Investigation of Climatological Onset and Withdrawal of the Rainy Season in Panama Based on a Daily Gridded Precipitation Dataset with a High Horizontal Resolution

1APRIL 2015 N A K A E G A W A E T A L . 2745

TOSIYUKI NAKAEGAWA,OSAMU ARAKAWA, AND KENJI KAMIGUCHI Climate Research Department, Meteorological Research Institute, Tsukuba, Japan

(Manuscript received 28 March 2014, in ﬁnal form 19 December 2014)

ABSTRACT

The present study investigated the onset and withdrawal dates of the rainy season in Panama by using newly developed, gridded, daily precipitation datasets with a high horizontal resolution of 0.058 based on ground precipitation observations. The onset and withdrawal dates showed very complicated geographical features, although the country of Panama is oriented parallel to latitude lines, and the geographical patterns of the onset and withdrawal dates could simply reflect the latitudinal migration of the intertropical convergence 2 zone, as seen in other regions and countries. An absolute threshold value of 3 mm day 1 (pentad mean precipitation) was used to determine the onset and withdrawal dates. The onset and withdrawal dates obtained from the gridded daily precipitation dataset clearly depicted the migration of the rainy season. The rainy season starts suddenly in pentad 21 (11–15 April) in most of eastern Panama and in pentad 22 (16–20 April) in most of western Panama. The termination of the rainy season begins in Los Santos Province during pentad 67 (27 November–1 December) and expands to both the eastern and western surrounding areas. There is no dry season in the western part of the Caribbean coastal zone. Water vapor fluxes and topography suggest dynamical causes, such as a topographically induced upward mass flux accompanied by high humidity, for the complicated geographical features of the onset and withdrawal dates. An assessment was made of uncertainties in the timing of the onset and withdrawal associated with the definition of these terms.

1. Introduction based on the Köppen climate classification system: group A includes tropical climates and group C includes mild, The country of Panama is located in the tropics be- temperate climates (UNESCO 2008). The climate of tween latitudes and longitudes of 78–108N and 77–838W, more than 90% of Panama is tropical: the climate of the respectively. It is bordered on the south by the Pacific coastal zone of west Panama is a tropical rain forest Ocean and on the north by the Caribbean Sea (Fig. 1a). climate (Af), the climate of the provinces of Chiriquí The climate of Panama is affected by its proximity to the and Veraguas and the coastal zone of east Panama is Atlantic Ocean and the fact that it is bordered by the a tropical monsoon climate, and the climate of the Caribbean Sea and the Pacific Ocean. Together these coastal zone surrounding the Gulf of Panama is a tropi- bodies of water create a maritime atmospheric envi- cal wet and dry climate (Aw). Mild, temperate climates ronment characterized by high humidity that is typical of are confined to the Cordillera de Talamanca and Cor- much of Central America and the Caribbean region dillera Central Mountains. Both of these climates are (Hastenrath 1978; Ropelewski and Halpert 1987; characterized by high humidity with much precipitation. Nakaegawa et al. 2014b). The northeast trade winds, In most of Panama, except for a small part of the which contribute to the Caribbean low-level jet (CLLJ; Caribbean side, there are two seasons, namely a dry Amador 1998, 2008), are directed over the country by season and a rainy season (Taylor and Alfaro 2005; the North Atlantic subtropical high (Taylor and Alfaro UNESCO 2008). This seasonality is primarily associated 2005). The climates of Panama fall into two main groups with the migration of the intertropical convergence zone (ITCZ), although the latitudinal migration of the ITCZ is minimal near Panama, and other regional phenomena Corresponding author address: T. Nakaegawa, Climate Research Department, Meteorological Research Institute, 1-1 Nagamine, affect the seasonality (e.g., Magaña et al. 1999; Alfaro Tsukuba, Ibaraki 305-0052, Japan. 2002; Wang and Enfield 2003; Amador et al. 2006). The E-mail: [email protected] dry season typically occurs from December to April,

DOI: 10.1175/JCLI-D-14-00243.1

Ó 2015 American Meteorological Society Unauthenticated | Downloaded 09/30/21 06:48 PM UTC 2746 JOURNAL OF CLIMATE VOLUME 28

FIG. 1. (a) Provinces and Cordillera in Panama with topography (shaded, m). (b) Distribution of precipitation gauge stations (red squares) and the eight stations used in this study (enclosed by blue circles). See also Table 3. when the ITCZ is moving south from Panama toward country, because the area of Panama is smaller than that the equator, and the rainy season occurs from May to of the synoptic-scale phenomena that usually trigger the November, when the ITCZ is passing over Panama to- onset and withdrawal. Alternatively, the onset can ward the Caribbean Sea, just to the north of Panama proceed from the Pacific to the Caribbean coast with the (Mitchell and Wallace 1992). The rainy season can be northward migration of the ITCZ, and the withdrawal divided into three phases: rainfall is heavy during the can proceed from the Caribbean to the Pacific coast as onset period from late April to early May, when the the ITCZ moves south. Analysis of such a small area ITCZ is passing northward over Panama; after the ITCZ requires a precipitation dataset with high horizontal has moved over the Caribbean Sea, rainfall weakens resolution to delineate the detailed spatial features of slightly (Magaña et al. 1999); and as the ITCZ is moving the onset and withdrawal. Enfield and Alfaro (1999) back toward the south over Panama, the heaviest rain- computed annual onset and withdrawal dates based on fall of the rainy season occurs in some regions, including data from fewer than 20 gauge stations in the western the Panama Canal, before the withdrawal period (e.g., part of Panama to elucidate the effect of the tropical Murphy et al. 2014). North Atlantic and El Niño–Southern Oscillation on the The climatological onset and withdrawal of the rainy interannual variability of the dates and showed that the season have been investigated in many regions: India variability of the sea surface temperature (SST) anom- (IMD 1943); the Indochina Peninsula (Matsumoto aly in the tropical Atlantic influences rainfall over the 1997); Japan (JMA 2014); the Asia–Pacific region Caribbean and Central America more than the vari- (Wang and Linho 2002); the tropical Americas, in- ability of the SST in the tropical eastern Pacific. They did cluding Panama (Gramzow and Henry 1972; Horel et al. not, however, provide climatological onset and with- 1989; Alfaro 2002); the Amazon region (Marengo et al. drawal dates. No comprehensive investigations of the 2001); and South America (Liebmann et al. 2007). These climatological onset and withdrawal dates of the rainy transitions are of great interest because they determine season have been carried out for Panama. In the present the timing of agricultural practices (Yavitt et al. 2004), study we used a gridded precipitation dataset with high and the length of the dry season determines the type of horizontal resolution developed from gauge-station ecosystem in a region (Sombroek 2001). These studies observations to investigate the climatological onset have documented distinct spatial and temporal differ- and withdrawal dates of the rainy season in Panama. ences in the onset and withdrawal dates of the rainy season because the target areas have been relatively large. For such large areas, datasets obtained from 2. Data sparsely located gauge stations or satellite-based data- a. Gridded data production sets with a low horizontal resolution are insufficient for an informed analysis. Records of precipitation have been maintained by the The area of Panama is 75 416 km2, only about 0.05% Panama Canal Authority (ACP) for the Panama Canal of Earth’s land area. The onset or withdrawal of the region and by the Empresa de Transmisión Eléctrica, rainy season can occur simultaneously over the entire S.A. (ETESA) for the remainder of Panama. Both

Unauthenticated | Downloaded 09/30/21 06:48 PM UTC 1APRIL 2015 N A K A E G A W A E T A L . 2747

TABLE 1. Stations with less than 80% of acceptable observations used in this analysis. Distance is the distance to the nearest stations where more than 80% of the observations are acceptable. (Note that the distances denote not the one between the exact two station locations but the one between the two grid points closest to the observation stations for each.)

Longitude Latitude Altitude Available Distance Stations Province (8W) (8N) (m)Data source data (%) (km) Bocas del Toro Bocas del Toro 82.225 9.375 2 GSOD 3 5.6 Howard Z.C. Panama 7983600000 885500000 137 ACP 4 11.1 Changuinola 1 Bocas del Toro 8282900000 982900000 10 ETESA 5 20.0 Balboa Heights Panama 7983300000 885700000 30 ACP 8 5.6 La Humedad Colon 8080202100 980205400 30 ACP 8 5.6 Fort Sherman Colon 7985700000 982200000 9 ACP 11 7.8 Fortuna Chiriqui 8281405800 884403800 1040 ETESA 22 12.4 Los Canones Panama 8080400000 885700000 188 ETESA 36 11.1 La Mesa De Macaracas Los Santos 8083700000 783800000 180 ETESA 38 15.7 Boca Del Rio Mamoni Panama 7980400000 980700000 10 ETESA 38 5.6 Los Canones Panama 8080304500 885605600 186 ACP 46 12.4 Cerro Verde Veraguas 8085000000 883100000 800 ETESA 73 7.8 datasets have been archived by the ETESA. There are these stations from the nearest station at which more officially 151 meteorological gauge stations in Panama than 80% of the total observations were acceptable (ETESA 2014). However, we used the data from the 118 (Table 1) was less than 20 km in all cases and less than gauge stations that were available in digital form 8 km in six cases. Therefore, the low percentage of ac- (Fig. 1b). ceptable observations at these 12 stations did not sub- First, all the time series of daily precipitation data stantially affect the gridded precipitation dataset. were plotted for visual analysis and manually checked. To improve the interpolation near the borders with We found several types of doubtful data: artificial rep- Costa Rica and Colombia, we also used two additional etition of the same values (seven gauge stations), un- datasets: the Global Historical Climatology Network- natural time series (four instances), and values that were Daily (Klein Tank et al. 2002) and the Global Surface integer multiples of 10 (three instances). Extremely Summary of the Day (GSOD; http://www.ncdc.noaa.gov/ large values (more than 10 instances) were found and cgi-bin/res40.pl). As the base climatological monthly were regarded as missing values; omission of these large precipitation for the spatial interpolation, we adopted values may have excluded actual extremes. These data WorldClim, which is a dataset of interpolated climate are under review at the ETESA. Second, we applied the surfaces at a spatial resolution of 30 arc s (Hijmans et al. automated, objective quality control used in the Asian 2005). These data were obtained in 2010. We confirmed Precipitation–Highly Resolved Observational Data In- that the data in these two datasets were independent of tegration toward Evaluation of Water Resources the ETESA dataset used for Panama. (APHRODITE) high-resolution, daily precipitation We estimated daily precipitation using the same sta- product (Kamiguchi et al. 2010; Yatagai et al. 2012)to tistical methods used in APHRODITE. The analysis the daily precipitation data. The nation and the eleva- period was 40 years, from 1 January 1970 to 31 December tion of stations in the metadata were compared with the 2009. The temporal resolution was 1 day, and the hori- digital maps. Time series of daily precipitation data from zontal resolution was 0.058,or3min. both single stations and multiple stations were subjected Daily climatology was produced as follows. We com- to this quality control. A total of 104 gauge stations were puted monthly climatologies from the ETESA dataset. used for the final analysis. The density of gauge stations We used the Mountain Mapper method (Schaake 2004) was nine stations per 104 km2; the corresponding aver- to compute the ratio of the monthly climatology be- age area per station was 1109 km2. This density of sta- tween WorldClim and the ETESA dataset at each gauge tions is comparable to the highest density of rain gauge station. The Sheremap scheme (Willmott et al. 1985), an stations in the world. For comparison, in Japan there are angular-distance-weighting method, was used to spa- 34 stations per 104 km2, the corresponding area per sta- tially interpolate the ratios in Panama. We obtained tion being 290 km2 (JMA 2013). Table 1 lists gauge a new monthly climatology by multiplying the World- stations used in this analysis at which fewer than 80% of Clim climatology by these ratios at a horizontal resolu- the total observations were acceptable after the quality tion of 0.058. The new monthly climatology at each control. There were 12 such stations, corresponding to observation station was consistent with the climatology 11.5% of the total gauge stations. The distance of each of obtained from the ETESA dataset. We constructed the

Unauthenticated | Downloaded 09/30/21 06:48 PM UTC 2748 JOURNAL OF CLIMATE VOLUME 28

TABLE 2. Climatological annual precipitation averaged over the entire land area of Panama estimated with ﬁve different precipitation datasets.

Annual Resolution precipitation 2 Temporal (left, mm day 1; Estimation 2 Datasets Horizontal Temporal coverage right, mm yr 1) period MRI 0.058 1 day 1970–2009 6.97 2545 1971–2002 7.06 2578 1971–2009 7.43 2715 1998–2009 UNESCO — Climatology 1971–2002 8.01 2924 1971–2002 CRU 0.508 1 month 1901–2009 7.19 2625 1971–2002 GPCP1DD 1.008 1 day 1997–present 7.69 2810 1998–2009 TRMM 0.258 3 h 1998–present 6.85 2501 1998–2009 daily climatology at each grid by adding the first six Measuring Mission (TRMM) 3B42 dataset (Huffman et al. Fourier components of the annual cycle of the new 2007), and the Global Precipitation Climatological Project monthly climatology. This construction assumes that the 1-day and 1-degree dataset (GPCP1DD; Huffman et al. daily climatology changes smoothly with time when 2001). The characteristics of these datasets are summarized sufficient observations are available. We produced high- in Table 2. frequency variability in the following way. We used the Japanese 55-yr Reanalysis produced by Daily gridded precipitation data were produced as fol- the JMA (JRA-55; Kobayashi et al. 2015) to depict the lows. We spatially interpolated the ratios of daily pre- mesoscale circulation of water vapor fluxes and their cipitation between observations and climatology with the convergences. JRA-55 has a horizontal resolution of Sheremap scheme. The ratio used for this interpolation about 55 km and covers the period of time from 1958 to was the daily precipitation divided by the sum of the cli- the present. Climatological values were computed for 2 matology and 1 mm day 1 to avoid unnatural behavior of the 40 years from 1970 to 2009. A high-resolution the interpolated ratio. In this interpolation, we applied blended analysis of daily SST [the National Oceanic a small weight to a target grid on the opposite side of a high and Atmospheric Administration (NOAA) Optimum ridgeandalargeweighttoatargetgridonthesamesideof Interpolation Daily Sea Surface Temperature (OISST) a high ridge. We refer to this newly developed, daily version 2; Reynolds et al. (2007); http://www.esrl.noaa.gov/ gridded precipitation dataset as the Meteorological Re- psd/data/gridded/data.noaa.oisst.v2.highres.html] was also search Institute (MRI) dataset. The current version of this used to depict the geographical distribution of SST sur- dataset should not be used for meteorological analysis rounding Panama. The dataset has a horizontal resolution because extreme precipitation events may have been ex- of 0.258 and covers the period of time from September cluded during the quality control, as mentioned in section 1981 to the present. Climatological values were computed 2a. However, this dataset can be used for a climate analysis for the 28 years from 1982 to 2009. such as the present study. b. Observations 3. Definition of onset and withdrawal Four other observation-based precipitation datasets Many methods of determining the onset and with- were used to compare the climatological annual pre- drawal dates of the rainy season have been proposed. cipitation for the entire country of Panama against the Two methods are widely used: the uniform threshold MRI dataset. A United Nations Educational, Scientific method and the distributed threshold method. The and Cultural Organization (UNESCO) dataset was de- uniform threshold method defines the onset to be the veloped by applying a simple spatial interpolation scheme first pentad when the mean pentad precipitation exceeds to the climatological annual mean of surface gauge station a certain uniform value over a target domain (e.g., data in the ETESA (UNESCO 2008). The climatological Gramzow and Henry 1972; Ogallo 1989; Marengo et al. annual mean was the only available value in the ETESA, 2001; Alfaro 2002). Zhang and Wang (2008) modified and the geographical distribution was in the form of a pa- the definition for global monsoon rainfall. They defined 2 per map. The three other datasets contained gridded data the uniform value to be 3 mm day 1 during the rainy from different periods of time and with different temporal season, the interval of time when greater than 55% of and horizontal resolutions: the Climate Research Unit the annual precipitation occurs. The Indian Meteorolog- dataset (CRU; Harris et al. 2014), the Tropical Rainfall ical Department (IMD) uses the distributed threshold

Unauthenticated | Downloaded 09/30/21 06:48 PM UTC 1APRIL 2015 N A K A E G A W A E T A L . 2749

21 FIG. 2. Geographical distribution of climatological annual precipitation (mm day ) in Panama determined with the (a) MRI, (b) CRU, (c) GPCP, and (d) TRMM datasets. method and defines the onset date to be the first pentad The threshold value in the present study was set to 2 when the mean pentad precipitation exceeds the clima- 3 mm day 1 based on Zhang and Wang (2008). The tological annual mean precipitation in at least three threshold value of Wang and LinHo (2002) is also similar consecutive pentads (IMD 1943). We calculated the to this threshold based on the definition of the threshold 2 threshold value of the climatological annual mean pre- used in the present study. The threshold of 3 mm day 1 2 2 cipitation throughout the target domain (Fig. 2). Almost corresponds to 91 mm month 1 or 1095 mm yr 1.In the same methods have been used in many studies (e.g., general, a period of time with precipitation greater than 2 Matsumoto 1997; Liebmann et al. 2007). Wang and 3mmday 1 cannot be regarded as a dry season. Most LinHo (2002) have modified this definition for Asian wheat is harvested in areas with annual precipitation of 2 monsoon rainfall by equating the onset to the first pentad 500–1000 mm yr 1, and most rice is harvested in areas 2 when the maximum pentad precipitation exceeds the with annual precipitation greater than 1000 mm yr 1 2 mean precipitation for January by 5 mm day 1. In a similar (MAFF 2003). Hence this threshold value is useful from manner, the distributed threshold method defines the an agricultural standpoint. Previous studies have used 2 withdrawal date to be the last pentad when the mean 5 mm day 1 (e.g., Gramzow and Henry 1972; Alfaro pentad precipitation is less than the climatological an- 2002) for the definition of the rainy season, but in a nual mean in at least three consecutive pentads. In slightly different context such as clear identification of contrast, the uniform threshold method defines the the rainy pentad. withdrawal date to be the last pentad when the mean pentad precipitation is less than a certain value. We 4. Comparison of rainfall datasets determined the onset and withdrawal dates primarily with the uniform threshold method. We used the dis- Table 2 compares the average climatological annual tributed threshold method for comparative purposes to precipitation over the entire land area of Panama esti- address the uncertainty in the estimates due to defini- mated with different datasets. The climatological aver- 2 tional differences. age of the MRI datasets was 2578 mm yr 1 for the period

Unauthenticated | Downloaded 09/30/21 06:48 PM UTC 2750 JOURNAL OF CLIMATE VOLUME 28

TABLE 3. Geographical information about the eight precipitation gauge stations. See Fig. 1b for the geographical locations in Panama.

Station name City Province Longitude (8W) Latitude (8N) David David Chiriqui 82.67 9.38 Seiyic Changuinola Bocas del Toro 82.42 8.40 Los Santos Los Santos Los Santos 80.42 7.95 Cocle del Norte Cocle del Norte Colon 80.57 9.06 Amador Panama City Panama 79.82 9.40 San Pedro Cativa Colon 79.53 8.92 Garachine Garachine Darien 78.37 8.08 Mulatupo Mulatupo Kuna Yara 77.75 8.93

2 1971–2009 and varied from 2545 to 2715 m yr 1,de- local maxima are apparent. Such maxima are often pending on the time interval. The values determined artifacts of the use of a simple interpolation method. from the CRU and MRI datasets were closest for the These comparisons indicate that the MRI dataset time interval 1971–2002, and the averages determined provides good estimates of the spatial patterns and from the GPCP and MRI datasets were closest for the mean values of climatological annual precipitation time interval 1998–2009. The largest and smallest throughout the entire country of Panama. The MRI values were derived from the UNESCO and TRMM dataset should therefore be a reliable source of in- datasets, respectively; those values differed by about formation for determining the onset and withdrawal of 17%, which is comparable to the differences between the rainy season in Panama. different Japanese rainfall datasets (Utsumi et al. 2008). Figure 2 compares the spatial patterns of climatolog- 5. Results ical annual mean precipitation among the four datasets. a. Onset and withdrawal during the seasonal cycle The spatial patterns derived from the MRI and TRMM datasets were similar (Figs. 2a,d): high precipitation in We compared onset and withdrawal dates and the the Caribbean coastal zone and in the provinces of differences of the seasonal cycle of pentad precipitation Chiriquí and Veraguas, and low precipitation in the at eight gauge stations in Panama (Fig. 1b). The eight western coastal zone of the Gulf of Panama from Los stations consisted of four pairs of gauge stations that Santos Province to Panama Province. However, large were located at a similar longitude on the Caribbean and differences were apparent in the eastern part of Panama, Paciﬁc sides of Panama. Table 3 provides information where the distribution of gauge stations is sparse about the eight gauge stations. (Fig. 1b). In addition, small-scale features of the spatial Figure 3a depicts the time series of pentad pre- distribution were apparent only in the MRI dataset. The cipitation at station David. The rainy season starts at spatial patterns apparent from the CRU, GPCP, and pentad 22 and terminates at pentad 69 (7–11 Decem- TRMM datasets were similar, but the patterns of rain- ber). A clear seasonal cycle of precipitation is apparent fall derived from the CRU and GPCP (Figs. 2b,c) in Fig. 3a. A sharp peak of pentad precipitation occurs in were more uniform. The spatial pattern in the UNESCO late May, and a broad peak is evident in September and dataset [Fig. 10 of UNESCO (2008)] resembles that of October. Similar seasonal cycles are apparent at stations the MRI dataset but differs from the MRI dataset with Los Santos and Amador, but with sharp maxima in respect to small-scale features. October and November (Figs. 3c,e). The break of the The MRI dataset showed high precipitation rates rainy season in July and August is a result of passage of 2 (.10 mm day 1) in two areas, Chiriquí Province and the the ITCZ to the north, a time of year referred to as western part of Colón Province (Fig. 2a), but the TRMM ‘‘Veranito (de San Juan)’’ [UNESCO 2008; in English, dataset did not (Fig. 2d). The dominant vegetation in the ‘‘the mid-summer drought’’ (Magaña et al. 1999)]. The coastal zone between the provinces of Coclé and Los rainy season at Amador persists for 47 pentads, from Santos is dry forest (Condit et al. 2010), and in that region pentad 23 to pentad 70 (21 April to 16 December). At the MRI dataset indicates that the rate of pre- Los Santos a late onset is apparent at pentad 27 (11–15 2 cipitation is less than 4 mm day 1.TheCRUand May), and withdrawal occurs at pentad 67. San Pedro GPCP datasets indicate that precipitation is high has a similar seasonal cycle, but with a single sharp peak 2 (.10 mm day 1) only in the eastern coastal zone of before withdrawal at pentad 2 (6–10 January) (Fig. 3f). Darién Province (Figs. 2b and 2c, respectively). In the Stations Garachine and Mulatupo have small seasonal UNESCO dataset [Fig. 10 of UNESCO (2008)], many variations of pentad precipitation and similar seasonal

Unauthenticated | Downloaded 09/30/21 06:48 PM UTC 1APRIL 2015 N A K A E G A W A E T A L . 2751

FIG. 3. Seasonal variations of climatological pentad precipitation (solid line) 6 1 standard deviation (gray shading). 2 The horizontal green line denotes the threshold value of 3 mm day 1. See Fig. 1b and Table 3 for detailed information on each station location.

Unauthenticated | Downloaded 09/30/21 06:48 PM UTC 2752 JOURNAL OF CLIMATE VOLUME 28

FIG. 4. Geographical distributions of the (a) onset and (b) withdrawal dates of the rainy season in Panama. The onset and withdrawal dates have been defined with the uniform threshold value method and a threshold value of 2 3 mm day 1. Areas with no dry season have pentad 1 for the onset date and pentad 73 for the withdrawal date. See also Fig. 5 for the areas. cycles (Figs. 3g,h). The onset and withdrawal dates at province of Bocas del Toro to the province of Colon Garachine occur at pentad 23 (21–25 April) and pentad (Figs. 3b,d). The earliest onset of the rainy season in 71 (17–21 December), respectively. The onset and pentads less than 16 occurs on the Pacific side of the withdrawal dates at Mulatupo occur during pentads 23 Cordillera de Talamanca Mountains, close to the na- and 70 (2–6 December), respectively. tional border with Costa Rica. The second earliest onset, At the two other gauge stations on the Caribbean in pentad 20 (6–10 April), occurs in Chiriquí Province, side, Seiyic and Cocle del Norte, precipitation exceeds the southern part of Veraguas Province, and the eastern 2 3 mm day 1 even during the period from January to part of Colon Province and western part of San Blas April (Figs. 3b,d), the indication being that there is no Province. The rainy season starts suddenly in pentad 21 dry season at these two gauge stations. The intense rainy in most of the eastern part of Panama, including San season at Seiyic continues from pentad 26 to pentad 71 Miguel Island, although it does so in a very confined area (6 May to 21 December), during which time the rate in the western part of Panama. The onset occurs at 2 of precipitation is about 9 mm day 1, with no distinct pentad 22 in Veraguas Province and the northwestern peak. At Coclé del Norte, the intense rainy season starts part of Darién Province. The latest onset, in pentad 26 in pentad 23 and terminates during pentad 3 (11–15 (6–10 May), occurs in the coastal zone of the Gulf of January). There is a local minimum of pentad pre- Panama in Los Santos Province, following the onset in cipitation from pentad 47 to pentad 57 (19 August–12 adjacent areas in pentads 23 to 25 (21 April–5 May). October). Peak pentad precipitation occurs during pentad Even in a country as small as Panama, the onset varies 65 (17–21 November). from pentads 16 to 26, a time interval of 55 days or about The heaviest rainfall in the late rainy season occurs at 2 months. Los Santos, Cocle del Norte, Amador, San Pedro, and Figure 4b shows the geographical distribution of the Garachine (Murphy et al. 2014);adistinctpeakofthe withdrawal dates of the rainy season. As mentioned in rainfall during the rainy season is not seen at Seiyic and the previous paragraph, there is no withdrawal date in Mulatupo. Veranito, the midsummer drought (Magaña the Caribbean coastal zone from the province of Bocas et al. 1999), is clearly apparent at David, Los Santos, del Toro to Colon Province because there is no dry and Amador in the Pacific coastal zone; three peaks of season. The earliest withdrawal of the rainy season rainfall with two midsummer droughts are apparent occurs during pentad 67 in the northern coastal zone of at Cocle del Norte and San Pedro in the Caribbean the Gulf of Panama in Los Santos Province. The coastal zone. withdrawal extends to the surrounding areas and reaches Veraguas Province and the region from the mouth b. Geographical distribution of onset and withdrawal ofthePanamaCanaltotheGulfofPanamáinpentad Figure 4a shows the geographical distribution of the 68 (2–6 December), when the withdrawal also occurs in onset dates of the rainy season in Panama. There is no Chiriquí Province. In the remainder of the western part dry season in the Caribbean coastal zone from the of Panama, the rainy season terminates on the Pacific

Unauthenticated | Downloaded 09/30/21 06:48 PM UTC 1APRIL 2015 N A K A E G A W A E T A L . 2753

c. Mesoscale circulations In the dry season, January to March, water vapor fluxes are divergent over Panama (Figs. 6a,b). Weakly convergent water vapor fluxes cover Panama in April (Fig. 6c) and migrate to the north in May (Fig. 6d). At this time, the rainy season starts over all of Panama (Fig. 3). A northeasterly water vapor flux, accompanied by a branch of the CLLJ (e.g., Amador 1998; Nakaegawa et al. 2014a) or the Panama jet (Xie et al. 2005; Amador et al. 2006), prevails from January to April and weakens in May. The water vapor flux turns toward the east in June (Fig. 6e) as it blows toward the ITCZ north of Panama. These changes in the source of water vapor in Panama have been identified with a backward water- vapor-tracking analysis (Durán-Quesada et al. 2010). FIG. 5. Geographical distributions of the length of the rainy The convergence is intensified by the southern migra- season in Panama (pentads). Areas with no dry season have 73- pentad rainy seasons. tion of the ITCZ in September (Fig. 6f). This convergent water vapor flux field is associated with SSTs greater than the threshold value of 26.58C, which is a diagnostic for active convective development (Gray 1968). These side during and after Pentad 69, except for the Cordil- high SSTs spread to the eastern Caribbean in May, cover lera de Talamanca mountain range near the national the entire Caribbean by August, and extend to the entire border with Costa Rica, where the latest withdrawal tropical North Atlantic by October (Fig. 7)(Taylor et al. occurs in pentad 73 (27–31 December). In contrast, the 2002; Wang and Enfield 2003). The seasonal cycles of withdrawal migrates from the Panama Canal to the the SSTs and the CLLJ are closely related. After the national border with Colombia in pentads 70 and 71. In second precipitation peak with the passage of the ITCZ the southern part of Darién Province near the national over Panama, the convergence field migrates to the border with Colombia, the latest withdrawal occurs in south in November and diminishes in December. Most pentad 1 (1–5 January). of Panama experiences a concurrent withdrawal of the The Caribbean coastal zone from the province of rainy season in December (Fig. 4b). Bocas del Toro to the province of Colon, except for the These mesoscale circulation patterns can explain the eastern part of Bocas del Toro Province and the onset and withdrawal of the rainy season in Panama on northern part of Veraguas Province, has the longest a country-wide scale, but they cannot account for the rainy season, the entire year (73 pentads), because details of the spatial features depicted in Figs. 3 and 4. there is no dry season. Areas with rainy seasons longer than 60 pentads are confined to the Cordillera de The earliest onset and the latest withdrawal should oc- Talamanca Mountains near the national border with cur on the Pacific side based on the northward and CostaRica(Fig. 5). In contrast, the shortest rainy southward migration, respectively, of the convergent season, less than 43 pentads, is found in the coastal water vapor fluxes; the second earliest onset occurs in zone of the Gulf of Panama in the provinces of Los Chiriquí Province and the southern part of Veraguas fi Santos, Herrera, and Cocle. The rainy season persists Province on the Paci c side, as well as in the eastern part for 43–46 pentads in adjacent areas. The remaining of Colon Province and the western part of San Blas areas have rainy seasons that last 47–52 pentads. Province on the Caribbean side. This discontinuous The geographic variability in the length of the rainy pattern reflects the low horizontal resolution of the JRA- season is larger than the variability of the onset and 55, which cannot represent the detailed spatial features withdrawal dates. The geographical distributions of the of topography and land–sea distributions that determine onset and withdrawal dates resemble each other (Fig. 4). the onset and withdrawal. However, we can infer a pos- The fact that the spatial correlation coefficient between sible cause of the detailed spatial features of the onset the two is 20.89 suggests that early onset dates often and withdrawal dates from the water vapor flux fields. correspond to late withdrawal dates, and vice versa. The For example, during January to March a topographically difference between the onset and withdrawal dates, or induced upward mass flux accompanied by a strong the length of the rainy season, therefore shows the water vapor flux produces a local-scale convergence in largest geographical variability. the Caribbean coastal zone from the province of Bocas

Unauthenticated | Downloaded 09/30/21 06:48 PM UTC 2754 JOURNAL OF CLIMATE VOLUME 28

21 21 FIG. 6. Climatological monthly means of mesoscale column total water vapor ﬂuxes (vector; kg kg ms )and 2 2 2 2 their convergences (shading; 10 5 kg kg 1 ms 1 m 1). The abbreviation P. in the top right of each panel denotes pentad.

Unauthenticated | Downloaded 09/30/21 06:48 PM UTC 1APRIL 2015 N A K A E G A W A E T A L . 2755

FIG.7.AsinFig. 6, but for SST (8C). del Toro to the province of Colon (Taylor and Alfaro Cordillera Central mountain range. The northeasterly 2005), the result being the absence of a dry season water vapor ﬂuxes tend to converge on the Caribbean (Figs. 6a,b). In addition, the coastal zone of the Gulf of side of the highlands because of the topographically Panama in Los Santos Province is surrounded by the induced upward mass ﬂux, which prohibits humid air

Unauthenticated | Downloaded 09/30/21 06:48 PM UTC 2756 JOURNAL OF CLIMATE VOLUME 28

FIG. 8. (a) Minimum of the actual number of acceptable observations in each grid in the 73 pentads during the 40 years. A value of 20 indicates that 50% of the records available out of 40 years were acceptable, and a value greater than 40 indicates that there were multiple stations in the same grid. (b) Normalized minima by dividing the numbers in Fig. 8a by the 73-pentad mean number of acceptable observations in that grid. from reaching the coastal zone of the Gulf of Panama in orography. The existence of dry season in San Blas may Los Santos Province. The rainy season can therefore be the result of orographic effects on precipitation since start only when cyclonic water vapor fluxes come from the Cordillera de Talamanca and Cordillera Central the south in May (Fig. 6d). Mountains in Bocas del Toro is much higher than the mountains in San Blas (Fig. 1a). d. SST The SST contrast on the Pacific side (Fig. 7h) is not as distinct during the withdrawal season (December) as Because of the temperature dependence of evapora- during the onset season because of the small upwelling tion from the sea surface, the geographical distribution of of cold water by the Panama jet. Relatively warm SSTs SST may partly account for the timing of the onset and off the coastal zone of the provinces of Chiriqui and withdrawal dates. On the Pacific side, the northeasterly Veraguas may be associated with withdrawal dates that wind over the Gulf of Panama, the Panama jet, becomes are later than withdrawal dates in the coastal zone of the strongest in March and April and induces a lowering of Gulf of Panama, where the SSTs are relatively cold. SST (Figs. 7b,c) only in the Gulf of Panama. Xie et al. However, the fact that there is no SST contrast on the (2005) have shown that the suppression of atmospheric Caribbean Sea side suggests that the SST difference convection to the south of 58N produces a local band of does not influence the withdrawal dates on the Carib- minimal precipitation within the ITCZ. This cooling may bean Sea side, and other factors must therefore be re- delay the onset dates around the Gulf of Panama. In sponsible for the differences in withdrawal dates on the contrast, SSTs off the coastal zones of the Pacific coast of Caribbean Sea side. western Panama (Chiriqui and Veraguas) are warmer than the SSTs off the coastal zones of the Caribbean side of western Panama (Bocas del Toro and Veraguas), 6. Discussion and the southeastern edge of the eastern North Pacific/ a. Spatiotemporal inhomogeneity in the availability western Hemisphere warm pool (Xie et al. 2005; Wang of observation data and Enfield 2001; Wang and Enfield 2003) may advance the onset of the rainy season. On the Caribbean Sea side, The availability of observation data for the gridded only San Blas Province has a dry season. The SST is lower dataset varied both temporally and spatially. Systematic in March off the coastal zone of San Blas than off Bocas variations of data availability can affect the spatial dis- del Toro, but this lower SST is unlikely to delay the onset. tribution of the climatological onset and withdrawal The SST off San Blas is higher than 26.58C(Fig. 7b)andis dates. First, we counted the actual number of acceptable therefore sufficiently high that the active convection and observations in each grid for each pentad in the 40 years strong water vapor fluxes are similar in the two regions and then determined the minimum of the 40-yr mean (Fig. 6b). The water vapor fluxes in JRA-55 show di- number of acceptable observations during the 73 pen- vergence fields over Panama mainly due to the low hor- tads for each grid (Fig. 8a). This figure reflects the length izontal resolution and insufficient representation of the of the observation period. If the observation period was

Unauthenticated | Downloaded 09/30/21 06:48 PM UTC 1APRIL 2015 N A K A E G A W A E T A L . 2757 short, the minimum of acceptable observations was western part of Panama. Figure 9 shows the geo- small. In several grids the minimum was less than 4. graphical distributions of onset and withdrawal dates as These grids were randomly distributed and not clustered a function of threshold values. The qualitative features together; there was always another station with a large of the geographical distributions are unaffected by the minimum number of acceptable observations within choice of different threshold values. However, as the 20 km of the stations in these grids (Table 1). In some threshold value increases, the onset is delayed, and grids the minimum number of acceptable observations the withdrawal date advances over the entire country, exceeded 40 because there was more than one station in especially in the coastal zones of the provinces of Los the grid. Each number in Fig. 8a was normalized by di- Santos and Herrera. This result can be understood by viding it by the 73-pentad mean number of acceptable increasing the threshold value in Fig. 3. At the same observations in that grid. The normalized minima at time, areas with no dry season become smaller, because most of the grids with low minima in Fig. 8a were about a large threshold value can produce a dry season. The 1.0, except for three grids (Fig. 8b): two in the Caribbean geographical contrast in the onset dates seems to de- coastal zone near the border with Costa Rica, Bocas del crease because of the disappearance of areas with no dry Toro and Changuinola 1, and one on the Pacific side of the season, whereas the differences of the withdrawal dates Panama Canal, Howard Z.C. The incompleteness of the seems to increase because of the appearance of areas data record in these three grids was due to their stations with withdrawal dates earlier than pentad 50 (3–7 Sep- being relatively new stations, to termination of the station tember). Most of these features are apparent in the with a record shorter than 40 years, and to missing data. analysis by Marengo et al. (2001) of the sensitivity to the Our gridding method weighted and interpolated the pre- threshold value of the onset and withdrawal dates of cipitation data based on the number of observations and the rainy season in the Brazilian Amazon basin. This was probably not affected by the inhomogeneity displayed similarity suggests that sensitivity to the threshold value is in Fig. 8. A comparison of the geographical distribution of a general characteristic of the onset and withdrawal dates. the onset and withdrawal dates in Fig. 4 with the spatio- The pioneering work of Gramzow and Henry (1972) temporal inhomogeneity of the available observation data concerning the onset and withdrawal of the rainy season in Fig. 8 does not reveal any particular geographical pat- in Central America is generally in good agreement with tern in Fig. 4 that corresponds to the station locations with the results in Figs. 9e and 9f. They used observation data low minima in Fig. 8. We conclude that the spatiotem- at 61 stations for all of Central America and a threshold 2 poral inhomogeneity in our gridded data did not influence value of 5 mm day 1 to depict the geographical distri- the geographical distribution of the calculated climato- butions of the onset and withdrawal dates. On the one logical onset and withdrawal dates. hand, features similar to our results are apparent in their The uncertainties in the onset and withdrawal dates results: early onset in the Caribbean coastal zone, east- are large in areas where the observation stations were ward migration of the onset from the national border sparsely distributed, such as Bocas del Toro and Darien, with Costa Rica to the Panama Canal, and late onset in relative to areas with dense distributions of observation the coastal zone of the Gulf of Panama in Los Santos stations. In the spatial interpolation scheme described in Province. On the other hand, the onset migrates from section 2a, weighting depended on which side of high the Panama Canal to the east in Gramzow and Henry ridges the station was located. Mulaputo is the only station (1972), whereas it occurs in mid-April in most of the in San Blas, and the geographical distribution of the onset eastern part of Panama in Fig. 9e. The withdrawal occurs and withdrawal dates in San Blas was strongly influenced earlier in the coastal zone of the Pacific than of the by the data from Mulaputo. The data from Garachine had Caribbean in both results; the withdrawal migrates a similar influence on estimates for Darien. Therefore, the westward in Gramzow and Henry (1972) but occurs actual geographical distribution of onset and withdrawal concurrently in Fig. 9f. In general, the onset (with- dates may be more inhomogeneous. drawal) dates in Fig. 9e (Fig. 9f) are earlier (later) by about 15 days than those of Gramzow and Henry (1972). b. Uniform threshold value method with different These differences are due to the small number of station values data available in Panama and the subjective inter- There are logical reasons for varying the threshold polation of the onset and withdrawal dates in Gramzow value used to define the rainy season. For agricultural, and Henry (1972). hydrological, and/or ecological purposes, 4, 5, or The station-based work of Alfaro (2002) with 2 2 6 mm day 1 might work better as the threshold value. a threshold value of 5 mm day 1 showed spatially dis- Enfield and Alfaro (1999), Alfaro and Cid (1999), and crete geographical distributions of the onset and with- 2 Alfaro (2002) used 5 mm day 1 for Costa Rica and the drawal dates that were similar to those depicted in

Unauthenticated | Downloaded 09/30/21 06:48 PM UTC 2758 JOURNAL OF CLIMATE VOLUME 28

FIG. 9. Comparison of the geographical distributions of the (left) onset and (right) withdrawal dates with different 2 threshold values: (a),(b) 3, (c),(d) 4, (e),(f) 5, and (g),(h) 6 mm day 1. Note that (a) and (b) are the same as in Figs. 4a and 4b, but with a different color scale.

Figs. 9e and 9f, although the dates in Alfaro (2002) and the western part of the Caribbean coastal zone of the present study differ by one or two pentads. Because Panama. the Alfaro (2002) analysis involved only one station c. Distributed threshold value method on the Caribbean side near the Panama Canal, that analysis indicated that all areas of Panama experience It would be interesting to know how sensitive the re- a dry season. Therefore, Figs. 9e and 9f are the ﬁrst to sults are to the choice of the uniform or distributed demonstrate that an area with no dry season exists in threshold value methods. The choice between the two

Unauthenticated | Downloaded 09/30/21 06:48 PM UTC 1APRIL 2015 N A K A E G A W A E T A L . 2759

FIG. 10. As in Fig. 4, but calculated with the distributed threshold value method. The distributed threshold values are seen in Fig. 2a. methods depends on the purpose of the study. The dis- Panama in Los Santos Province, whereas only the distributed threshold value method may be useful for tributed threshold value method shows that the latest qualitative comparative studies because it uses anoma- onset occurs simultaneously in the Cordillera Central lous pentad precipitation data from climatological an- Mountains in Veraguas. nual mean precipitation values. Therefore, onset and Based on the distributed threshold method, the earliest withdrawal dates always exist in most places. withdrawal occurs during pentad 63 (7–11 November) in Based on the distributed threshold method, the rainy two areas isolated from surrounding areas: the Cordillera season starts in the eastern part of Panama during pen- de Talamanca Mountains close to the national border tad 22 from the eastern part of Colon Province and the with Costa Rica in Chiriquí Province and the Cordillera eastern inland part of Darien Province. The rainy season Central Mountains in Veraguas Province (Fig. 10b). The expands to the surrounding areas during pentad 23 and withdrawal of the rainy season migrates from Costa Rica occurs concurrently in the eastern part of the provinces during pentad 65 to Los Santos Province and through the of San Blas and Darien. The remaining eastern part of Pacific side of Panama during pentad 66 (22–26 Novem- Panama experiences the onset during pentad 24 (26–30 ber) to the coastal zone of the Gulf of Panama and West April; Fig. 10a). In the western part of Panama, the rainy Panama Province. Three remote areas simultaneously season starts from the Caribbean coastal zone near the experience withdrawal during pentad 68: the coastal zone western part of Colon Province during pentad 22, ex- of the Gulf of Panama in Los Santos Province, the tends to the south, and then stalls. In the western part of provinces of Darién and San Blas, and the Caribbean ChiriquíProvinceonthePacific side, the rainy season coastal zone. The latest withdrawal occurs in the eastern starts during pentad 24 and extends to the north on the part of Colon Province during pentad 69. Both of the Caribbean side. The onset occurs simultaneously in most areas with the earliest and latest withdrawals of the rainy of the remaining areas during pentad 26. The latest onset season are found in the Caribbean coastal zone. occurs during pentad 27 in two remote areas: the Cordillera Table 4 compares the onset and withdrawal dates Central mountains in Veraguas Province and the coastal determined with the two different methods at eight zone of the Gulf of Panama in Los Santos Province. gauge stations. Onset and withdrawal dates estimated The onset dates estimated with the distributed thresh- with the distributed threshold value method were gen- old method are concentrated during pentads 22–27, erally within the range of values estimated with the whereas the onset dates based on the uniform threshold uniform threshold value method. In wet areas, the onset method occur during pentads 16–26. The earliest onset and withdrawal dates estimated with the distributed occurs in different areas: the Pacific side of the Cordillera threshold value method were closer to those estimated Talamanca Mountains close to the national border by using the uniform threshold value method with 2 with Costa Rica based on the uniform threshold value a threshold value of 6 mm day 1; in dry areas, these method and in Colón Province and the eastern inland dates were closer to those estimated by using the uni- part of Darién Province based on the distributed form threshold value method with a threshold value of 2 threshold value method. Both methods show that the 3 mm day 1. These comparisons reveal that the different latest onset occurs in the coastal zone of the Gulf of methods provided different estimates of the onset and

Unauthenticated | Downloaded 09/30/21 06:48 PM UTC 2760 JOURNAL OF CLIMATE VOLUME 28

TABLE 4. Comparison of the onset and withdrawal dates of the rainy season estimated with different methods.

Onset date (pentad) Withdrawal date (pentad) 21 21 Uniform threshold (mm day ) Distributed Uniform threshold (mm day ) Distributed Method 3456threshold 3456 threshold David 22 23 24 24 25 69 69 67 67 66 Seiyic — — 20 20 26 — — 9 9 69 Los Santos 26 28 29 54 26 67 65 65 62 67 Cocle del — — 17 17 23 — — 14 5 73 Norte Amador 23 23 26 27 23 70 69 67 67 69 San Pedro 20 22 23 23 26 5 3 71 71 70 Garachine 23 25 27 57 25 71 70 66 64 70 Mulatupo 23 24 26 26 23 70 67 66 66 67

withdrawal dates, although the different methods re- Western Hemisphere warm pool and warms in the bo- tained the fundamental features of the onset and with- real summer after the eastern North Pacific warm pool drawal dates. A comparison between the two different does so in the boreal spring (Wang and Enfield 2003). In methods has not yet been made in other regions; de- the spring, the fact that the warm pool in the eastern termining whether these conclusions are general or not North Pacific (ENP) is warmer than the Atlantic warm will require additional comparisons in different regions. pool intensifies the CLLJ, whereas cooler conditions in the ENP during the summer weakens it (Enfield and d. Low-horizontal-resolution atmospheric dataset Alfaro 1999). The CLLJ is closely related to the seasonal SST contrast between the two pools through the activity The JRA-55, with a grid spacing of 55 km, does not of the North Atlantic subtropical high (Wang 2007). The have a horizontal resolution sufficient to capture the eastern North Pacific warm pool influences midsummer small-scale features of atmospheric circulation. How- droughts (Magaña et al. 1999) by changing the intensity ever, its horizontal resolution is the highest of the global of convective activity. Thus synoptic-scale atmospheric reanalysis datasets. A higher horizontal resolution data- circulation influences the onset and withdrawal dates set requires understanding the dynamical causes of the through its effect on mesoscale atmospheric circulation. detailed features of the geographical distributions, such As mentioned above, the CLLJ is a unique feature of as we inferred from the water vapor fluxes. Conversely, Central America and the Caribbean. The semiannual the low horizontal resolution atmospheric datasets nor- cycles of the CLLJ and accompanying water vapor mally used for this type of analysis cannot reveal the fluxes evidence a local minimum during the onset period detailed geographical features of the onset and within April and May when the North Atlantic subtropical drawal depicted in Figs. 4, 9,and10. high is weak. The CLLJ and accompanying water vapor We also examined the water vapor fluxes and con- fluxes are intensified during the withdrawal period. vergence fields in the North American Regional Re- These relationships suggest that the seasonal variations analysis, which has a high horizontal resolution of 32 km of the CLLJ influence at least the onset dates and per- (Mesinger et al. 2006). However, we could not obtain haps weakly influence the withdrawal dates. reasonable seasonal cycles or migrations, perhaps be- The eastern North Pacific warm pool reaches a maxi- cause the regional lateral boundaries were too close to mum temperature of .298C during the period from Panama. April to June and activates deep atmospheric convec- tions that trigger the onset of the rainy season in Central e. Synoptic-scale circulations America. When the temperature of the eastern North The mesoscale circulations and the local SST distri- Pacific warm pool drops below 288C, the midsummer butions presented above are embedded in large drought occurs in the ITCZ, although 288C is higher regional-scale or synoptic-scale atmospheric circulation. than the threshold for active atmospheric convection. The seasonal cycle of SST in the Western Hemisphere This suggests that synoptic-scale circulation such as the warm pool is a key to defining the climatological fea- convergence of water vapor fluxes maintains the deep tures of synoptic-scale atmospheric circulation in Cen- convection of the ITCZ in addition to high SST (Magaña tral America and the Caribbean (Amador et al. 2006). et al. 1999). These processes determine the timing on The Atlantic warm pool is the eastern part of the a monthly time scale (April to May) of the onset of the

Unauthenticated | Downloaded 09/30/21 06:48 PM UTC 1APRIL 2015 N A K A E G A W A E T A L . 2761 rainy season on a synoptic scale, but the detailed geo- pentad 22 in most of the western part. The onset moves graphical features in Figs. 4, 9, and 10 suggest the im- toward and finally reaches Los Santos Province during portance of mesoscale processes. Subtropical cold pentad 26. surges or Nortes bring synoptic-scale stratiform pre- The rainy season terminates in Los Santos Province cipitation to Central America during the withdrawal during pentad 67 (Fig. 4b). The withdrawal expands to period corresponding to the boreal winter (Cortez the surrounding areas to the east and west and reaches 2000). This type of precipitation can delay the with- Veraguas Province and the Panama Canal during pen- drawals until December and January, as seen in Figs. 4, tad 68, when it occurs simultaneously in Chiriquí Prov- 9, and 10. However, subtropical cold surges or Nortes ince. The withdrawal occurs during and after pentad 69 almost never reach Panama and persist only a week or in the remaining western part of Panama on the Pacific less (Garreaud 2001). Such phenomena can modulate side, whereas it migrates from the Panama Canal to the the withdrawal dates on an interannual time-scale, but national border with Colombia by pentad 71. The latest not on a climatological mean time scale. withdrawal occurs in the Cordillera de Talamanca Mountains during pentad 73 and in the southern part of Darien Province during pentad 1. The length of the rainy 7. Concluding summary season varies from 40 pentads in Los Santos to 73 pen- In the present study we investigated the onset and tads in the western part of the Caribbean coastal zone, withdrawal dates of the rainy season in Panama. First, where there is no dry season (Fig. 5). we used ground precipitation observations to develop Although the low horizontal resolution of the JRA-55 gridded daily precipitation datasets with a high hori- cannot directly reveal the cause of the high variability of zontal resolution of 0.058 to make it possible to analyze the onset and withdrawal dates and of the length of the the geographical distributions of the onset and with- rainy season, the water vapor fluxes (Fig. 6) and topog- drawal dates. raphy (Fig. 1) suggest dynamical causes, such as topo- The country of Panama is oriented parallel to latitude graphically induced upward mass fluxes accompanied by lines, and onset and withdrawal dates could therefore strong water vapor fluxes. Differences of SST on the adhere to a simple geographical pattern, with the onset Pacific side seem to be associated with the timing of dates moving from south to north and the withdrawal the onset and withdrawal, but this is not the case on the dates from north to south concurrently with the migra- Caribbean Sea side. tion of the ITCZ, a pattern evident in other regions and As we have discussed, the optimal choice of the thresh- countries. However, the geographical distributions old value for the uniform threshold value method depends showed very complicated features in Panama. on the purpose. A similar conclusion applies to the choice The rainy season starts in pentads 22 and 23 at four of between the uniform threshold and distributed threshold the eight gauge stations listed in Table 3 and terminates methods. Figures 9 and 10 may facilitate that choice. in pentads 69–71 at the same four gauge stations (Fig. 3). The uncertainty in precipitation datasets apparent in The first onset and latest withdrawal occur at San Pedro Fig. 2 creates uncertainty in the estimated onset and in pentads 20 and 5, respectively, whereas the latest withdrawal dates. There is need for creation of more onset and the first withdrawal occur at Los Santos in reliable gridded daily precipitation datasets by com- pentads 26 and 67, respectively. There is no dry season at bining satellite and ground observations to cover areas the two gauge stations at Seiyic and Cocle del Norte in where there are no ground observation data available, the Caribbean coastal zone if we use a uniform threshold such as Darien Province and the Caribbean coastal zone. 2 value of 3 mm day 1. During the rainy season, a single In this study, we were not able to clarify the causes of the peak of rainfall or bimodal peaks separated by Veranito high variability of the onset and withdrawal dates in are apparent at each gauge station. Panama. A regional climate simulation with high hori- The geographical distribution of the onset dates of the zontal resolution is needed to facilitate understanding of rainy season show a wide range of about 2 months from this variability with respect to atmospheric dynamics pentads 16 to 26 (Fig. 4a). The earliest onset of the rainy and local SST effects. season occurs in the Cordillera de Talamanca Moun- The present study focused on climatological onset and tains in pentads less than 16. The second-earliest onset withdrawal dates. Atmospheric circulation over Central occurs during pentad 20 in Chiriquí Province, the America and the Caribbean is modulated by the in- southern part of Veraguas Province, the eastern part of terannual variability of atmospheric circulation over the Colon Province, and the western part of San Blas Pacific and Atlantic (Ropelewski and Halpert 1987; Province. The rainy season starts suddenly during pen- Enfield and Alfaro 1999). The CLLJ influences the in- tad 21 in most of the eastern part of Panama and in terannual variability of wet and dry years (Méndez and

Unauthenticated | Downloaded 09/30/21 06:48 PM UTC 2762 JOURNAL OF CLIMATE VOLUME 28

Magaña 2010), and the water vapor flux accompanied by Durán-Quesada, A. M., L. Gimeno, J. A. Amador, and R. Nieto, the CLLJ is projected to be intensified in the future 2010: Moisture sources for Central America: Identification climate (Nakaegawa et al. 2014a). This topic has been of moisture sources using a Lagrangian analysis technique. J. Geophys. Res., 115, D05103, doi:10.1029/2009JD012455. investigated in other regions (e.g., Marengo et al. 2001). Enfield, D. B., and E. J. Alfaro, 1999: The dependence of Carib- The interannual variability of the onset and withdrawal bean rainfall on the interaction of the tropical Atlantic and dates of the rainy season in Panama is therefore an im- Pacific Oceans. J. Climate, 12, 2093–2103, doi:10.1175/ portant research topic. Future changes in the onset and 1520-0442(1999)012,2093:TDOCRO.2.0.CO;2. withdrawal dates are also important because of their ETESA, cited 2014: National network of meteorological and hydrological stations. Empresa de Transmisión Eléctrica, S.A. [Avail- potential effect on anthropogenic activities such as wa- able online at http://www.hidromet.com.pa/red_nacional.php.] ter resource and flood managements (e.g., Fábrega et al. Fábrega, J., and Coauthors, 2013: Hydroclimate projections for 2013; Nakaegawa et al. 2014b,c). The present study has Panama in the late 21st Century. Hydrol. Res. Lett., 7, 23–29, provided information that will facilitate these further doi:10.3178/hrl.7.23. studies. Garreaud, R. D., 2001: Subtropical cold surges: Regional aspects and global distribution. Int. J. Climatol., 21, 1181–1197, doi:10.1002/ joc.687. Acknowledgments. The present study was initiated by Gramzow, R. H., and W. K. Henry, 1972: The rainy pentads of the technical cooperation scheme of the Japan In- Central America. J. Appl. Meteor., 11, 637–642, doi:10.1175/ , . ternational Cooperation Agency. The precipitation 1520-0450(1972)011 0637:TRPOCA 2.0.CO;2. Gray, W. M., 1968: Global view of the origin of tropical distur- dataset was provided by ETESA. The authors thank bances and storms. Mon. Wea. Rev., 96, 669–700, doi:10.1175/ Professors Fábrega and Pinzón at the Technological 1520-0493(1968)096,0669:GVOTOO.2.0.CO;2. University of Panama for their kind support during TN’s Harris, I., P. D. Jones, T. J. Osborn, and D. H. Lister, 2014: Updated stay in Panama and Professor Nakayama of the Kitami high-resolution grids of monthly climatic observations—The Institute of Technology, who paved the way for this re- CRU TS3.10 dataset. Int. J. Climatol., 34, 623–642, doi:10.1002/ joc.3711. search. The authors wish to acknowledge Dr. Rosana Hastenrath, S., 1978: On modes of tropical circulation and climate Nieto-Ferreira as editor and three anonymous reviewers anomalies. J. Atmos. Sci., 35, 2222–2231, doi:10.1175/ for their constructive and meaningful comments on this 1520-0469(1978)035,2222:OMOTCA.2.0.CO;2. manuscript. This research was conducted as part of an Hijmans, R. J., S. E. Cameron, J. L. Parra, P. G. Jones, and MRI project entitled ‘‘Research on the mechanisms and A. Jarvis, 2005: Very high resolution interpolated climate surfaces for global land areas. Int. J. Climatol., 25, 1965–1978, the predictability of extreme climate events.’’ TN was doi:10.1002/joc.1276. financially supported by KAKENHI Grant 23226012 Horel, J. D., A. N. Hahmann, and J. E. Geisler, 1989: An in- from the Japan Society for the Promotion of Science, and vestigation of the annual cycle of convective activity over the OA and KK by the SOUSEI program from the Ministry tropical Americas. J. Climate, 2, 1388–1403, doi:10.1175/ of Education, Culture, Sports, and Science, Japan. 1520-0442(1989)002,1388:AIOTAC.2.0.CO;2. Huffman, G. J., R. F. Adler, M. M. Morrissey, D. T. Bolvin, S. Curtis, R. Joyce, B. McGavock, and J. Susskind, 2001: Global precipitation at one-degree daily resolution from REFERENCES multisatellite observations. J. Hydrometeor., 2, 36–50, Alfaro, E. J., 2002: Some characteristics of the annual precip- doi:10.1175/1525-7541(2001)002,0036:GPAODD.2.0.CO;2. itation cycle in Central America and their relationships with ——, and Coauthors, 2007: The TRMM multi-satellite pre- its surrounding tropical oceans. Tóp. Meteor. Oceanogr., 9, cipitation analysis: Quasi-global, multiyear, combined-sensor 88–103. precipitation estimates at fine scalea. J. Hydrometeor., 8, 38– ——, and L. Cid, 1999: Análisis de las anomalías en el inicio y el 55, doi:10.1175/JHM560.1. término de la estación lluviosa en Centroamérica y su relación IMD, 1943: Climatological atlas for airmen. India Meteorological con los océanos Pacífico y Atlántico. Tóp. Meteor. Oceanogr., Department, 100 pp. 6, 1–13. JMA, 2013: Introduction to meteorological observation statistics Amador, J. A., 1998: A climatic feature of the tropical Americas: (in Japanese). Guidelines for Meteorological Observation The trade wind easterly jet. Top. Meteor. Oceanogr., 5, 91–102. Statistics, Japan Meteorological Agency, 126 pp. [Available ——, 2008: The Intra-Americas Seas low-level jet (IALLJ): online at http://www.data.jma.go.jp/obd/stats/data/kaisetu/ Overview and future research. Ann. N. Y. Acad. Sci., 1146, index.html.] 153–188, doi:10.1196/annals.1446.012. ——, cited 2014: Historical record of the onset and withdrawal date ——, E. J. Alfaro, O. G. Lizano, and V. O. Magaña, 2006: Atmo- of the rainy season, Baiu, in Japan (in Japanese). [Available spheric forcing in the eastern tropical Pacific: A review. Prog. online at http://www.data.jma.go.jp/fcd/yoho/baiu/index.html.] Oceanogr., 69, 101–142, doi:10.1016/j.pocean.2006.03.007. Kamiguchi, K., O. Arakawa, A. Kitoh, A. Yatagai, A. Hamada, and Condit, R., R. Pérez, and N. Daguerre, 2010: Forest of Panama and N. Yasutomi, 2010: Development of APHRO_JP, the first Costa Rica. Trees of Panama and Costa Rica, Princeton Uni- Japanese high-resolution daily precipitation product for more versity Press, 11–16. than 100 years. Hydrol. Res. Lett., 4, 60–64, doi:10.3178/hrl.4.60. Cortez, M., 2000: Variaciones intraestacionales de la actividad con- Klein Tank, A. M. G., and Coauthors, 2002: Daily dataset of 20th- vectiva en México y América Central. Atmósfera, 13, 95–108. century surface air temperature and precipitation series for

Unauthenticated | Downloaded 09/30/21 06:48 PM UTC 1APRIL 2015 N A K A E G A W A E T A L . 2763

the European Climate Assessment. Int. J. Climatol., 22, 1441– for sea surface temperature. J. Climate, 20, 5473–5496, 1453, doi:10.1002/joc.773. doi:10.1175/2007JCLI1824.1. Kobayashi, S., and Coauthors, 2015: The JRA-55 Reanalysis: Ropelewski, C. F., and M. S. Halpert, 1987: Global and regional General specifications and basic characteristics. J. Meteor. scale precipitation patterns associated with the El Niño/ Soc. Japan, doi:10.2151/jmsj.2015-001, in press. Southern Oscillation. Mon. Wea. Rev., 115, 1606–1626, Liebmann, B., S. J. Camargo, A. Seth, J. A. Marengo, L. M. V. doi:10.1175/1520-0493(1987)115,1606:GARSPP.2.0.CO;2. Carvalho, D. Allured, R. Fu, and C. S. Vera, 2007: Onset and Schaake, J., 2004: Application of PRISM climatologies for hydrologic end of the rainy season in South America in observations and modeling and forecasting in the western U.S. 18th Conf. on the ECHAM 4.5 atmospheric general circulation model. Hydrology, Amer. Meteor. Soc., Seattle, WA, 5.3. [Available J. Climate, 20, 2037–2050, doi:10.1175/JCLI4122.1. online at https://ams.confex.com/ams/84Annual/techprogram/ MAFF, 2003: Diversity of irrigation by hydrological conditions, paper_72159.htm.] etc. The global diversity of irrigation for achieving sustainable Sombroek, W., 2001: Spatial and temporal patterns of Amazon water use and forming healthy water cycles, Japanese Institute rainfall: Consequences for the planning of agricultural occu- of Irrigation and Drainage, 15–32. [Available online at http:// pation and the protection of primary forests. Ambio, 30, 388– www.maff.go.jp/j/nousin/keityo/mizu_sigen/pdf/panf01_e.pdf.] 396, doi:10.1579/0044-7447-30.7.388. Magaña, V., J. A. Amador, and S. Medina, 1999: The mid- Taylor, M. A., and E. Alfaro, 2005: Climate of Central America summer drought over Mexico and Central America. J. Cli- and the Caribbean. The Encyclopedia of World Climatology, mate, 12, 1577–1588, doi:10.1175/1520-0442(1999)012,1577: J. Oliver, Ed., Springer Press, 183–189, doi:10.1007/ TMDOMA.2.0.CO;2. 1-4020-3266-8. Marengo, J. A., B. Liebmann, V. E. Kousky, N. P. Filizola, and I. C. ——, D. B. Enfield, and A. A. Chen, 2002: The influence of the Wainer, 2001: Onset and end of the rainy season in the Bra- tropical Atlantic vs. the tropical Pacific on Caribbean rainfall. zilian Amazon Basin. J. Climate, 14, 833–852, doi:10.1175/ J. Geophys. Res., 107, 3127, doi:10.1029/2001JC001097. 1520-0442(2001)014,0833:OAEOTR.2.0.CO;2. UNESCO, 2008: Balance hídrico superficial de Panamá, período Matsumoto, J., 1997: Seasonal transition of summer rainy season 1971–2002. Documento Técnicos del PHI-LAC 9, 133 pp. over Indochina and adjacent monsoon region. Adv. Atmos. [Available online at unesdoc.unesco.org/images/0015/001591/ Sci., 14, 231–245, doi:10.1007/s00376-997-0022-0. 159103s.pdf.] Méndez, M., and V. Magaña, 2010: Regional aspects of prolonged Utsumi, N., S. Kanae, H. Kim, S. Seto, T. Oki, T. Nitta, and meteorological droughts over Mexico and Central America. Y. Hirabayashi, 2008: Importance of wind-induced undercatch J. Climate, 23, 1175–1188, doi:10.1175/2009JCLI3080.1. adjustment in a gauge-based analysis of daily precipitation Mesinger, F., and Coauthors, 2006: North American Regional over Japan. Hydrol. Res. Lett., 2, 47–51, doi:10.3178/hrl.2.47. Reanalysis. Bull. Amer. Meteor. Soc., 87, 343–360, doi:10.1175/ Wang, B., and LinHo, 2002: Rainy season of the Asian–Pacific BAMS-87-3-343. summer monsoon. J. Climate, 15, 386–398, doi:10.1175/ Mitchell, T. P., and J. M. Wallace, 1992: The annual cycle in equatorial 1520-0442(2002)015,0386:RSOTAP.2.0.CO;2. convection and sea surface temperature. J. Climate, 5, 1140–1156, Wang, C., 2007: Variability of the Caribbean low-level jet and its doi:10.1175/1520-0442(1992)005,1140:TACIEC.2.0.CO;2. relations to climate. Climate Dyn., 29, 411–422, doi:10.1007/ Murphy, M. J., K. P. Georgakakos, and E. Shamir, 2014: Clima- s00382-007-0243-z. tological analysis of December rainfall in the Panama Canal ——, and D. B. Enfield, 2001: The tropical Western Hemisphere watershed. Int. J. Climatol., 34, 403–415, doi:10.1002/joc.3694. warm pool. Geophys. Res. Lett., 28, 1635–1638, doi:10.1029/ Nakaegawa, T., A. Kitoh, Y. Ishizaki, S. Kusunoki, and H. Murakami, 2000GL011763. 2014a: Caribbean low-level jets and accompanying moisture ——, and ——, 2003: A further study of the tropical Western fluxes in a global warming climate projected with CMIP3 multi- Hemisphere warm pool. J. Climate, 16, 1476–1493, doi:10.1175/ model ensemble and fine-mesh atmospheric general circulation 1520-0442-16.10.1476. models. Int. J. Climatol., 34, 964–977, doi:10.1002/joc.3733. Willmott, C. J., C. M. Rowe, and Y. Mintz, 1985: Climatology of the ——, ——, S. Kusunoki, H. Murakami, and O. Arakawa, 2014b: terrestrial seasonal water cycle. J. Climatol., 5, 589–606, Hydroclimate change over Central America and the Ca- doi:10.1002/joc.3370050602. ribbean in a global warming climate projected with 20-km Xie, S.-P., H. Xu, W. S. Kessler, and M. Nonaka, 2005: Air–sea and 60-km mesh MRI atmospheric general circulation interaction over the eastern Pacific warm pool: Gap winds, models. Pap. Meteor. Geophys., 65, 15–33, doi:10.2467/ thermocline dome, and atmospheric convection. J. Climate, mripapers.65.15. 18, 5–20, doi:10.1175/JCLI-3249.1. ——, ——, H. Murakami, and S. Kusunoki, 2014c: Annual maxi- Yatagai, A., K. Kamiguchi, O. Arakawa, A. Hamada, N. Yasutomi, mum 5-day rainfall total and maximum number of consecutive and A. Kitoh, 2012: APHRODITE: Constructing a long-term dry days over Central America and the Caribbean in the late daily gridded precipitation dataset for Asia based on a dense 21st century projected by an atmospheric general circulation network of rain gauges. Bull. Amer. Meteor. Soc., 93, 1401– model with three different horizontal resolutions. Theor. 1415, doi:10.1175/BAMS-D-11-00122.1. Appl. Climatol., 116, 155–168, doi:10.1007/s00704-013-0934-9. Yavitt, J. B., S. J. Wright, and R. K. Wieder, 2004: Seasonal drought Ogallo, L. A., 1989: The spatial and temporal patterns of the East and dry season irrigation influence leaf litter nutrients and soil African seasonal rainfall derived from principal component enzymes in a moist, lowland forest in Panama. Austral Ecol., analysis. Int. J. Climatol., 9, 145–167, doi:10.1002/joc.3370090204. 29, 177–188, doi:10.1111/j.1442-9993.2004.01334.x. Reynolds, R. W., T. M. Smith, C. Liu, D. B. Chelton, K. S. Casey, Zhang, S., and B. Wang, 2008: Global summer monsoon rainy and M. G. Schlax, 2007: Daily high-resolution-blended analyses seasons. Int. J. Climatol., 28, 1563–1578, doi:10.1002/joc.1659.

Unauthenticated | Downloaded 09/30/21 06:48 PM UTC