Scribd
Upload
12 views13 pages

Rainfall Time Series Data Analysis

anlyzing

Uploaded by

abdulseid1998
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
12 views13 pages

Rainfall Time Series Data Analysis

anlyzing

Uploaded by

abdulseid1998
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

See discussions, stats, and author profiles for this publication at: [Link]

net/publication/312943347

Time Series Data Analysis of Rainfall

Technical Report · January 2017

CITATIONS READS
0 900

1 author:

Asid Ur Rehman
Newcastle University
13 PUBLICATIONS 108 CITATIONS

SEE PROFILE

All content following this page was uploaded by Asid Ur Rehman on 27 January 2017.

The user has requested enhancement of the downloaded file.


Time Series Data Analysis of Rainfall

Term Project Report


Data Analysis for Earth Sciences

Asad-Ur-Rehman | MS Remote Sensing & GISc


Institute of Space Technology | Islamabad, Pakistan
Table of contents

1 INTRODUCTION ____________________________________________________ 1

2 DATA ANALYSIS ____________________________________________________ 2

2.1 STUDY AREA _________________________________________________________ 2

2.2 DATA _______________________________________________________________ 2

2.3 METHODOLOGY ______________________________________________________ 2


2.3.1 Time Series Data Pattern _____________________________________________ 4
2.3.2 Mean Rainfall and Standard Deviation ___________________________________ 4
2.3.3 Mean Annual Rainfall ________________________________________________ 5
2.3.4 Decadal Analysis ____________________________________________________ 5
2.3.5 Histogram and Shape ________________________________________________ 6
2.3.6 Data Distribution Examination _________________________________________ 7
2.3.7 Hypothesis Testing for Data Distribution _________________________________ 7
2.3.8 Proof of Central Limit Theorem ________________________________________ 8
2.3.9 Autospectral Analysis ________________________________________________ 8
2.3.10 Frequency Analysis __________________________________________________ 9
List of Figures

Figure 1: Google Earth Map of Islamabad ___________________________________________ 2


Figure 2: Methodology for Project _________________________________________________ 3
Figure 3: Rainfall Pattern ________________________________________________________ 4
Figure 4: Mean Rainfall and Standard Deviation ______________________________________ 4
Figure 5: Average Annual Rainfall__________________________________________________ 5
Figure 6: Decade wise Rainfall Pattern ______________________________________________ 6
Figure 7: Histogram and Shape ____________________________________________________ 6
Figure 8: Visual inspection of Data Distribution _______________________________________ 7
Figure 9: Hypothesis Testing ______________________________________________________ 7
Figure 10: Proof of Central Limit Theorem ___________________________________________ 8
Figure 11: Autospectrum without Removing Trend ____________________________________ 9
Figure 12: Autospectrum with Trend Removed _______________________________________ 9

List of Tables

Table 1: Maximum Rainfall _______________________________________________________ 4


1 INTRODUCTION
Time series analysis comprises methods for analyzing time series data in order to extract
meaningful statistics and other characteristics of the data. Time series forecasting is the
use of a model to predict future values based on previously observed values. A time series
analysis is a valuable tool to get information about analyzed data structures and their
components, which provides a good basis for successful future predictions

Time series data analysis varies from simple to complex in nature. It may include trend
analysis, sessional analysis and different types of modeling on the basis of available data.
For time series data analysis there are many preliminary phases which should be
undertaken before moving forward. Such phases consists of data consistency check,
measuring data central tendency and shape, examining the data distribution (histogram)
and others. Thus in order to perform precise time series analysis we will need to consider
preliminary stages first.

Data used in project contains daily rainfall data for 32 years. Univariate Statistics,
Hypothesis Testing and Autospectral Analysis techniques were performed to analyze the
data.

1|Page
2 DATA ANALYSIS

2.1 STUDY AREA


Name: Islamabad, Pakistan

Location: 73° 2'41.21"E, 33°41'3.06"N

Height: 530m

Figure 1: Google Earth Map of Islamabad

2.2 DATA

Daily Rainfall data obtained from Pakistan Meteorology Department

2.3 METHODOLOGY
Figure 8 shows the detailed methodology adopted for this project. It mainly contains
Univariate Statistics, Hypothesis Testing and Autospectral Analysis techniques.

2|Page
Daily Rainfall Data

Mean Rainfall

Standard Deviation
Central Tendency & Shape
Analysis
Mean Annual Rainfall

Decadal Analysis

Histogram and Shape

Hypothesis Testing for Data


Distribution Hypothesis Testing

Removing NaN Values

Fourier Transformation with


Trend

Autospectral Analysis
Fourier Transformation
without Trend

Frequency Analysis

Figure 2: Methodology for Project

3|Page
2.3.1 Time Series Data Pattern
Below figure shows rainfall data pattern over 32 years. From graph we can see
maximum rainfall days belong to year 1997 & 2001.

Figure 3: Rainfall Pattern

Year Month Date Rainfall (mm)


1997 8 27 200
2001 7 24 200
Table 1: Maximum Rainfall

2.3.2 Mean Rainfall and Standard Deviation


Minimum mean rainfall observed in the month of May, October and November, while
mean rainfall is maximum in July and August which is true as these two months
represents monsoon season in Pakistan with maximum rainfall. Standard deviation also
follow the same pattern as of mean rainfall.

Figure 4: Mean Rainfall and Standard Deviation

4|Page
2.3.3 Mean Annual Rainfall
Maximum Annual Mean Rainfall = 1983 with 4.694mm rainfall

Figure 5: Average Annual Rainfall

2.3.4 Decadal Analysis


Data was divided into three decades i.e. 1976-1985, 1986-1995, and 1996-2005. Mean
rainfall for each month in given decades was calculated. Below graph showed us a
decrease in rainfall with time in monsoon session. However an increase in rainfall was
observed for the month of February. Thus we can conclude with the passage of time
there is a shift in rainfall pattern.

5|Page
Figure 6: Decade wise Rainfall Pattern

2.3.5 Histogram and Shape


Data was found highly skewed with high Kurtosis

Skewness = 6.6596
Kurtosis = 64.3528

Figure 7: Histogram and Shape

6|Page
2.3.6 Data Distribution Examination
Below figure was depicting data distribution as lognormal distribution.

Figure 8: Visual inspection of Data Distribution

2.3.7 Hypothesis Testing for Data Distribution


Hypothesis examination showed that data distribution did not belong to lognormal
distribution.

Hypothesis Stats:
The Chi-square test statistic value = 989.3627
The Chi-square critical value is = 168.613
Null Hypothesis is rejected at 5% significance level
Rainfall data does not belong to a Log-normal distribution

Figure 9: Hypothesis Testing

7|Page
2.3.8 Proof of Central Limit Theorem
Central limit theorem was proved i.e. the means of the samples tends to form normal
distribution as number of sample increases.

Figure 10: Proof of Central Limit Theorem

2.3.9 Autospectral Analysis


In order to perform Autospectral Analysis, NaN values were treated. There were two
option to treat NaN values, first option was to remove NaN values from data and second
one was to interpolate NaN values. I chose first option as NaN values were only
occurring as extra days for the month having 30 or less number of days.

8|Page
Figure 11: Autospectrum without Removing Trend

f1 f2

f3

Figure 12: Autospectrum with Trend Removed

2.3.10 Frequency Analysis


From graphs above we have three main frequency components i.e.

f1 = 0.002747 f2 = 0.005493 f3 = 0.00824

Now we can find cycles in data

Cycle 1 = 1/f1 = 364.0335 days, which shows yearly cycle in data

Cycle 2 = 1/f2 = 182.0499 days, which shows 6 month cycle in data

Cycle 3 = 1/f3 = 121.3592 days, which presents 4 month cycle in data

9|Page

View publication stats

Common questions

Powered by AI

Univariate Statistics help summarize the rainfall data by providing descriptive measures like mean and standard deviation, which are crucial for understanding central tendency and dispersion . Hypothesis Testing is used to determine if the data follows a lognormal distribution, where a Chi-square test rejected the null hypothesis that the rainfall data belonged to a lognormal distribution . Autospectral Analysis identifies the frequency components of the rainfall data, helping in detecting cycles such as annual and semi-annual variations . Together, these techniques provide a comprehensive understanding of the data's characteristics and underlying patterns.

In conducting hypothesis testing for data distribution, the document outlines steps including formulating null and alternative hypotheses, selecting an appropriate statistical test (e.g., Chi-square test), and comparing the test statistic against a critical value . In this analysis, the null hypothesis that the rainfall data belongs to a lognormal distribution was rejected, indicating deviations from assumed distributions. This process is systematic and essential for assessing distribution characteristics and ensuring suitable model assumptions.

Histograms and shape analysis are significant as they provide visual and statistical summaries of the rainfall data distribution's skewness and kurtosis. A skewness value of 6.6596 and a high kurtosis of 64.3528 indicate a highly skewed distribution with a long tail . Understanding these aspects of distribution helps in selecting appropriate statistical methods and models for analysis, such as transformations that could normalize the data or adjust for outliers.

Frequency analysis is conducted by applying Fourier transformations to identify periodic components in the rainfall data. This process isolates key frequencies, revealing cycles of yearly, biannual, and four-month occurrences based on identified components (f1, f2, f3). Recognizing these cycles is important for understanding seasonal patterns, anticipating high or low rainfall periods, and aiding in the development of predictive models that could inform agricultural planning and disaster management.

Decadal analysis involves dividing the rainfall data into three decades to observe changes in rainfall patterns over time. The analysis showed a noticeable decrease in monsoon rainfall with time, while February experienced an increase, denoting a non-uniform shift potentially impacted by environmental changes or climatic variations . This approach highlights long-term trends and anomalies, allowing for a comprehensive evaluation of potential climate change effects and providing evidence for further research into causes and mitigation strategies.

The Central Limit Theorem (CLT) was utilized in the analysis to demonstrate that the sampling distribution of the sample mean approaches a normal distribution as the sample size increases. This holds even if the original rainfall data distribution is not normal, thus allowing for the application of parametric statistical methods, which assume normality in the sampling distribution of the mean . CLT is pivotal in justifying the use of these statistical techniques in further analyses of the rainfall data.

Mean rainfall and standard deviation are used as summary statistics to identify and interpret variations in seasonal rainfall. High mean rainfall in July and August confirms the monsoon peak , while standard deviation indicates variability around this mean, highlighting the unpredictable nature of rainfall within these months. These metrics provide a clear quantitative basis for understanding the extent of variability and supporting forecasts and risk assessments related to seasonal weather changes.

The analysis of mean annual rainfall helps in identifying historical rainfall trends by examining yearly data aggregates. This analysis showed that the year with maximum annual mean rainfall was 1983, revealing variability and allowing for comparisons over time . This provides insights into rainfall consistency, variability, and potential outliers, contributing to understanding long-term climatic trends and supporting future climatic modeling and prediction efforts.

Identifying and addressing NaN values is critical because they can distort analyses, such as Autospectral Analysis, which requires complete data sets for accurate frequency detection. The presence of NaN values, largely due to shorter months, can impact the continuity of time series data. Removing NaN values is preferred over interpolation here to maintain integrity as these values are artifacts of calendar month lengths rather than missing data . Correct handling ensures robust results in statistical evaluations and pattern analyses.

Changes in decadal rainfall patterns may indicate shifts in climate, such as changes in seasonal intensity or distribution. The analysis showed a decrease in monsoon rainfall over decades alongside an increase in February's rainfall, suggesting a possible change in climatic conditions or variability. This could imply alterations in monsoon dynamics or anthropogenic impacts affecting regional rainfall patterns . Such patterns are critical for understanding long-term climate change impacts and for water resource management.

You might also like