This study focuses on track and intensity changes of three tropical cyclones that, during the season of 2006, developed in the eastern North Pacific basin and made landfall over northwestern Mexico. Observational datasets, including satellite and radar imagery and a rain gauge network, are used to document regional-scale structures. Additionally, gridded fields are applied to determine the large-scale environment. John made landfall as a category-2 hurricane on the Saffir–Simpson scale and moved along the Baja California Peninsula during more than 40 h, resulting in total rainfall of up to 506 mm. The largest accumulations were located over mountains and set new records with respect to daily rates from the 1969–2005 period. Later in the season, Lane and Paul made landfall over the mainland and brought moderate rainfall over the coastal plains. Lane became a category-3 hurricane and was the third strongest hurricane to make landfall since 1969. In contrast, Paul followed a recurving track to reach the coastline as a weakening tropical depression. Strong wind shear, associated with a midlatitude trough, is found to be related to the intensity change. Examination of the official forecasts reveals that first inland positions were predicted several days before the actual landfall events. An assessment of the forecasts issued 1–3 days prior to landfall shows large track errors associated with some of the above tropical cyclones and this resulted in a westward bias. It is suggested that the track errors are due to an inadequate representation of the large-scale environment that steered the tropical cyclones.
The west coast of North America routinely experiences landfalling tropical cyclones (TCs) from the eastern North Pacific basin, located east of 140°W. During recent decades, several studies have focused on TCs that made landfall over the coastal areas of Mexico. Serra (1971) analyzed track information from the period 1921–69 to document that TCs crossing onshore had maximum frequency from August through October. Jáuregui (2003) found that, between 1951 and 2000, 54% of the basin landfalls occurred over the northwest coast of the country. Some of these TCs continued moving northward and, eventually, they influenced the weather conditions in the southwestern United States (Smith 1986; Garza 1999).
Northwestern Mexico is defined as the geographical region located north of 20°N and west of 104°W (Fig. 1), including Nayarit, Sinaloa, and Sonora on the mainland as well as the Baja California Peninsula. In 2005, Mexico had 103 million inhabitants and 10% of them lived in the northwest (INEGI 2011b). Table 1 provides the top five communities in Sinaloa and Baja California Sur (BCS, south of 28°N) and their geographical locations are shown in Fig. 1. Sinaloa had a population over 2.5 million and the highest density (45 inhabitants km−2) while BCS had both the lowest population and density (7 inhabitants km−2) in the region. Therefore, upon TC landfall, major differences should be expected between the impacts upon the population from mainland tracks and from tracks moving across the peninsula. Since BCS has coastlines along both the Pacific Ocean and the Gulf of California, this makes it highly susceptible to TC impacts from the surrounding seas.
During the 2006 season in the eastern North Pacific, from 27 May until 20 November, 18 named TCs developed and 10 of them reached hurricane strength. These numbers are slightly above the 1971–2005 average estimated by Arnt et al. (2010). In 2006, two hurricanes made landfall in northwestern Mexico. Strong winds and heavy rainfall affected coastal communities by causing extensive property damage and nine deaths (Pasch et al. 2009). Later in the season, Paul arrived as a weakening tropical depression and this event was associated with four deaths. Comprehensive information on the TC evolution is provided by the U.S. National Hurricane Center (NHC), through the corresponding reports and season summary (Pasch et al. 2009). However, our study strives for a more detailed discussion of TC evolution prior to and at landfall with emphasis on the impact over populated areas.
The present study focuses on TC impact over Baja California and the mainland during the relatively active season of 2006. This includes three landfall events and our specific objectives are the following:
to discuss track and structure changes that occurred during landfall; this discussion is based on satellite and radar observations as well as on upper-air soundings and gridded data;
to determine the associated rainfall patterns from a regional network operated by the Servicio Meteorológico Nacional (SMN) of Mexico; a comparison is made with respect to records from other events during the period 1969–2005; and
to assess the ability of official forecasts, issued by NHC from 1–3 days before landfall, in the prediction of the actual track and intensity; an extended survey, covering the seasons from 1988 through 2009, was also performed.
This paper is divided into six sections. Section 2 describes observational sources used to document TC evolution. A climatological perspective and individual case studies are described in section 3 while official forecast performance is discussed in section 4. Finally, a discussion is presented in section 5 while section 6 provides a summary and our conclusions.
2. Methodology and datasets
Tropical cyclone position and intensity are taken from the (NHC) best-track database while the development of convective features and moisture patterns are derived from satellite imagery. During the last few decades, satellite surveillance has become an important source of observations and, in many cases, is the only available method for estimating TC structure, motion, and intensity (Velden et al. 2003).
Digital imagery from the Geostationary Operational Environmental Satellite-11 (GOES-11) infrared and water vapor channels is provided by the Space Science and Engineering Center at the University of Wisconsin—Madison. The imagery is available at 30-min intervals and is used to determine cloud-cover structures and moisture patterns. Infrared images have 4-km resolution and they supply a measure of cloud-top temperature with the coldest (highest) tops associated with deep convection. The water vapor imagery (8-km resolution) is useful for monitoring changes in the moisture patterns during the evolution of mid- to upper-level systems such as troughs, ridges, and cutoff circulation patterns (Dvorak and Mogil 1994). In addition, a set of animations is available online (http://met-bcs.cicese.mx/2006) in which the infrared images cover 3-day periods while the water vapor images display the whole TC life cycles.
To identify the extent of accumulated TC rainfall over sea and land, data from the Tropical Rainfall Measuring Mission (TRMM) are used. The data are from product 3B42 version 6, which derives precipitation estimates from various satellite systems and, when possible, from rain gauge data over land (Huffman et al. 2007). The gridded output has spatial resolution of 0.25° × 0.25°. This product provides good coverage over the tropics and has proven to be useful in determining TC contributions over the global basins (e.g., Jiang and Zipser 2010). Radar imagery from stations in BCS and Sinaloa is used to document the reflectivity and rain rate, respectively. These radars are operated by SMN and they were extremely useful for TC monitoring during the season of 2006 (Pasch et al. 2009). To determine spatial coverage and intensity patterns of selected precipitation episodes, a ground-based network of rain gauges is used. The network records are available as 24-h accumulations and are archived by SMN. To set a climatological reference, records from the period 1969–2005 are examined and they are compared against those from the landfalls in 2006.
The three-dimensional structure of the flow is derived from gridded reanalyses issued by the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR; Kalnay et al. 1996) and analyses from the Global Forecast System (GFS). They are used to resolve large-scale circulation patterns and provide a reasonable representation of the environment surrounding the TCs. Isobaric plots and vertical cross sections are constructed to identify relevant circulations through the atmosphere. Upper-air soundings, from SMN and the National Oceanic and Atmospheric Administration (NOAA), are examined to identify relevant characteristics in the environment over Mexico and the United States.
a. Climatological overview
Examination of records from the eastern North Pacific reveals the development of 579 named TCs during the period 1969–2006 and, approximately, 10% of them made landfall in northwestern Mexico. The majority of these landfalls were along the coasts of BCS and Sinaloa with almost equal numbers of strikes per state. Figure 2 displays the tracks that were identified to make first landfall over either the peninsula or the mainland. Inserts show the temporal frequency of incidence that was determined by dividing each month into three periods of 10 or 11 days.
Figure 2a shows 28 tracks that first moved over Baja California. Incoming systems approached the peninsula from the south and almost 90% of them occurred from late August through early October. Thirteen tracks had a second strike, either over the northern peninsula or the mainland. During the same period, 27 TCs tracked only over the mainland (Fig. 2b). This group developed later in the season, with the highest frequencies occurring in the last third of September and most of October. The tracks are similar to distinct groups identified by Romero-Vadillo et al. (2007; see their Fig. 9) and by Ritchie et al. (2011; their groups 1 and 2) for late season tracks affecting the northwest. The change in direction from northwest through north to northeast, into the midlatitude westerlies, is known as recurvature (e.g., Dobos and Elsberry 1993).
A relatively large set of mainland TCs moved across the state of Sinaloa, and Fig. 1 shows distinct clusters around Mazatlán (site B) and Los Mochis (site C). This result agrees with findings from Blake et al. (2009) of a significant concentration of hurricane-strength landfalls along Sinaloa. Further investigation suggests that, when compared with peninsular hits, stronger TCs tend to make landfall over the mainland. This is likely due, in part, to warmer sea surface temperature in the Gulf of California with respect to the cooler California Current in the Pacific. Information on relatively strong TCs at landfall is shown in the lower-left inserts of Fig. 2. Four out of 12 hurricanes were classified as major (category 3 or stronger on the Saffir–Simpson scale) hurricanes while moving across the mainland. In contrast, only one of 10 hurricanes crossed the peninsula with category-3 strength.
Farfán (2004) examined general features of the large-scale environment associated with a sample of six TCs that made landfall over Baja California. The author found that the mean 500-hPa flow showed a midlatitude trough traveling eastward across the western United States and an anticyclonic circulation was located over the Gulf of Mexico. A similar analysis was performed for the landing tracks during the period 1969–2005. Figure 3a shows that when landfall occurs over the peninsula, there is a weak trough within 110°–130°W, the flow over western Mexico is from the south, and, in agreement with the steering concept, there is northward TC motion. In contrast, during mainland landfall, there is a more intense trough and northeastward motion (Fig. 3b). As in Ritchie et al. (2011), the 584-dm contour is chosen as a practical aid to demarcate the base of a midlevel trough and is located close to 20°N. This is accompanied by a distinct region of large wind shear (>20 m s−1) northwest of the anticyclone. Examination of composites 24 and 48 h prior to landfall indicates eastward propagation of the trough. These facts are consistent with results discussed by Corbosiero et al. (2009) on the large-scale environment that steered TCs affecting the southwestern United States.
b. The season of 2006
Tracks of TCs that made landfall over northwestern Mexico are shown in Fig. 4. The tropical depression, which eventually became Hurricane John, was first detected near the Gulf of Tehuantepec. John moved parallel to the coast and experienced slow weakening over BCS. This is related to lower terrain elevations, the reduced width of the peninsular mountains, and the marginal presence of the relatively warm Gulf of California at the right side of the low-level circulation. Later, Lane was associated with a relatively short track during its life cycle and moved over the southern gulf to cross the mainland in Sinaloa. As soon as Lane made landfall, the intensity decreased from category-3 strength (57 m s−1) to become a tropical storm (18 m s−1) in only 12 h. This is likely due to the passage over the Sierra Madre Occidental. Finally, similar to the cases in Fig. 2b, Paul followed a recurving track just west of Isla Socorro and quick dissipation occurred during landfall.
1) Synoptic-scale circulations
Figure 5 displays the GFS analyses at landfall. The fields shown in Fig. 5a indicate that, during John’s arrival, the large-scale flow was controlled by a midlevel anticyclone over the U.S.–Mexico border. The anticyclone’s southern edge provided steering flow from the southeast, which is consistent with TC displacement toward Baja California. The wind shear was relatively low, in the range of 5–10 m s−1; therefore, the environment was favorable for TC intensification (e.g., Kaplan et al. 2010). The corresponding precipitable water field (not shown) depicts a maximum around John and a secondary moisture concentration that is located to the west, associated with the development of another TC (Kristy). With the 584-dm contour about 2000 km to the northwest of the landfall suggests no direct influence on John from the midlatitude environment. Note that, simultaneously, an Atlantic system (Ernesto) was developing over the eastern United States.
In contrast to the above setting, during Lane’s landfall (Fig. 5b) the 584-dm contour is closer to the TC, there is moderate shear over northern Mexico, and these features are associated with the passage of a midlatitude trough. Additionally, there is a relatively strong anticyclone (geopotential height > 590 dm) over the Gulf of Mexico that is associated with southerly flow over central Mexico. Figure 5c is useful in determining that Paul’s weakening trend was due, in part, to the presence of an unfavorable environment including large wind shear, along with a midlatitude system off Baja California, north of the TC center. This system is accompanied by northeastward, humid flow from the equatorial Pacific into western Mexico.
More information on the three-dimensional structure of the environment, during the landfall of the three case studies, is determined from a set of vertical cross sections constructed over the Gulf of California and southern California. The examination was also performed during TC approach (24 and 48 h prior to landfall), although, for the sake of brevity, the corresponding figures are not shown. For the environment associated with Lane and Paul, the cross sections depicted northward advection of humid air below the 700-hPa level and trough passage with strong westerly winds at upper levels. The cross sections capture well-defined jets within the 400–100-hPa layer, along with a significant decrease in low-level humidity over southern California, after the trough passage. These structures are consistent with the lack of cloud cover and rainfall over the northern gulf, a subject to be discussed below.
2) Satellite and radar imagery
At landfall, infrared imagery from John (Fig. 6a) and Lane (Fig. 6b) shows well-defined areas of high cloud tops around the circulation centers and minimum temperatures below −60°C. The corresponding water vapor imagery (Figs. 6d and 6e) shows the extent of the dry air over the Pacific Ocean and a portion of northwestern Mexico. High-resolution (1 km) visible images are used to identify spiral bands and an eye, which are consistent with their hurricane strength. In contrast, Paul arrived as a weakening low-level circulation with a reduced area of high cloud tops (Fig. 6c) although a more humid environment was present over northern Mexico (Fig. 6f).
While interacting with the peninsula, John’s infrared imagery indicates that intense clusters developed around the circulation center (http://met-bcs.cicese.mx/2006/john_ir.gif). A set of convective systems developed farther north, in the southwestern United States, within a moderately moist environment (precipitable water from 25 to 40 mm) detected by the sounding network. Note the missing data from Mexican stations in Fig. 6, an issue to be discussed in section 5. As the storm approached BCS, the Cabo San Lucas radar range and temporal frequency were changed from 540 to 240 km and from 15 to 5 min, respectively. This fact allows us to identify a 15-km diameter eye surrounded by a ring of relatively intense reflectivity and a distinct spiral band located south of the circulation center. In addition, intense cells developed over the eastern flank of the core and the potential for major damage was also present in a region 100–200 km off the coast (Fig. 7a). The mainland radar is used to provide complementary monitoring of John’s passage over La Paz. This imagery (not shown) displays several features, including curved bands over the gulf and intense clusters along the eastern flank of the core.
During Lane’s development, the animation of the water vapor imagery (http://met-bcs.cicese.mx/2006/lane_wv.gif) is convenient tool for identifying the arrival of a large area of dry air from the southwestern United States while the infrared images indicate that, prior to landfall, deep convection developed over the southern gulf. The Guasave radar allowed more detailed monitoring of an eye as the TC moved near Mazatlán and curved bands of heavy rainfall developed over central Sinaloa (Fig. 7b). The GOES infrared imagery (Fig. 6d), along with the corresponding visible image (not shown), indicate that the core had a 300-km width. This is consistent with the 345-km diameter of tropical storm winds estimated, 4 h prior to landfall, by NHC.
The animation of satellite imagery determines relevant features from Paul’s life cycle. Prior to landfall, high cloud tops developed ahead of the TC. This activity occurred on 24–25 October over the southern gulf, while the circulation was south of Cabo San Lucas and a set of cloud clusters moved northeastward into Sinaloa and Nayarit. The Guasave radar detected heavy precipitation along the mainland plains and a feature of interest is the rainfall and cluster location to the east (ahead) of the circulation center. Since cloud tops near the TC center were in the range of 10°–15°C, it is assumed that the circulation is representative of the layer below 1500 m. This height assignment is derived from a temperature comparison between dropsonde observations taken during a reconnaissance mission from the U.S. Air Force and estimates from the GOES infrared imagery. The low-level circulation remained active as it moved across the gulf and reached the mainland near Culiacán (Fig. 7c).
3) Accumulated rainfall
Figure 8 shows the distribution of accumulated rainfall from TRMM. While moving over the ocean and south of the Mexican coast, accumulations above 200 mm are estimated around John’s core (Fig. 8a). In contrast, amounts below 150 mm are estimated along most of BCS. No rainfall over the peninsula or over Sonora was associated with the passage of Lane but an enhanced area is confined to the southern Gulf of California, west of Nayarit (Fig. 8b). Paul constituted a source of limited rainfall for northwestern Mexico (Fig. 8c), associated with weakening and recurving motion while the circulation approached the mainland.
Actual rainfall amounts from John’s passage over land, based on SMN data, are shown in Fig. 9, and Table 2 presents the 10 largest accumulations. The pattern of enhanced accumulation along the mountains and lack of rainfall over the central peninsula, west of 113°W, is consistent with TRMM estimates (Fig. 8a). However, in situ observations are used to determine that substantial amounts were underestimated by the TRMM-based retrieval algorithm. While maximum estimates are below 200 mm, in situ observations reported totals between 287 and 506 mm. The reasons for this difference are unclear and beyond the scope of this study, however.
Extraordinary rainfall was recorded at stations 1–4, with less than 300 inhabitants each. In contrast, the five largest communities listed in Table 1 received totals in the range 49–265 mm. This fact suggests that moderate rainfall affected the majority (67%) of the BCS population that these communities comprise. Another outstanding aspect is that maximum rain rates from John (270–449 mm day−1) set new records at several stations with observations since 1969. Previous maxima were associated with TC approach or landfall between 1976 and 2003. Positions of the strongest landfalls in southern BCS (Fig. 9) indicate preferred arrival south of 24°N, across the Gulf coast. This represents an important weather threat to a small area that, during the last few decades, has been affected by four category-2 and one category-3 hurricanes. John moved over land for more than 40 h to become a unique case among TCs that moved over the peninsula (Fig. 2a).
The combined information from GOES and TRMM satellites, as well as SMN radars, suggests favorable conditions for heavy precipitation from Lane’s landfall. Data collected by the rain gauge network (Fig. 10a and Table 3) show maxima above 250 mm (stations 1 and 3) in southern Sinaloa, while secondary maxima (46–211 mm) occurred over the central portion of the state (stations 2 and 4–10). The largest precipitation rates were in the range 220–260 mm day−1, while the center of circulation was making landfall. Inspection of individual soundings released at Mazatlán reveals that the thermodynamic conditions supported deep convection. These conditions include 1) high precipitable water, >55 mm; 2) nearly saturated air, with relative humidity >75%, below the 900-hPa level; and 3) an unstable atmosphere given by the lifted index <−1°C.
According to TRMM (Fig. 8c), limited rainfall occurred over most of the mainland, and scattered areas of light precipitation developed over the southern Gulf and east of Paul’s best track. Despite its limited strength, considerable accumulations were collected by in situ stations around the landfall area. According to Fig. 10b and Table 4, more than 200 mm were reported by stations 1–4, on the coastal plains of north-central Sinaloa. The largest rate was 200 mm day−1 at the closest station to the landfall position. Inspection of daily records indicates that most of the rainfall was collected on 25 October, when intense cloud clusters were developing ahead of Paul’s low-level center of circulation.
4. Official forecasts
Forecasts are the result of human experience and careful assessment of the model guidance. Track and intensity forecasts are routinely issued during the entire life cycle of every TC and they are derived from a detailed examination of simple prediction techniques as well as from deterministic and ensemble models (Heming and Goerss 2010). Current guidance is based on products from meteorological centers around the world. At NHC, an official forecast is released at four initial times (0000, 0600, 1200, and 1800 UTC) and contains projections valid at 12, 24, 36, 48, 72, 96, and 120 h (Rappaport et al. 2009). This product is an important tool available from hurricane specialists with responsibility in the Atlantic and eastern North Pacific basins.
a. Forecast selection
Section 3 described observations from the evolution of TCs that had an impact in northwestern Mexico in 2006. Now, and by using official forecasts, we wish to provide an examination of the operational products issued prior to landfall. A survey of all the official forecasts indicates that the first available predictions with a position over land were issued 62 h (John), 60 h (Lane), and 114 h (Paul) before the actual events. Therefore, we choose to inspect a sample of forecasts issued within 5 days from their respective landfall times. In particular, as shown in Fig. 11, only the following forecast cycles are inspected: 1) 0000 UTC from 30 August to 1 September, 2) 1800 UTC from 13 to 15 September, and 3) 0000 UTC from 23 to 25 October. These forecasts were issued between 1 and 3 days from landfall and they were associated with timely warnings as well as valuable information for emergency management decisions before strong winds and heavy rainfall arrived at populated areas.
Figure 11a shows that the first of John’s official forecasts generated a track that was misplaced over the Pacific Ocean, away from land. John was expected to become a category-4 hurricane with no impact over northwestern Mexico and, with respect to the best track, there was a westward bias in the motion. Subsequent forecasts resulted in a weaker hurricane with brief landfall and potential impact over the relatively large communities in southern BCS. The first forecast for Lane had a northwestward track while later predictions suggested a northward path over the Gulf and landfall, as a category-1 hurricane, over the peninsula or over northern Sinaloa (Fig. 11b). Again, the first forecast had a westward bias. Finally, Paul was expected to follow a recurving track, near Isla Socorro (Fig. 11c). The corresponding forecasts had some position changes and kept the landfall site over central Sinaloa and, in contrast to the previous cases, these tracks appeared to have no lateral bias.
As a baseline level of accuracy to determine forecast skill, predictions from the Climatology and Persistence (CLIPER) model are incorporated and compared with the above set of forecasts. In general, CLIPER tracks followed a northwestward displacement and only a few of them indicated passage across the Mexican coast. In addition, tracks from three global-scale models were examined from the operational forecasts issued 3 days prior to landfall (Figs. 11d–f). These tracks are derived from the GFS model, the Geophysical Fluid Dynamics Laboratory (GFDL) model, and the Navy Operational Global Atmospheric Prediction System (NOGAPS) model. According to Rappaport et al. (2009), they are among the leading models in accuracy and are a key component of several ensemble models at the NHC. Most of the corresponding tracks are in agreement with the official forecasts for John, Lane, and Paul; however, the GFS tracks take the TCs farthest west (left) away from the corresponding best tracks. Therefore, we choose to examine forecasts from this model only.
b. Forecast errors
We used an NHC database to examine track and intensity errors from the official forecasts shown in Figs. 11d–f. This database is available online (http://www.nhc.noaa.gov/verification/verify7.shtml) and track error is defined as the distance between the forecast and best-track positions at a given verification time. Table 5 shows that, as expected, the forecast errors increase with time. Beyond 72 h, the errors from John’s forecast were above 400 km and since the official forecast errors are larger than those from CLIPER, this is considered to be a relatively poor forecast with no skill at 96 and 120 h. A similar situation occurred during Lane’s forecast at 48 and 72 h in the verification. In contrast, improved skill is associated with Paul’s forecast that followed a recurving track around Isla Socorro. Beyond 72 h, Lane and Paul have no error assignments because the observed TCs had already dissipated.
For intensity, as a counterpart to CLIPER, the Statistical Hurricane Intensity Forecast (SHIFOR) is a simple intensity model that assesses forecast skill. Intensity errors, defined as the difference between forecast and best-track maximum winds, were also analyzed but not shown. Relatively large (>25 m s−1) errors occurred at 96 and 120 h of John’s official prediction, with no skill relative to SHIFOR, due to the failure to forecast landfall in BCS. Similarly, Lane’s forecast was not able to anticipate the category-3 strength reached just prior to landfall and this resulted in underpredictions in the forecasts. In contrast, Paul’s official forecast did a better job of predicting the intensity just prior to landfall with errors under the corresponding SHIFOR forecast.
c. GFS forecasts
In an attempt to understand the nature of the track errors for John and Lane and, since the GFS had the largest departures among the selected models (Table 5 and Figs. 11d–f), we perform an examination of the output fields associated with forecasts issued 3 days before landfall. The GFS fields are taken from the National Operational Model Archive and Distribution System (NOMADS; available online at http://nomads.ncdc.noaa.gov) on a 1° global grid and 6-hourly intervals. The examination includes 500- and 200-hPa heights as well as 850–200-hPa wind shear and sea level pressure.
Figure 12a shows that John’s forecast initiated with a midlevel anticyclone (A) over northern Mexico and a couple of cyclonic circulations over the Pacific, south of 20°N. One of them corresponds to John (J) and another, to the west, is consistent with the early stages of Kristy (K). An Atlantic TC (Ernesto, E) was present far to the east of the anticyclone and heading north into the Florida Peninsula. During the first 24–48 h into the GFS run, the anticyclone remained essentially in place and its southern flank favored westward displacement of the Pacific TCs. At 72 h (Fig. 12d), the circulation associated with John is located approximately 400 km south-southwest of the peninsula. The forecast location was determined from the sea level pressure minimum, below the 500-hPa circulation, and these fields are shown in Figs. 13a and 13b at full resolution. Note that, with respect to the best track, John’s circulation is in the wrong location. The circulation associated with Kristy weakened and was shifted toward the display’s western edge, 450 km away from the actual position.
A similar situation occurred when Lane’s forecast was initialized as a tropical storm close to the coast and with an elongated anticyclone to the north, from 140° to 75°W (Fig. 14a). This feature did not change significantly during the next few days (Figs. 14b and 14c). At 72 h into the model run, the height configuration resulted in steering flow from the southeast, which kept the circulation associated with Lane south of 20°N, over open sea (Fig. 14d). The more detailed fields (Fig. 13c) show that the contour orientation, over the actual landfall area, is not from south to north, as is shown in the analysis (Fig. 5b). A weak circulation that resembles Miriam developed and it is located at about 125°W (Figs. 13c and 13d), 1000 km west of the actual position. The westward motion remained active and at 120 h the circulation associated with Lane was located farther away from the continent, at 20°N, 120°W.
To identify specific patterns in the forecast fields, which may be critical to determining the correct TC track, a comparison with respect to the corresponding analyses is performed. This examination involved several forecast times, vertical levels, and scalar fields; however, as shown in Figs. 12 and 14, the discussion concentrates on the 500-hPa geopotential heights at 24, 48, and 72 h. Difference heights from John’s forecast indicate that, during the first 48 h into the model run, moderate departures developed west of 100°W (Figs. 12b and 12c). These differences, up to 40 m, suggest that the anticyclone over northern Mexico extends farther to the west with respect to the analyses. The differences cover a large area from the west coast of North America to the central Pacific, centered at 40°N, 130°W. At 72 h, the well-defined maximum over southern BCS is related to John’s missed landfall from the GFS. A similar distribution of height differences was also detected in the heights at the 400-, 700-, and 850-hPa levels as well as in the sea level pressure.
The spatial distribution from Lane’s forecast has a relatively large area of positive departures west of 100°W (Figs. 14b and 14c), as well as an error maximum around the landfall site in Sinaloa (Fig. 14d). These features are consistent with the westward extension of the anticyclone in the zonal band 20°–30°N and, when compared with the analysis (Fig. 5b), the westward steering flow is a likely error source in the missed track. Therefore, it appears that the westward track is likely due to an inaccurate representation by the GFS model of the midlevel anticyclone extent over northern Mexico, reducing the northward component of the steering flow and, therefore, missing the landfall position over northwestern Mexico. Since similar forecast tracks were derived from the NOGAPS and GFDL models (Fig. 11e), we suggest that the problem may be related to the initial fields and, as discussed in section 5, this may originate from a limited number of upper-air observations.
d. Additional seasons (1988–2009)
To have a larger dataset of official forecasts, we conducted an extended survey to cover landfall events from other seasons. Besides the 2006 events, we identified 21 cases from 1988 through 2005 and 6 additional cases from the period 2007–09. This makes a total of 30 cases that were classified, depending on the motion prior to landfall, into recurving or (approximately) straight-moving tracks. This classification resulted in 13 tracks that completed recurvature while 10 of the straight tracks had a large component of northward motion, such as Lane in 2006.
Table 6 lists specific information, including TC intensity and rainfall rates at landfall. It is interesting to note that 70% of these cases moved onshore through BCS. In agreement with the information displayed in Fig. 2, most of the landfalls occurred in the second half of the season. Some recurving tracks experienced extreme intensification or weakening, such as Kenna in 2002 and Rick in 2009, during their approach to the mainland. Maximum rainfall rates, in the 360–449 mm day−1 range, are associated with straight tracks while most of the rates below the average (242 mm day−1) were supplied by recurving or weak TCs moving north. For example, the lowest rate (35 mm day−1) is from Nora of 2003, a recurving tropical depression than moved more than 500 km during the day prior to landfall.
Plots of the official forecasts issued 1–3 days before landfall were created and they were useful in identifying the occurrence of distinct patterns with respect to the best track. One pattern is associated with westward departures such as those from John (Fig. 11a) and Lane (Fig. 11b), and they tend to be larger in the first prediction. To determine performance, for the above tracks, we selected the 72-h verification time from the forecast issued 3 days prior to landfall. Then, the NHC database was used to compute average errors from the official forecasts and a summary is shown in Fig. 15. Relatively large errors (180–1230 km) are from recurving TC forecasts, while the straight TCs have lower errors (44–818 km). Errors from John and Lane were below and above the long-term average, respectively. A comparison to the corresponding CLIPER forecast errors suggests that the official forecasts had, in most cases and including Paul, some skill in predicting track changes due to recurvature. In contrast, straight tracks were easier to predict, as the official and CLIPER forecast errors have similar averages.
The large-scale environment was examined on an individual basis and fields from cases with the largest track errors are shown in Fig. 16. The 32-km North American Regional Reanalysis (NARR; Mesinger et al. 2006) was used to determine the setting of tropical and midlatitude systems over the study area. Note that anticyclones centered over northern Mexico or the southwestern United States are associated with TCs tracks predicted to move away from the continent but, when compared with the best track, they were incorrect. The steering flow seems to favor northwestward displacement of the TCs and inspection of the model guidance, available when issuing the official forecasts, suggests that some of the models did not anticipate environmental changes that resulted in a larger component of northward motion. Additionally, some of the recurving tracks (Figs. 16b and 16f) occurred during the passage of midlatitude systems and large errors may be due to an inaccurate prediction of the along-track component of motion.
This study analyzes landfalling TCs over northwestern Mexico and provides documentation from existing regional datasets. Satellite and radar imagery are used to identify convective features while environmental flow patterns are determined from gridded analyses and upper-air soundings. Previous studies identified the key role of the topography in Baja California and the large-scale flow in the motion, intensity, and structural changes of the incident TCs. However, the present research found that, during the most active portion of the 2006 season, there were TCs hitting the peninsula and the mainland. These TCs made landfall over regions with different environmental conditions and contrasting population densities.
Assuming availability of the observations, an extended study may be performed to document 1969–2009 landfalls, when more than 50 TCs made landfall in the northwest, and to derive more general results. However, there are some limitations that are important to consider. For example, 1) satellite imagery from GOES platforms is available from the 1974 season with relatively low spatial and temporal resolutions; 2) the Cabo San Lucas and Guasave radars started operations in the early 1990s while their digital imagery is archived only since 2000; 3) upper-air observations have been variable, as only one station has been in operation since 1948 while the most recent site started in 1991; and 4) the GFS analysis–forecast system is available from 2002. These facts suggest that the documentation of events prior to 2000 is more difficult to fully accomplish.
The density and reliability of upper-air observations in Mexico may play a critical role in the correct initialization of the large-scale environment in the operational models. As part of this study, we made a simple survey of the observations taken during selected periods from the 2006 season. For example, prior to John’s landfall, only 20% of the maximum number of standard soundings (66) was released at the 11 operational sites west of 98°W. Few soundings were taken at inland stations and there were no releases at Isla Socorro, La Paz, or Guaymas (Sonora). The sounding coverage increased to 38% and 41% before Lane’s and Paul’s landfall, respectively, with more stations active along the west coast. An increased number of soundings should better represent the location, extent, and intensity of the large-scale circulations that are used to initialize the model forecasts. However, there are still questions about the quality control and model assimilation from these observations that need to be carefully examined, and this is recommended as part of a subsequent research study.
We recognize that our results are derived from the analysis of a few TCs and they do not include enough seasons to lead us to expect that the findings will become representative of long-term statistics. To generalize these results and, as part of an ongoing research study, we are performing an extension of the official forecast error analysis, with some of the results discussed at the end of section 4. However, an important element is to realize that during the last few decades the performance of numerical models has improved and it is difficult to compare model versions (such as those from the GFS) that have been upgraded over the course of several years. This is an important factor to be considered when assessing a model’s ability to simulate TC tracks.
The official forecasts are also useful for estimating the fraction of the population that may be directly affected by the TC approach or landfall. We computed the population fraction, with respect to the total population, from communities located along the forecast tracks. The influence area is estimated from the radius of specific wind speeds prior to and during landfall, which is among the NHC products released in real time. Based on the population census from 2005, we determined the population to be affected by the selected set of official forecasts for John (Fig. 11a): nobody would have been affected by the first forecast, while subsequent forecasts ranged from 30% to 35% of the total BCS population (Table 1). Likewise, no impact was derived from Lane’s first forecast and 15%–40% of the state of Sinaloa’s population would have been affected by the next forecasts (Fig. 11b). This particular application is still under development but we envision it to be an extremely useful tool to be used by emergency managers and decision makers while designing protection plans.
Information derived from the most recent census (INEGI 2011a) reveals that the population in Mexico has increased dramatically and, currently, there are nine million more inhabitants than in 2005. The fastest-growing state in the northwest is BCS and the largest rates are concentrated in San José del Cabo and Loreto (see Fig. 1 and Table 1 for locations). These facts, plus the relatively large coastal length and the frequent approach of eastern Pacific TCs, suggest that these cities require special attention to prevent major storm impacts on the population and infrastructure. One promising way to accomplish this task is by providing better long term (3–5 days) forecasts, which are essential for giving advance warning of landfall events. However, an important human dimension aspect to consider is the correct interpretation of graphical products widely adopted by the local media to communicate forecasts. Incorrect interpretations may lead to incorrect conclusions with respect to track, size, intensity, and impact region (Broad et al. 2007). We propose training actions focused on awareness and education for specific groups, such as the effort described by Farfán et al. (2010) in which Latin American students are offered short courses on scientific aspects of TC landfall as well as their corresponding social and economic impacts.
6. Summary and conclusions
The goal of this study was to document TC track and structure changes, before and during landfall in northwestern Mexico, from the relatively active 2006 season. Based on the best-track dataset, three of the eastern North Pacific TCs were selected for analysis: John, Lane, and Paul. Two of them became the third-strongest TCs hitting the northwest since 1969. They brought strong winds and heavy rainfall that affected populated areas, and our inspection of data from satellites, radars, gridded fields, and a rain gauge network resulted in the following set of key findings:
John had a track similar to that of TCs developing during late August or the month of September. It was steered by an anticyclone, made landfall as a category-2 hurricane, and moved along the peninsula for several days. This resulted in large amounts of total precipitation over the mountains with maximum rates that were comparable to, and in some cases larger than, previous records.
Lane developed as a midlatitude trough approached the western United States and an anticyclone was in place over the Gulf of Mexico. The TC moved over the southern Gulf of California and made landfall near large communities in Sinaloa. Its passage provided moderate accumulations of rainfall and daily rates in the coastal plains.
Paul was the only recurving TC and its track is consistent with past TCs that developed late in the season. Convective activity, ahead of the storm center, resulted in moderate rainfall over north-central Sinaloa. Its recurvature was linked to the development of a midlatitude system and strong wind shear that was, in part, responsible for weakening over the Gulf of California.
Our results indicate that the first official forecast, with an inland position, was issued as early as over 4 days and as late as 60 h before the actual landfall. A selection of the forecasts issued 3 days prior to landfall was examined and an assessment revealed general characteristics of the track errors. Predicted tracks for John and Lane showed a westward bias, which suggested reduced impact on the population of northwestern Mexico. By inspecting the GFS model, we showed that the incorrect spatial extent of anticyclones throughout the model run is associated with westward flow and a limited component of northward motion to steer the TCs into the continent. A survey examining additional seasons revealed that similar environmental conditions were present during the approaches of other TCs for which the official forecasts were also unable to anticipate landfall.
In conclusion, our analysis of case studies that made landfall during the season of 2006 represents a contribution to the knowledge of TC impact on northwestern Mexico. More recent landfall events, from 2007 through 2009, confirm the need to continue examining TC development along with an evaluation of the official forecasts. Further events are likely to occur and their impact will become more important given the rapid growth of coastal populations and their corresponding infrastructures.
This work was carried out with the aid of the Inter-American Institute for Global Change Research (IAI, Grant CRNII-2048) which is supported by the U.S. National Science Foundation (Grant GEO-0452325), and from the National Council on Science and Technology in Mexico (CONACYT, Grant 23448). SMN personnel supplied the rainfall observations (Alejandro González and Adolfo Portocarrero), radar imagery (Modesto Mendoza), and upper-air soundings (Victor Ramos). The NHC error database was provided by James Franklin. The authors wish to thank the three anonymous reviewers for comments that improved the manuscript.