The document analyzes the relationship between the Atlantic Multidecadal Oscillation (AMO) and Atlantic hurricane counts. It finds that warmer AMO phases are correlated with increased hurricane activity. Statistical models show the AMO's 10-year moving average helps forecast hurricane counts up to 25 years into the future. Based on these relationships, the analysis provides forecasts of increased hurricane activity through 2017 as the AMO is currently in a warm phase near its peak.
The document discusses the recent strong El Niño event and its implications for seasonal rainfall distribution in Belg (FMAM) benefiting areas of Ethiopia. It finds that previous strong El Niño years saw normal to above normal rainfall during Belg, with particularly enhanced rainfall in 1983 and 2010. Based on this, the seasonal rainfall in 2016 is expected to have normal to above normal performance, potentially compensating for low rainfall during the Kiremt season. Further analysis using additional statistical forecasting methods is recommended.
Quantile regression ensemble for summer temperaturesManuel Herrera
This document presents research on using quantile regression and ensembles to model summer temperatures in London from 1961-2010. It finds that quantile regression can better capture relationships between temperature and weather variables like wind speed and cloud cover at extreme quantiles. An ensemble approach is also proposed, combining 50 quantile regression models to generate synthetic weather data for use in built environment studies rather than using complete observed years. Further work would expand these quantile methods and test their use with other weather data.
This document summarizes research on modeling extreme hurricane winds and insured losses in the United States. The researchers use extreme value theory and Bayesian statistical methods to estimate return levels of hurricane winds for different regions and time periods. They find higher return levels are associated with warmer climate conditions. The same methods are applied to model insured hurricane losses and estimate extreme potential losses under different climate scenarios.
Ten most popular software for prediction of cyclonic stormsMrinmoy Majumder
In recent years the frequency and intensity of cyclones and hurricanes have been increased manifold compared to the last decade. As a result, the necessity for the development of computer models to predict the track, intensity, and time of occurrence of cyclonic storms has increased to avoid loss of life and prevention of damages to public properties. In this presentaion I had tried to highlight the ten most used models in this aspect which are responsible for saving millions of life and their livelihood.
This document provides an overview of wind systems and their effects on wildland fire behavior. It begins by defining wind and discussing how wind is the most critical factor influencing fire behavior. It then describes general winds caused by high and low pressure systems, as well as local winds like slope and valley winds. Critical winds that can drastically impact fires, such as cold fronts, foehn winds, and thunderstorm winds, are also outlined. The document provides details on characteristics and fire impacts of different wind types to help understand how winds affect wildland fire.
1) The document analyzes seasonal forecasts for global wind energy availability in autumn.
2) It identifies several key regions where wind resource is both abundant and highly variable between years, making them most suitable for seasonal wind forecasting.
3) The forecasts are evaluated against past data and found to have the highest skill in predicting wind resource variability, magnitude, and uncertainty in certain regions like Patagonia, parts of Africa, Asia, and Australia.
The document analyzes sea surface temperature data in the Arctic from 1982 to 2006. It describes using imaging software to measure the area of water below 15 degrees Celsius over time. The results found the area fluctuated without a clear decreasing or increasing trend. This disproved the hypothesis that the area would consistently decrease due to global warming. Future small changes are possible but human impacts are minimal currently.
The document discusses the recent strong El Niño event and its implications for seasonal rainfall distribution in Belg (FMAM) benefiting areas of Ethiopia. It finds that previous strong El Niño years saw normal to above normal rainfall during Belg, with particularly enhanced rainfall in 1983 and 2010. Based on this, the seasonal rainfall in 2016 is expected to have normal to above normal performance, potentially compensating for low rainfall during the Kiremt season. Further analysis using additional statistical forecasting methods is recommended.
Quantile regression ensemble for summer temperaturesManuel Herrera
This document presents research on using quantile regression and ensembles to model summer temperatures in London from 1961-2010. It finds that quantile regression can better capture relationships between temperature and weather variables like wind speed and cloud cover at extreme quantiles. An ensemble approach is also proposed, combining 50 quantile regression models to generate synthetic weather data for use in built environment studies rather than using complete observed years. Further work would expand these quantile methods and test their use with other weather data.
This document summarizes research on modeling extreme hurricane winds and insured losses in the United States. The researchers use extreme value theory and Bayesian statistical methods to estimate return levels of hurricane winds for different regions and time periods. They find higher return levels are associated with warmer climate conditions. The same methods are applied to model insured hurricane losses and estimate extreme potential losses under different climate scenarios.
Ten most popular software for prediction of cyclonic stormsMrinmoy Majumder
In recent years the frequency and intensity of cyclones and hurricanes have been increased manifold compared to the last decade. As a result, the necessity for the development of computer models to predict the track, intensity, and time of occurrence of cyclonic storms has increased to avoid loss of life and prevention of damages to public properties. In this presentaion I had tried to highlight the ten most used models in this aspect which are responsible for saving millions of life and their livelihood.
This document provides an overview of wind systems and their effects on wildland fire behavior. It begins by defining wind and discussing how wind is the most critical factor influencing fire behavior. It then describes general winds caused by high and low pressure systems, as well as local winds like slope and valley winds. Critical winds that can drastically impact fires, such as cold fronts, foehn winds, and thunderstorm winds, are also outlined. The document provides details on characteristics and fire impacts of different wind types to help understand how winds affect wildland fire.
1) The document analyzes seasonal forecasts for global wind energy availability in autumn.
2) It identifies several key regions where wind resource is both abundant and highly variable between years, making them most suitable for seasonal wind forecasting.
3) The forecasts are evaluated against past data and found to have the highest skill in predicting wind resource variability, magnitude, and uncertainty in certain regions like Patagonia, parts of Africa, Asia, and Australia.
The document analyzes sea surface temperature data in the Arctic from 1982 to 2006. It describes using imaging software to measure the area of water below 15 degrees Celsius over time. The results found the area fluctuated without a clear decreasing or increasing trend. This disproved the hypothesis that the area would consistently decrease due to global warming. Future small changes are possible but human impacts are minimal currently.
This Project is helpful for Time Series Analysis Forecasting. Better accuracy and metrics
in short-term forecasting are provided for intermediate planning for the target to reduce
CO2 emissions. Implementing different models like Exponential techniques, Linear
statistical modeling, and Autoregressive are used to forecast the emissions and finally
deployed on Stream lit.
Applied Statistics Chapter 2 Time series (1).pptswamyvivekp
Time series analysis is used to forecast future activity and trends. It involves analyzing trends, seasonal variations, cyclical variations, and irregular fluctuations in data over time. There are several techniques for analyzing trends, including semi-averages, moving averages, least squares regression, and exponential smoothing. Accurately identifying trends can help organizations plan for changes but forecasting also carries risks of uncertainty.
Climate Change Scenarios for Tourist Destinations in the Bahamas: Eluthreaintasave-caribsavegroup
This document discusses gathering and analyzing climate change data at regional, national, and destination scales. It describes observing historical climate data from weather stations and satellites, and projecting future climate using global and regional climate models under different emissions scenarios. The models simulate changes in temperature, precipitation, hurricanes, sea level rise and other climate variables. The results can identify potential climate impacts and vulnerabilities to inform further studies.
Recent Developments in Predicting El Nino and Insurance ImplicationsAlexander Pui
1) Recent developments in predicting El Nino events allow for improved forecasting up to 9 months in advance.
2) Climate cycles like El Nino and the Indian Ocean Dipole influence Australian rainfall patterns and weather-related insurance claims costs.
3) Better understanding of climate cycles could help insurers with technical pricing, explaining periods of high claims, and allowing for climate influences in catastrophe models. However, directly varying premiums or capital in response to cycles may add unwanted volatility.
This document summarizes a presentation analyzing hourly temperature and energy data from Barrow, Alaska between 1985-2017. A model was created that found: 1) CO2 concentrations played an important role in explaining the positive net inward energy imbalance. 2) Temperature data was best explained by including CO2 concentrations as an explanatory variable, rather than alternative drivers. 3) The model accurately predicted out-of-sample temperature and energy data, supporting the role of CO2 in increasing temperatures. The analysis provides evidence that contradicts views of climate change deniers.
The document summarizes the process of analyzing and modeling annual tornado count data for Illinois from 1950 to 2015. It begins by describing the data and outlining the steps that will be taken, which include checking for stationarity, identifying potential models, estimating parameters, and validating the models. The training data from 1950 to 2009 is found to be non-stationary based on a Dickey-Fuller test. Taking the difference of the logarithm of the data achieves stationarity. Scatter plots of the transformed training data suggest a potential time series model of order 1. Formal model selection and validation on the training and validation data will follow.
Seasonal climate forecasts in colombia apr widescreen2_no_animationCIAT
This document outlines a study evaluating seasonal climate forecasting skill in Colombia. It discusses using global climate models and statistical modeling to generate probabilistic forecasts of precipitation and temperature up to 6 months in advance. The forecasts are verified against observations at 4 sites, showing some skill at 1-month leads but declining skill at longer leads. Areas for improving the forecasts are identified, such as incorporating additional climate predictors like the tropical Atlantic. Collaboration with Colombia's climate agency IDEAM is recommended to coordinate research efforts on understanding climate drivers in Colombia.
The aim of the work is to find, combine and explore relevant fishing activities data with a focus on activities in Norway and develop data-specific tools for visualization, observing, accessing, forecasting, and managing fisheries. The area of study to which the project belongs is the intersection of Arctic engineering and data science. The problem we try to solve is the predictions of conditions that can lead to maximum catch by the vessel so that both organizations and people can benefit from it. Since the task is a prediction of catch hence we use regression algorithms (tree-based). The approach is to model for specific vessel groups, geographical locations, and species. The study explores the relationship between the physical parameters of the vessels for each year and uses that relation for further analysis. The study reflects the dependence of catch on physical parameters of vessels like length, tonnage, and power as well as the impact of geographic locations(latitude and longitude), species of fish targeted, and tools with which fishing is done. the preliminary results demonstrate that results are good for southern locations and the northern region still needs improvement.
However, it does not explore fluctuations in catch caused by environmental variation or any political interference. In the rapidly warming region, it is of vital importance to understand how stocks may be further affected by climate change in addition to fishing pressure.
Final Time series analysis part 2. pptxSHUBHAMMBA3
The document discusses key concepts in time series analysis including stationarity, trend, seasonality, autocorrelation, and partial autocorrelation. It defines stationarity as a time series having constant mean and variance, and that it is important for analysis and accurate predictions. Trend refers to the general tendency of data to increase or decrease over a long period. Seasonality describes regular, predictable changes that occur each calendar year. Autocorrelation measures the correlation between values of a time series at different points in time, while partial autocorrelation measures the direct correlation between two time points excluding intermediate values.
This document summarizes research on monsoon rainfall forecasting in India. It discusses:
1) The importance of monsoon prediction and approaches to long-term and short-term forecasting. Long-term prediction models use statistical correlations with ocean and atmospheric parameters, while short-term relies on numerical weather prediction models.
2) Factors used in the Indian Meteorological Department's long-term statistical forecasts in March/April and May/June, which include sea surface temperatures and pressures.
3) Evidence that short-term daily rainfall shows a scale-invariant power law distribution, making it difficult to predict precisely at a single location but easier when averaged over multiple locations.
4) The use of
2nd CSP Training series : solar resource assessment (2/2)Leonardo ENERGY
Fifth session of the 2nd Concentrated Solar Power Training dedicated to solar resource assessment.
DNI Variability, Frequency Distributions
Typical Meteorological Years
DNI measurements: broadband vs. spectral, and their limitations
What is circumsolar radiation and why should we care in CSP/CPV?
How much diffuse irradiance can be used in concentrators?
How to measure and model the circumsolar irradiance?
Spectral irradiance standards and their use for PV/CPV rating
The AM1.5 direct standard spectrum: Why did it change? Why AM1.5?
Use of the SMARTS radiative code to evaluate clear-sky spectral irradiances
Sources of measured spectral irradiance data
Spectral effects on silicon and multijunction cells and their dependence on climate
Climate change is projected to impact drastically in southern African during the 21st century
under low mitigation futures (Niang et al., 2014). African temperatures are projected to rise
rapidly, in the subtropics at least at 1.5 times the global rate of temperature increase (James
and Washington, 2013; Engelbrecht et al., 2015). Moreover, the southern African region is
projected to become generally drier under enhanced anthropogenic forcing (Christensen et
al., 2007; Engelbrecht et al., 2009; James and Washington, 2013; Niang et al., 2014). These
changes in temperature and rainfall patterns will plausibly have a range of impacts in South
Africa, including impacts on energy demand (in terms of achieving human comfort within
buildings and factories), agriculture (e.g. reductions of yield in the maize crop under higher
temperatures and reduced soil moisture), livestock production (e.g. higher cattle mortality as
a result of oppressive temperatures) and water security (through reduced rainfall and
enhanced evapotranspiration) (Engelbrecht et al., 2015).
The document provides an overview of time series analysis, including definitions, components, and methods for measuring trends, seasonal variations, cyclical variations, and irregular variations in time series data. It discusses adjusting raw time series data, measuring linear and nonlinear trends, converting annual trends to monthly trends, and different methods for measuring seasonal, cyclical, and irregular variations, including indexes and averages. Examples are provided to illustrate calculating seasonal variations using the monthly average method.
Storm Prediction data analysis using R/SASGautam Sawant
• Performed data cleaning and analysis using R, SAS to predict financial loss caused due to storms also predict when a storm will occur depending upon previous storm data
• Implemented algorithms like Logistic Regression, Multiple Regression, Linear Discriminant Analysis, PCA to obtain insights from the Storm Dataset from 1950-2007
The document discusses frequency analysis and return periods of hydrologic extremes like floods and droughts. It aims to relate the magnitude of events to their frequency of occurrence using probability distributions. Return period is the average time interval between occurrences of an event of a defined magnitude. Two storms are considered for calibration: the 2007 storm had moderate antecedent conditions while the 2010 storm had very dry conditions. Results show runoff is inconsistent between storms likely due to differences in antecedent moisture conditions between the events. Adjusting for antecedent conditions improves consistency in modeled versus observed runoff results.
This Project is helpful for Time Series Analysis Forecasting. Better accuracy and metrics
in short-term forecasting are provided for intermediate planning for the target to reduce
CO2 emissions. Implementing different models like Exponential techniques, Linear
statistical modeling, and Autoregressive are used to forecast the emissions and finally
deployed on Stream lit.
Applied Statistics Chapter 2 Time series (1).pptswamyvivekp
Time series analysis is used to forecast future activity and trends. It involves analyzing trends, seasonal variations, cyclical variations, and irregular fluctuations in data over time. There are several techniques for analyzing trends, including semi-averages, moving averages, least squares regression, and exponential smoothing. Accurately identifying trends can help organizations plan for changes but forecasting also carries risks of uncertainty.
Climate Change Scenarios for Tourist Destinations in the Bahamas: Eluthreaintasave-caribsavegroup
This document discusses gathering and analyzing climate change data at regional, national, and destination scales. It describes observing historical climate data from weather stations and satellites, and projecting future climate using global and regional climate models under different emissions scenarios. The models simulate changes in temperature, precipitation, hurricanes, sea level rise and other climate variables. The results can identify potential climate impacts and vulnerabilities to inform further studies.
Recent Developments in Predicting El Nino and Insurance ImplicationsAlexander Pui
1) Recent developments in predicting El Nino events allow for improved forecasting up to 9 months in advance.
2) Climate cycles like El Nino and the Indian Ocean Dipole influence Australian rainfall patterns and weather-related insurance claims costs.
3) Better understanding of climate cycles could help insurers with technical pricing, explaining periods of high claims, and allowing for climate influences in catastrophe models. However, directly varying premiums or capital in response to cycles may add unwanted volatility.
This document summarizes a presentation analyzing hourly temperature and energy data from Barrow, Alaska between 1985-2017. A model was created that found: 1) CO2 concentrations played an important role in explaining the positive net inward energy imbalance. 2) Temperature data was best explained by including CO2 concentrations as an explanatory variable, rather than alternative drivers. 3) The model accurately predicted out-of-sample temperature and energy data, supporting the role of CO2 in increasing temperatures. The analysis provides evidence that contradicts views of climate change deniers.
The document summarizes the process of analyzing and modeling annual tornado count data for Illinois from 1950 to 2015. It begins by describing the data and outlining the steps that will be taken, which include checking for stationarity, identifying potential models, estimating parameters, and validating the models. The training data from 1950 to 2009 is found to be non-stationary based on a Dickey-Fuller test. Taking the difference of the logarithm of the data achieves stationarity. Scatter plots of the transformed training data suggest a potential time series model of order 1. Formal model selection and validation on the training and validation data will follow.
Seasonal climate forecasts in colombia apr widescreen2_no_animationCIAT
This document outlines a study evaluating seasonal climate forecasting skill in Colombia. It discusses using global climate models and statistical modeling to generate probabilistic forecasts of precipitation and temperature up to 6 months in advance. The forecasts are verified against observations at 4 sites, showing some skill at 1-month leads but declining skill at longer leads. Areas for improving the forecasts are identified, such as incorporating additional climate predictors like the tropical Atlantic. Collaboration with Colombia's climate agency IDEAM is recommended to coordinate research efforts on understanding climate drivers in Colombia.
The aim of the work is to find, combine and explore relevant fishing activities data with a focus on activities in Norway and develop data-specific tools for visualization, observing, accessing, forecasting, and managing fisheries. The area of study to which the project belongs is the intersection of Arctic engineering and data science. The problem we try to solve is the predictions of conditions that can lead to maximum catch by the vessel so that both organizations and people can benefit from it. Since the task is a prediction of catch hence we use regression algorithms (tree-based). The approach is to model for specific vessel groups, geographical locations, and species. The study explores the relationship between the physical parameters of the vessels for each year and uses that relation for further analysis. The study reflects the dependence of catch on physical parameters of vessels like length, tonnage, and power as well as the impact of geographic locations(latitude and longitude), species of fish targeted, and tools with which fishing is done. the preliminary results demonstrate that results are good for southern locations and the northern region still needs improvement.
However, it does not explore fluctuations in catch caused by environmental variation or any political interference. In the rapidly warming region, it is of vital importance to understand how stocks may be further affected by climate change in addition to fishing pressure.
Final Time series analysis part 2. pptxSHUBHAMMBA3
The document discusses key concepts in time series analysis including stationarity, trend, seasonality, autocorrelation, and partial autocorrelation. It defines stationarity as a time series having constant mean and variance, and that it is important for analysis and accurate predictions. Trend refers to the general tendency of data to increase or decrease over a long period. Seasonality describes regular, predictable changes that occur each calendar year. Autocorrelation measures the correlation between values of a time series at different points in time, while partial autocorrelation measures the direct correlation between two time points excluding intermediate values.
This document summarizes research on monsoon rainfall forecasting in India. It discusses:
1) The importance of monsoon prediction and approaches to long-term and short-term forecasting. Long-term prediction models use statistical correlations with ocean and atmospheric parameters, while short-term relies on numerical weather prediction models.
2) Factors used in the Indian Meteorological Department's long-term statistical forecasts in March/April and May/June, which include sea surface temperatures and pressures.
3) Evidence that short-term daily rainfall shows a scale-invariant power law distribution, making it difficult to predict precisely at a single location but easier when averaged over multiple locations.
4) The use of
2nd CSP Training series : solar resource assessment (2/2)Leonardo ENERGY
Fifth session of the 2nd Concentrated Solar Power Training dedicated to solar resource assessment.
DNI Variability, Frequency Distributions
Typical Meteorological Years
DNI measurements: broadband vs. spectral, and their limitations
What is circumsolar radiation and why should we care in CSP/CPV?
How much diffuse irradiance can be used in concentrators?
How to measure and model the circumsolar irradiance?
Spectral irradiance standards and their use for PV/CPV rating
The AM1.5 direct standard spectrum: Why did it change? Why AM1.5?
Use of the SMARTS radiative code to evaluate clear-sky spectral irradiances
Sources of measured spectral irradiance data
Spectral effects on silicon and multijunction cells and their dependence on climate
Climate change is projected to impact drastically in southern African during the 21st century
under low mitigation futures (Niang et al., 2014). African temperatures are projected to rise
rapidly, in the subtropics at least at 1.5 times the global rate of temperature increase (James
and Washington, 2013; Engelbrecht et al., 2015). Moreover, the southern African region is
projected to become generally drier under enhanced anthropogenic forcing (Christensen et
al., 2007; Engelbrecht et al., 2009; James and Washington, 2013; Niang et al., 2014). These
changes in temperature and rainfall patterns will plausibly have a range of impacts in South
Africa, including impacts on energy demand (in terms of achieving human comfort within
buildings and factories), agriculture (e.g. reductions of yield in the maize crop under higher
temperatures and reduced soil moisture), livestock production (e.g. higher cattle mortality as
a result of oppressive temperatures) and water security (through reduced rainfall and
enhanced evapotranspiration) (Engelbrecht et al., 2015).
The document provides an overview of time series analysis, including definitions, components, and methods for measuring trends, seasonal variations, cyclical variations, and irregular variations in time series data. It discusses adjusting raw time series data, measuring linear and nonlinear trends, converting annual trends to monthly trends, and different methods for measuring seasonal, cyclical, and irregular variations, including indexes and averages. Examples are provided to illustrate calculating seasonal variations using the monthly average method.
Storm Prediction data analysis using R/SASGautam Sawant
• Performed data cleaning and analysis using R, SAS to predict financial loss caused due to storms also predict when a storm will occur depending upon previous storm data
• Implemented algorithms like Logistic Regression, Multiple Regression, Linear Discriminant Analysis, PCA to obtain insights from the Storm Dataset from 1950-2007
The document discusses frequency analysis and return periods of hydrologic extremes like floods and droughts. It aims to relate the magnitude of events to their frequency of occurrence using probability distributions. Return period is the average time interval between occurrences of an event of a defined magnitude. Two storms are considered for calibration: the 2007 storm had moderate antecedent conditions while the 2010 storm had very dry conditions. Results show runoff is inconsistent between storms likely due to differences in antecedent moisture conditions between the events. Adjusting for antecedent conditions improves consistency in modeled versus observed runoff results.
1. Statistical Analysis of Sea Surface Temperature
(SST) and Hurricanes in the Atlantic Basin
Bryan Butler
2. Outline of Analysis
• Overview of the sea surface temperature analysis and connection to the
Atlantic Multidecadal Oscillation (AMO).
• View of the time series for the AMO and it’s relationship to hurricane counts.
• Transforming the series to 10-year moving average.
• Developing a model
– Diagnostics
• Forecasts of the AMO
• Using the AMO forecast to forecast hurricane counts.
• Refinements
3. Introduction
• North Atlantic sea surface temperatures for 1871 to present contain a 65 –
80 year cycle called the Atlantic Multidecadal Oscillation (AMO).
• Fluctuates between warmer and cooler phases.
• Often measured as an index relative to the normal.
• Year-to-year fluctuations reflect more volatility and the index is more
commonly shown as a 10-year moving average to smooth out the spikes.
• There is a link between the 10-year moving average of the AMO and the 10-
year average of the total number of Atlantic storms.
• Potential to refine the interaction
– Total events (hurricanes and tropical storms)
– Hurricanes only
– Large hurricanes (CAT 4, CAT 5)
– Landfall hurricanes
4. Comparison of Annual Average to Hurricane
Season Average of AMO
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1871 1876 1881 1886 1891 1896 1901 1906 1911 1916 1921 1926 1931 1936 1941 1946 1951 1956 1961 1966 1971 1976 1981 1986 1991 1996 2001
Season Avg
6. Series Properties
• Annual AMO average and the average during hurricane season (Jun – Sep)
very closely aligned.
• Has long swings away from zero and then crosses back over, persistent
cycle.
• When compared to the hurricane counts, both series move in similar
directions but contain volatility that is difficult to predict.
• Both series show that there is a relationship between past levels on future
levels at one and two year intervals.
7. Granger Tests for Causality
• Granger test is a statistical test that measures how well data in one series can be
used to help forecast another series.
– Measured at different time intervals (1 year, 2 year, 5 year, etc.) called lags.
– Example: performing a 1 lag analysis implies that over the entire series the
previous year of one series helps to predict the next year of another series.
• Tests are bi-directional with two null hypotheses, results are to reject or fail to reject.
– AMO does not help forecast the number of hurricanes (Canes)
– Number of hurricanes does not help forecast the AMO
• In all cases (1 year, 2 year, 5 year) the AMO does help forecast the number of
hurricanes.
– AMO and Hurricane landfall counts do not show granger causality.
– Ex: 2005 had most hurricanes, but below average landfall (4 vs. an avg. of 5).
– Significant for CAT4 & CAT 5 counts.
• There is feedback and at the 1 year and 2 year interval, the number of hurricanes
helps forecast the AMO.
– Series are inter-related, however the impacts of the AMO on the number of
hurricanes is the dominant relationship.
8. Transforming the Series (Using Moving Averages)
• To reduce the noise in the series and reflect the longer term impact and
cycle, the 10-year moving average is often used
– 10 year moving average for hurricanes is also used to maintain the
analysis.
• Aggregating the series does induce some other properties that need to be
taken into consideration.
• When looking at the cross correlations between the two series the
persistence and feedback are evident:
– AMO lags impact hurricane counts for up to 25 years.
– Hurricane count lags affect AMO for about 8 years.
– There is strong persistence and feedback between the 10 year moving
average of the AMO and the 10 year moving average of CAT 4 and
CAT 5 storms.
9. Comparison of AMO Index to Hurricane Counts
-0.4
-0.2
0.0
0.2
0.4
3
4
5
6
7
8
9
1880 1900 1920 1940 1960 1980 2000
AMO10 CANES10
AMOIndex
10YearMovingAverageofHurricanes
10. AMO Index and 10 Year Moving Average of CAT 4
& Cat 5 Hurricanes
-0.4
-0.2
0.0
0.2
0.4
0.0
0.5
1.0
1.5
2.0
2.5
1880 1900 1920 1940 1960 1980 2000
AMO10 AVG45
AMOIndex
MovingAvgofCAT4&5
11. Forecast of AMO Moving Average (10 Year)
Dependent Variable: AMO10
Method: Least Squares
Date: 12/21/05 Time: 10:11
Sample(adjusted): 1882 1995
Included observations: 114 after adjusting endpoints
Convergence achieved after 4 iterations
Variable Coefficient Std. Error t-Statistic Prob.
C -0.016881 0.185720 -0.090892 0.9277
AR(1) 1.415656 0.086995 16.27286 0.0000
AR(2) -0.429409 0.087256 -4.921278 0.0000
R-squared 0.979886 Mean dependent var -0.110193
Adjusted R-squared 0.979524 S.D. dependent var 0.163233
S.E. of regression 0.023358 Akaike info criterion -4.649815
Sum squared resid 0.060560 Schwarz criterion -4.577810
Log likelihood 268.0395 F-statistic 2703.813
Durbin-Watson stat 2.191815 Prob(F-statistic) 0.000000
Inverted AR Roots .98 .44
• Model reflects 1 year and 2 year influence of past years
• High Adjusted R-Squared
• All critical variables are significant
– Constant is essentially zero
• All structure has been removed to the best ability as evidence by the D-W statistic
being close to 2.00.
• Use data to 1995, forecast to 2005 to test model performance against known data.
12. Performance of AMO Model Against Whole Series
-0.4
-0.2
0.0
0.2
0.4
1880 1900 1920 1940 1960 1980 2000 2020
AMO10F1 AMO10
13. Comparison of Forecast to Known Values of AMO
Using 1 Year Increments
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
80 85 90 95 00 05 10 15 20
AMO10F1S AMO10
15. Performance of Combined AMO and Hurricane
Model Forecast on 10 Year Moving Average
4
5
6
7
8
9
65 70 75 80 85 90 95 00 05 10 15 20
CANES10 CANES10F
16. Forecast of Moving Average of Total Number of
Events
8
10
12
14
16
65 70 75 80 85 90 95 00 05 10 15 20
EVENTS10 EVENTS10F
17. Evaluation of Models
• 2005 was a very rare year and reflects a shock to the system that cannot be
forecasted.
• The 10 year moving average of the AMO appears to follow a cycle that is
easier to forecast.
• After forecasting out the AMO, forecasts have been developed for two
additional series.
– 10 Year moving average of hurricanes.
– 10 Year moving average of all events.
• There is less precision in forecasting both hurricanes and events since the
correlation is not direct.
– Correlation is very strong and may be the most dominant factor.
– There are other variables that may impact the number of hurricanes that
are not included in the model.
18. Forecasting CAT 4 & CAT 5 Making Landfall in the
US
0.0
0.5
1.0
1.5
2.0
2.5
65 70 75 80 85 90 95 00 05 10 15 20
AVG45F7ALLS AVG45
22. Composition of FL Hurricanes
• Long run mean indicates that there are currently some imbalances.
– CAT 2 and CAT 3 are behind the long run mean.
– CAT 5 currently ahead of the long run mean.
• CAT 1 and CAT 4 storms are close to long run means.
Period
TS CAT1 CAT2 CAT3 CAT4 CAT5
All Years 2.15 2.59 5.67 4.64 6.95 19.13
1956 - 2005 1.67 2.78 16.67 10.00 7.14 8.33
1906 - 1955 2.38 3.85 7.14 5.00 4.17 4.17
1852 - 1905 2.60 1.86 3.06 2.89 17.33 #DIV/0!
FL Return (Years)
23. Composition of Gulf of Mexico (GoM)
• CAT 2 and CAT 3 storm counts appear out of balance with the long-run mean.
• Due to events of Rita and Katrina, CAT 5 return period is running ahead of the long
run average.
Period
TS CAT1 CAT2 CAT3 CAT4 CAT5
All Years 1.24 2.61 3.35 5.50 8.11 15.40
1952 - 2002 1.04 2.63 16.67 7.14 7.14 6.25
1902 - 1952 1.14 2.27 3.85 4.55 5.56 25.00
1852 - 1902 1.69 3.00 1.80 5.40 18.00 #DIV/0!
Return (Years)
24. Conclusions
• Each year approximately 5 hurricanes (of varying strength) are expected to
form in the basin.
• The average has highs and lows that is correlated to the AMO.
• Complete AMO cycle lasts approximately 50 years.
• Current forecast indicates that we are near a peak and that the AMO should
start to decrease slowly.
– This is the feedback influence of the hurricanes.
• Expect a period of increased hurricane activity due to the increased AMO
and is expected to last until approximately 2017.
• Recent activity has brought the short run (50 year cycle) of large storms
(CAT 4, CAT 5) into closer alignment with the long run average.
• Expect more activity in the CAT 2 and CAT 3 storms for both FL and the
GoM
25. Additional Work
• Improve forecasting of CAT 4 and CAT 5 storms making landfall in the US.
– Link to AMO or other index.
• Develop better forecast.
– Current models forecast moving averages which are difficult to interpret.
• Continue more detailed analyses for other areas of the US.
– Northeast
– Carolinas