.A., Sahol Hamid, A.B., Maisarah, W. W. I., 2011. Quantitative
https://doi.org/10.1175/2008MWR2556.1. precipitation forecast using MM5 and WRF models for Kelantan River Basin. Int. J.
Moya-Álvarez, A.S., Martínez-Castro, D., Flores, J.L., Silva, Y., 2018a. Sensitivity study on Geol. Environ. Eng. 5, pp. 712–716, https://waset.org/publications/7831/
the influence of parameterization schemes in WRF_ARW model on short- and quantitative-precipitation-forecast-using-mm5-and-wrf- models-for-kelantan-river-
medium-range precipitation forecasts in the Central Andes of Peru. Adv. Meteorol. basin.
2018, 16. https://doi.org/10.1155/2018/1381092. 1381092. Wisse, J.S.P., de Arellano, J.V.G., 2004. Analysis of the role of the planetary boundary
Moya-Álvarez, A.S., Gálvez, J., Holguín, A., Estevan, R., Kumar, S., Villalobos, E., layer schemes during a severe convective storm, Ann. Geophys 22, 1861–1874 (SRef-
Martínez-Castro, D., Silva, Y., 2018b. Extreme rainfall forecast with the WRF-ARW ID: 1432–0576/ag/2004-22-1861).
model in the Central Andes of Peru. Atmosphere 9, 362. https://doi.org/10.3390/ Xu, L.R., Zhao, M., 2000. The influences of boundary layer parameterization schemes on
atmos9090362. mesoscale heavy rain system. Adv. Atmos. Sci. 17, 458–472.
Moya-Álvarez, A.S., Martínez-Castro, D., Estevan, R., Kumar, S., Silva, Y., 2019. Response Zeyaeyan, S., Fattahi, E., SaadatAbadi, A.R., Azadi, M., Vazifedoust, M., 2017. Evaluating
of the WRF model to different resolutions in the rainfall forecast over the complex the effect of physics schemes in WRF simulations of summer rainfall in North West
Peruvian orography. Theor. Appl. Climatol. 1–15. https://doi.org/10.1007/s00704- Iran. Climate 5, 48. https://doi.org/10.3390/cli5030048.
019-02782-3.

19

You might also like