1 Climatology of Tropical Cyclone Rainfall Over

CLIMATOLOGY OF TROPICAL CYCLONE RAINFALL OVER PUERTO RICO: PROCESSES, PATTERNS AND IMPACTS

By JOSÉ JAVIER HERNÁNDEZ AYALA

A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2016

To my beloved Puerto Rico, its atmosphere, environment and people

ACKNOWLEDGMENTS

The main ideas behind the development of this dissertation came from multiple experiences with tropical cyclones while living in Puerto Rico. Those life experiences motivated me to explore the climate of the tropics, with special attention to the rainfall associated with those extreme events and their role in Puerto Rico’s physical geography. The research conducted in this dissertation was possible from support of Dr.

Corene Matyas, Associate Professor and Graduate Coordinator at the Department of

Geography at the University of Florida. Her magnificent mentoring and continuous support enable me to invest the necessary effort and time to complete this dissertation. I am truly grateful for her exceptional role as my committee chair. I want to thank Dr.

Peter Waylen for his continuous support and all of the inspiring conversations we’ve had about research and life in general that gave me even more strength to continue in this journey towards the PhD. I thank Dr. Timothy Fik for sharing his expertise in quantitative methods through two great courses and for inspiring me to aspire to more in life. Thanks to Dr. Zhong-Ren Peng for being my external committee member and teaching me more about the human dimension of climate related phenomena. I am truly grateful to

Dr. Michael Binford, Desiree Price and Rhonda Black for their support in departmental procedures. I want to thank Dr. David Keellings for his expertise, support and mentoring in developing the methods section of Chapter 4. I want to thank my family for all of their support during this process, specially my mother Sandra Ayala for always believing in all of my capabilities and encouraging me to dream big. At last, I truly want to thank my wife Katiria Quiles for her continuous emotional support throughout this entire process

4 and for always maintaining a positive attitude towards life that has given me the energy to finish the dissertation and move on to our next dreams in life.

TABLE OF CONTENTS

Page

ACKNOWLEDGMENTS ...... 4

LIST OF TABLES ...... 8

LIST OF FIGURES ...... 9

ABSTRACT ...... 11

CHAPTER

1 INTRODUCTION ...... 13

Tropical Cyclones, Rainfall and the Case of Puerto Rico...... 13 Processes: Tropical Cyclone Rainfall over Puerto Rico and its Relations to Environmental and Storm Specific Factors ...... 15 Patterns: Spatial Distribution of Tropical Cyclone Rainfall and its Contribution to the Climatology of Puerto Rico ...... 16 Impacts: Extreme Floods and their Relationship with Tropical Cyclones in Puerto Rico ...... 17 Importance of Study ...... 18

2 TROPICAL CYCLONE RAINFALL OVER PUERTO RICO AND ITS RELATIONS TO ENVIRONMENTAL AND STORM SPECIFIC FACTORS ...... 20

Factors Influencing TC Rainfall ...... 22 Data and Variable Construction ...... 25 Methods ...... 28 Results and Discussion...... 30 Tropical Cyclone Characteristics ...... 30 Correlation Analyses ...... 33 Principal Component Regression ...... 36 Concluding Remarks...... 38 Chapter 2 Limitations ...... 40

3 SPATIAL DISTRIBUTION OF TROPICAL CYCLONE RAINFALL AND ITS CONTRIBUTION TO THE CLIMATOLOGY OF PUERTO RICO ...... 53

Data ...... 58 Geo-statistical Methods ...... 61 Results and Discussion...... 64 Characteristics of TC Groups ...... 64 Spatial Distribution of TCR ...... 65 TCR Contribution ...... 70 6

Concluding Remarks...... 72 Chapter 3 Limitations ...... 74

4 EXTREME FLOODS AND THEIR RELATIONSHIP WITH TROPICAL CYCLONES IN PUERTO RICO ...... 84

Data and Methods ...... 87 Floods and Tropical Cyclone Data ...... 87 Extreme Value Analysis Point Process Approach ...... 89 Results ...... 91 Descriptive Statistics ...... 91 EVA Point Process Model Results ...... 93 Concluding Remarks...... 102 Chapter 4 Limitations ...... 105

5 CONCLUSION ...... 117

Tropical cyclone Rainfall over Puerto Rico and its Relations to Environmental and Storm Specific Factors ...... 117 Spatial Distribution of Tropical Cyclone Rainfall and tts Contribution to the Climatology of Puerto Rico ...... 119 Extreme Floods and their Relationship with Tropical Cyclones in Puerto Rico ..... 122 Dissertation Contributions ...... 124 Future Directions ...... 126

LIST OF REFERENCES ...... 128

BIOGRAPHICAL SKETCH ...... 138

LIST OF TABLES

Table page

2-1 Storm specific characteristics and environmental factor variables used in this study...... 41

2-2 Descriptive statistics of storm specific and environmental factors associated with the 86 TCs analyzed...... 42

2-3 Tropical cyclones that produced more than 50 mm rainfall over the island of Puerto Rico...... 44

2-4 Spearman’s correlation coefficients for each of the predictor’s relationship with mean and maximum TCR...... 46

2-5 Spearman’s correlation coefficients of variables that were found to be significantly correlated with mean and maximum TCR...... 47

2-6 Varimax rotated principal component analysis (PCA) results. Data includes the number of components its % of variance...... 48

2-7 Forward principal component regression model results for mean and maximum TCR...... 49

2-8 Statistics of the 23 highest/lowest mean and maximum TCR events...... 51

2-9 Mann-Whitney U tests results for the 23 highest/lowest mean and maximum TCR events...... 51

3-1 Rain gauges with daily and monthly data for Puerto Rico for the period of 1970- 2010...... 77

3-2 Tropical cyclone groups based on different PRX and TPW values...... 78

4-1 Stations with complete mean daily discharge for the 1970-2010 period...... 107

4-2 Descriptive statistics of mean daily discharge for the entire series, the series with TCs removed and the three maximum flood events...... 108

4-3 Characteristics of TCs associated with the highest number of flood peaks over the twelve mean discharge stations...... 114

LIST OF FIGURES

Figure page

2-1 Topography of the main island of Puerto Rico and selected weather stations. . 41

2-2 Tracks of all 86 TCs within a 500 km radius and 100 km range rings of Puerto Rico from 1970 to 2010...... 42

2-3 Total precipitable water for September 21, 1998 during the passage of Hurricane Georges...... 43

2-4 Kriging interpolated surfaces for visualization purposes of the storms with the highest mean TCR values and their respective 100 km...... 45

2-5 Scatter plot of the relationship between mean and maximum TCR ...... 46

2-6 Scatter plot of the relationship between mean and maximum TCR ...... 47

2-7 Scatter plot of the relationship between mean and maximum TCR...... 48

2-8 Scatter plot of the relationship between mean TCR...... 49

2-9 Scatter plot of the relationship between maximum TCR ...... 50

2-10 Tracks of the 23 mean and maximum TCR events...... 50

2-11 Kriging interpolated surface for visualization purposes of average total precipitable water (TPW) for the 23...... 52

3-1 The elevation of the island of Puerto Rico and the rain gauges used in the study...... 76

3-2 Tracks of all 86 TC events that passed within a 500 km radius of Puerto Rico. 76

3-3 Scatter plot of the four tropical cyclone groups based on similarities between TPW (y) and PRX (x) values...... 79

3-4 Tracks of tropical cyclones groups divided...... 79

3-5 Kriging predicted surfaces of TCR...... 80

3-6 Kriging predicted surfaces of TCR...... 81

3-7 Kriging predicted surfaces of TCR...... 81

3-8 Ordinary kriging surfaces of TCR of storms located north...... 82

3-9 Ordinary kriging predicted surfaces...... 82

3-10 Ordinary kriging predicted surfaces of average rainfall for the months of July. . 83

3-11 Ordinary kriging predicted surfaces of average rainfall for the months of September...... 83

4-1 The island of Puerto Rico and the stations with complete daily discharge data for the 1970-2010 period and their respective water drainage basins...... 106

4-2 Tracks of all 86 TCs that passed within a 500 km radius of the island’s coast with the TCs that caused extreme floods (99th percentile) in red...... 106

4-3 Scatter plot series of mean daily discharge data...... 107

4-4 Percentage of mean daily discharge values above the 99th percentile threshold (floods) that were associated with the passage of TCs...... 108

4-5 Point process approach diagnostics probability plot...... 109

4-6 Point process approach diagnostics probability plot...... 109

4-7 Daily mean discharge above the 99th percentile of the entire series...... 110

4-8 GEV parameters of the entire series in x and the series with TCs removed in y for Location...... 111

4-9 GEV location parameter for the entire series...... 112

4-10 GEV scale parameter for the entire series...... 112

4-11 GEV shape parameter for the entire series...... 113

4-12 Flood frequency probability for the entire series...... 114

4-13 Tracks of the TCs that were associated with extreme flood peaks (99th percentile) over Puerto Rico...... 115

4-14 Maps of the standard deviation ...... 116

Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy

CLIMATOLOGY OF TROPICAL CYCLONE RAINFALL OVER PUERTO RICO: PROCESSES, PATTERNS AND IMPACTS

José Javier Hernández Ayala

May 2016

Chair: Corene Matyas Major: Geography

Although tropical cyclone rainfall (TCR) is common over Puerto Rico, the factors that cause this rain to vary from one storm to another, its spatiotemporal distribution and relationship with extreme flood events have not been studied. This dissertation focuses on the climatology of tropical cyclone rainfall over the island of Puerto Rico. Several aspects of the rainfall associated with the passage of tropical cyclones are explored.

The first problem focuses on understanding the environmental and storm specific factors that control tropical cyclone rainfall variability over the island. The second research question examines the spatial distribution of precipitation associated with tropical cyclones and the storms contribution to the rainfall climatology of Puerto Rico.

The third and final problem deals with understanding the relationship between extreme floods and tropical cyclones that impacted Puerto Rico.

Results from correlation analyses of the individual predictors, principal component regression (PCR) procedures and Mann-Whitney U tests identified precipitable water, storm center proximity to land, mid-level relative humidity, duration, horizontal translation speed and longitude as the predictors with the strongest influence on tropical cyclone rainfall. Results from ordinary kriging and cokriging techniques show

11 that tropical cyclone rainfall tends to be cluster in the eastern, southeastern and central regions of the island with a decrease in values as we move west and northwest. The month with the largest contributions (>20%) for most of the stations was August followed by September and October while the months with the lowest contributions were

June and July. Results from an extreme value analysis (EVA) point process approach suggest that TCs play a major role in the flood peak distribution of Puerto Rico, especially in stations in the eastern interior and the northcentral region which exhibited a strong relationship with tropical cyclones.

CHAPTER 1 INTRODUCTION

Tropical Cyclones, Rainfall and the Case of Puerto Rico

Tropical cyclones (TCs) are rotating low pressure systems that develop in the warm waters of the North Atlantic and other tropical oceans. In terms of maximum sustained winds TCs include weaker tropical depressions (17 m/s), tropical storms (18-

32 m/s) and the more intense hurricanes (33 m/s or >) that could range from category 1

(33 – 42 m/s) to category 5 (> 70 m/s). These tropical weather systems obtain their energy from warm ocean waters and latent heat release. TCs have a spiral shape with a low pressure in the center of the system, in an area known as the eye. The most intense winds and rainfall tends to be concentrated in the eyewall of the TC, an area dominated by intense convective activity associated with towering cumulonimbus clouds. TCs also have outer rainbands that can produce significant rainfall at a farther distance from the thunderstorms in the eyewall. One of the places on Earth that is highly exposed to TCs is the Caribbean island of Puerto Rico. TCs are an important aspect of the physical geography of Puerto Rio, yet little is known of their role as heavy rainfall producing processes over the island.

The island of Puerto Rico is highly exposed to the passage of TCs during the hurricane season months of June to November, yet most of the storms that have impacted the island occurred in the months of August and September (Caviedes, 1991;

Colon 2008). Easterly waves that develop from unorganized bands of thunderstorms off the coast of Northwest Africa can become tropical cyclones and bring intense winds and heavy precipitation to the island (Pico, 1974). The heavy rainfall produced by the passage of nearby tropical cyclones can trigger deadly mudslides or extreme floods,

13 while at the same time that precipitation can be beneficial to ecological systems and socioeconomic activities in the island. Puerto Rico has experienced heavy rainfall from multiple TCs over the last decades, which include storms like tropical depression Eloise

(1975), Hurricane David (1979), tropical depression Isabel (1985), hurricanes Hugo

(1989), Hortense (1996), Georges (1998) and tropical storm Jeanne (2004). One of the most devastating mudslide events in North American history was triggered by heavy rain from tropical depression Isabel (1985) which took the lives of more than 130 people in Puerto Rico. The interesting case about Isabel is that it was a weak TC in terms of wind speed and its center was located more than 200 km away from land, yet it managed to produce more than 500 mm of rainfall in some areas in the central region of the island. This case shows that even weak TCs that are located farther from land can still produce significant rainfall in the island.

An extensive review of the literature on TCs and rainfall in Puerto Rico revealed that most of the studies conducted only focus on single events and mostly on wind or floods. Some researchers have focused on individual TCs like Hurricane Hugo (1989) and its impacts on forest ecosystems (Scatena and Larsen, 1991; Boose et al., 1994).

Others have focused on the role storms play in extreme flooding events over the island

(Larsen and Simon 1993; Larsen and Torres-Sanchez 1998) with special attention on

Hurricane Hortense (1996) (Torres-Sierra, 1997) and Hurricane Georges (1998) (Larsen and Santiago-Roman 2001, Smith et al., 2005). Even though some studies have explored TC impacts and the historical record of storms on the island (Boose et al.,

2004) an examination of the climatology of tropical cyclone rainfall with a focus on

14 understanding its storm to storm variability, its spatial distribution and its impacts when it comes to floods has not been conducted.

This dissertation presents an analysis of TCR over Puerto Rico, with the specific aim of understanding several aspects of storm rainfall over the island with a special focus on its variability, its spatial distribution and contribution and its relation to extreme flood events. This dissertation is divided in three main chapters in which each of those research questions is addressed.

Processes: Tropical cyclone rainfall over Puerto Rico and its relations to environmental and storm specific factors

Chapter 2 presents an analysis of tropical cyclone rainfall (TCR) over Puerto

Rico, with the specific aim of understanding how environmental factors and storm specific characteristics affect storm precipitation over the island. Guided by the results of TCR analyses in other locations, moisture distribution, vertical wind shear, intensity, translation speed, storm duration, center location and proximity to storm’s center are examined to determine their contributions to TCR variability over Puerto Rico. Rain gauge data are utilized to calculate a storm-mean total rainfall value for 86 TCs during the 1970-2010 period. The maximum observed precipitation value is obtained for each storm.

Both mean and maximum TCR were used as the dependent variables in the different statistical procedures, to test if the same factors affecting average TCR over the whole island were also associated with the extreme precipitation values.

Spearman’s correlation coefficients are calculated to explore individual relationships between the different factors and the two TCR measures. Principal components analysis (PCA) is utilized to reduce the environmental and storm-specific factors to four

15 components. Then principal component regression (PCR) is employed to construct models to explain the mean and maximum TCR received over the island. Mann-Whitney

U tests are also implemented to compare the characteristics of the events with the highest and lowest TCR values.

Patterns: Spatial Distribution of Tropical Cyclone Rainfall and its Contribution to the Climatology of Puerto Rico

The aim of chapter 3 is to explore and understand the spatial distribution of TCR and its contribution to the rainfall climatology of Puerto Rico. The first question deals with understanding the spatial distribution of rainfall over the island, is TCR randomly distributed or is it clustered in some specific regions? The first hypothesis is that high

TCR is concentrated in the eastern area of the island since most of the storms move east to west and that is the region that first encounters the cyclones. TCR relations to topography over the island are also examined. The main idea here is that high elevation areas, mostly in the eastern region of the island, are going to exhibit higher TCR values.

As warm moist air associated with the passage of a nearby TC starts to go upslope over the eastern facing hills of the island more rain is going to fall in those regions due to the orographic enhancement of precipitation. Understanding the spatial distribution of TCR is important for identification of areas exposed to heavy rainfall that could lead to extreme floods and mudslides.

The second problem in Chapter 3 aims to understand how much TCs contribute to the rainfall climatology of the island. This problem was explored both spatially and temporally by examining the percentage of rainfall contributed by TCs for each of the hurricane season months (June-Nov). Given that rainfall from tropical cyclones is an indispensable source of water for ecosystems, communities, agricultural and industrial

16 activities is important to understand how dependent the different regions of the island are to TCR. The questions here are; do TCs have different contributions to the rainfall climatology of the island in different regions and do some hurricane season month’s exhibit higher or lower rainfall contributions from storms? The main hypothesis here is that the eastern and southern regions of the island have the largest contribution percentages in the peak hurricane season months of August, September and October.

The east is the first area that interacts with much of the TCs, while the south is the driest region of the island, so any precipitation generating process would have a substantial contribution to its rainfall climatology.

Impacts: Extreme Floods and Their Relationship with Tropical Cyclones in Puerto Rico

The main purpose of chapter 4 is to examine the relationship between extreme flood events over Puerto Rico and tropical cyclones. Mean discharge data from 12 stations over twelve different water drainage basins in Puerto Rico for the 1970-2010 period were used. Tropical cyclone six-hourly track data from 86 TCs was used in order to identify the floods associated with the different storms that impacted the island in the time period. Floods were defined as the 99th percentile of the mean discharge data for the whole series over the twelve stations. An extreme value analysis (EVA) point process (PP) approach WAS to determine if TCs strongly affected the statistical properties of the GEV distribution parameters location (central tendency) scale

(variance) and shape (skewness) of the mean discharge time series for 12 stations with

41 years of data. First, the PP model was used to fit the entire mean discharge data of all the stations and then the model was implemented again with the series that had the flood peaks associated with TCs removed. The GEV parameters of the entire discharge

17 series and the series without TCs were retrieved in order to compare them and determine if there were any statistical differences between the parameters in the series that included the storms and the series without them. Maps of percentage change between the GEV parameters were generated to visually compare the differences between the location, scale and shape of the entire series and the one with TCs removed. The maps also served to examine the spatial characteristics of the floods in the different stations over the island. Flood frequencies were also calculated from the

GEV parameters to examine the differences between the magnitude and frequency of floods between the two series. The TCs that produce the largest floods in most of the stations were identified in order to examine the relationship between extreme floods and tropical cyclone rainfall.

Importance of Study

This dissertation provides a detailed explanation of the physical geography behind the processes, patterns and impacts associated with tropical cyclones that produce heavy rainfall over Puerto Rico, their distribution and contribution to the precipitation climatology of the island and the hydrologic response to those extreme events. This study incorporates multiple datasets, different statistical and geo-statistical procedures to understand certain characteristics of rainfall from tropical cyclones and their impact on the island of Puerto Rico with applications to other areas in the tropics.

By identifying the environments and characteristics of the storms that produce heavy rainfall in the island, where the high precipitation values are located in relation to topography and the different hydrologic responses this study significantly improves our

18 understanding of the processes that take place in small island environments in the tropics during extreme rainfall events associated with the passage of tropical cyclones.

As a whole this dissertation improves our understanding of the physical geography of tropical cyclone rainfall on small island environments where complex topography, varying environmental conditions and storm characteristics can all play important roles when it comes to the amount and distribution of precipitation associated with these events. The results of these analyses also identify characteristics of storms that produced heavy rainfall and floods, and key thresholds that could later improve the precipitation forecasting of these extreme events. The findings in this dissertation advance our understanding of tropical cyclone rainfall prediction and mitigation in the tropics. With this new knowledge on tropical cyclone rainfall over Puerto Rico it is expected that the awareness and preparedness towards these events will significantly improve, which will then translate in fewer fatalities and lower economic losses associated with the passage of these storms. The findings of this study will be published and will greatly benefit the scientific community that studies the physical geography of tropical cyclones in the Caribbean and in the vast tropical areas of the planet.

CHAPTER 2 TROPICAL CYCLONE RAINFALL OVER PUERTO RICO AND ITS RELATIONS TO ENVIRONMENTAL AND STORM SPECIFIC FACTORS

Understanding the processes that cause high rainfall totals as tropical cyclones

(tropical depressions, tropical storms and hurricanes) move over or in close proximity to land is of crucial importance for all societies in their path. Given the variation in size and rainband structures and that a variety of environmental conditions promote or hinder rainfall production, it is difficult to predict how much precipitation could be received during the passage of an individual storm. Yet, identifying the conditions under which large amounts of rainfall occur during the passage of a tropical cyclone (TC) is very important since these events cause significant losses of life and property when heavy precipitation leads to flooding and mudslides, particularly on small islands with high terrain (Rappaport, 2000; Scatena and Larsen, 1991). The island of Puerto Rico has experienced heavy rainfall from numerous TCs that led to devastating floods and mudslides. For example, heavy rainfall associated with the passage of an easterly wave that later developed into tropical depression Isabel in October of 1985 triggered one of the deadliest mudslide events in North American history killing 130 people (Jibson,

1989; Larsen and Simon, 1993). The fact that Isabel’s circulation center was at a distance of 221 km at its closest approach to the island indicates that flood-producing rainfall can occur even when the core of the storm does not pass over the island.

The main island of Puerto Rico is located at 18.25 ˚ N and 66.39˚ W covering an area of 8897 km2 (Figure 2-1). The climatological distribution of rainfall over the island mainly responds to interactions between the predominant easterlies and the central mountains (Colón, 2009). Rainfall over an annual basis varies from more than 4000 mm

20 at the Yunque National Forest in the northeast to 750 mm or less in the Guanica Dry

Forest located in the southwest (Birdsey and Weaver, 1987). Puerto Rico is highly exposed to tropical storms, especially during the months of August to November

(Caviedes, 1991). The long term average for TC landfalls in the northern Caribbean where Puerto Rico is located is one per year (Pielke et al., 2003). However, as occurred with Isabel (1985) a TC does not need to make landfall to produce heavy rainfall over a location.

A careful review of the literature reveals a lack of research in the TC rainfall climatology of Puerto Rico. Some studies have focused on individual storms like

Hurricane Hugo (1989) and its impacts on forest ecosystems (Scatena and Larsen,

1991; Boose et al., 1994). Others have focused on the role TCs play in extreme flooding events over the island (Larsen and Simon 1993; Larsen and Torres-Sanchez, 1998) with special attention on Hurricane Hortense (1996) (Torres-Sierra, 1997) and Hurricane

Georges (1998) (Larsen and Santiago-Roman, 2001; Smith et al., 2005). Even though some studies have explored TC impacts and the historical record of storms on the island (Boose et al., 2004) an examination of storm rainfall variability based on environmental conditions and TC characteristics has not been conducted.

This study presents an analysis of tropical cyclone rainfall (TCR) over Puerto

Rico, with the specific aim of understanding how environmental factors and storm specific characteristics affect storm precipitation over the island. Guided by the results of TCR analyses in other locations, we examined moisture distribution, vertical wind shear, intensity, translation speed, storm duration, center location and proximity to storm’s center to determine their contributions to TCR variability over Puerto Rico. Rain

21 gauge data are utilized to calculate a storm-mean total rainfall value for 86 TCs. The maximum observed precipitation value is obtained for each storm. Both mean and maximum TCR were used as the dependent variables in the different statistical procedures, to test if the same factors affecting average TCR over the whole island were also associated with the extreme precipitation values. Spearman’s correlation coefficients are calculated to explore individual relationships between the different factors and the two TCR measures. Principal components analysis (PCA) is utilized to reduce the environmental and storm-specific factors to four components. Then principal component regression (PCR) is employed to construct models to explain the mean and maximum TCR received over the island. Mann-Whitney U tests are also implemented to compare the characteristics of the events with the highest and lowest TCR values.

Factors Influencing TC Rainfall

Multiple factors contribute to TCR variability over a region, including environmental moisture distribution, proximity to the storm's center, storm intensity, translation speed and local topographic effects (Anthes, 1982). Several studies have noted the relationship between environmental moisture and TCR. High values of relative humidity in the environment around the storm, especially in the lower and middle troposphere, lead to higher precipitation production (Bosart and Carr, 1978; DiMego and

Bosart, 1982). Large horizontal moisture convergence and total precipitable water are decisive in initiating and maintaining heavy precipitation before and during TC landfalls

(Jiang et al., 2008). Other studies have pointed out that precipitation is controlled by the availability of atmospheric water vapor and that the change in rainfall extremes would enhanced or be constrained by changes in precipitable water (Dai et al., 1999; Allen

22 and Ingram, 2002). A study looking at the relationship between moisture and TCs in

Baja California found that precipitable water values of 40 mm or more in the environment around the storms were associated with moderate and heavy precipitation over land (Farfán and Fogel, 2007). Konrad and Perry, (2010) found that the area in which precipitable water exceeds 50.8 mm is strongly associated with high precipitation totals in the Carolina region of the US. The size of a TC also has important implications on the potential areas to be affected, and researchers have found that higher amounts of moisture surrounding the storm are associated with larger raining areas (Hill and

Lackmann, 2009; Matyas, 2010), an important consideration for TCs that do not make direct landfall over Puerto Rico yet may produce high rainfall totals.

Another environmental factor that affects TCs and their associated rainfall is vertical wind shear. Weak shear promotes storm development and strengthening by releasing latent heat from condensation directly above the surface low, while strong shear causes the asymmetric displacement of convection into the outer regions of TCs

(Corbosiero and Molinari, 2002; Rogers et al., 2003). Rain rates are enhanced in the downshear left direction when shear is high, while the highest rain rates are usually in front of the TC in the outer band regions when shear is low (Chen et al., 2006; Cecil

2007). Matyas, (2010) found that vertical wind shear was one of the most important predictors of rain field size during hurricane landfall as strong southwesterly wind shear was highly correlated to a large extent toward the northeast side of the storm. However,

Lonfat, (2007) found that the effect of vertical wind shear on TCR asymmetries is more dominant before rather than during landfall, since the shear factor is overwhelmed when interaction with land and topography occurs. It’s important to note that the mean

23 vertical wind shear in the Atlantic for the hurricane season peak months is westerly with a magnitude greater than 8 ms -1 over much of the basin (Goldenberg and Shapiro,

1996).The El Niño Southern Oscillation and Saharan air layer can influence the vertical wind shear in the Atlantic by changing the direction and intensity of the wind which can then affect TC development and intensification (Gray, 1984; Dunion and Velden 2004).

The vertical wind shear in the Caribbean is also highly influenced by the tropical upper- tropospheric trough (TUTT), which can have both detrimental and favorable effects on

TCs cyclogenesis and intensification (Sadler, 1976; Yaukey, 2011) In its average position the TUTT is located north of Puerto Rico with its southern portion associated with stronger vertical wind shear affecting TCs over the eastern Caribbean region.

(Fitzpatrick et al., 1995).

Characteristics of the storm itself also contribute to TCR variability, including intensity, storm motion and location relative to land. Several studies have noted a strong linear relationship between rainfall and TC intensity (Alliss et al.,1992; Rao and

MacArthur,1994) and others have found that TC intensification was accompanied not only by increases in the average rain rate, but also in the relative contribution of the heavy rainfall (Simpson and Riehl, 1981; Shepherd et al., 2007). Cerveny and Newman,

(2000) found that if TCs are categorized into classes based on their intensity and those categories are used in a regression model, 94% of the variability in rainfall can be explained by the storms maximum sustained wind speeds. An important distinction here is that these results were obtained when examining rainfall that accumulated at the surface as the TC passed directly overhead.

Proximity to a storm’s inner core eyewall and its associated ring of towering cumulonimbus clouds plays a major role in TCR rates and total accumulations on the surface (Rogers et al., 2009). Several studies have noted that as distance from the center of the storm decreases higher rainfall rates and accumulations are experienced on the ground (Riehl and Malkus, 1961; Huff, 1970; Simpson and Riehl, 1981).

Goodyear, (1968) identified a threshold of 40-70 km from the storm center as the areas where the highest rainfall amounts were recorded. The outer rainbands of TCs can also contain convective cells that produce high rain rates (Matyas, 2009). Lonfat et al.,

(2004) also noted that at longer distances of 300 km from the center high rainfall peaks can still be observed. Therefore, it is most likely that high rainfall totals would occur for a

TC that passes close to a location, however rainfall could vary if only the outer rainbands affected a location.

Other researchers have also considered TC specific factors such as translation speed to explain rainfall variability. Gao et al., (2009) found that higher average rain potentials before landfall are mainly associated with larger vortex size and slower translation speeds. Chen et al., (2006) found that translation speed and direction of TCs are important for precipitation asymmetries when the wind shear is weak. A more recent study on typhoon Morakot in Taiwan found that when translation speeds were increased by 55% under simulated environments, 33% less rainfall occurred over the island (Yen et al., 2011).

Data and Variable Construction

The first step of the analysis was to determine which TCs might have produced rainfall over Puerto Rico. Six-hourly TC positions were obtained from the International

Best Track Archive for Climate Stewardship (IBTrACS) (Knapp et al., 2010) for the years 1970-2010.These data were entered into a GIS where a 500 km radius around the main island of Puerto Rico was employed to identify the TCs for this study. This radius has been utilized by previous researchers to define the distance at which a TC may bring rainfall to a given region (Georgiou et al., 1983; Boose et al., 2004; Zhu and

Quiring, 2013). This delimitation yielded 86 TCs that were analyzed further (Figure 2-2).

Daily rainfall totals were obtained from the National Climatic Data Center

(NCDC), for 32 weather stations located on the main island of Puerto Rico for the period of 1970-2010 (Figure 2-1). Storm total rainfall was defined by the amount of time each storm was within the 500 km radius, which ranged from twelve hours to four days. The rainfall for the entire day was considered for storms that spent 12 to 24 hours within the

500 km radius. The station with the highest storm total rainfall was identified and an areal mean TCR value was calculated. We chose to characterize TCR using these variables to better understand conditions under which extreme localized rainfall as well as wide-spread high precipitation totals occur.

For the mean TCR, rainfall associated with each TC was averaged over the 32 stations. The mean TCR for storms that spent more than one day inside the 500 km radius was calculated by summing up the average daily rainfall of their respective days.

For example, if a TC spent 1.5 days within the radius the data for the two days were considered when calculating the averaged total rainfall. It is important to note that a rainfall day according to the NCDC data begins at 7:00 am LST (1100 UTC). Thus, there is a mismatch with the IBTrACS data whose observations are available every six hours beginning at 00 UTC. Due to this mismatch in times, we performed the analysis

26 by adding or removing a day of data as the TC approached or exited the study region and we found that the results did not significantly vary from those presented here.

Another approach to estimate the rainfall associated with TCs uses the radius of outer closed isobar (ROCI) in a method known as the moving ROCI buffer technique (MRBT) to identify daily TCR from a network of weather stations (Zhu and Quiring, 2013).

However, we could not employ this technique as ROCI data are only available beginning in 1988.

Attributes of each TC as well as the environment surrounding Puerto Rico were averaged over the same number of days as were employed in the mean TCR calculation. However, TC center proximity to the island was calculated from the storm’s point of closest approach to land. These factors were explored to determine which exhibited the strongest association with TCR over the island (Table 2-1). Storm specific factors including the average center location (LAT, LON), mean maximum sustained wind speed (VMX) and duration (DUR) were calculated from the IBTrACS dataset.

Other TC specific factors such as storm center proximity to the island (PRX) and average horizontal translation speed (HTS) were calculated after entering the IBTrACS dataset into a GIS.

Daily values of total precipitable water (TPW), mid-level relative humidity averaged over the layer between 700 hPa and 500 hPa (MRH), and westerly and southerly wind shear values over 850-200 hPa (WSU and WSV) were obtained from the

National Center for Environmental Prediction (NCEP) and the National Center for

Atmospheric Research (NCAR) reanalysis data set (Kalnay et al., 1996). Mid-level relative humidity averaged over the 700 hPa and 500 hPa layers was selected for

27 analysis since other researchers have found that drier mid-levels might impact TC development (Thorncroft and Pytharoulis, 2001), therefore affecting the TC’s rainfall potential. As this study is concerned with rainfall over Puerto Rico rather than that produced in the entire area surrounding the TC, the spatial extent of the environmental variables was limited to the area over and near Puerto Rico using the GIS. The values available for the closest four cells to the island for each of the environmental predictors

(Figure 2-3) were averaged over the time the TC was within the study region. The area associated with the four chosen cells surrounding Puerto Rico is approximately 290,000 sq.km.

It is important to note that the NCEP/NCAR reanalysis data set derives many of its variables from atmospheric soundings. The environmental variables TPW and MRH where compared with atmospheric soundings from San Juan, PR to determine the suitability for further statistical analysis of the NCEP/NCAR reanalysis data set. A

Pearson correlation test was employed to compare the reanalysis data and the soundings from San Juan for 20 (10 highest and lowest mean TCR events) of the 86

TCs. The results of the Pearson correlation test between TPW data from the soundings and mean NCEP/NCAR show a statistically significant (0.866) relationship (p < 0.001).

MRH also exhibits a statistically significant relationships (0.701) between radiosonde and mean NCEP/NCAR data. These results confirm that the NCEP/NCAR reanalysis dataset is useful to study the environment during the passage of TCs over or near PR.

Methods

The first step in this study was to explore how the individual environmental and storm-specific factors are associated with TCR variability. This problem was addressed

28 by employing a Spearman’s rank correlation test, which is a non-parametric rank statistic used to measure the amount of association between two variables (Hauke and

Kossowski, 2011). The Spearman’s rank correlation test was performed between each independent variable as well as with mean and maximum TCR.

A principal component analysis (PCA) procedure was implemented to reduce the number of predictors and account for independent variable collinearity. The purpose of the PCA is to extract the important information from the independent variables and reduce them to a new set of orthogonal variables called principal components

(Hotelling, 1933). PCA is a method widely used to study different characteristics of many climatological factors over both spatial and temporal scales (Barnston and

Livezey, 1987; Janowiak, 1988; Ogallo, 1989). This study also employed a Varimax orthogonal rotation which allowed a more simplified interpretation of the PCA results since each component represents only a small number of the independent variables

(Abdi and Williams, 2010). All components with Eigen-values greater than one were retained.

A forward principal component regression (PCR) model was implemented to identify the components making statistically significant contributions to TCR variability over Puerto Rico. The mean and maximum TCR values that serve as predictands in the

PCR model did not followed a normal distribution, so both dependent variables were normalized by transforming them to base 10 logs. Similar to a multiple linear regression model the PCR assumes that there is a relationship between the logged mean and maximum TCR and each of the principal component scores. The PCR procedure has been employed in studies looking at TC motion prediction in all ocean basins (Bessafi et

29 al., 2002) and rainfall forecasts in the US Pacific islands (Yu et al., 1997). No specific study has used PCR as a method for examining TCR variability, however several studies have used linear regression models to look at how different factors contribute to

TC precipitation (Cerveny and Newman, 2000; Jiang et al., 2008; Matyas, 2010). This study employs a PCR procedure to provide a more robust method than multiple linear regression since it accounts for multivariate correlations between the independent variables that could aid in explaining TCR variability over Puerto Rico. After the PCR procedure, several Mann-Whitney U tests were implemented to compare the characteristics of the TCs with highest and lowest TCR values.

Results and Discussion

Tropical Cyclone Characteristics

Before examining rainfall totals, it is useful to discuss the statistics associated with the storm characteristics and environmental factors for all 86 TCs (Table 2-2).

When looking at longitude and latitude together we find that TC circulation centers were located at an average arithmetic position that is offshore near southeast Puerto Rico. In terms of the storm center proximity to Puerto Rico, TC centers averaged a distance of

239.7 km from the island. The mean duration for TCs within the 500 km radius was 41.5 hours. Storms had an average translation speed of 6.3 ms-1 which is close to the 6 ms-

1 mean for most TCs found in the Atlantic basin by DeMaria and Kaplan, (1994) and average maximum sustained winds of 28.3 ms-1, which falls into the tropical storm category. The global scale precipitable water values tend to be between 40-50 mm in areas where TCs develop (Chu, 2002; Inoue et al., 2002; Matyas, 2015), and values over Puerto Rico mainly fell into this range, while mid-level relative humidity averaged

49.8% . The primary component of vertical wind shear was westerly, with a very small meridional component.

Of the 86 TCs analyzed in this study the highest mean TCR associated with a storm was for Tropical Depression Eloise (1975) (Table 2-3). This storm dropped heavy rainfall throughout the island (Figure 2-4) with peaks above 500 mm in the central mountains (Colón, 2009), and flash floods caused 34 fatalities (Hebert, 1976). Eloise

(1975) caused losses of 458 million dollars to the island (Pielke et al., 2003). The second largest rainfall producer in terms of mean TCR was Hurricane Georges (1998), with accumulations of more than 500 mm in the central mountains (Figure 2-4) and causing more than 2 billion dollars in losses (Bennet and Mojica, 1998; Pasch et al.,

2001; Smith et al., 2005) The third and fourth largest mean TCR values (Table 3) were for Hurricane David (1979) and Hurricane Hortense (1996). David (1979) was responsible for 200 million dollars in losses and 7 fatalities in the island while torrential rains associated with Hortense (1996) caused flash floods that ended the lives of 18 people and had associated damages of 128 million dollars (Herbert, 1979; Pasch and

Avila, 1999). The four major TCs exhibited different tracks and patterns of mean TCR over the island (Figure 2-4). Eloise and Georges had higher rainfall accumulations in the central and western areas of the island while Hortense exhibited larger TCR values in the eastern and David in the southeast and southwest regions of the island. All four TCs exhibited unique tracks, with Hortense and Georges making landfall in different locations of the island, while Eloise and David had farther tracks located north and south of the island.

The storm with the highest maximum observed precipitation in this study was

Tropical Depression Isabel (1985), which is responsible for dropping the torrential rains that caused the Mameyes landslide that ended the lives of 130 people (Jibson, 1989).

The other three maximum TCR events were Eloise (1975), Georges (1998) and

Hortense (1996) (Table 2-3). It is important to note that the maximum observed precipitation for some TCs does not coincide with the values in the official reports of the storms. This study used 32 stations with records from 1970-2010 and in some instances the actual maximum precipitation value was not recorded in the selected rain gauges.

For example, Hurricane Georges (1998) had accumulations exceeding 750 mm in the central mountains, yet the max TCR in this study was lower (577 mm).

Before discussing the correlation analyses and PCR results it is important to note that this study employed a 50 mm daily rainfall threshold to define heavy precipitation events (Groisman et al., 2004). This threshold has been used in Puerto Rico by Jury and Sanchez, (2009) who defined a flood event as when the average of all rain gauges exceeds 50 mm of rainfall over the daily basis. While 23 TCs were associated with mean TCR values of 50 mm or higher, only nine produced more than 100 mm over the island (Table 2-3). Fifteen of the 23 TCs were associated with maximum TCR values of

200 mm or more. Seven of the eight TCs that made landfall were associated with mean

TCR values of more than 50 mm, however only Georges (1998), Hortense (1996) and

Jeanne (2004) produced values of 150 mm or more. Four of the eight TCs that made landfall were associated with max TCR values of 300 mm or more. An important finding to note was that 38.3 % of all maximum TCR values were recorded in stations located in

32 the eastern region of Puerto Rico, which suggests that TCR over the island is not evenly distributed, but rather is highly concentrated in specific regions.

Correlation Analyses

Correlation coefficients from a Spearman’s rank test identify the factors TPW,

PRX, MRH, DUR, LON and HTS as the individual predictors with the highest correlations with both measures of TCR (Table 2-4). Larger mean and maximum TCR values were associated with high moisture environments around the island and closer, slower moving TCs. The Spearman’s correlation coefficient calculated between proximity and both measures of moisture was highly significant and negative (Table 2-

5), indicating that TCs passing closer to the island can bring more moisture over Puerto

Rico which can lead to higher rainfall totals.

Of the TCs associated with mean TCR values of 50 mm or more 91 % were within a 233 km radius from the island at their closest approach, while 55% of those associated with 100 mm or more came within 70 km of Puerto Rico (Figure 2-5a).

Maximum TCR values for 90% of the storms with 300 mm or more were associated with

TCs that passed within a distance of 220 km from the island. Closer moving storms spent more time within 500 km of the island which explains why 73% of the cases with durations of 42 hours or more exhibited mean TCR values of 50 mm or higher (Figure 2-

5b). These results also show similarities with other studies that reported a decrease in rainfall as distance from the cyclone center becomes larger (Simpson and Riehl, 1981) with higher rainfall amounts of 150 mm or more located approximately 40 to 80 km from the center (Goodyear, 1968). Storm centers at a distance of 233 km or more exhibit a decreasing trend in both TCR measures over the island (Figure 2-5a). However,

Tropical Storm Odette (2003) was associated with mean TCR values of 50 mm or more while its center was located 326 km from the island. Also Hurricane Floyd (1999) had a maximum TCR value that exceeded 150 mm, with its center located 489.25 km from land. These cases support the results found by Lonfat et al., (2004) which show that even at 300 km from the storm center, high rain rates can still be observed.

When it comes to the moisture distribution in the surrounding environment it was found that TPW values of exceeding 44.5 mm and MRH percentages of 44% or above were related to storms that produced mean TCR values of 50 mm or more over the island (Figure 2-6a, b). Similar results were found with the maximum TCR of storms associated with 300 mm or more, with the ten of them exhibiting TPW and MRH environments exceeding 44.5 mm and 44% respectively. About 66% of the storms associated with mean TCR values of 100 mm or more had TPW values exceeding 46 mm and MRH percentages of above 51%. Similar to our findings, Konrad and Perry,

(2010) found that the area in which precipitable water exceeds 50.8 mm or more was strongly associated with high TCR values over the Carolina region in the US. Another study looking at TCs in Southern Baja California found that nearby passing storms were associated with total precipitable water environments of 40 mm or above that lead to moderate and heavy precipitation events (Farfán and Fogel, 2007).

The Spearman’s correlation test also found LON to have a significant negative correlation with TCR over Puerto Rico (Table 2-4). The TCs that had their centers located between the averaged longitudes of -65 º and -66 º showed the largest mean

TCR amounts over the island (Figure 2-7a). TCs associated with maximum TCR values of 300 mm or higher were located close to the longitudes of the island, with 70 % of

34 them located close to -66 º or farther west. About 69% of the storms associated with mean TCR values of 50 mm or more were located at the -65.4 º degree of longitude or to the west close to -66.5 degree of longitude. When it comes to storms related to mean

TCR values of 100 mm or more it was found that 66% of them were located along the -

66.2 degree which falls within the longitudes of Puerto Rico.

Horizontal translation speed (HTS) was also identified as statistically significant by the Spearman’s statistic. Of the storms that produced 100 mm or more of average total rainfall 88% had speeds of 6.4 ms-1 or less (Figure 2-7b). Similar results were found with the maximum TCR, with 90% of the storms associated with extreme events of 300 mm or more exhibiting lower translation speeds of 6 ms-1 or less. The TCs with

HTS values of 7.5 ms-1 or higher show a decreasing trend for both the mean and maximum TCR (Figure 2-7b).

Other studies have found TC intensity and vertical wind shear to be significant factors affecting the amount and distribution of rainfall associated with a storm (Cerveny and Newman, 2000; Matyas, 2010). However, this study found no statistically significant relationship between these factors and TCR variability over Puerto Rico. Of the 86 TCs analyzed in this study more than 72% had averaged intensities associated with tropical depressions and tropical storms and 65 % of the TCs that produced average totals of 50 mm or more fell into these two categories. We also performed calculations utilizing the highest value of the wind speed for each TC, yet no significant relations were found with mean and maximum TCR over the island. When we decomposed vertical wind shear into speed and direction components, we found a negative relationship between

35 northerly shear and mean and maximum TCR, however it was not statistically significant.

Principal Component Regression

The Varimax rotated PCA produced four components with Eigen-values of 1 or more. Each component accounted for 10-25% of the variance in the dataset, with a cumulative total of 71%. Two or three variables loaded onto each component, which are discussed below.

Component one grouped the variables DUR, PRX and HTS. This grouping validates the results of the Spearman’s correlation tests demonstrating that slower moving storms passing close to the island will spend more time within the study region.

As MRH, TPW and LON were found to be highly correlated to one another, it is not surprising that component two groups these factors (Table 2-6). TCs with average longitudes close to or west of the island are associated with higher levels of moisture around and over Puerto Rico. Component three showed that westerly vertical wind shear was associated with TC tracks located south of the island, while component four was associated with higher storm intensity with a stronger southerly component of the vertical wind shear. These four components were entered as predictors in the forward principal component regression (PCR) models of mean and maximum TCR.

The forward PCR procedure identified components one and two as the predictors with the strongest correlations with both TCR measures (Table 2-7).When taken together both components account for 70% of the variability in mean TCR and

52% in the maximum rainfall observations over the island. Component two was the most significant predictor in both models. TCs like Isabel (1985), Grace (1997), Debby

(1982) and Odette (2003) where among the storms associated with high moisture environments that passed farther away from the island (>200 km) and were associated with mean TCR values of 50 mm or more and maximum TCR values of 230 mm or higher (Figure 2-8a, Figure 2-9a). Thus, storm total rainfall does not necessarily decrease if a storm remains farther offshore. Rather, the pool of moisture within which the storm is embedded is the most important factor. Similar to the findings of Jiang et al.

(2008), our results suggest that adding more parameters related to environmental moisture to our current forecast models can improve the estimates of rainfall associated with TCs.

Component one shows that as the distance between the island and the center of the cyclone decreases the duration increases and if storms also exhibit low translation speeds, more rainfall will occur over Puerto Rico during the passage of a TC. The two largest rainfall producers, Eloise (1975) and Georges (1998) and other storms associated with mean TCR values of 50 mm or higher including Frederic (1979),

Hortense (1996), and Jeanne (2004) were associated with circulation centers relatively close to the island, longer duration times and lower translation speeds (Figure 2-8b).

These TCs were also associated with maximum TCR values of 300 mm or more (Figure

2-9b).

Comparisons between the highest and lowest 23 mean and maximum TCR events support the results of both the correlation analyses and PCR models. The twenty three TCs with the highest mean and maximum TCR values where associated with mean TPW values nearly 9 mm higher and MRH values 30% higher than the 23 lowest rainfall producers (Table 2-8). The top mean TCR storms were located at a mean

37 distance of 93.57 km from the island while the average distance of the storms with the lowest precipitation accumulations was 363.11 km (Table 2-8). The mean distance of the maximum TCR events is higher than the average distance of the mean TCR storms by 37.25 km (Figure 2-10a). A visual comparison of the tracks for the TCs with the highest maximum TCR confirms this result (Figure 2-10a, b). When looking at the TPW in the environment around the island it is evident that the top rainfall producers brought more moisture to the area than the lowest precipitation producers. (Figure 2-11a, b). TC duration for both mean and maximum TCR is almost twice as high as the DUR of the lowest 23 rainfall producing storms. The LON for the top rainfall producers is farther to the west and closer to Puerto Rico than the LON of the lowest TCs, which is farther east of the island (Table 2-8). When it comes to HTS both the mean and maximum TCR events have a lower average speed for the high precipitation producing storms and higher speeds of almost 2 ms-1 more for the low rainfall storms. To confirm that the differences in the 23 highest and lowest mean and maximum TCR events were statistically significant, Mann-Whitney U tests were employed to compare the group means of the seven key predictors and all factors were found to be significantly different from one another (Table 2-9).

Concluding Remarks

This chapter presented an analysis of TCR over Puerto Rico to understand how characteristics of the environment and each storm contributed to the variability of mean and maximum storm precipitation. Eighty-six TCs passing within a 500 km radius of

Puerto Rico during 1970-2010 were analyzed. Daily rainfall from 32 stations were used to calculate a mean TCR value for the island and the maximum observed precipitation

38 for each storm. Correlation analyses, principal component regression (PCR) procedures and Mann-Whitney U tests were employed to identify which environmental and storm specific factors were the most significant contributors to mean and maximum TCR variability over the island.

In terms of the precipitation analysis, 23 (9) of the 86 TCs were associated with mean TCR values of 50 (100) mm or more over the island. Four TCs were associated with maximum TCR values of 500 mm or more. Spearman’s statistical tests results show that the individual factors that were mostly associated with mean and maximum

TCR variability over Puerto Rico were TPW, PRX, MRH, DUR, LON and HTS. Results from PCR models show that the component combining moisture and longitude accounts for most of the variability in mean and maximum TCR over the island. Key thresholds for high rainfall production were environments with TPW values greater than 44 mm and/or

MRH values greater than 44% averaged over the 700-500 hPa layer, with storm tracks located predominantly over or west of Puerto Rico in their average longitude position.

TCs that passed close to or over the island that moved slowly and had long durations also accounted for a large portion of TCR variability. High mean (> 50 mm) and maximum (> 300 mm) TCR values resulted from TCs whose centers came within 233 km of land, moved at speeds less than 6.4 ms-1, and/or spent at least 42 hours in the study region. Taken together both of these components accounted for 70% of the variability in the mean TCR model and 52% in the maximum TCR model.

The main contribution of this chapter was the identification of important thresholds associated with heavy rainfall for each of the statistically significant environmental and storm specific factors, which could aid in future TCR forecasting for

Puerto Rico and other islands in the tropics. We found that TPW and PRX account for much of the variability in mean and maximum TCR, yet including the factors MRH,

DUR, LON and HTS through principal components analysis increased our ability to account for TCR variability. Another important finding was that TPW and PRX were grouped in different components in the PCA procedure, which suggests that the amount of moisture and its spatial dispersion varies among TCs and that moisture convergence into the outer rainbands can bring high rainfall similar to convergence in the storm’s core. One more important finding was that TCR variability over Puerto Rico was not found to be associated with storm intensity and wind shear, which is different from the results of other studies. This might be due to the fact that this study was based on a location rather than storm-relative analysis.

Chapter 2 Limitations

The main limitation of this chapter is the sample of rain gauges used to calculate the average total rainfall for each TC that passed within a 500 km from Puerto Rico. The island has a vast network of weather stations that extend back to the 1890’s, however only a few stations have complete data. The time period for this chapter is from 1970-

2010 so radar and satellite based rainfall data can’t be used for a long-term climatology since both are only available for the island from the late 1990’s to present. Another limitation is the low spatial resolution of the environmental variables precipitable water, mid-level relative humidity and vertical wind shear. It may be possible for future studies to utilize data from the North American Regional Reanalysis with its higher spatial resolution to assess atmospheric conditions in the vicinity of TCs (Zick and Matyas

2015a, b).

Figure 2-1. Topography of the main island of Puerto Rico and selected weather stations.

Table 2-1. Storm specific characteristics and environmental factor variables used in this study. Storm Specific Characteristics Abbreviation Units Circulation Center Latitude LAT º Circulation Center Longitude LON º Storm Center Proximity PRX km Storm Duration DUR hrs. Maximum Sustained Winds VMX ms-1 Horizontal Translation Speed HTS ms-1 Environmental Factors Abbreviation Units Total Precipitable Water TPW mm Mid-level Relative Humidity (Avg. 500-700 hPa) MRH % Wind Shear (850-200 hPa) West To East WSU ms-1 Wind Shear (850-200 hPa) South To North WSV ms-1

Figure 2-2. Tracks of all 86 TCs within a 500 km radius and 100 km range rings of Puerto Rico from 1970 to 2010.

Table 2-2. Descriptive statistics of storm specific and environmental factors associated with the 86 TCs analyzed. Minimum Maximum Mean LAT 13.5 22.8 17.9 LON -70.6 -61 -65.5 PRX* 1 499.9 239.7 DUR 12 102 41.5 HTS 3.15 13.17 6.3 VMX 11.11 72.28 28.3 TPW 30.31 53.74 44.7 MRH 23.45 74.32 49.8 WSU -10.79 32.08 9.9 WSV -16.30 12.46 .22 *1 means landfall 42

Figure 2-3. Total precipitable water for September 21, 1998 during the passage of Hurricane Georges.

Table 2-3. Tropical cyclones that produced more than 50 mm rainfall over the island of Puerto Rico. TC intensity is based on the Saffir-Simpson scale which classifies the storms by maximum sustained winds with the tropical depression (TD) been the weakest and hurricane category 5 (H5) been the most intense. Max Mean TCR Maximum TCR TC Month/Year Intensity (mm) (mm) Eloise TD 09/1975 279.15 591.80 Georges* H3 09/1998 271.43 577.80 David H5 08/1979 237.56 382.60 Hortense* H1 09/1996 209.74 552.20 Jeanne* TS 09/2004 190.35 370.80 Isabel TS 10/1985 186.72 690.10 Chris TD 08/1988 158.91 304.50 Grace TS 10/1997 122.35 257.60 Frederic* TS 09/1979 106.28 360.20 Olga* TS 12/2007 99.89 209.80 Lenny H3 11/1999 99.37 235.50 Claudette* TD 07/1979 98.93 232.00 Debby H1 08/2000 94.86 212.10 Debby TD 09/1982 85.96 235.00 Hugo* H4 09/1989 84.14 285.80 Marilyn H2 09/1995 76.35 132.10 Dean TS 08/2001 73.21 177.10 Klaus TS 11/1984 72.01 179.50 Mindy TS 10/2003 66.92 173.30 Carmen TD 08/1974 64.76 150.10 Odette TS 12/2003 58.43 217.90 Earl H3 08/2010 57.35 106.40 Luis H4 09/1995 53.2 120.40 *Tropical cyclones that made landfall.

Figure 2-4. Kriging interpolated surfaces for visualization purposes of the storms with the highest mean TCR values and their respective 100 km radii around the island. A) Tropical depression Eloise (1985), B) Hurricane Georges (1998), C) Hurricane David (1979) and D) Hurricane Hortense (1996).

Table 2-4. Spearman’s correlation coefficients for each of the predictor’s relationship with mean and maximum TCR. Correlation Coefficients Correlation Coefficient Factors Mean TCR Significance Maxi TCR Significance TPW 0.694 0.000 0.700 0.000 PRX -0.703 0.000 -0.537 0.000 MRH 0.548 0.000 0.479 0.000 DUR 0.556 0.000 0.342 0.001 LON -0.383 0.001 -0.359 0.001 HTS -0.278 0.010 -0.228 0.035 VMX 0.112 0.306 0.071 0.514 WSV -0.099 0.364 -0.191 0.078 WSU 0.057 0.599 0.071 0.517 LAT 0.015 0.887 0.027 0.807

Figure 2-5. Scatter plot of the relationship between mean and maximum TCR and (a) center proximity to the island and (b) duration.

Table 2-5. Spearman’s correlation coefficients of variables that were found to be significantly correlated with mean and maximum TCR. TPW PRX MRH DUR HTS LON TP 1 W PR -0.334** 1 X MR 0.545** -0.262* 1 H DU 0.274* -0.624** 0.188 1 R HT -0.268* 0.138 -0.139 -0.529** 1 S LO -0.385** 0.010 -0.436** 0.124 -0.122 N 1 ** Correlation is significant at the 0.01 level. * Correlation is significant at 0.05 level.

Figure 2-6. Scatter plot of the relationship between mean and maximum TCR and (a) total precipitable water and (b) mid-level relative humidity.

Table 2-6. Varimax rotated principal component analysis (PCA) results. Data includes the number of components its % of variance, cumulative variance and variable loadings. % of Cumulative Components Variables Loadings Variance % DUR 0.89 1 25.39 25.39 PRX -0.74 HTS -0.66 MRH 0.79 2 20.20 45.60 LON -0.77 TPW 0.77 LAT -0.88 3 15.09 60.70 WSU 0.87 VMX 0.85 4 10.64 71.34 WSV 0.69

Figure 2-7. Scatter plot of the relationship between mean and maximum TCR and (a) longitude and (b) horizontal translation speed.

Table 2-7. Forward principal component regression model results for mean and maximum TCR. Model R R2 Adj. R2 Comp Factors Coeff. Sig. MRH 2 LON .605 0.000 Mean TCR .84 .70 TPW .701 2 8 DUR 1 PRX .585 0.000 HTS MRH 2 LON .600 0.000 Max TCR .72 .53 TPW .519 8 0 DUR 1 PRX .403 0.000 HTS

Figure 2-8. Scatter plot of the relationship between mean TCR and (a) component 2 (DUR, PRX and HTS) and (b) component 1 (MRH, LON and TPW).

Figure 2-9. Scatter plot of the relationship between maximum TCR and (a) component 2 (DUR, PRX and HTS) and (b) component 1 (MRH, LON and TPW).

Figure 2-10. Tracks of the (a) top and (b) bottom 23 mean and maximum TCR events.

Table 2-8. Statistics of the 23 highest/lowest mean and maximum TCR events. Highest Mean TCR Highest Maximum TCR Minimum Maximum Mean Minimum Maximum Mean TCR 53.20 279.15 123.82 157.50 690.10 313.82 TPW 40.00 53.74 48.56 40.00 53.74 48.59 PRX 1.00 326.60 93.57 1.00 489.25 130.82 MRH 43.91 74.32 55.85 37.78 74.32 55.42 DUR 12.00 102.00 55.57 12.00 102.00 53.73 LON -70.60 -63.60 -66.38 -70.6 -64.1 -66.49 HTS 3.44 11.87 5.81 3.44 8.28 5.69 Lowest Mean TCR Lowest Maximum TCR TCR 0.65 10.52 4.64 5.60 59.70 37.74 TPW 30.31 47.85 40.07 30.31 44.25 39.90 PRX 104.65 499.90 363.11 104.65 499.90 346.61 MRH 23.45 56.64 41.36 23.45 59.87 42.98 DUR 12.0 60.0 23.73 12.00 72.00 30.00 LON -68.0 -61.0 -64.28 -68.0 -61.8 -64.45 HTS 3.15 13.17 7.39 3.15 13.17 7.25

Table 2-9. Mann-Whitney U tests results for the 23 highest/lowest mean and maximum TCR events.

Mean TCR Factors TCR TPW PRX MRH DUR LON HTS Mann- 0.00 24.00 29.00 42.50 80.00 122.50 144.00 Whitney U Z-score -5.81 -5.28 -5.18 -4.87 -4.07 -3.12 -2.64 Significance .000 .000 .000 .000 .000 .002 .008 P-value Maximum TCR Mann- .000 24.00 65.00 89.50 128.50 119.50 165.00 Whitney U Z-score -5.81 -5.28 -4.39 -3.84 -3.00 -3.18 -2.18 Significance .000 .000 .000 .000 .003 .001 .029 P-value

Figure 2-11. Kriging interpolated surface for visualization purposes of average total precipitable water (TPW) for the 23 (a) top and (b) bottom mean TCR events.

CHAPTER 3 SPATIAL DISTRIBUTION OF TROPICAL CYCLONE RAINFALL AND ITS CONTRIBUTION TO THE CLIMATOLOGY OF PUERTO RICO

Rainfall associated with the passage of tropical cyclones (TCs) is one of the most complex geophysical phenomena to estimate or model. When it comes to understating tropical cyclone rainfall (TCR) patterns at any given area, factors like storm location, environmental moisture distribution and local topographic effects need to be considered

(Anthes, 1982). When nearby or landfalling TCs interact with steep terrain the distribution of rainfall over space is even more complex and this causes a lot of issues in areas that often lack the measurements to develop rain estimates (Prudhomme and

Reed, 1998). Tropical depressions, storms and hurricanes bring a lot of rainfall which might lead to the loss of life and property when floods and mudslides are triggered by heavy rains. In fact, in the US the leading cause of death from storms is associated with flash floods caused by torrential rain from TCs (Rappaport, 2000). On the other hand, understanding the spatial distribution of rainfall associated with TCs and their contribution to the overall precipitation is important, since these storms are an indispensable source of water for ecosystems, communities and economic activities

(Dare et al., 2012). One location that exhibits those characteristics is Puerto Rico, which is mostly exposed to TCs during the months of August to October and is characterized as being an area where topography plays a key role in how rainfall is spatially distributed over the island (Pico, 1974). Puerto Rico has experienced multiple floods and mudslides that were triggered by heavy rain associated with TCs. For example,

Tropical Storm Eloise (1975) and Hurricanes David (1979), Hortense (1996) and

Georges (1998) triggered flash floods and mudslides that caused combined losses of

53 more than 2.5 billion dollars and more than 50 fatalities over the island (Hebert, 1976;

Hebert, 1980; Bennet and Mojica, 1998, Pasch and Avila, 1999).

Studies that look at the spatial distribution of extreme rainfall events associated with storms (non-TC and TCs) and their relation to topography have been conducted in places like the United Kingdom and Taiwan which like Puerto Rico are also characterized for their complex topography. Prudhomme and Reed, (1998) found that the spatial distribution of extreme rainfall events in Scotland does not reflect a simple relationship with elevation; they found more complex relations with topography and location relative to sources of moisture. Brunsdon et al., (2001) study on the relationship between topography and precipitation in the UK shows that there seems to be a steep increment in the rainfall-elevation coefficient when you move from the southeast of

England through the Midlands to the north and northwest. These studies show that the precipitation-elevation relationship is not a simple one, since the patterns of rainfall over a given area depend on multiple factors unique to the study area’s geography. A study conducted by Cheung et al., (2008) found that all 62 TC observations on Taiwan during

1989-2002 showed significant relations with topography, since the orographic effect on rainfall was present in all storms. Yu et al., (2002) study on Typhoon Herb in Taiwan shows that the existence of the central mountain range plays an important role in significantly increasing the total rainfall amounts by enhancing the lifting mechanism of humid moist air over the rising mountains. Like Taiwan, the island of Puerto Rico exhibits complex interactions between topography, the predominant easterly winds and rainfall from different precipitation generating processes in which TCs play an important role (Colon, 2008). For that reason it is important to examine the spatial distribution of

54 precipitation associated with TCs and their overall contribution to the rainfall climatology of the island.

Other TCR studies have focused on the contribution of these storms to the overall precipitation of specific regions. Studies looking at TCs contribution to rainfall in the north Pacific and north Atlantic using satellite-based estimations found that precipitation associated with storms accounted for 40% of the rain off the coast of Baja

California and 30% northeast of Puerto Rico during the months of June to November

(Rodgers et al., 2000; Rodgers et al., 2001). Several studies using weather station networks have focused their attention on TCR contribution in the southeastern US, where it has been found that extreme precipitation events associated with TCs are exhibiting an increasing trend over the last decades (Knight and Davis, 2009). Others have found that 15% of the hurricane season rain in the Carolinas and 10% over Florida can be attributed to TCs impacting the region (Knight and Davis, 2007). Konrad and

Perry, (2010) conducted a study using rain gage data that looked at storm rainfall contribution and its relations with atmospheric conditions in the Carolinas and found that

90% of the heavy precipitation in the eastern part of the region was associated with

TCs. Another study that focused in the Eastern US found that September is the month with the highest TCR contribution (30%), especially in the coastal areas of Alabama,

Virginia and North Carolina (Nogueira and Keim, 2011). Dare et al., (2012) found that

TCs contribution to the rainfall climatology of Australia varies widely over space and time, with some locations receiving 10% and other getting as much as 60%. It’s important to note that TCR contribution is spatially unique to each area affected by these phenomena, therefore understanding the spatial and temporal characteristics of

55 that contribution is important for drought monitoring and water management in Puerto

Rico and other places in the tropics.

No study has looked at the spatial distribution of TCR over Puerto Rico or to the storms contribution to the overall rainfall climatology of the island. Some studies have focused on the role that individual TCs like hurricanes Hugo (1989), Hortense (1996) and Georges (2008) have played in extreme flood events (Torres-Sierra, 1997; Larsen and Santiago-Roman, 2001; Smith et al., 2005). However, researchers have explored statistical methods for forecasting convective rainfall and have also employed factor analysis regionalization to identify convective precipitation regions over the island

(Carter and Elsner 1996, Carter and Elsner 1997). Another study looked at mapping the climate of Puerto Rico by employing the parameter elevation regressions on independent slopes model (PRISM) in which they found a complex relationship between rainfall, elevation, upslope exposure to predominant moisture bearing winds, and proximity to the ocean (Daly et al., 2003). A more recent study conducted by Hernandez and Matyas, (2015) found that TCR variability over Puerto Rico is affected by environmental moisture distribution, storm center distance from the island’s coast, storm duration and its horizontal translation speed. These findings suggest that closer, slower moving TCs over high moisture environments tend to be associated with heavy rainfall over Puerto Rico. However, the spatio-temporal characteristics of rainfall associated with the passage of TCs and their contribution to the overall precipitation was not examined in that study.

The aim of this chapter is to explore and understand the spatial distribution of

TCR and its contribution to the rainfall climatology of Puerto Rico. The first question

56 seeks to understand the spatial distribution of rainfall over the island, is TCR randomly distributed or is it clustered in some specific regions? The first hypothesis is that high

As warm moist air associated with the passage of a nearby TC starts to go upslope over the eastern facing hills of the island more rainfall is going to fall in those regions due to the orographic enhancement of precipitation. Understanding the spatial distribution of

TCR is important for identification of areas exposed to heavy rainfall that could lead to extreme floods and mudslides. The second problem aims to understand how much TCs contribute to the rainfall climatology of the island. This problem is explored both spatially and temporally by examining the percentage of rainfall contributed by TCs for each of the hurricane season months (June-Nov). Given that rainfall from tropical cyclones is an indispensable source of water for ecosystems, communities, agricultural and industrial activities is important to understand how dependent the different regions of the island are to TCR. The questions here are; do TCs have different contributions to the rainfall climatology of the island in different regions and do some hurricane season months exhibit higher or lower rainfall contributions from storms? The main hypothesis here is that the eastern and southern regions of the island have the largest contribution percentages in the peak hurricane season months of August, September and October.

The east is the first area that interacts with much of the TCs, while the south is the

57 driest region of the island, so any precipitation generating process would have a substantial contribution to its rainfall climatology.

Data

The first step was to identify the TCs that might have brought rainfall over Puerto

Rico. TC tracks data were obtained from the International Best Track Archive for

Climate Stewardship (IBTrACS) (Knapp et al., 2010) for the time period of 1970 to 2010.

All six-hourly positions of TCs that were at least tropical depressions and spent 12 hours or more within a 500 km radius around the island were considered for analysis, and based on those parameters 86 storms were identified (Figure. 3-2). The 500 km radius has been implemented in recent studies that examined rainfall associated with the passage of TCs in Texas and Puerto Rico (Boose et al., 2004; Zhu and Quiring,

2013; Hernandez and Matyas, 2015). Daily and monthly rainfall totals were obtained from the National Climatic Data Center (NCDC) for 32 rain gauges (Table 3-1) located on the main island of Puerto Rico for the years 1970-2010 (Figure. 3-1). Several of the stations that were used in this study were operational until 2010 and for that reason the study was limited to consider data until that year. The 32 stations have at least 80% of the daily/monthly rainfall observations for all TCs for the 1970-2010 period. Digital elevation model (DEM) data for Puerto Rico were obtained from the United States

Geological Survey (USGS). The percent slope (change in elevation) and aspect

(downslope direction) of the DEM were calculated in a Geographic Information System

(GIS).

An accumulated TCR value was calculated for all 32 weather stations for each of the 86 TCs that at any given point in their track passed within a 500 km radius of Puerto

Rico. If a storm spent 12 to 24 hours within the 500 km radius, only one day of rainfall was considered in the TCR calculation for that storm. If a TC spent more than one day inside the radius of influence, then another rainfall day was added to the accumulated

TCR value. TCs ranged between 12 hours to 4 days within the radius, and those times were used to calculate the accumulated TCR value for each storm over all 32 weather stations in the island. A similar calculation was implemented in Hernandez and Matyas,

(2015), yet in that case, an island average value was calculated from the accumulated

TCR of all stations, which completely eliminated the spatial component of storm rainfall over the island. Despite the one-hour discrepancy between when daily rainfall data is recorded (1100 UTC) and the nearest synoptic time for TC data (1200 UTC), results did not vary significantly if one day was added or subtracted from either side of the study window.

Previous studies have shown that environmental moisture has a strong relationship with storm size and the amount of rainfall that a TC can produce (Hill and

Lackmann 2009, Matyas and Cartaya 2009, Zick and Matyas 2015). Therefore it is important to examine the moisture content of the atmosphere for TCs that produce heavy rainfall that could lead to floods and mudslides. For that reason the focus in this study is to understand the spatial distribution of rainfall and it’s variability by examining the precipitation of TC events at different storm center distances from the island’s coast and over varying moisture environments. Total precipitable water (TPW) data were obtained from the National Center for Environmental Prediction and National Center for

Atmospheric Research (NCEP/NCAR) reanalysis data set for all 86 TCs that impacted

Puerto Rico during the time period (Kalnay et al., 1996). An average (TPW) value was

59 calculated from the closest four cells to the island for each TC day. These closer cells were chosen in order to limit the environment to the areas that might have influenced

TCR over Puerto Rico, not over the entire area encompassing the TCs. Storm center proximity to land (PRX) was calculated in a GIS by measuring the distance between the island’s coast and the TC tracks at their closest approach to the island. Both of these variables were used to place the into four TCs in different groups that had varying moisture environments and storm centers located at different distances from the island’s coast. This was done to test the idea that different spatial patterns of TCR will be observed as TCs with different TPW and PRX values impact the island.

The third step was to calculate five different means for all 32 stations for the 86

TCs. Proximity to the center of the storm and moisture are the most important factors when it comes to overall TCR variability over Puerto Rico, and for that reason the groups of TCs were divided considering those two key variables. Before dividing the

TCs in different groups all 86 events were considered, this was done to examine the overall spatial distribution of rainfall associated with all the TCs that impacted Puerto

Rico during the 1970-2010 period. The next four groups had fifteen unique TCs, this was done to limit the groupings to storms that had similar PRX and TPW values. The first mean calculation grouped the TCs that were associated with closer proximity (PRX) to the island’s coast (< 230 km) and high total precipitable water (TPW) values (> 44.5 mm). Those PRX and TPW thresholds were identified in Hernandez and Matyas,

(2015) as the break points between high and rainfall producing TCs for Puerto Rico.

Two additional studies also support the selection of 230 km as a threshold distance.

Matyas (2010) found that hurricane rain fields extended an average of 223 km from the

60 circulation center as measured by land-based radar, and Guo and Matyas (2016) found that the wind fields of TCs extended an average of 226 km from the storm center while over the ocean.

The second mean value considered the TCs that were farther away from the island’s coast with PRX values of > 230 km, yet had TPW values of > 44.5 mm. The third mean value grouped the TCs that had their circulation centers within 230 km from the island’s coast and that were over lower TPW environments of less than 44.5 mm.

The last group consists of the mean of the TCs that were associated with farther PRX values of >230 km and lower TPW of < 44.5 mm. The five mean calculations for each of the 32 stations were then used as the input data in the next step that dealt with using spatial interpolation techniques to generate modeled TCR surfaces from the TC groups.

Geo-statistical Methods

In order to examine the spatial distribution of TCR over Puerto Rico this study employed two spatial interpolation techniques that belong to the kriging family of geostatistical models. The first method is ordinary kriging (ORK), a technique that generates an estimated surface from a given number of data points with known values that also assumes that the mean is constant but unknown (Isaaks and Srivastava, 1989;

Oliver and Webster, 1990). Similar to other members of this geo-statistical family, ORK deals with a two-step process. First it estimates the spatial autocorrelation of the data by fitting a model in the empirical semivariogram and then it uses the resulting calculations to predict the unknown values (McBratney and Webster, 1986). This study chose the widely used spherical model to fit to the empirical semivariogram of TCR over

Puerto Rico since this function assumes linear behavior at the origin and it has been

61 found to be a good fit in other rainfall studies (Goovaerts, 2000; Mair and Fares 2011).

Deterministic methods like natural neighbor (NN) and inverse distance weighted interpolation (IDW) have also been employed in several studies that have estimated rainfall at unknown locations (Dirks et al. 1998; Zimmerman et al. 1999), however, they do not offer the advantage of having calculations of prediction errors like kriging. Also, the NN and IDW methods have been found to perform better in locations with a high density of weather stations, while kriging has been found to provide better estimates in areas with lower observations (Ly et al., 2013). ORK provides prediction errors and better estimates at low station densities, for that reason this study didn’t employ simpler methods like NN and IDW to estimate TCR over Puerto Rico. ORK with first order trend removal was implemented on all five TC mean groups. These was done since exploratory analysis identified the presence of linear trends in all TC groups and data stationarity is an assumption that needs to be satisfy when using ordinary kriging (Ly et al., 2013).

The other technique employed in this study is ordinary cokriging (OCK), a multivariate extension of kriging (Goovaerts, 1997) that was implemented to examine the TCR-topography relationship. This model was chosen since it attempts to predict in this case rainfall, just like ordinary kriging, yet it uses information in the covariate

(elevation, slope and aspect) in an attempt to do a better job of predicting TCR over

Puerto Rico. A study that examined annual and monthly rainfall relations to elevation in

Portugal found that OCK and other multivariate geo-statistical algorithms outperformed

IDW and ORK (Goovaerts, 2000). Another study that focused on the relationship between topography and rainfall patterns in southern Italy found that adding elevation

62 data as an auxiliary variable improved the estimates when compared to prediction done by ORK (Diodato, 2005). Some researchers suggest that OCK should be implemented when the correlation between rainfall and elevation is 0.75 or higher (Goovaerts, 2000;

Mair and Fares 2011). However, another study in the Basin of Mexico found that the interpolation of daily climatic variables improved even when the correlation with elevation was low (Carrera-Hernandez et al., 2007). Other studies looking at interpolating daily rainfall found that adding radar data as an auxiliary variable in the

OCK method improved the model estimates (Creutin et al., 1988; Schuurmans et al.

2007; Ly et al., 2013). This study did not incorporate radar as an auxiliary variable in the

OCK model since data only goes back to 1997 for Puerto Rico. Both geo-statistical interpolation methods were checked by a cross-validation procedure that removes one observation at a time and then estimates the value at that point using data from the other locations. This process was repeated with all 32 stations and the resulting root mean square error (RMSE) values were used to determine the accuracy of the methods. Values closer to zero indicate the more accurate the model. Comparing

RMSE statistics from the different models provided the necessary information to determine which model is giving the best estimate of TCR over Puerto Rico.

After exploring the spatial distribution of rainfall associated with TCs, the study focused on understanding how much do these events contribute to the precipitation climatology of Puerto Rico. An accumulated TCR value for each of the 32 stations during the hurricane season months (June-November) over the 1970-2010 period was calculated. The rainfall for the entire day was considered for storms that spent 12 to 24 hours within the 500 km radius. All the rain that fell during the time of passage of the

63 storm within the 500 km radius was considered as part of the precipitation associated with the TCs. This station accumulated TCR was then used in combination with the monthly total rainfall to calculate a percentage of contribution to the precipitation climatology of the island. The contribution of rainfall associated with TCs was estimated for each month of the hurricane season by calculating a percentage between monthly total rainfall and accumulated TCR for all storms over the 1970-2010 period. For example, if a station had a total monthly value of 10,000 mm for the given time period and an accumulated TCR of 800 mm, the percentage of contribution of TCs to overall precipitation on that rain gauge is 8%. A similar method was employed to study the contribution of TCs to the rainfall climatology of the southeast US (Knight and Davis,

2007) and Australia (Dare et al., 2012). This step was repeated for all 32 weather stations over all Hurricane season months (June-November). After the percentage for each month was calculated for all weather stations the next step was to generate interpolated surfaces of the percentages for each month.

Results and Discussion

Characteristics of TC Groups

The TCs were divided in four groups of 15 unique storms that had similar characteristics in PRX and TPW values (Fig 3-3.) before implementing the geostatistical interpolation techniques (Table 3-2). The first group combines the TCs with the closer circulation centers to the island’s coast and the highest precipitable water values. Heavy rainfall producing TCs like Hortense (1996) and Georges (1998) were among the storms in the first group. Combined, these TCs caused damages of more than 3 billion dollars over the island and were responsible for 20 fatalities (Bennet and Mojica, 1998; Pasch

64 and Avila, 1999). Group two brings together the TCs that also had high TPW values of >

44.5 mm, yet were farther away from the islands coast with PRX values of > 220 km

(Figure 3-3). In group three the TCs that were relatively close to the island < 270 km, yet had TPW values of less than 44.5 mm were combined (Figure 4). Tropical depression

Eloise (1975) which was identified by Hernandez and Matyas, 2015 as the TC with the highest mean TCR for the whole island belonged to this group of storms that were relatively close to land in their closest approach, while been associated with low TPW environments. The fourth group brings together the TCs that were the farthest away from the island’s coast (> 275 km) and had lower moisture environments of 43.5 mm or less (Figure 3-3). None of the TCs belonging to this last group have been identified as heavy rainfall producers in the island by previous studies. All of the TC groups exhibited similar trajectories, with the vast majority moving southeast to northwest or east to west around or over the island (Figure 3-4). Only tropical storms Klaus (1984), Iris (1995) and

Odette (2003) showed tracks that did not move in the typical east to west trajectory that most storms exhibited.

Spatial Distribution of TCR

Results from the ordinary kriging interpolation show that high values of mean

TCR over Puerto Rico are concentrated in the east, southeast and central mountains region of the island, when all 86 events are considered (Figure 3-5). As most TCs move from east to west when passing within 500 km of Puerto Rico, this spatial pattern reflects that most of the rainfall is concentrated in the areas that first interact with the storms rainbands or convective cells in the eyewall, for the ones that move closer to land. A general decrease in mean TCR is observed as we move to the north, west and

65 southwestern regions of the island. The OCK predicted surface shows a more detailed spatial pattern of TCR, with highly elevated and steep slope areas in the east, southeast and central mountains exhibiting pockets of higher values (Fig 3-5b). The orographic component is present, specifically in the eastern region close to El Yunque National

Rainforest and towards southeastern mountains (Fig 3-5b). When comparing the ORK and OCK methods in terms of its RMSE, it is evident that adding elevation and slope as auxiliary variables improved the model estimates of mean TCR when all 86 events are considered since the OCK method yielded a lower RMSE. These results are similar to the ones found in other studies that suggest that adding auxiliary data to estimate rainfall improves the model estimates (Goovaerts, 2000; Mair and Fares, 2011).

In group 1 the eastern and east central regions of the island were shown again as the areas that received much of the rainfall associated with those TCs (Figure 3-6a).

These TCs had higher TCR values when compared to other storm groups considered in this study, with the highest value being 181 mm which was found in the highly elevated areas of the southeast. Again, this supports the idea that TCs that move east to west tend to drop most of the rainfall over the area of the island that they encounter first. The eastern part of the central mountains also exhibits some high rainfall values over steep eastern facing slopes (Figure 3-6b). Is not until reaching the west that low rainfall values over both high elevation areas and the coastal plains are found, and this is due to the fact that most of the rain associated with the storm has already been deposited in the east, southeast and central mountains region. Cross-validation results suggest that the

OCK method outperformed the ORK, which shows that adding elevation, slope and

66 aspect as covariates enhances prediction of rainfall associated with closer circulation centers and higher moisture TCs over Puerto Rico.

The group that combined the TCs with high moisture environments and farther circulation centers (group 2) exhibits some differences in spatial patterns of TCR and its amounts over the island when compared to the results of the higher moisture and closer circulation (group1) center storms (Figure 3-6c). This group exhibits lower predicted

TCR amounts, with 111 mm being the highest rainfall value found in the southeastern region of the island. The highly elevated areas in El Yunque National Rain forest also exhibit lower values when compared to the results of the high moisture and close proximity TC group (Figure 3-6c, d). In general group 2 exhibits high TCR values in the southeast, while values lower than 61.2 mm dominate in the rest of the island. It is evident that farther TCs, even with high moisture environments over Puerto Rico, tend to be associated with lower rainfall amounts over land when compared to the closer and higher moisture TCs. Cross-validation results based on RMSE comparison between models also suggest that adding elevation and slope as covariates enhances model prediction of rainfall associated with higher moisture environments and TCs with farther circulation centers from land (Figure 3-6d).

The TCs that were close to the island’s coast and were over low TPW environments exhibit larger values of predicted rainfall amounts when compared to the

TC group that had farther distances and high moisture values (Figure 3-7). The highest rainfall values for this group of TCs were found in the east, central southeast and the western interior region of the island. These closer TCs over lower moisture environments exhibit a more random spatial distribution of precipitation over the island,

67 instead of the usual cluster of high-low values found in the groups that considered all 86 events and the high moisture-close proximity storms. Outside those three areas in which high values are concentrated, the rest of the island shows lower values that range from 40.6-63 mm of rainfall, which were only previously found in the northwest and southwest regions of the island when all 86 TCs and close-high moisture storms were considered. Cross-validation results yielded a lower RMSE value for the OCK method which once again suggests that adding elevation and slope enhances the prediction of

TCR (Figure 3-7b).

The TCs that had farther circulation centers from land and low moisture environments exhibited the lowest TCR values of all of the TC groups (Figure 3-7c, d).

The OCK model of TCR for the farther and lower moisture events was found to perform poorly when compared to RMSE of the ORK model, showing that adding aspect as an auxiliary variable did not improve the prediction of TCR for these events. This might be due to the fact that the rainfall associated with those storms had a higher spatial variability and was not showing a strong relationship with topography. Even though all four interpolated surfaces of the TC groups exhibited varying spatial patterns of TCR and different high to low ranges one observation is clear, the highest rainfall values associated with the passage of TCs were found in the southeast and central regions of the island, while a general decrease is evident towards the west. It is important to note that the OCK method performed better than the ORK in most TC groups with exception of the group of farther and lower moisture storms, which coincides with the literature that suggests that adding information about topography (elevation, slope and aspect) as covariates improves the models prediction capabilities in some of the cases (Ly et al.,

2013). It is important to note that all possible combination of the covariates were implemented in the OCK models, yet only the ones that yielded the lowest RMSE scores for each of the storm groups in the cross-validation procedures were reported here.

Another factor that might cause different spatial patterns of TCR over the island is the storm tracks location relative to Puerto Rico. Regardless of their TPW and PRX values, TCs were divided in two group means that combined 22 unique storms that were located to the north and to the south of Puerto Rico and that were at least 100 km from the island’s coast at their closest approach. ORK surfaces of the TCs that were located to the north (Figure 3-8a) show the southeast and the western interior as the areas with the higher TCR values. The mean TCR values for the TCs that were located south of the island exhibit the highest rainfall values in the southeastern region of the island, higher than the values for the same region in the north (Figure 3-8b). When the south TCs are subtracted from the north TCs two different patterns emerge (Figure 3-

8c). Storm tracks south of Puerto Rico produce more rainfall (13.6-24 mm) in the southeastern interior, northeast and north west while TCs north of the island are associated with higher values (4.28-19.5 mm) in the western interior region of the island. These patterns could be the product of the counter clockwise circulation of the

TCs and their interactions with the island’s topography. These results identify similar high TCR regions when compared to the ones in Group 2 and 3, with the southeast exhibiting higher rainfall values in Group 2 and the western interior exhibiting higher values in Group 3.

TCR Contribution

When it comes to the TCs contribution to the rainfall climatology of the island it is evident that some regions are more dependent on TCR than others. The spatial pattern of average rainfall over Puerto Rico during the hurricane season months (June-Nov) for the 1970-2010 period exhibits two areas of the island with high precipitation values, these are the east and the western interior (Figure 3-9a). From our spatial distribution analysis we know that the southeast tends to be the area were most of the high values of TCR tend to be concentrated for much of the TCs that impacted Puerto Rico.

However, we found that much of the western region of the island tends to exhibit low values of TCR, so this pattern of high rainfall values in the west is not necessarily connected to TCs. The high precipitation values in the western interior might be due to a combination of processes that include localized convective rainfall, sea breeze dynamics, west to east moving troughs and easterly waves that never developed to become a TC (Colon, 2008). The hurricane season (June-Nov) contribution percentage surface exhibits values between 10.8-15.5% for some areas located in the southern region of Puerto Rico (Fig 3-9b). This is expected since the south is the driest area in the island, so any precipitation generating process is important for its hurricane season rainfall climatology. The south and the southern slopes of the central mountains exhibit

TCR contribution percentages of 10.8 to 13.4% (Figure 3-9b). The highest contribution percentage of 15.5 % is found in a small area in the southern region of the island, which is the heart of the agricultural industry of the island.

When it comes to average rainfall and TCR contribution patterns of the months of

July and August there are similarities and differences in spatial patterns from one month

70 to another (Figure 3-10). The average rainfall surfaces for both months show clusters of high precipitation values in the east and western interior region of the island (Figure 3-

10a, c). It is important to note that June was not reported here since only one TC impacted the island in that month during the 1970-2010 period. July tends to be one of the drier months of the hurricane season over much of Puerto Rico because of the

Caribbean mid-summer drought which deals with the position of the North Atlantic high pressure (NAHP) (Gamble et al., 2008). July is the month with the lowest rainfall contribution of TCs for Puerto Rico (Figure 3-10b) since not a lot of storms develop in the central Atlantic during that time of the hurricane season (Elsner and Kara, 1999).

Similar to the patterns found for the hurricane season in general, July shows an area in the central south with contribution percentages between 10-14% (Figure 3-9d). August shows the highest contribution of any of the months with some areas in the southern coastal plains exhibiting percentages of 28.8-33.3% (Figure 3-10d). These areas in the south are of great importance for the agricultural industry in the island, and here is found that they are strongly dependent on TCR for 1/3 of the total precipitation that falls in August. These contribution percentage values of 30% were similar to the ones found by Rodgers et al., (2001) in the Atlantic sector close to where Puerto Rico is located and to the results found by Nogueira and Keim (2011) for some coastal areas of the Eastern

US. Other areas that exhibited high contributions in August (>20% or more) were the southern slopes, the central mountains and a significant portion of the eastern region of the island.

In terms of the spatial distribution of average rainfall, it is evident that September and October are the rainiest months of the hurricane season (Pico, 1974), with the

71 southeast and western interior regions exhibiting the highest precipitation values (Figure

3-11a, c) The month of September shows that almost the entire island received more than 10% of its rainfall from TCs for the 1970-2010 period (Figure 3-11b). Contribution percentages between 19.5-28.7 % are found in the eastern area where El Yunque

National Rainforest is located and in the southern region of the island for September.

Similar to August, a general decrease in contribution percentages is observed as we move west for the month of September. October was identified as the rainiest month in terms of average precipitation, yet its contribution percentages were found to be lower than those found for August and September (Figure 3-11d). October exhibits a small region in the south with percentages around 10-14%. It is clear that TCs are not the main contributor of rainfall in the rainiest month, this is due to the fact that during this time of the year Puerto Rico receives precipitation from multiple processes that include localized convective thunderstorms, west to east moving troughs and cold fronts (Colon,

2008). The month of November was not reported here since only four TCs impacted

Puerto Rico in that month during the 1970-2010 period.

Concluding Remarks

This study examined the spatial distribution of rainfall associated with TCs over

Puerto Rico and their overall contribution to the precipitation climatology of the island.

Daily and monthly rainfall data at 32 sites where used to compute a tropical cyclone rainfall (TCR) value and a contribution percentage for 86 TCs that passed within a radius of 500 km around the island. In the spatial distribution analysis of the study the focus was on understanding the patterns of four TCR groups based on their moisture

(TPW) and proximity to the island’s coast (PRX) values which included; a) high TPW

72 and close PRX TCs, b) high TPW and far PRX TCs, c) low TPW and close PRX TCs and d) low TPW and far PRX TCs. Geostatistical interpolation models that included ordinary kriging (ORK) and ordinary cokriging (OCK) were used to generate predicted surfaces of TCR for all TC groups in order to examine similarities or differences in the spatial distribution of rainfall associated with TCs with different TPW and PRX characteristics. The ORK method was used to produce predicted surfaces of percentage contributions for the hurricane season months in order to explore the spatiotemporal characteristics of rainfall contributed by TCs.

This study found that the highest rainfall values associated with the passage of

TCs are concentrated in the eastern, southeastern and central regions of the island. A general east to west decrease in rainfall values was also found. The highest rainfall values were attributed to TCs over high TPW environment and to storms that were within 230 km of the island’s coast. Farther TCs, even with high moisture environments over Puerto Rico, tend to be associated with lower rainfall amounts over land when compared to the closer and higher moisture TCs. The closer TCs over lower moisture environments exhibit a more random spatial distribution of precipitation over the island, instead of the usual cluster of high-low values found in the groups that considered all 86 events and the high moisture-close proximity storms. The TCs that had farther circulation centers from land and low moisture environments exhibited the lowest TCR values of all of the TC groups. The OCK method performed better than the ORK in three of the four groups, which coincides with the literature that suggests that adding information about topography (elevation, slope and aspect) as covariates improves the models prediction capabilities. When southern and northern TCs were compared we

73 found that storms south of Puerto Rico produce more rainfall (13.6-24 mm) in the southeastern interior, northeast and northwest while TCs north of the island were associated with higher values (4.28-19.5 mm) in the western interior region of the island.

When all hurricane season months are considered we find that TCs contributed around 10-15% of the rainfall in the south and southeastern regions of the island. July showed an area in the central south with contribution percentages between 10-14%.

The months in which TCs contributed the most to the rainfall climatology of Puerto Rico for the period of 1970-2010 were August and September. Some areas in the south, east and central regions of the island had rainfall contribution percentages of more than 25% and in some cases more than 30% for the month of August. The southeast and much of the agricultural south were found to be strongly dependent on TCR, with some areas receiving 1/3 of their hurricane season rainfall from TCs during the months of August and September. The month of September shows that almost the entire island received more than 10% of its rainfall from TCs for the 1970-2010 period October was identified as the rainiest month in terms of average precipitation, yet its contribution percentages were found to be lower than those found for August and September. October exhibited low rainfall contributions from storms, which suggests that heavy rainfall events over

Puerto Rico during that month are not necessarily associated with TCs.

Chapter 3 Limitations

Similar to the limitations discussed in chapter 2 this chapter is limited by the use of only 32 stations with data for the 1970-2010 period. Some of the 32 stations have data all the way into 2015, however more than 10 of the 32 rain gauges do not extend

74 until more recent years and for that reason the 1970-2010 is chosen. The stations are randomly distributed over the island, however there are some areas of the island that are not well represented. For example there is only one weather station in the north central coastal plains and in some portions of the central mountains region of the island there are no rain gauges. Not having sufficient observations in the central mountains is a huge limitation since topography plays a big role when it comes to the rainfall patterns in the island. These areas that are not represented in the station network and present issues when it comes to the interpolation techniques that are employed to estimate TC rainfall amounts and contribution percentages.

Figure 3-1. The elevation of the island of Puerto Rico and the rain gauges used in the study.

Figure 3-2. Tracks of all 86 TC events that passed within a 500 km radius of Puerto Rico.

Table 3-1. Rain gauges with daily and monthly data for Puerto Rico for the period of 1970-2010. Station ID Name Lat Lon Elev (m) GHCND:RQC00660040 Aceituna Water Treatment 18.15 -66.50 663.9

GHCND:RQC00660061 Adjuntas Substation PR 18.18 -66.80 556.9

GHCND:RQC00660152 Aguirre PR 17.97 -66.22 9.1

GHCND:RQC00661345 Calero Camp PR 18.48 -67.12 75.9

GHCND:RQC00661590 Canovanas PR 18.38 -65.90 9.1

GHCND:RQC00662801 Coloso PR 18.38 -67.15 11.9

GHCND:RQC00663023 Corral Viejo PR 18.07 -66.65 121

GHCND:RQC00663431 Dos Bocas PR 18.33 -66.67 60

GHCND:RQC00663532 Ensenada 1 W PR 17.97 -66.93 3

GHCND:RQC00663904 Guajataca Dam PR 18.40 -66.93 200.9

GHCND:RQC00664126 Guayabal PR 18.07 -66.48 111.9

GHCND:RQC00664276 Gurabo Substation PR 18.25 -66.00 47.9

GHCND:RQC00664330 Hacienda Constanza PR 18.22 -67.08 146

GHCND:RQC00664702 Isabela Substation PR 18.47 -67.05 128

GHCND:RQC00664867 Jajome Alto PR 18.08 -66.15 726.9

GHCND:RQC00665020 Juana Diaz Camp PR 18.05 -66.50 79.9

GHCND:RQC00665064 Juncos 1 Se PR 18.23 -65.91 64.9

GHCND:RQC00665097 Lajas Substation PR 18.03 -67.07 27.4

GHCND:RQC00665807 Manati 2 E PR 18.43 -66.47 82

GHCND:RQC00665911 Maricao Fish Hatchery PR 18.17 -66.98 456.9

GHCND:RQC00666083 Mayaguez Airport PR 18.25 -67.15 11

GHCND:RQC00666073 Mayaguez City PR 18.22 -67.13 29.9

GHCND:RQC00666361 Mora Camp PR 18.47 -67.03 125

GHCND:RQC00666390 Morovis 1 N PR 18.33 -66.40 219.2

GHCND:RQC00666805 Paraiso PR 18.27 -65.72 109.7

GHCND:RQC00667292 Ponce 4 E PR 18.02 -66.53 21

GHCND:RQW00011641 San Juan L M Marin AP 18.43 -66.00 21

GHCND:RQC00668815 San Lorenzo 3 S PR 18.15 -65.97 155.1

GHCND:RQC00668940 Santa Isabel PR 17.97 -66.38 9.1

GHCND:RQC00668955 Santa Rita PR 18.01 -66.88 22.9

GHCND:RQC00669521 Trujillo Alto 2 SW PR 18.33 -66.02 41.1

GHCND:RQC00669774 Villalba 1 E PR 18.13 -66.48 157.9

Table 3-2. Tropical cyclone groups based on different PRX and TPW values. Categories (CAT) are based on the mean Saffir-Simpson Hurricane Wind Scale for the TC. High Moisture Close Proximity High Moisture Far Proximity Close Proximity Low Moisture Far Proximity Low Moisture

PRX TPW PRX TPW PRX TPW PRX TPW TC Year Cat (km) (mm) TC Cat Year (km) (mm) TC Cat Year (km) (mm) TC Year Cat (km) (mm) Carmen 1974 TD 105.0 45.2 Gloria H1 1985 324.8 44.5 Doria TD 1971 33.0 42.4 Elaine 1974 TD 384.3 42.0 Frederic 1979 TS 0.0 49.5 Iris TD 1989 426.1 50.5 Christine TD 1973 34.0 43.3 Emmy 1976 TS 408.6 37.9 Claudette 1979 TD 0.0 47.5 Eduard H3 1996 446.3 45.0 Eloise TD 1975 68.0 43.6 Anna 1979 TD 435.2 36.5 Klaus 1984 TS 4.4 52.9 Dennis TD 1999 336.5 47.9 Floyd TS 1981 156.5 38.8 Arthur 1984 TD 357.3 39.5 Hugo 1989 H4 0.0 52.9 Floyd H2 1999 489.3 45.6 Gert TS 1981 0.0 43.1 Daniel 1986 TS 447.3 36.2 Marylin 1995 H2 38.4 47.2 Lili TS 2002 448.2 45.3 Lili TS 1984 158.6 30.3 Arthur 1990 TS 347.8 42.2 Hortense 1996 H1 0.0 49.7 Odette TS 2003 326.6 49.3 Kate H1 1985 211.4 42.2 Iris 1995 TS 379.6 35.4 Bertha 1996 H2 62.8 47.4 Dennis TS 2005 419.1 49.3 Emily H2 1987 221.1 42.7 Fran 1996 H1 499.9 41.6 Georges 1998 H3 0.0 49.6 Gamma TS 2005 384.1 44.8 Chris TD 1988 31.7 43.7 Chantal 2001 TS 400.7 43.6 Jose 1999 TS 81.1 49.6 Ernesto TS 2006 420.0 46.5 Cindy TS 1993 132.3 43.5 Jerry 2001 TS 380.0 41.6 Debby 2000 H1 50.0 47.2 Dean H4 2007 293.3 45.4 Luis H4 1995 185.4 44.0 Olga 2001 TD 490.9 30.3 Dean 2001 TS 43.9 47.1 Gustavo TS 2008 403.7 49.4 Frances H4 2004 221.8 44.4 Claudette 2003 TS 392.9 37.5 Jeanne 2004 TS 0.0 51.4 Hanna TS 2008 374.8 48.3 Chris TS 2006 192.7 42.6 Charley 2004 TS 426.1 40.1 Fay 2008 TS 14.2 44.7 Ike H3 2008 376.1 48.1 Olga TS 2007 0.0 40.0 Emily 2005 H3 486.4 42.4 Earl 2010 H3 120.6 48.8 Kyle TS 2008 449.9 48.4 Erika TD 2009 104.7 41.0 Tomas 2010 TS 425.8 43.2

Figure 3-3. Scatter plot of the four tropical cyclone groups based on similarities between TPW (y) and PRX (x) values.

Figure 3-4. Tracks of tropical cyclones groups divided in a) high moisture and close proximity, b) high moisture and far proximity, c) low moisture and close proximity and d) low moisture and far proximity.

Figure 3-5. Kriging predicted surfaces of TCR for a) ORK of all 86 TCs and b) OCK of all 86 events with elevation and slope as covariates.

Figure 3-6. Kriging predicted surfaces of TCR for a) high moisture and close proximity events, b) high moisture and close proximity storms with elevation, slope and aspect as covariates, c) high moisture and farther storms and d) high moisture and farther storms with elevation and slope as covariates.

Figure 3-7. Kriging predicted surfaces of TCR for a) low moisture and close proximity events, b) low moisture and close proximity storms with elevation and slope as covariates, c) low moisture and farther storms and d) low moisture and farther storms with aspect as a covariate.

Figure 3-8. Ordinary kriging surfaces of TCR of storms located north (a) and south (b) of Puerto Rico and the differences between both surfaces (c).

Figure 3-9. Ordinary kriging predicted surfaces of a) average rainfall for the hurricane season months (June-November) and b) percentage of contribution of TCs to the rainfall climatology of the island for the hurricane season months.

Figure 3-10. Ordinary kriging predicted surfaces of average rainfall for the months of July (a) and August (c) and the contribution of tropical cyclones to the rainfall climatology of the months of July (b) and (d) August.

Figure 3-11. Ordinary kriging predicted surfaces of average rainfall for the months of September (a) and October (c) and the contribution of tropical cyclones to the rainfall climatology of the months of September (b) and (d) October.

CHAPTER 4 EXTREME FLOODS AND THEIR RELATIONSHIP WITH TROPICAL CYCLONES IN PUERTO RICO

The island of Puerto Rico is known to experience extreme floods associated with heavy rainfall from the passage of tropical cyclones (TCs) that include tropical depressions, tropical storms and hurricanes. As a tropical island, Puerto Rico receives a lot of precipitation especially in the north and central regions of the island where it can range from 3000 mm to 4300 mm on the yearly basis (Pico, 1974). Orographic influences causes rainfall to be unevenly distributed in the island, with the northern region experiencing more rainfall through the year than the southern region (Colon,

2008). This variability of rainfall over space and time causes different hydrologic responses to extreme precipitation events in the island. The flood hydrology of Puerto

Rico is of concern and interest because of the high frequency of extreme unit discharge flood peaks relative to other locations in the United States (O’Connor and Costa, 2004).

During the hurricane season months (June to November) TCs play an important role in the flood hydrology of extreme events in the island (Scatena and Larsen, 1991; Larsen and Simon, 1993; Larsen and Torres-Sanchez, 1998). Many of the record peak measurements in Puerto Rico were associated with tropical cyclones like Hurricane

Donna of September 1960 (Barnes and Bogart, 1961), Hurricane Hortense of

September 1996 (Torres-Sierra, 1997) and Hurricane Georges of September 1998

(Larsen and Santiago-Roman, 2001). The extreme floods associated with these TCs have led to devastating impacts that have translated into billions of dollars in losses.

The results of one study suggest that peak flood response in Puerto Rico for Hurricane

Georges was strongly dependent on peak 15–60 min rainfall rates and less influenced

84 by high storm total accumulations (Smith, et al., 2005). Even though some researchers have explored the hydrologic response associated with tropical cyclone rainfall over the island for a few intense hurricanes, no study has looked at a climatology of TCs that includes tropical depressions and tropical storms that can also produce extreme flood events over the island. For that reason, the main purpose of this study is to examine the statistical characteristics of extreme flood events over Puerto Rico and its relationship with tropical cyclones by implementing extreme value analysis (EVA).

Studies that looked at the relationship between TCs and floods vary in their scope and scale, with some looking at the regional hydrologic response to these events

(O’Connor and Costa, 2004; Villarini and Smith, 2010) and others focusing on a more local scale (Waylen, 1992; Kostaschuk et.al; 2009). O’Connor and Costa, (2004) examined the spatial distribution of floods in the United States and found that Puerto

Rico and Hawaii together account for 32% of the observed record flows in the 99th percentile of unit discharges although they only represent less than 5% of all observations in the study. These tropical island locations exhibit such high unit discharge characteristics, especially in their windward areas, since they are in the path of the northeasterly winds, where mountains force the moist air to rise and enhance the rainfall associated with local convective storms and tropical cyclones (Colon, 2008).

One of the main methods implemented to examine the relationship between floods and

TCs is the general extreme value (GEV) distribution (Coles, 2001) which allows the examination of upper tail properties of flood peaks. Smith and Morrison, (2002) found that a number of flood records in central Appalachia have anomalously large values of the estimated GEV shape parameter. Recent studies on floods in the Eastern US and

Texas have implemented the use of GEV probability distributions to investigate whether

TCs control the upper tail of the flood peak distribution (Villarini and Smith, 2010;

Villarini and Smith 2013). One of those studies focused on the relationship between floods and TCs in the Eastern US, here the GEV distribution was implemented in order to compare the shape (skewness) parameter of the entire discharge series and the series without tropical cyclones and found that when peaks associated with TCs are removed there’s a reduction in the shape parameter, which shows that floods associated with TCs tend to have heavier tails (Villarini and Smith, 2010). They also found that TCs had a larger influence on the upper tails of the flood distribution east of

Appalachia and in coastal areas between Florida and New England. However, a more recent study on TCs and floods in Texas found that tropical cyclones tend to have a weaker influence on flood peaks when compared to the GEV shape parameter of other locations in the eastern US (Villarini and Smith, 2013).

The main purpose of this study is to examine the relationship between extreme flood events over Puerto Rico and tropical cyclones. Mean discharge data from 12 stations over twelve different water drainage basins in Puerto Rico for the 1970-2010 period were used. Tropical cyclone six-hourly track data from 86 TCs was used in order to identify the floods associated with the different storms that impacted the island in the time period. Floods were defined as the 99th percentile of the mean discharge data for the whole series over the twelve stations. An extreme value analysis (EVA) point process approach was implemented to determine if TCs strongly affected the properties of the GEV distribution parameters location (central tendency) scale (variance) and shape (skewness) of the mean discharge time series for 12 stations with 41 years of

86 data. First, the point process model was used to fit the entire mean discharge data of all the stations and then the model was implemented again with the series that had the flood peaks associated with TCs removed. The GEV parameters of the entire discharge series and the series without TCs were retrieved in order to compare them and determine if there were any statistical differences between the parameters in the series that included the storms and the series without them. Maps of percentage change between the GEV parameters were generated to visually compare the differences between the location, scale and shape of the entire series and the one with TCs removed. The maps also served to examine the spatial characteristics of the floods in the different stations over the island. Flood frequencies were also calculated from the

GEV parameters to examine the differences between the probabilities of floods between the two series. The TCs that produce the largest floods in the most of the stations were identified in order to examine the relationship between extreme floods and tropical cyclone rainfall.

Data and Methods

Floods and Tropical Cyclone Data

The first step of the analysis was to obtained daily mean discharge data for twelve stations with complete data (14,975 observations) for the time period of the

1970-2010 (41 years) for the island of Puerto Rico (Figure 4-1). The daily mean discharge data was obtained from the United States Geological Survey (USGS)

National Water Information System and it was approved for analysis for all of the stations in this study (Table 4-1). The 12 stations were located in twelve different water drainage basins over the island that extended east to west with an area size range of

517.1-20.71 square kilometers. The size of the drainage basin is an important factor when it comes to the stream flow characteristics of each station, with larger basins capturing more water from rainfall hence having higher discharge values than smaller basins (O’Connor and Costa, 2004). This factor is relevant when it comes to defining a flood in stations over water basins of different sizes. In the absence of a practical physical threshold that could be applied to all the twelve stations in the different basins this study defines a flood as a mean discharge value in the 99th percentile of the entire data distribution. By implementing the mathematical instead of the physical approach when defining floods over different basins we treated each daily mean discharge series equally since only the most extreme flood peaks (99th percentile) were chosen for modeling. The scatter plots of one of the stations shows the entire series of mean daily discharge and the extreme flood peaks in the 99th percentile of the flood distribution with the ones associated with TCs highlighted (Figure 4-3).

The second step, was to obtain tropical cyclone track data for all the TCs that might have impacted the island of Puerto Rico during the 1970-2010 period. Six-hourly

TC positions were obtained from the International Best Track Archive for Climate

Stewardship (IBTrACS) (Knapp and Kruk, 2010) for the years 1970–2010. The TC track data were entered in a Geographic Information System (GIS) where a 500 km radius was implemented to select the storms that could have brought significant rainfall to the island and caused floods over different locations. These radius has been used by other studies that examined tropical cyclones and their associated rainfall over Puerto Rico

(Boose et al., 2004; Hernandez and Matyas, 2015). The next step was to identify the floods related to TCs, this was done by associating a flood peak to a cyclone if the

88 center of the storm is within 500 km of the island’s coast during a time frame of two days before and seven days after the passage of the TC (Villarini and Smith, 2013). The implementation of this radius and time window allows us to capture the full spatiotemporal extent of the rainfall associated with a TC, its specific characteristics (center proximity to land, intensity, and translation speed) and the environment surrounding the storm.

Extreme Value Analysis Point Process Approach

Extreme value analysis (EVA) using a point process approach is selected to model the statistical characteristics (mean, variance an skewness) of extreme flood events (99th percentile) over Puerto Rico for twelve entire daily mean discharge time series and a series in which flood peaks associated with TCs have been removed. This method combines existing approaches in the modeling of extremes, specifically the annual maximum series (AMS) and the partial duration series (PDS), also known as the peak over thresholds approach (POT). The AMS and PDS approaches have been used extensively in different studies looking at hydrological or climatological events above or below a specific physical or mathematical threshold (Waylen, 1988; Waylen and

LeBoutillier, 1989; Waylen et al., 2012). The point process approach is devised in terms of the GEV parameters; scale μ (central tendency or mean), location σ (variance), and shape ξ (skewness) and as a result extreme properties of X variable are defined by those three parameters (Coles, 2001). The point process approach has advantages over both traditional approaches of modeling extremes, it can modeled the frequency and magnitude of events in a single point process and it can also include all of the

89 values above a specific threshold which coincides with more reliable results (Keelings and Waylen, 2014).

Some studies have used the shape parameter which is obtained from the GEV distribution to examine the upper tail properties of the flood distribution and its relationship with tropical cyclones. The shape parameter has been found to be useful in studies looking at flood peaks in the Eastern US and Texas, where it was found that high values of the shape parameter were associated with heavy tails of the flood distribution when tropical cyclones were present in the discharge series (Villarini and

Smith, 2010; Villarini and Smith 2013). Those researchers found that the shape parameter was large in locations east of Appalachia and in the eastern coast of the

United States and that when TCs were removed from the data series those shape values dropped, which shows the role that TCs play in shaping the upper tails of the flood distribution. However, in the case of Texas, which is the state that experiences the most floods in the continental US and the highest rate of flood related deaths

(Rappaport, 2000) they found that the shape parameter tends to be lower than other locations in the eastern US, which shows that TCs are not the dominant processes behind extreme floods in Texas (Villarini and Smith, 2013).

This study uses the location (central tendency or mean), scale (variance) and shape (skewness) parameters product of the point process model to examine the statistical characteristics of the floods associated with TCs over Puerto Rico. The first step before introducing the mean daily discharge series from the twelve stations individually into the point process model was to decluster the data. This study considered a flood to be a single event if it’s separated by more than three days of

90 below threshold (99th percentile) values. For example, a tropical cyclone could cause significant rainfall over a given area in a short time period, yet there’s a lag in the hydrologic response which can produce a couple of days of above threshold discharge and when those values go below that threshold for at least three days and then go back up in the fourth day they will be considered to be part of a different event. After declustering the entire series of each station and the ones with the TCs removed they were both introduced in the point process model. The location, scale and shape parameters of both series were retrieved and compared to determine if TCs had an influence on the central tendency, variance and skewness of the mean daily discharge.

Percentage differences between each of the GEV parameters were calculated to see if there was a reduction or an increase in their values when TCs were removed from the data. These percentage change calculations were also mapped to examine the spatial distribution of the different GEV parameters and their corresponding changes when TCs were removed from the data series. The top flood peak producing TCs were identified and their characteristics in terms of rainfall, moisture and proximity to land were examined in order to explore their relationship with the extreme floods. All of the calculations in this paper were computed in R (R Development Core Team, 2008) using the freely available extRemes and ismev packages (Cole, 2001).

Results

Descriptive Statistics

Of the 86 TCs that passed within a 500 km radius over Puerto Rico only 51

(59%) were associated with individual flood peaks at or above the 99th percentile threshold use to identify the most extreme discharge values on all of the stations (Figure

4-2). TCs were responsible for the maximum value of mean daily discharge in 75% of the stations (Table 4-2). As expected, larger water drainage basins had higher mean discharge values than smaller basins and for that reason the 99th percentile was useful in identifying changes from one series (TCs included) to another (TCs removed). All of the stations reflected a reduction in the mean discharge when TCs were removed from the series, yet this decrease varied from station to station with the ones located in the eastern area of the island showing the largest percentage decreases on their means

(10-14%). Similar TCs were responsible for multiple flood peaks in different stations around the island. Among the ones responsible for most of the extreme floods we find powerful hurricanes like Georges (1998), Hortense (1996), Hugo (1989), and David

(1979). These hurricanes were associated with heavy rainfall over the island since their centers, characterized by intense convective thunderstorm activity, were relatively close to land and the storms were also over high moisture environments that promoted higher precipitation accumulations (Hernandez and Matyas, 2015). Hurricanes were not the only ones responsible for the first, second and third maximum flood peaks in some of the stations, here tropical depressions and storms like Eloise (1975), Isabel (1985) and

Jeanne (2004) also produce extreme discharge events throughout the island (Table 4-2)

. Tropical depression Eloise dropped heavy rainfall throughout the island with values above 500 mm in the central mountains region (Colon, 2008) and with floods causing 34 fatalities (Hebert, 1976) with a total loss of 458 million dollars (Pielke et al.,2003).

Before discussing the results of the point process model it’s important to identify the stations in the island that exhibited the highest percentage of flood peaks associated with TCs. Here, all of the daily mean discharge values at or above the 99th percentile

92 threshold for each station were retrieved and the ones associated with TCs were identified. A spatial pattern of the percentage of contribution of flood peaks by TCs is evident, with a majority of stations in the eastern region of the island exhibiting higher percentages than the ones located farther west (Figure 4-4).This is expected since TCs move east to west around the island and they tend to deposit higher rainfall amounts in the windward facing mountains of the eastern region of the island (O’Connor and Costa,

2004, Hernandez and Matyas, 2016). The highest flood peak contribution percentages are found in the eastern interior and in the south central region of the island, while the lowest percentage is located in the station that extends the most to the west. The flood peaks in the east tend to be more associated with the passage of TCs than the ones in the west, which means that flood generating processes in the west might be the product of a combination of phenomena such as local convective thunderstorms, west to east moving troughs and cold fronts (Colon, 2008). It is important to note that this map only shows the percentage of the count of flood peaks associated with the passages of TCs and not the results of the point process model.

EVA Point Process Model Results

Point process approach diagnostics probability and quantile plots for four of the stations, two in the east and two in the west, exhibit different results when the data of the entire series are visually compared to the series with the TCs removed from them

(Figure 4-5, Figure 4-6). The quantile plot for the station in Rio Gurabo located in the eastern region of the island shows that when TCs are present in the data the model overestimates the empirical observations since the actual flood peaks associated with the storms are high even when compared to the other peaks at or above the 99th

93 percentile in the same station (Figure 4-5a, 4-5b). However, when TCs are removed from the data in Rio Gurabo the model improves and it’s able to fit a line through the actual observations, which means that flood peaks associated with TCs in this station are more complicated to model than non-TC floods (Figure 4-5e, 4-5f). When comparing the point process model results from Rio Gurabo in the east with those ones from Rio

Culebrinas near Moca located in the western most region of the island there’s a contrasting difference. In Rio Culebrinas near Moca both the probability and quantile plots exhibit a better model fit of the flood peak data for the entire series and the one with the TCs removed (Figure 4-5c, 4-5d, 4-5g and 4-5h). This is evidence that TCs do not play a major role in changing the flood distribution in this station in the west. A similar comparison can be made with Rio Grande de Loiza in the east and Rio Grande

De Añasco in the west, with the point process model performing efficiently in the series with TCs removed in the east and in both series on the west (Figure 4-6). Even though only the point process probability and quantile plots of four stations from four different drainage basins were visually compared here, it is pertinent to state that TCs have a stronger influence on the flood distribution on the stations in the east and almost a non- existing influence on the stations in the west. Villarini and Smith, (2010) found similar results for the eastern US, with areas east of Appalachia exhibiting strong TC influences on the upper tail properties of the flood peak distribution, while the areas to the west show a weal connection with TCs.

Before looking at the GEV parameters product of the point process approach it’s important to examine the spatial distribution of the 99th percentile, since this was the flood definition for all stations in this study. The 99th percentile of flood peaks ranged

94 from 3.3 to 97.6 m3/s, with the larger water drainage basins exhibiting higher values than the smaller ones (Figure 4-7). The largest 99th percentile threshold was found in the largest drainage basin in the study at El Rio La Plata near Toa Alta located in the eastern region of the island, while the lowest set threshold was found in the southern most water basin at Rio Inabon at Real Abajo. When TCs were removed from the data series it resulted in a reduction of the 99th percentile that ranged from of 1.6 to 24.7%.

The stations with the largest reduction (11-24.7%) in the 99th percentile threshold when

TCs were removed from the series are located in the eastern interior region of the island, which shows that the water basins that extended mostly to the southeast are the ones where TCs control the most extreme flood peaks. This might be due to the fact that heavy rainfall associated with TCs tends to be concentrated over those basins in the southeast region of the island. It is important to note that there are two stations in the northeast region of the island that exhibit low change percentages (1.8-7.8%), which shows once again that not all areas in the east have similar hydrologic responses to the passage of TCs. A general decrease in change percentages is evident as we move to the central and western region of the island. These changes in the percentage decrease of the 99th percentile threshold shows that TCs have a stronger influence in extreme floods in the southeast and tend to decrease as we move to the central and western region of the island.

The GEV parameters that were retrieved from the point process model show that some stations in the island exhibited changes when TCs were removed from the data series (Figure 4-8). The location parameter, which could be understood as the central tendency or mean of the flood peak distribution, shows 5 stations with high values in the

95 entire series when compared to the one with TCs remove. The station at Rio Grande de

Loiza near Caguas in the east shows the largest difference in the location parameter with a value of more than 350 m3/s for the entire series and one closer to 100 m3/s when TCs are removed (Figure 4-8a). A similar pattern is found in the scale (variance) parameter plot which shows that Rio Grande de Loiza near Caguas and other stations in the east experience a larger variance in their mean discharge values when TCs are included in the series (Figure 4-8b). The shape parameter plot between the entire series and the series without TCs also shows some large reductions when these are removed.

Here again the stations in the east show the largest reduction in the shape parameter, while the ones in the west exhibit lower declines in the values. The GEV shape parameter has been used in other studies to examine the upper tail properties of the flood distribution in which they have shown that when TCs are removed from the series there are changes in the tails of the flood peak distribution (Villarini and Smith 2010;

Villarini and Smith 2013).

Results from the point process model for the GEV location parameter reveal that all of the stations had a decrease in their mean daily discharge when the flood peaks associated with TCs were removed from the series (Figure 4-9). It is important to note that the larger drainage basins tend to have the higher location (mean) parameter values since these cover larger areas in which more rainfall will tend to fall. The reduction in the location parameter varied over the island with the largest decrease

(72.6%) found in the Rio Grande de Loiza near Caguas station located in the eastern interior region of the island and the lowest one found in the western most station Rio

Culebrinas near Moca (6.8%). The location percentage decreases for the stations in the

96 central parts of the island show a higher variance in their values ranging from 9.8% to

55.2%. This is expected since in Chapter 2 we identify this area of the island as a transitional area between the east and west when it comes to rainfall associated with

TCs. The two stations located in the northeast region of the island exhibit similarities in their location reduction with the two stations at the opposite side of the island in the northwest, this might be due to the fact that these locations receive lower amounts of rainfall during the passage of TCs. The stations with the largest decrease in the location parameter are in water drainage basins of different sizes, which reinforces the selection of the 99th percentile as the method for identifying extreme floods over all of the twelve stations.

All of the stations also exhibited a decrease in the scale (variance) parameter when the flood peaks associated with TCs were removed from the series (Figure 4-10).

Higher scale percentage decreases (62.6-89%) were also found in the stations located in the eastern interior, which means that when TCs are part of the series the variability in mean daily discharge over those places is higher than when they’re not included. The two stations in the northeast exhibit low scale percentage decreases, while the two sites in the west exhibit higher variance in their series when TCs are included. The scale percentage change for the stations in the central part of the island exhibits a higher variance in their values (25-89%), which shows once again that this region is a transitional area between the east and west in which the rainfall associated with TCs tends to vary more in their amounts. The station in Rio Tanama located in the western interior shows that when TCs are removed the variance in discharge values decreases by more than 60%, which shows that TCs have a large impact on the variability of floods

97 in that site. The scale shares some similarities with the results of the location parameter, since both identified the largest mean and variance decreases when TCs were removed from the series in stations in the eastern interior area of the island, while the lower percentage decreases of both GEV parameters were in the northeast and northwest regions of Puerto Rico.

The results for the GEV shape parameter which has been found to be useful in examining the upper tail properties of the flood distribution also exhibit some interesting changes when the entire series is compared to the one were TCs have been removed

(Figure 4-11). The largest shape parameter values, associated with thicker tails in the distribution are found once again in the stations Rio Grande de Loiza near Caguas and

Rio Gurabo which are located in the eastern interior region of the island, while the lowest shape parameters are found in the western and northeastern region of the island

(Figure 4-11a). The shape (skewness) parameter decreased in all of the stations when

TCs were removed from the series (Figure 4-11b). However, those changes in the upper tail properties of the flood peak distribution were different for the stations. Four of the stations located in the eastern region of the island exhibited large reductions (47.1-

88.4%) in their shape parameters when TCs were removed from the series, while the three stations located in the western region of the island exhibited lower decreases

(Figure 4-11c). Three of the stations in the central part of the island exhibited even lower percentage changes than the ones in the west, while the station Rio Espiritu

Santo near Rio Grande located in the northeast had an increase in the shape parameter when TCs were removed from the series. This means that TCs do not control the upper tail properties of the flood peak distribution in that station, so other precipitation

98 generating process are responsible for the most extreme floods in that water drainage basin. The GEV shape parameter has been used in other studies examining the upper tail properties of floods in Appalachia (Smith and Morrison, 2002) and in studies examining the role that TCs play in shaping the upper tail properties of the flood peak distribution in the Eastern US and Texas (Villarini and Smith, 2010; Villarini and Smith,

2013) were they found that similar to our findings the stations east of Appalachia tend to exhibit upper tail properties that are controlled by nearby TCs impacting the region.

Flood frequency probabilities were calculated from the lambda values that resulted from point process model of the GEV parameters location, scale and shape of both the entire series and the one with TCs removed. Visual comparisons between four stations, two in the east and two in the west exhibit changes in the probability of experiencing a given amount of floods over the yearly basis when TCs are part of the series and when they are removed from it. The flood frequency probabilities for the entire series of the station in Rio Grande de Loiza near Caguas (Figure 4-12a) and Rio

Gurabo (Figure 4-12c) located in the eastern region of the island exhibit a reduction in the probability of having one flood per year when TCs are removed from their series.

The probabilities of having one extreme flood per year decreases from 34% to 19% in

Rio Grande de Loiza in Caguas (Figure 4-12b) and from 32% to 17% in Rio Gurabo

(Figure 12d). Rio Gurabo also shows a decrease in the probability of two extreme floods per year of 16% to 9% when the TCs are removed from the series. These shows once again that the eastern interior region of the island is the area in which TCs tend to control more of the flood peaks, which coincides with the findings of Costa and

O’Connor, (2004). The flood frequency probabilities for the two stations in the west

99 show a different result for the probability of having one flood per year in each of the series. Instead of decreasing probabilities of one flood per year when TCs are removed, there are increasing probabilities in Rio Grande de Añasco and Rio Culebrinas (Figure

4-12e, 4-12f, 4-12g and 4-12h). Rio Grande de Añasco shows an increasing probability of one flood per year that went from 21% to 31%, while Rio Culebrinas exhibits an increment that went from 24% to 36% when TCs were removed from the series. This is more evidence to suggest that flood peaks in the western region of the island are not controled by TCs, so other flooding generating processes like local convective thunderstorms, west to east moving troughs and cold fronts are probably responsible for the high stream flows in this region of the island.

Recent findings by Hernandez and Matyas, (2015) suggest that tropical cyclone rainfall over Puerto Rico tends to be associated with the variability related to the storm’s center distance to land and the surrounding moisture environments. They found that

TCs that come within 230 km of the island’s coast within moisture environments of total columnar precipitable water of 44.5 mm or more tend to be associated with mean tropical cyclone rainfall values of more than 50 mm for the entire island, with that value been considered an extreme event of precipitation over the daily basis (Jury, 2009).

When examining the individual daily flood peaks at or above the 99th percentile used to define extreme floods in all stations the results coincide (Table 4-3) with those of

Hernandez and Matyas, (2015). The TC responsible for most of the individual flood peaks was Klaus (1984), followed by David (1979), Jeanne (2004) and Isabel (1985).

The top 15 individual flood producing TCs had a mean tropical cyclone rainfall value of

147 mm for the entire island and in average their circulation centers came in 85.84 km

100 of the island’s coast at their closest approach with humidity environments averaging

48.64 mm. These findings suggest that the TCs associated with the largest number of individual flood peaks over Puerto Rico produce significant rainfall over the island since their circulation centers associated with intense convection were relatively close to land and embedded in high moisture environments. By observing the tracks of the 51 TCs that were associated with at least one flood over the island it is evident that the storms that were closer to land caused a higher number of flood peaks than the ones located farther away (Figure 4-13). These closer TCs that include hurricanes like Hugo (1989),

Hortense (1996) Georges (1998) and tropical storms like Klaus (1984) and Jeanne

(2004) were also located over high moisture environments exceeding 49.5 mm.

The rainfall associated with the passage of the top fifteen flood peak producing

TCs explains the spatial distribution of floods that resulted when mapping the percentage changes in the GEV parameters location, scale and shape between the entire series and the series without TCs. Averaging the rainfall associated with those

TCs reveals a clear pattern of the spatial distribution of TC precipitation over Puerto

Rico (Figure 4-14). The southeast and the eastern interior region are identified as the areas of the island in which tropical cyclone rainfall tends to be highly concentrated with mean values ranging from 155.8 to 241.4 mm (Figure 4-14c). These were the areas where TCs had the strongest influence on the flood statistical properties which were identified by comparing the GEV parameters location, scale and shape of the entire series and the series without TCs. As we move to the west rainfall associated with the passage of those top flood peak producing TCs tends to decrease until it reaches a minimum in the western region of the island. These spatial pattern of rainfall also

101 explains the weak effect that TCs have on the upper tail properties of the flood distribution in the western region of the island, which simply means that rain associated with the passage of TCs is not the main force behind the largest flood peaks in that region (Figure 4-14c). Standard deviation and coefficient of variation maps were also generated to examine the variability of rainfall associated with those top flood producing

TCs (Figure 4-14a, 4-14b). The standard deviation values of mean rainfall associated with TCs tend to be higher in the south and central regions of the island while the coefficient of variation exhibits higher percentages of variability in much of the western, central and southern region of the island. These high variations in rainfall also explain the behavior of the stations in the central region of the island which also showed large variability in the GEV parameters location, scale and shape.

Concluding Remarks

This study focused on examining the relationship between extreme flood events over Puerto Rico and tropical cyclones. Mean daily discharge data for twelve stations with 41 years of observation for the time period of 1970-2010 were selected to study the relationship. Floods were defined using the mathematical approach, with the 99th percentile chosen to extract all of the extreme flood peaks in all of the twelve series.

Three days of below percentile values of mean daily discharge meant that if values went above threshold in the fourth day they would be considered to be part of a different event. All tropical cyclones that passed within a 500 km of the island’s coast were considered in this study, this resulted in 86 storms. A flood event was associated with a tropical cyclone if the peak was observed two days before or seven days after the day of closest approach to land. An extreme value analysis (EVA) point process approach

102 was chosen to model the statistical properties of the flood distribution of the different series by identifying changes in the GEV parameters location (central tendency or mean), scale (variance) and shape (skewness) that resulted from the model. The point process model was implemented to the entire series of extreme flood peaks and the series without TCs. This was done to examine if any changes took place in the GEV parameters when TCs were removed from the data.

Of the 86 TCs identified in this study only 51 produce at least one flood event over Puerto Rico over the 1970-2010 period. TCs tend to produce the highest count of individual flood peaks (>30%) in the eastern interior region of the island while a general decrease in the percentage is observed as move to the western region. The point process was able to model the extreme flood peaks in stations in the western region of the island more efficiently since TCs were not the main producers of floods in that area.

The point process was not able to efficiently model the extreme flood peaks of the entire series in some of the stations in the eastern interior since TCs were responsible for their highest flood peaks. However, when the point process was fitted to the series without

TCs for those stations in the east, the model estimates improved and a reduction in all of the GEV parameters was evident. The GEV parameters that resulted from the point process model of the entire series were compared to the results obtained when TCs were removed from the series. The location (mean) exhibits a decrease in all of the sites, yet decreases in the east are larger than the ones in the west and central areas of the island. This suggests that TCs tend to have a stronger effect on the central tendency of the flood peak distribution in stations on the eastern region of the island.

Similar results were found when the scale (variability) of both series were compared,

103 with variance increasing in the east when TCs are present and declining when they are remove from the data. The shape parameter which has been used in previous studies to study the upper tail properties of the flood distribution was also found to change when

TCs were removed from the series. The stations on the eastern interior region of the island tend to have larger shape values when TCs are included in the series, while stations in the northeast and west tend to have a lower value of the shape parameter.

The TCs that were responsible for much of the individual flood peaks had mean rainfall values of 147 mm over the whole island, were located at an average distance of

85.84 from the island’s coast and embedded in moisture environments of 48.64 mm of columnar precipitable water. Hurricanes like Hugo (1989), Hortense (1996) and

Georges (1998) and tropical storms and depressions like Eloise (1975), Klaus (1984) and Isabel (1985) were among the closer TCs to land that were over high moisture environments and produce a high number of individual flood peaks at the 99th percentile threshold. The spatial distribution of rainfall associated with the top individual flood peak producing TCs exhibits a high concentration of precipitation in the southeast and interior region of the island, while a general decrease in rain is evident as we move west. This pattern of rainfall explains the point process model results. The higher values of the

GEV parameters location, scale and shape tend to concentrate in the eastern interior region of the island, while lower values are found in the west. These results suggest that rainfall from TCs have a strong effect on the extreme flood properties in stations in the eastern interior, yet the storms have a weaker influence on the flood peaks in the western and northeastern regions of the island.

104

Chapter 4 Limitations

The main limitation of this chapter is that not all water basins have stations with discharge data for the time period of 1970-2010. For example, there are no stations with data available for the period of study in the metropolitan area of San Juan where much of the population affected by floods lives. Other areas in the northwest and south don’t have stations with available data. Land use and land cover change is also an important factor that is not considered in this study since changes in the cover and use of land can modify the flow of water over the surface, hence affecting the amount of rainfall need it to produce a flood.

105

Rio Culebrinas Near Moca

Rio Grande De Añasco

Rio Gurabo At Gurabo

Rio Grande De Loiza Caguas

Figure 4-1. The island of Puerto Rico and the stations with complete daily discharge data for the 1970-2010 period and their respective water drainage basins.

Figure 4-2. Tracks of all 86 TCs that passed within a 500 km radius of the island’s coast with the TCs that caused extreme floods (99th percentile) in red.

106

Table 4-1. Stations with complete mean daily discharge for the 1970-2010 period. Site # Site ID Daily Mean Discharge Station Lat Lon Elev (m) 50061800 1 Rio Canovanas Near Campo Rico 18.19 -65.53 225 50147800 2 Rio Culebrinas Near Moca 18.21 -67.05 45 50063800 3 Rio Espiritu Santo Near Rio Grande 18.21 -65.48 40 50071000 4 Rio Fajardo Near Fajardo 18.17 -65.41 136 50144000 5 Rio Grande De Anasco Near San Sebastian 18.17 -67.03 103.7 50031200 6 Rio Grande Manati Near Morovis 18.17 -66.24 440 50055000 7 Rio Grande De Loiza At Caguas 18.14 -66 164.04 50035000 8 Rio Grande De Manati At Ciales 18.19 -66.27 140 50057000 9 Rio Gurabo At Gurabo 18.15 -65.58 150 50112500 10 Rio Inabon At Real Abajo 18.05 -66.33 410 50046000 11 Rio De La Plata At HWY 2 Near Toa Alta 18.24 -66.15 9.15 50028000 12 Rio Tanama Near Utuado 18.18 -66.46 948

Figure 4-3. Scatter plot series of mean daily discharge data (a) and the 99th percentile series with the floods (b) associated with TCs for the station at Rio Canovanas near Campo Rico.

107

Table 4-2. Descriptive statistics of mean daily discharge for the entire series, the series with TCs removed and the three maximum flood events and their respective TCs or dates. Entire Series Series No TCs Max Mean Max Mean Site (mm) (mm) (mm) (mm) Max Event 2nd Max 3rd Max 50061800 119.78 0.80 89.48 0.71 Hortense (96) 10/9/70 David (79) 50147800 481.39 8.49 376.61 8.19 Georges (98) 10/21/72 Jeanne (04) 50063800 73.62 1.65 73.62 1.58 12/7/87 Georges (98) Hugo (89) 50071000 249.19 1.88 81.27 1.77 Hugo (89) Eloise (75) 11/12/03 50144000 1979.35 10.06 540.85 9.42 Georges (98) Eloise (75) 5/18/85 50031200 484.22 2.55 484.22 2.40 5/18/85 10/9/70 Hortense (96) 50055000 566.34 6.28 506.87 5.76 Hortense (96) 10/9/70 11/27/87 50035000 1209.13 7.12 1209.13 6.61 5/18/85 Isabel (85) Hortense (96) 50057000 741.90 3.71 560.67 3.26 Hortense (96) 11/27/1987 10/9/70 50112500 70.79 0.51 22.60 0.47 Eloise(75) Isabel (85) 10/11/05 50046000 1928.37 7.65 1132.67 6.52 Hortense (96) Georges (98) 10/7/70 50028000 92.31 1.42 53.52 1.36 Georges (98) Isabel (85) Eloise (75) *Landfalling TCs are in bold.

Figure 4-4. Percentage of mean daily discharge values above the 99th percentile threshold (floods) that were associated with the passage of TCs.

108

Figure 4-5. Point process approach diagnostics probability plot (a) and quantile plot (b) for the entire series of Rio Gurabo. Probability plot (c) and quantile plot (d) for the entire series of Rio Culebrinas near Moca. Probability plot (e) and quantile plot (f) for the series with TCs removed of Rio Gurabo. Probability plot (g) and quantile plot (h) for the series with TCs removed of Rio Culebrinas near Moca.

Figure 4-6. Point process approach diagnostics probability plot (a) and quantile plot (b) for the entire series of Rio Grande De Loiza near Caguas. Probability plot (c) and quantile plot (d) for the entire series of Rio Grande De Añasco. Probability plot (e) and quantile plot (f) for the series with TCs removed of Rio Grande De Loiza near Caguas. Probability plot (g) and quantile plot (h) for the series with TCs removed of Rio Grande De Añasco.

109

Figure 4-7. Daily mean discharge above the 99th percentile of the entire series (a), the series with the TCs removed (b) and the percentage change between the series (c).

110

Figure 4-8. GEV parameters of the entire series in x and the series with TCs removed in y for Location (a), Scale (b) and Shape (c).

111

Figure 4-9. GEV location parameter for the entire series (a), for the series with TCs removed (b) and the percentage change between both series (c).

Figure 4-10. GEV scale parameter for the entire series (a), for the series with TCs removed (b) and the percentage change between both series (c).

112

Figure 4-11. GEV shape parameter for the entire series (a), for the series with TCs removed (b) and the percentage change between both series (c).

113

Figure 4-12. Flood frequency probability for the entire series (a) and the series without TCs (b) for Rio Grande De Loiza at Caguas. Flood frequency probability for the entire series (c) and the series without TCs (d) for Rio Gurabo. Flood frequency probability for the entire series (e) and the series without TCs (f) for Rio Añasco. Flood frequency probability for the entire series (g) and the series without TCs (h) for Rio Culebrinas.

Table 4-3. Characteristics of TCs associated with the highest number of flood peaks over the twelve mean discharge stations. Flood Max Precip Mean Precip TC Distance Moisture TCs Year Month/Days Cat Days (mm) (mm) (km) (mm) Klaus 1984 11/06-11/08 TS 34 179.50 72.01 4.43 52.85 David 1979 08/29-08/31 H5 30 382.60 237.56 125.00 46.52 Jeanne 2004 09/15-09/17 TS 29 370.80 190.35 0.00 51.39 Isabel 1985 10/06-10/08 TD 28 690.10 186.72 221.20 48.07 Lenny 1999 11/17-11/19 H3 28 235.50 99.37 123.80 53.74 Georges 1998 09/21-09/23 H3 24 577.80 271.43 0.00 49.56 Ike 2008 09/25-09/27 H3 23 111.30 28.67 376.10 48.10 Hortense 1996 09/09-09/11 H1 22 552.20 209.74 0.00 49.72 Hugo 1989 09/17-09/19 H4 21 285.80 84.14 0.00 52.88 Eloise 1975 09/15-09/17 TS 18 591.80 279.15 68.00 43.58 Frederic 1979 08/30-09/01 TS 17 360.20 106.28 0.00 49.48 Chris 1988 08/24-08/26 TD 15 304.50 158.91 31.70 43.65 Olga 2007 12/10-12/12 TS 14 209.80 99.89 0.00 40.00 Debby 1982 09/13-09/14 TD 13 212.10 94.86 50.00 47.19 Debby 2000 08/22-08/24 H1 12 235.00 85.96 287.41 52.85 Means TS 21.87 353.27 147.00 85.84 48.64 114

Figure 4-13. Tracks of the TCs that were associated with extreme flood peaks (99th percentile) over Puerto Rico.

115

Figure 4-14. Maps of the standard deviation (a), coefficient of variation (b) and mean rainfall (c) for the 15 TCs associated with the highest numbers of flood peaks

116

CHAPTER 5 CONCLUSION

The main purpose of this dissertation was to explore the climatology of tropical cyclone rainfall in the Caribbean island of Puerto Rico. Three main questions related to

TCR were asked in order to improve our understanding of the role that rainfall from TCs play in the physical geography of Puerto Rico. The first question focused on why some

TCs bring more rainfall than others to the island, by examining the environmental conditions and storm-specific factors of more than 86 storms during the 1970-2010 period that were within a 500 km radius of the island’s coast. The second question looked at the spatial distribution of rainfall associated with the passage of TCs and its contribution to the overall climatology of the island in order to find areas of high TCR concentration and also identify regions that strongly depend on storms for their precipitation. The third question dealt with exploring the relationship between extreme flood events in Puerto Rico and the passage of tropical cyclones in order to determine if

TCs control the flood peaks in different regions of the island. All of these questions were explored in three different chapters and their main findings are summarized below.

Tropical Cyclone Rainfall over Puerto Rico and its Relations to Environmental and Storm Specific Factors

The first question was examined in Chapter 2, there an analysis of TCR over

Puerto Rico was done to understand how characteristics of the environment and each storm contributed to the variability of mean and maximum storm rainfall. Daily rainfall from 32 stations were used to calculate a mean TCR value for the island and the maximum observed precipitation for each storm. Correlation analyses, principal component regression (PCR) procedures and Mann-Whitney U tests were employed to

117 identify which environmental and storm specific factors were the most significant contributors to mean and maximum TCR variability over the island.

TCR variability over Puerto Rico were total precipitable water, proximity to the center of the storm, mid-level relative humidity, duration, longitude and horizontal translation speed. Results from PCR models show that the component combining moisture and longitude accounts for most of the variability in mean and maximum TCR over the island. Key thresholds for high rainfall production were environments with TPW values greater than 44 mm and/or MRH values greater than 44% averaged over the 700-500 hPa layer, with storm tracks located predominantly over or west of Puerto Rico in their average longitude position. TCs that passed close to or over the island that moved slowly and had long durations also accounted for a large portion of TCR variability. High mean (> 50 mm) and maximum (> 300 mm) TCR values resulted from TCs whose centers came within 233 km of land, moved at speeds less than 6.4 ms-1, and/or spent at least 42 hours in the study region. Taken together both of these components accounted for 70% of the variability in the mean TCR model and 52% in the maximum

TCR model.

The main contribution of this chapter was the identification of important thresholds associated with widespread rainfall for each of the statistically significant environmental and storm specific factors, which could aid in future TCR forecasting for

118

Puerto Rico and other islands in the tropics. We found that TPW and PRX account for much of the variability in mean and maximum TCR, yet including the factors MRH,

Spatial Distribution of Tropical Cyclone Rainfall and its Contribution to the Climatology of Puerto Rico

The second question was the focused of Chapter 3, here the spatial distribution of rainfall associated with TCs over Puerto Rico and their overall contribution to the precipitation climatology of the island was examined. The findings of Chapter 2 suggest that there are some important factors that explain around 70% of the TCR variability in the island, yet the spatial distribution of the rain associated with the storms was not examined. Here daily and monthly rainfall data at 32 sites where used to compute a tropical cyclone rainfall (TCR) value and a contribution percentage. In the spatial distribution analysis of the study the focus was on understanding the patterns of five

TCR groups based on their moisture (TPW) and proximity to the island’s coast (PRX) values which included; a) all 86 TCs, b) high TPW and close PRX TCs, c) high TPW and far PRX TCs, d) low TPW and close PRX TCs and e) low TPW and far PRX TCs.

119

Geostatistical interpolation models that included ordinary kriging (ORK) and ordinary cokriging (OCK) were used to generate predicted surfaces of TCR for all five TC groups in order to examine similarities or differences in the spatial distribution of rainfall associated with TCs with different TPW and PRX characteristics. The ORK method was used to produce predicted surfaces of percentage contributions for the hurricane season months in order to explore the spatio-temporal characteristics of rainfall contributed by TCs.

This study found that the highest rainfall values associated with the passage of

TCs are concentrated in the eastern, southeastern and central regions of the island. A general east to west decrease in rainfall values was also found. The highest rainfall values were attributed to TCs over high TPW environment and to storms that were within 230 km of the island’s coast. Most of the TCs in this group were also identified in the top 23 rainfall producing storms on Chapter 2. Farther TCs, even with high moisture environments over Puerto Rico, tend to be associated with lower rainfall amounts over land when compared to the closer and higher moisture TCs. The closer TCs over lower moisture environments exhibit a more random spatial distribution of precipitation over the island, instead of the usual cluster of high-low values found in the groups that considered all 86 events and the high moisture-close proximity storms. The TCs that had farther circulation centers from land and low moisture environments exhibited the lowest TCR values of all of the TC groups. Most of these were also identified as the lowest mean rainfall TCs in Chapter 2. The OCK method performed better than the

ORK in four of the five TC groups, with the exception of the low TPW and close PRX storms, which coincides with the literature that suggests that adding information about

120 topography (elevation, slope and aspect) as covariates improves the models prediction capabilities. When southern and northern TCs were compared we found that storms south of Puerto Rico produce more rainfall (13.6-24 mm) in the southeastern interior, northeast and northwest while TCs north of the island were associated with higher values (4.28-19.5 mm) in the western interior region of the island.

September shows that almost the entire island received more than 10% of its rainfall from TCs for the 1970-2010 period. October was identified as the rainiest month in terms of average precipitation, yet its contribution percentages were found to be lower than those found for August and September. October exhibited low rainfall contributions from storms, which suggests that heavy rainfall events over Puerto Rico are not necessarily associated with TCs.

121

Extreme Floods and their Relationship with Tropical Cyclones in Puerto Rico

The third question was explored in chapter 4, which focused on examining the relationship between extreme flood events in Puerto Rico and tropical cyclones. Mean daily discharge data for twelve station with 41 years of observation for the time period of

1970-2010 were selected to study the relationship. Floods were defined using the mathematical approach, with the 99th percentile chosen to extract all of the extreme flood peaks in all of the twelve series. Three days of below percentile values of mean daily discharge meant that if values when above threshold in the fourth day they would be considered to be part of a different event. A flood event was associated with a tropical cyclone if the peak was observe two days before or seven days after the day of closest approach to land. An extreme value analysis (EVA) point process approach was chosen to model the statistical properties of the flood distribution of the different series by identifying changes in the GEV parameters location (central tendency), scale

(variance) and shape (skewness) that resulted from the model. The point process model was implemented to the entire series of extreme flood peaks and the series without

TCs. This was done to examine if any changes took place in the GEV parameters when

TCs were removed from the data.

Of the 86 TCs identified in this study only 51 produce at least one flood event over Puerto Rico over the 1970-2010 period. The highest rainfall producing TCs in

Chapters 1 and 2 were also part of the 51 storms that were associated with flood peaks.

TCs tend to produce the highest count of individual flood peaks (>30%) in the eastern interior region of the island while a general decrease in the percentage is observed as move to the western region. The point process was able to model the extreme flood

122 peaks in stations in the western region of the island more efficiently since TCs were not the main producers of floods in that area. The point process was not able to efficiently model the extreme flood peaks of the entire series in some of the stations in the eastern interior since TCs were responsible for their highest flood peaks. However, when the point process was fitted to the series without TCs for those stations in the east, the model estimates improved and a reduction in all of the GEV parameters was evident.

This coincides with the findings of Chapter 3 which shows that TC rainfall tends to be concentrated in the southeastern region of the island. The GEV parameters that resulted from the point process model of the entire series were compared to the results obtained when TCs were removed from the series. The location (mean) exhibits a decrease in all of the sites, yet decreases in the east are larger than the ones in the west and central areas of the island. This suggests that TCs tend to have a stronger effect on the central tendency of the flood peak distribution in stations on the eastern region of the island. Similar results were found when the scale (variability) of both series were compared, with variance increasing in the east when TCs are present and declining when they are remove from the data. The shape parameter which has been used in previous studies to study the upper tail properties of the flood distribution was also found to change when TCs were removed from the series. The stations on the eastern interior region of the island tend to have larger shape values when TCs are included in the series, while stations in the northeast and west tend to have a lower value of the shape parameter. The changes in the GEV parameters (location, scale and shape) when TCs are removed from the entire series agree with the general findings of

Chapters 1 and 2. The combined results of the three chapters show that closer and

123 higher moisture TCs produce more rainfall in the southeastern region of the island and for that reason the storms tend to have a strong connection with floods in the eastern interior and weak influence with the peaks in the west.

The TCs that were responsible for much of the individual flood peaks had mean rainfall values of 147 mm over the whole island, where located at an average distance of 85.84 from the island’s coast and embedded in moisture environments of 48.64 mm of columnar precipitable water. Those thresholds were similar to the ones found for the top rainfall producing TCs in Chapters 1 and 2. Hurricanes like Hugo (1989), Hortense

(1996) and Georges (1998) and tropical storms and depressions like Eloise (1975),

Klaus (1984) and Isabel (1985) were among the closer TCs to land that were over high moisture environments and produce a high number of individual flood peaks at the 99th percentile threshold. The spatial distribution of rainfall associated with the top individual flood peak producing TCs exhibits a high concentration of precipitation in the southeast and interior region of the island, while a general decrease in rain is evident as we move west. This pattern of rainfall explains the GEV point process model results. The higher values of the GEV parameters location, scale and shape tend to concentrate in the eastern interior region of the island, while lower values are found in the west. These results suggest that rainfall from TCs closer higher moisture storms have a strong effect on the extreme flood properties in stations in the eastern interior, yet they have a weak influence on the flood peaks in the western and northeastern regions of the island.

Dissertation Contributions

This dissertation made important contributions to the fields of physical geography, climatology and hydrology. By examining a climatology of tropical cyclones

124 and their associated rainfall in Puerto Rico this dissertation added substantial information about the role that tropical cyclones play in the rainfall climatology of Puerto

Rico. Most of the TC studies in the island focused on individual cases, while this dissertation examined the characteristics of the rainfall from a climatological perspective that included a vast number of storms. This dissertation contributed information about which TCs brought more rainfall to the island and identified important thresholds in environmental moisture, storm center proximity to land, duration and horizontal translation speed that could be beneficial to rainfall forecasts of future TCs. This dissertation contributed new knowledge about the areas in which much of the rainfall associated with TCs tends to be concentrated, which is the southeast and central mountains region of the island, while it also improved our understanding of the contribution of TCs to the overall precipitation by identifying the south as the area that depends the most on rainfall from storms. This dissertation improved our understanding of the hydrology of extreme floods associated with TCs, by showing that storms tend to have a higher influence on flood peaks in the eastern interior region of the island while a general decrease in their influence is evident as we move to the west and northeast.

Overall, this dissertation improved our understanding and produced new knowledge about the spatiotemporal characteristics of rainfall associated with TCs and their relationship to floods in small island environments like Puerto Rico.

125 with these events. The results of these analyses also identify characteristics of storms that produced heavy rainfall and floods, and key thresholds that could later improve the precipitation forecasting of these extreme events in Puerto Rico and other islands in the tropics. The findings in this dissertation advance our understanding of tropical cyclone rainfall prediction and mitigation in the tropics. With this new knowledge on tropical cyclone rainfall over Puerto Rico it is expected that the awareness and preparedness towards these events will significantly improve, which will then translate in fewer fatalities and lower economic losses associated with the passage of these storms. The findings of this dissertation provide a new dimension to our understanding of tropical cyclone rainfall over small islands and it has significant implications in future urban and environmental planning projects in Puerto Rico.

Future Directions

Future work will use satellite based rainfall data from the Tropical Rainfall

Measuring Mission (TRMM) and Global Precipitation Measuring Mission (GPM) to study all TCs, especially those that were at a significant distance from land and yet ended up producing significant precipitation over Puerto Rico. Another project will focus on the changing contribution of TCs over time to examine if rainfall associated with storms has been decreasing, increasing or remain unchanged over the last 60 years in Puerto Rico.

Another future research project will examine the role that TCs play in alleviating drought conditions in Puerto Rico and other islands in the Caribbean. Future work will use rain gauge-satellite based data from the Climate Hazards Group InfraRed Precipitation with

Stations data (CHIRPS) to examine rainfall characteristics during severe drought

126 periods in the Caribbean to determine if lower than normal frequency of TCs was the main factor behind a decrease in rainfall amounts in the region.

127

LIST OF REFERENCES

Abdi H, Williams LJ. 2010. Principal component analysis. Wiley Interdisciplinary Reviews: Computational Statistics 2: 433-459. DOI: 10.1002/wics.101.

Allen MR, Ingram WJ. 2002. Constraints on future changes in climate and the hydrologic cycle. Nature 419: 224-232. DOI: 10.1038/nature01092.

Alliss RJ, Raman S, Chang, SW. 1992. Special sensor microwave/imager (SSM/I) observations of Hurricane Hugo (1989). Monthly Weather Review 120: 2723- 2723. DOI: 10.1175/1520-0493(1992)120<2723: SSMOOH>2.0.CO; 2.

Anthes RA. 1982. Tropical cyclones: their evolution, structure and effects. Meteorological Monographs, American Meteorological Society: Boston, USA.

Barnes HH, Bogart DB. 1961. Floods of September 6, 1960, in Eastern Puerto Rico (Vol. 451). US Government Printing Office.

Barnston AG, Livezey R.E. 1987. Classification, seasonality and persistence of low-frequency atmospheric circulation patterns. Monthly Weather Review 115: 1083- 1126. DOI: 10.1175/1520-0493(1987)115<1083: CSAPOL>2.0.CO; 2.

Bennet SP, Mojica R. 1998. Hurricane Georges preliminary storm report: from the tropical Atlantic to the United States Virgin Islands and Puerto Rico. A NOAA National Weather Service NWSFO San Juan Official Report 1-15.

Bessafi M, Lasserre-Bigorry A, Neumann CJ, Pignolet-Tardan F, Payet D, Lee- Ching-Ken M.2002. Statistical prediction of tropical cyclone motion: An analog–CLIPER approach. Weather & Forecasting 17: 821-831. DOI: 10.1175/1520- 0434(2002)017<0821: SPOTCM>2.0.CO; 2.

Birdsey RA, Weaver PL. 1987. Forest area trends in Puerto Rico. United States Department of Agriculture. Southern Forest Experiment Station. New Orleans, LA.

Boose ER, Foster DR., Fluet M. 1994. Hurricane impacts to tropical and temperate forest landscapes. Ecological Monographs 64: 369-400. DOI: 10.2307/2937142 . Boose ER, Serrano MI, Foster DR. 2004. Landscape and regional impacts of hurricanes in Puerto Rico. Ecological Monographs 74: 335-352. DOI: 10.1890/02-4057.

Bosart LF, Carr FH. 1978. A case study of excessive rainfall centered around Wellsville, New York, 20-21 June 1972. Monthly Weather Review 106: 348-362. DOI: 10.1175/1520-0493(1978)106<0348: ACSOER>2.0.CO; 2.

128

Brunsdon C, McClatchey J, Unwin DJ. 2001. Spatial variations in the average rainfall–altitude relationship in Great Britain: an approach using geographically weighted regression. International Journal of Climatology, 21(4): 455-466.

Carrera-Hernández JJ, Gaskin SJ. 2007. Spatio temporal analysis of daily precipitation and temperature in the Basin of Mexico. Journal of Hydrology, 336(3): 231- 249.

Carter MM, Elsner JB. 1996. Convective rainfall regions of Puerto Rico. International Journal of Climatology, 16(9): 1033-1043.

Caviedes CN. 1991. Five hundred years of hurricanes in the Caribbean: their relationship with global climatic variabilities. GeoJournal 23: 301-310. DOI: 10.1007/BF00193603.

Cecil DJ. 2007. Satellite‐derived rain rates in vertically sheared tropical cyclones. Geophysical Research Letters 34. DOI: 10.1029/2006GL027942.

Cerveny RS, Newman LE. 2000. Climatological relationships between tropical cyclones and rainfall. Monthly Weather Review 128: 3329-3336. DOI: 10.1175/1520- 0493(2000)128<3329: CRBTCA>2.0.CO; 2.

Coles S, Bawa J, Trenner L, Dorazio P. 2001. An introduction to statistical modeling of extreme values (Vol. 208). London: Springer.

Colon T, José A. 2009. Climatología de Puerto Rico. La Editorial Universidad de Puerto Rico: San Juan.

Corbosiero KL, Molinari J. 2003. The relationship between storm motion, vertical wind shear, and convective asymmetries in tropical cyclones. Journal of the Atmospheric Sciences 60: 366-376. DOI: 10.1175/1520-0469(2003)060<0366: TRBSMV>2.0.CO; 2.

Chen SS, Knaff JA, Marks Jr FD. 2006. Effects of vertical wind shear and storm motion on tropical cyclone rainfall asymmetries deduced from TRMM. Monthly Weather Review 134: 3190-3208. DOI: 10.1175/MWR3245.1.

Cheung K. Huang LR, Lee CS. 2008. Characteristics of rainfall during tropical cyclone periods in Taiwan. Natural Hazards & Earth System Sciences, 8(6): 1463-1474.

Chu PS. 2002. Large-scale circulation features associated with decadal variations of tropical cyclone activity over the central North Pacific. Journal of Climate 15: 2678-2689. DOI: 10.1175/1520-0442 DOI:/10.1175/1520-0442 (2002)015<2678: LSCFAW>2.0.CO; 2.

129

Dai A, Giorgi F, Trenberth KE. 1999. Observed and model‐simulated diurnal cycles of precipitation over the contiguous United States. Journal of Geophysical Research: Atmospheres (1984–2012) 104: 6377-6402. DOI: 10.1029/98JD02720.

Dare RA, Davidson NE, McBride JL. 2012. Tropical cyclone contribution to rainfall over Australia. Monthly Weather Review, 140(11): 3606-3619.

DeMaria M, Kaplan J. 1994. Sea surface temperature and the maximum intensity of Atlantic tropical cyclones. Journal of Climate 7: 1324-1334. DOI: 10.1175/1520-0442. DOI: 10.1175/1520-0442(1994)007<1324: SSTATM>2.0.CO; 2.

DiMego GJ, Bosart LF.1982. The transformation of tropical storm Agnes into an extratropical cyclone. Part II: Moisture, vorticity and kinetic energy budgets. Monthly Weather Review 110: 412-433. DOI: 10.1175/1520-0493(1982)110<0412: TTOTSA>2.0.CO; 2.

Diodato N, Ceccarelli M. 2005. Interpolation processes using multivariate geostatistics for mapping of climatological precipitation mean in the Sannio Mountains (southern Italy). Earth Surface Processes and Landforms, 30(3): 259-268.

Dirks KN, Hay JE, Stow CD, Harris D. 1998. High-resolution studies of rainfall on Norfolk Island: Part II: Interpolation of rainfall data. Journal of Hydrology, 208(3): 187- 193.

Dunion JP, Velden CS. 2004. The impact of the Saharan air layer on Atlantic tropical cyclone activity. Bulletin of the American Meteorological Society 85(3): 353-365. DOI: 10.1175/BAMS-85-3-353.

Elsner JB, Kara AB. 1999. Hurricanes of the North Atlantic: Climate and society. Oxford University Press.

Farfán LM, Fogel I. 2007. Influence of tropical cyclones on humidity patterns over Southern Baja California, México. Monthly Weather Review 135: 1208-1224. DOI:10.1175/MWR3356.1.

Fitzpatrick PJ, Knaff, JA, Landsea CW, Finley SV.1995. Documentation of a systematic bias in the aviation model's forecast of the Atlantic tropical upper- tropospheric trough: Implications for tropical cyclone forecasting. Weather and forecasting 10(2): 433-446. DOI: 10.1175/1520- 0434(1995)010<0433:DOASBI>2.0.CO;2.

Gamble DW, Parnell DB, Curtis S. 2008. Spatial variability of the Caribbean mid‐ summer drought and relation to north Atlantic high circulation. International Journal of Climatology, 28(3): 343-350.

130

Gao S, Meng Z, Zhang F, Bosart LF. 2009. Observational analysis of heavy rainfall mechanisms associated with severe Tropical Storm Bilis (2006) after its landfall. Monthly Weather Review 137: 1881-1897. DOI:10.1175/2008MWR2669.1.

Georgiou PN, Davenport AG, Vickery BJ. 1983. Design wind speeds in regions dominated by tropical cyclones. Journal of Wind Engineering and Industrial Aerodynamics 13:139-152. DOI: 10.1016/0167-6105(83)90136-8.

Goldenberg SB, Shapiro LJ. 1996. Physical mechanisms for the association of El Niño and West African rainfall with Atlantic major hurricane activity. Journal of Climate 9(6): 1169-1187. http://dx.doi.org/10.1175/1520-0442 (1996)009<1169: PMFTAO>2.0.CO; 2.

Goodyear HV. 1968. Frequency and areal distributions of tropical storm rainfall in the United States coastal region on the Gulf of Mexico. US Department of Commerce, Environmental Science Services Administration, Weather Bureau.

Gray WM. 1984. Atlantic seasonal hurricane frequency. Part I: El Niño and 30 mb quasi-biennial oscillation influences. Monthly Weather Review 112(9): 1649-1668. http://dx.doi.org/10.1175/1520-0493 (1984)112<1649: ASHFPI>2.0.CO; 2.

Groisman PY, Knight RW, Karl TR., Easterling DR., Sun B, Lawrimore JH. 2004. Contemporary changes of the hydrological cycle over the contiguous United States: Trends derived from in situ observations. Journal of Hydrometeorology 5: 64-85. DOI: 10.1175/1525-7541(2004)005<0064: CCOTHC>2.0.CO; 2.

Goovaerts P. 2000. Geostatistical approaches for incorporating elevation into the spatial interpolation of rainfall. Journal of Hydrology, 228(1): 113-129.

Goovaerts P. 1997. Geostatistics for natural resources evaluation. Oxford university press.

Guo Q, Matyas CJ. 2016. Comparing the spatial extent of Atlantic basin tropical cyclone wind and rain fields prior to land interaction. Physical Geography. DOI: 10.1080/02723646.2016.1142929.

Hauke J, Kossowski T. 2011. Comparison of values of Pearson's and Spearman's correlation coefficients on the same sets of data. Quaestiones Geographicae 30(2): 87-93 . Hebert PJ. 1976. Atlantic Hurricane Season of 1975. Monthly Weather Review, 104(4): 453-465. http://dx.doi.org/10.1175/15200493(1976)104<0453:AHSO>2.0.CO;2

Hebert, PJ. 1980. Atlantic hurricane season of 1979. Monthly Weather Review, 108(7): 973-990. http://dx.doi.org/10.1175/1520-0493(1980)108<0973:AHSO>2.0.CO;2

131

Hernández Ayala JJ and Matyas CJ. 2015. Tropical cyclone rainfall over Puerto Rico and its relations to environmental and storm-specific factors. International Journal of Climatology. . doi: 10.1002/joc.4490

Hill KA, Lackmann GM. 2009. Influence of environmental humidity on tropical cyclone size. Monthly Weather Review, 137: 3294–3315. DOI: 10.1175/2009MWR2679.1. Hotelling H. 1933. Analysis of a complex of statistical variables into principal components. Journal of Educational Psychology 24: 417-441. DOI: 10.1037/h0071325.

Huff FA. 1970. Sampling errors in measurement of mean precipitation. Journal of Applied Meteorology 9: 35-44. DOI: 10.1175/1520-0450(1970)009<0035: SEIMOM>2.0.CO; 2.

Inoue M, Handoh IC, Bigg GR. 2002. Bimodal distribution of tropical cyclogenesis in the Caribbean: characteristics and environmental factors. Journal of Climate 15: 2897-2905. DOI: 10.1175/1520-0442 (2002)015<2897: BDOTCI>2.0.CO; 2.

Isaaks EH, Srivastava RM. 1989. An introduction to applied geostatistics. New York: Oxford University Press.

Janowiak JE. 1988. The global climate for March-May 1988: The end of the 1986-87 Pacific warm episode and the onset of widespread drought in the United States. Journal of Climate 1: 1019-1040.

Jibson RW. 1989. Debris ﬂows in southern Puerto Rico. Landslide processes of the eastern United States and Puerto Rico. Geological Society of America special paper 236: 29-55.

Jiang H, Halverson JB, Simpson J, Zipser EJ. 2008. Influence of environmental moisture on TRMM‐derived tropical cyclone precipitation over land and ocean. Geophysical Research Letters 35. DOI: 10.1029/2008GL034658.

Jury MR, Sanchez DM. 2009. Composite meteorological forcing of Puerto Rican springtime flood events. Weather & Forecasting 24: 262-271. DOI: 10.1175/2008WAF2222151.1.

Kalnay E, Kanamitsu M, Kistler R, Collins W, Deaven D, Gandin L, Joseph D. 1996. The NCEP/NCAR 40-year reanalysis project. Bulletin of the American Meteorological Society 77: 437-471. DOI: 10.1175/1520-0477(1996)077<0437: TNYRP>2.0.CO; 2.

Keelings D, Waylen PR. 2014. Investigating teleconnection drivers of bivariate heat waves in Florida using extreme value analysis. Climate Dynamics. DOI 10.1007/s00382-014-2345-8

132

Knapp KR, Kruk MC. 2010. Quantifying interagency differences in tropical cyclone best-track wind speed estimates. Mon. Weather Rev. 138: 1459–1473, doi: 10.1175/2009MWR3123.1.

Knight DB, Davis RE. 2007. Climatology of tropical cyclone rainfall in the southeastern United States. Physical Geography, 28(2): 126-147.

Knight DB, Davis RE. 2009. Contribution of tropical cyclones to extreme rainfall events in the southeastern United States. Journal of Geophysical Research: Atmospheres (1984–2012), 114(D23).

Konrad CE, Perry LB. 2010. Relationships between tropical cyclones and heavy rainfall in the Carolina region of the USA. International Journal of Climatology, 30(4): 522-534.

Kostaschuk R, Shugar D, Best J, Parsons D, Lane S, Hardy R, Orfeo O. 2009. Suspended sediment transport and deposition over a dune: Río Paraná, Argentina. Earth Surface Processes and Landforms 34: (12) 1605-1611.

Laing, AG.(2004. Cases of heavy precipitation and flash floods in the Caribbean during El Nino winters. Journal of Hydrometeorology 5:(4) 577-594.

Larsen MC, Santiago-Roman AS. 2001. Mass wasting and sediment storage in a small montane watershed: an extreme case of anthropogenic disturbance in the humid tropics. Water Science and Application 4: 119-138.

Larsen MC, Simon A. 1993. A rainfall intensity-duration threshold for landslides in a humid-tropical environment, Puerto Rico. Geografiska Annaler A 75: 13-23.

Larsen MC, Torres-Sanchez AJ. 1998. The frequency and distribution of recent landslides in three montane tropical regions of Puerto Rico. Geomorphology 24: 309- 331. DOI: 10.1016/S0169-555X (98)00023-3.

Lonfat M, Marks FD, Chen SS. 2004. Precipitation distribution in tropical cyclones using the Tropical Rainfall Measuring Mission (TRMM) Microwave Imager: A global perspective. Monthly Weather Review 132: 1645-1660. DOI: 10.1175/1520- 0493(2004)132<1645: PDITCU>2.0.CO; 2.

Lonfat M, Rogers R, Marchok T, Marks FD. 2007. A parametric model for predicting hurricane rainfall. Monthly Weather Review 135: 3086-3096. DOI: 101175MWR34331.

Ly S, Charles C, Degré A. 2013. Different methods for spatial interpolation of rainfall data for operational hydrology and hydrological modeling at watershed scale: A review. Biotechnology, Agronomy, Society and Environment. 17(2).

133

Mair A, Fares A. 2010. Comparison of rainfall interpolation methods in a mountainous region of a tropical island. Journal of Hydrologic Engineering.

McBratney AB, Webster R. 1986. Choosing functions for semi‐variograms of soil properties and fitting them to sampling estimates. Journal of Soil Science, 37(4): 617- 639. Matyas, C. J., and M. Cartaya. 2009. Comparing the rainfall patterns produced by Hurricanes Frances (2004) and Jeanne (2004) over Florida. Southeastern Geographer, 49 (2):132-156.

Matyas CJ. 2009. A spatial analysis of radar reflectivity regions within Hurricane Charley (2004). Journal of Applied Meteorology & Climatology 48: 130-142. DOI: 10.1175/2008JAMC1910.1.

Matyas CJ. 2010. Associations between the size of hurricane rain fields at landfall and their surrounding environments. Meteorology and Atmospheric Physics 106: 135-148. DOI: 10.1175/2008JAMC1910.1.

Matyas CJ. 2015. Tropical cyclone formation and motion in the Mozambique Channel. International Journal of Climatology. 35: 375–390. DOI: 10.1002/joc.3985.

Morrison JE, Smith JA, 2002. Stochastic Modeling of Flood Peaks Using the Generalized Extreme Value Distribution. Water Resources Research 38(12):1305, doi: 10.1029/ 2001WR000502.

Nogueira RC, Keim BD. 2011. Contributions of Atlantic tropical cyclones to monthly and seasonal rainfall in the eastern United States 1960–2007. Theoretical and Applied Climatology, 103(1-2): 213-227.

O'Connor JE, Costa JE. 2004. Spatial distribution of the largest rainfall‐runoff floods from basins between 2.6 and 26,000 km2 in the United States and Puerto Rico. Water Resources Research 40(1).

Ogallo LJ. 1989. The spatial and temporal patterns of the East African seasonal rainfall derived from principal component analysis. International Journal of Climatology 9: 145-167. DOI: 10.1002/joc.3370090204.

Oliver Margaret A, R Webster. 1990. Kriging: a method of interpolation for geographical information systems. International Journal of Geographical Information System 4.3 (1990): 313-332.

134

Pasch RJ, Avila LA. 1999. Atlantic hurricane season of 1996. Monthly Weather Review, 127(5): 581-610. http://dx.doi.org/10.1175/1520- 0493(1999)127<0581:AHSO>2.0.CO;2

Pasch RJ, Avila LA, Guiney JL. 2001. Atlantic hurricane season of 1998. Monthly Weather Review, 129 (12): 3085-3123. doi:http://dx.doi.org/10.1175/1520- 2 0493(2001)129<3085: AHSO>2.0.CO; 2. Picó R. 1974. The Geography of Puerto Rico. Chicago, Illinois: Aldine Publishing Company.

Pielke Jr RA, Rubiera J, Landsea C, Fernández ML, Klein R. 2003. Hurricane vulnerability in Latin America and the Caribbean: Normalized damage and loss potentials. Natural Hazards Review 4: 101-114. DOI: 10.1061/ (ASCE) 1527- 6988(2003)4:3(101).

Prudhomme C, Reed DW. 1998. Relationships between extreme daily precipitation and topography in a mountainous region: a case study in Scotland. International Journal of Climatology, 18(13): 1439-1453.

Prudhomme C, Reed DW. 1999. Mapping extreme rainfall in a mountainous region using geostatistical techniques: a case study in Scotland. International Journal of Climatology, 19(12): 1337-1356.

Rao GV, PD MacArthur. 1994. The SSM/I estimated rainfall amounts of tropical cyclones and their potential in predicting the cyclone intensity changes. Monthly Weather Review 122: 1568–1574. DOI: 10.1175/1520-0493(1994)122<1568: TSERAO>2.0.CO; 2.

Rappaport EN. 2000. Loss of life in the United States associated with recent Atlantic tropical cyclones. Bulletin of the American Meteorological Society 81: 2065- 2073. DOI: 10.1175/1520-0477(2000)081<2065: LOLITU>2.3.CO; 2.

Riehl H, Malkus J. 1961. Some aspects of Hurricane Daisy, 19581.Tellus 13: 181-213. DOI: 10.1111/j.2153-3490.1961.tb00077.x.

Rogers R, Chen S, Tenerelli J, Willoughby H. 2003. A numerical study of the impact of vertical shear on the distribution of rainfall in Hurricane Bonnie (1998). Monthly Weather Review 131: 1577-1599. DOI: 10.1175//2546.1.

Rogers R, Marks F, Marchok T. 2009. Tropical cyclone rainfall. Encyclopedia of hydrological sciences. DOI: 10.1002/0470848944.hsa030.

Sadler JC. 1976. A role of the tropical upper tropospheric trough in early season typhoon development. Monthly Weather Review 104: 1266-1278. DOI: 10.1175/1520- 0493(1976)104<1266: AROTTU>2.0.CO; 2.

135

Scatena FN, MC Larsen. 1991. Physical aspects of Hurricane Hugo in Puerto Rico. Biotropica 23: 317-323.

Schuurmans JM, Bierkens, MFP, Pebesma EJ, Uijlenhoet R. 2007. Automatic prediction of high-resolution daily rainfall fields for multiple extents: the potential of operational radar. Journal of Hydrometeorology, 8(6): 1204-1224.

Shepherd MJ, Grundstein A, Mote TL. 2007. Quantifying the contribution of tropical cyclones to extreme rainfall along the coastal southeastern United States. Geophysical Research Letters 34. DOI: 10.1029/2007GL031694.

Simpson RH, Riehl H. 1981. The hurricane and its impact. Baton Rouge: Louisiana State University Press 63: 251-252. DOI: 10.1029/EO063i017p00251-02.

Smith JA, Sturdevant‐Rees P, Baeck ML, Larsen MC. 2005. Tropical cyclones and the flood hydrology of Puerto Rico. Water Resources Research 41. DOI: 10.1029/2004WR00353.

Thorncroft C, Pytharoulis I. 2001. A dynamical approach to seasonal prediction of Atlantic tropical cyclone activity. Weather and forecasting 16(6): 725-734. DOI: 10.1175/1520-0434(2002)016<0725.

Torres-Sierra H. 1997. Hurricane Hortense: Impact on Surface Water in Puerto Rico. US Department of the Interior, US Geological Survey.

Villarini G, Smith JA. 2010. Flood peak distributions for the eastern United States. Water Resources Research 46(6).

Villarini G, Smith JA. 2013. Flooding in Texas: Examination of temporal changes and impacts of tropical cyclones. JAWRA Journal of the American Water Resources Association 49: (4)825-837.

Waylen PR. 1991. Modeling the effects of tropical cyclones on flooding in the Santa Fe River basin, Florida. GeoJournal 23(4) 361-373.

Waylen PR. 1988. Statistical analysis of freezing temperatures in Central and Southern Florida. Int J Climatol 8:607–628

Wu CC, Yen TH, Kuo YH, Wang W. 2002. Rainfall simulation associated with Typhoon Herb (1996) near Taiwan. Part I: The topographic effect. Weather & Forecasting, 17(5): 1001-1015

Zick SE, CJ. Matyas. 2015. Tropical cyclones in the North American Regional Reanalysis: An assessment of spatial biases in location, intensity, and structure. Journal of Geophysical Research- Atmospheres 120 (5):1651-1669.

136

Zick SE, CJ. Matyas. 2015. Tropical cyclones in the North American Regional Reanalysis: The impact of satellite derived precipitation over-ocean. Journal of Geophysical Research- Atmospheres 120 (17):8724-8742.

Zimmermann NE, Kienast F. 1999. Predictive mapping of alpine grasslands in Switzerland: species versus community approach. Journal of Vegetation Science, 469- 482.

Zhu L, Quiring SM. 2013. Variations in tropical cyclone precipitation in Texas (1950 to 2009). Journal of Geophysical Research: Atmospheres, 118(8): 3085-3096.

137

BIOGRAPHICAL SKETCH

José Javier Hernández Ayala, also known as Javy, Javier etc., is a physical geographer that focuses on extreme weather and climate events in his research at the

Department of Geography in the University of Florida. José Javier holds a Bachelor of

Arts in physical and environmental geography from the University of Puerto Rico Rio

Piedras Campus where in 2010 he graduated Summa Cum Laude for finishing his undergraduate degree with a 4.0 GPA. At that institution José worked on a research project that explored the urban heat island effect of the three largest cities in Puerto

Rico were he found that the three urban areas have been experiencing an increase in minimum and maximum temperature over the last 60 years, yet those increasing trends vary in relation to their respective urban area extent.

José also has a Master of Science in geographic information sciences from The

University of Akron in Ohio where he graduated with honors in 2012. In his master’s thesis titled “Spatial and Temporal Changes in Precipitation in Puerto Rico From 1956-

2010” José explored annual, seasonal and monthly trends in rainfall to determine if some regions of Puerto Rico have experienced increasing or decreasing trends. He found an increasing trend in rainfall in the eastern region of the island and a decreasing trend in some stations in the west during the rainy season that could be associated with atmospheric teleconnections. At The University of Akron José also taught two courses as a sole instructor, these were introduction to geography and geography of cultural diversity. After completing his master’s José was accepted in the PhD program at the

Department of Geography in the University of Florida.

138

At the University of Florida he’s been working on his dissertation titled

“Climatology of Tropical Cyclone Rainfall over Puerto Rico: Processes, Patterns and

Impacts”. His research deals with understanding why some tropical cyclones produce more rainfall over land than others, where is that precipitation concentrated, how much does it contribute to the climatology and how is it related to extreme flood events over the island. José has published one of the chapters of his dissertation in the International

Journal of Climatology and he has presented his research in multiple American

Association of Geographers (AAG) and Southeastern Division of the AAG meetings. He was one of the first students in the University of Florida to obtain the new graduate certificate in applied atmospheric sciences. José has also taught four courses as the sole instructor at the University of Florida that include classes like physical geography lab, physical geography, foundations of geographic information system (GIS) and extreme weather. For his future research José intends to explore the spatiotemporal characteristics of rainfall during severe drought periods in the Caribbean to understand how those extreme events are related to climate variability and change. José is on track to obtain his PhD in geography with a concentration in climate science in spring of 2016 and aims to find a job as an Assistant Professor.

139