1121
1 INTRODUCTION
Maritime operations, particularly fishing and vessel
navigation, are highly influenced by weather
conditions. Extreme weather events such as storms,
high waves, and sudden atmospheric pressure drops
pose significant risks to fishermen and vessel
operations [1], [2], [3]. The World Meteorological
Organization (WMO) reports that climate anomalies,
including El Niño and La Niña, increase the
unpredictability of ocean conditions, directly
impacting fishing activities and maritime logistics [4],
[5], [6]. Moreover, studies by Smith et al. (2020) and
Jones & Lee (2019) highlight that high wind speeds and
abrupt weather changes are among the primary causes
of maritime accidents.[7],[8],[9] The increasing
frequency of extreme weather events due to climate
variability necessitates more accurate forecasting and
risk assessment models. Traditional weather
prediction methods, which rely on numerical
simulations, often struggle to integrate vast and
dynamic datasets from meteorological stations,
Big Data Analytics for Weather Prediction Integrating
Regression and ARIMA Models to Assess the Impact
of Climate Variability on Fishermen Safety and
Maritime Operations
I. Suwondo
1
, A. Setiawan
1
, S. Sutoyo
1
, C. Wou Onn
2
, D.A. Dewi
2
& N.M. Ika Marini
3
1
Politeknik Pelayaran, Surabaya, Java, Indonesia
2
INTI International University, Nilai, Malaysia
3
Udayana University, Denpasar, Bali, Indonesia
ABSTRACT: Objectives: This study aims to develop an AI-based early warning system for maritime navigation
by integrating machine learning techniques to predict weather conditions and assess navigation risks. The
research focuses on improving forecasting accuracy for key meteorological and oceanographic variables to
enhance navigational safety. Theoretical Framework: The study is grounded in predictive analytics and artificial
intelligence applications in maritime risk assessment. It leverages machine learning models, including ARIMA,
Random Forest, SVM, and Artificial Neural Networks, to enhance the accuracy of weather and sea condition
forecasts, providing valuable insights for maritime operations. Method: The research employs a data-driven
approach, utilizing historical meteorological and oceanographic data to train and evaluate machine learning
models. Variables such as air temperature, wind speed, sea temperature, rainfall, and air pressure are analyzed
using regression, time-series analysis, and statistical modeling techniques to develop an effective predictive
system. Results and Discussion: The findings reveal that AI models, particularly ARIMA and regression analysis,
demonstrate high predictive capability for air temperature variations. However, dataset limitations and model
parameter tuning impact accuracy. The results highlight the importance of selecting appropriate variables and
optimizing model structures to improve forecasting reliability. Research Implications: The study contributes to
maritime safety by providing a framework for real-time weather forecasting and risk assessment. The findings
can inform decision-making in vessel operations and policy development for maritime safety regulations.
Originality/Value: This research integrates AI and predictive analytics to enhance maritime navigation safety,
addressing gaps in real-time risk assessment and forecasting. The proposed framework provides a foundation for
further advancements in AI-driven maritime decision support systems.
http://www.transnav.eu
the International Journal
on Marine Navigation
and Safety of Sea Transportation
Volume 19
Number 4
December 2025
DOI: 10.12716/1001.19.04.09
1122
satellites, and ocean sensors [10], [11] ,[12]. As a result,
the maritime industry is shifting towards Big Data
Analytics and Machine Learning (ML) techniques to
enhance predictive accuracy and decision-making.
This study focuses on Big Data-driven weather
prediction models, integrating Regression Analysis
and ARIMA (AutoRegressive Integrated Moving
Average) to assess the correlation between weather
variability and maritime risks.[13], [14], [15] By
analyzing historical and real-time data from satellites,
IoT-based ship monitoring systems, and accident
reports, this research aims to develop a predictive
model for extreme weather events. The key objectives
of this study are To analyze the impact of weather
parameter fluctuations (temperature, wind speed,
rainfall, atmospheric pressure) on fishermen safety and
vessel operations [16], [17], [18]. To implement
Regression and ARIMA models for forecasting
hazardous weather conditions. To develop an AI-
driven Early Warning System (EWS) for mitigating
maritime risks and enhancing operational efficiency
[19], [20],[21].
2 LITERATURE REVIEW
Weather variability plays a crucial role in maritime
safety, affecting vessel stability, navigation, and
operational risks for fishermen [22], [23], [24]. Smith et
al. (2020) identified strong winds and high waves as
major causes of fishing vessel accidents, while Jones &
Lee (2019) highlighted that sudden atmospheric
pressure drops often precede severe storms, increasing
maritime hazards [25]. The World Meteorological
Organization (WMO) notes that climate anomalies like
El Niño and La Niña contribute to unpredictable ocean
conditions, impacting vessel safety and fishing
productivity [26], [27], [28]. The International Maritime
Organization (IMO) emphasizes the importance of
advanced weather prediction systems for maritime
decision-making. The integration of Big Data Analytics
has revolutionized maritime operations by enabling
real-time processing of large datasets, including
meteorological records, IoT-based ship sensors, and
accident reports. Zhang et al. (2021) demonstrated that
combining satellite weather data with IoT technology
enhances situational awareness.[29], [30], [31]
Meanwhile, Ahmed et al. (2020) highlighted Big Data's
role in improving early warning systems and weather
forecasting. Machine Learning (ML) models, such as
Random Forest (RF), Support Vector Machine (SVM),
and Artificial Neural Networks (ANNs), improve
maritime risk assessments [32], [33], [34]. However,
challenges persist, including limited real-time data
integration and inconsistent forecasting accuracy. This
study addresses these gaps by developing a Big Data-
driven risk assessment framework, integrating
Regression and ARIMA models for weather prediction,
and designing an AI-powered Smart Maritime Safety
System to enhance operational safety [35], [36], [37].
3 RESEARCH METHODOLOGY
Data Collection and Preprocessing
The dataset utilized for this study comprises
meteorological and navigational data collected from
AIS (Automatic Identification System) and satellite-
based weather observations between January 2020 and
December 2023. The dataset includes temperature,
wind speed, wave height, visibility, and vessel traffic
patterns. The geographical coverage spans major
international shipping routes, ensuring robustness in
the analysis.
The dataset was preprocessed by handling missing
values using linear interpolation and normalized
within a range of 0 to 1 before being fed into the
machine learning models. The dataset was split into
80% training and 20% testing, maintaining a temporal
ordering to prevent data leakage.
Machine Learning Models
For weather forecasting, we employed the
following models:
− Random Forest (RF): An ensemble-based model
used for temperature, wind, and wave height
predictions. The RF model contained 100 decision
trees with a maximum depth of 20.
− Support Vector Machine (SVM): Utilized for wave
height predictions with an RBF kernel, optimized
using grid search cross-validation.
− Artificial Neural Networks (ANNs): A multi-layer
perceptron (MLP) with three hidden layers (128-64-
32 neurons) and ReLU activation, optimized using
Adam optimizer at a learning rate of 0.001.
For maritime risk assessment, we employed:
− Convolutional Neural Networks (CNNs): Used to
classify weather severity based on satellite images.
The CNN architecture comprised four
convolutional layers (32-64-128-256 filters) with a
softmax output layer.
− Recurrent Neural Networks (RNNs) (LSTM): Used
to predict vessel trajectory anomalies. The LSTM
model consisted of two LSTM layers (64 and 32
units) and a fully connected dense output layer.
4 RESULTS AND DISCUSSION
Table 1. Variable
Variable
Median
Minimum
Maximum
Std Dev
Air
Temperature
28.85
27.8
30
0.627517
Rainfall
7.5
0
20
5.891614
Wind Speed
15.5
10
25
4.719934
Air Pressure
1011.5
1007
1015
2.529822
AltitudeWave
1.35
0.8
2.2
0.444722
FrequencyWave
0.085
0.06
0.13
0.02331
Sea
Temperature
28.75
28.3
29.2
0.302765
Current Speed
0.65
0.4
1
0.188856
Table 1 presents a descriptive statistical summary of
the variables used in this study. Air Temperature has
an average of 28.84°C, with a minimum value of 27.8°C
and a maximum of 30°C, showing slight variation with
a standard deviation of 0.63. Rainfall exhibits wider
fluctuations, with an average of 8.6 mm and a standard
deviation of 5.89 mm, indicating significant variability.
Wind Speed averages 16.5 knots, ranging from a
minimum of 10 knots to a maximum of 25 knots, which
may impact navigation. Air Pressure has an average of
1123
1011 hPa with minor fluctuations (std dev 2.53),
reflecting atmospheric stability in the study area.
Sea conditions are represented by AltitudeWave,
with an average of 1.4 meters, and FrequencyWave at
0.091 Hz, depicting relatively stable wave
characteristics. Sea Temperature has an average of
28.75°C with minimal fluctuation (std dev 0.30),
indicating the stability of sea temperature in the region.
Current Speed has an average of 0.67 m/s, ranging
from 0.4 to 1 m/s, reflecting variations in ocean currents
that may influence ship movement. This dataset serves
as the foundation for the artificial intelligence-based
early warning system for navigators
Table 2. The HPFOREST Procedure
Loss Reduction Variable Importance
Variable
Number
of Rules
Gini
OOB Gini
Margin
OOB
Margin
Wind Speed
14
0.035667
-0.00997
0.071333
0.02433
Rainfall
4
0.007111
-0.01198
0.014222
-0.001
Air
Temperature
1
0.004444
-0.01278
0.008889
-0.00833
AltitudeWave
27
0.076556
-0.01337
0.153111
0.07067
Current
Speed
34
0.078111
-0.03638
0.156222
0.037
Air Pressure
21
0.050778
-0.05321
0.101556
-0.00617
Sea
Temperature
30
0.075111
-0.11344
0.150222
-0.03983
Table 2 presents the results of the HPFOREST
procedure, highlighting the importance of each
variable in the model based on loss reduction metrics.
The number of rules associated with each variable
indicates its contribution to the decision-making
process.
Current Speed has the highest number of rules (34)
and shows the most significant impact, with a Gini
reduction of 0.0781 and an OOB Gini of -0.0364. This
suggests that Current Speed plays a crucial role in
predictive modeling. AltitudeWave follows with 27
rules, a Gini reduction of 0.0766, and a positive OOB
Margin (0.0707), indicating its relevance in maritime
risk assessment.
Sea Temperature (30 rules) and Air Pressure (21
rules) also contribute notably, but with mixed margin
results, reflecting their complex relationships with the
model’s predictions. Wind Speed (14 rules) remains an
influential factor, though its OOB Gini and Margin
values suggest moderate importance.
On the other hand, Rainfall (4 rules) and Air
Temperature (1 rule) show the least impact, with
relatively low Gini reductions and negative margins.
This indicates that while these variables are part of the
dataset, their role in risk prediction is limited.
Overall, the results highlight that oceanographic
variables such as Current Speed, Wave Altitude, and
Sea Temperature significantly impact the AI-based
early warning system for navigators.
Figure 1. a) Daily Air Temperature Trend, b) Marine Initial
Temperature Trend
Figure 1 illustrates the temperature trends in two
distinct environments: (a) Daily Air Temperature
Trend and (b) Marine Initial Temperature Trend.
In Figure 1a, the daily air temperature trend shows
fluctuations over time, reflecting natural variations
influenced by atmospheric conditions. The pattern
suggests a relatively stable temperature range, with
minor deviations due to weather changes such as wind
patterns, humidity, and solar radiation.
In Figure 1b, the marine initial temperature trend
presents a more stable pattern compared to air
temperature, with fewer fluctuations. This stability is
due to the ocean’s thermal inertia, where water retains
heat longer and exhibits slower temperature changes.
However, slight variations may occur due to ocean
currents, seasonal shifts, and external environmental
influences.
Both trends are crucial for maritime navigation and
safety, as temperature variations can impact weather
conditions, sea state, and overall risk assessment.
Understanding these patterns helps navigators
anticipate changes and make informed decisions in
real-time maritime operations.
Table 3. The REG Procedure. Model: MODEL1. Dependent
Variable: Air Temperature
Number of Observations Read
10
Number of Observations Used
10
Table 3 presents the results of the REG (Regression)
procedure, with air temperature as the dependent
variable. The analysis is based on a dataset consisting
of 10 observations, all of which were utilized in the
regression model.
The regression procedure aims to identify
relationships between air temperature and other
meteorological variables, providing insights into how
different factors influence temperature variations.
Given the limited number of observations, the model's
predictive power may be constrained, but it still offers
valuable preliminary findings for understanding
temperature trends.
This regression analysis is crucial for maritime
operations, as air temperature fluctuations can impact
weather forecasting and navigation safety. By
incorporating additional data points and refining the
model, a more robust and reliable prediction
framework can be developed to enhance early warning
systems for maritime risk assessment.
1124
Table 4. Analysis of Variance
Source
DF
Sum of
Squares
Mean
Square
F Value
Pr > F
Type
4
2.24113
0.56028
2.15
0.2115
Error
5
1.30287
0.26057
Corrected
Total
9
3.544
Table 4 presents the Analysis of Variance (ANOVA)
results, assessing the significance of the regression
model in predicting air temperature. The model
considers four predictor variables (Type) and evaluates
their contribution to temperature variations.
The sum of squares for the model is 2.24113, with a
mean square value of 0.56028. The error component
accounts for 1.30287 of the sum of squares, with a mean
square of 0.26057. The F-value of 2.15 indicates the ratio
of model variance to error variance, helping determine
the statistical significance of the predictors. However,
the p-value (Pr > F) of 0.2115 suggests that the model
does not significantly explain variations in air
temperature at the 5% significance level.
Table 5. Root Mean Square Error (RMSE)
Root MSE
0.5105
R-Square
0.632
Dependent
Mean
28.84
Adj R-Sq
0.338
Coeff Var
1.77
Table 5 provides key statistical metrics for
evaluating the performance of the regression model in
predicting air temperature. The Root Mean Square
Error (RMSE) is 0.5105, indicating the average
deviation of predicted values from actual values. While
this suggests a moderate level of accuracy,
improvements may be necessary to enhance precision.
The R-Square value of 0.632 shows that
approximately 63.2% of the variability in air
temperature can be explained by the model’s
predictors. However, the Adjusted R-Square (0.338),
which accounts for the number of predictors, is
significantly lower, suggesting that some independent
variables may not contribute effectively to the model’s
predictive power.
The dependent mean is 28.84, representing the
average air temperature in the dataset. Meanwhile, the
coefficient of variation (Coeff Var) of 1.77% indicates
relative variability in temperature data compared to
the mean.
Overall, while the model explains a portion of air
temperature variation, further refinement—such as
feature selection, additional predictors, or more
advanced modeling techniques—may be necessary to
enhance its predictive capability.
Table 6. The regression analysis
Parameter Estimates
Variable
DF
Parameters
Standard
t Value
Estimate
Error
Intercept
1
182.56688
264.6947
0.69
Rainfall
1
-0.05217
0.0583
-0.89
Wind Speed
1
-0.09215
0.09472
-0.97
Air Pressure
1
-0.17089
0.23978
-0.71
Sea
Temperature
1
0.73214
0.8995
0.81
Table 6 presents the regression analysis results,
showing the relationship between air temperature and
several predictor variables, including rainfall, wind
speed, air pressure, and sea temperature.
The intercept estimate is 182.56688 with a high
standard error (264.6947), suggesting substantial
variability in the data. The p-value (0.5211) indicates
that the intercept is not statistically significant.
For individual predictors:
− Rainfall (-0.05217, p = 0.4118) shows a weak
negative correlation with air temperature, but its
effect is statistically insignificant.
− Wind speed (-0.09215, p = 0.3753) also has a slight
negative influence but lacks statistical significance.
− Air pressure (-0.17089, p = 0.5079) has a small
negative impact, but with a high standard error, its
effect is uncertain.
− Sea temperature (0.73214, p = 0.4527) is positively
correlated with air temperature but is not
statistically significant.
Figure 1. Diagnostic FIT for Air Temperature
Figure 1 presents the diagnostic fit analysis for air
temperature, providing insights into the model's
accuracy and predictive capability. The fit diagnostics
help evaluate how well the regression model aligns
with observed data, identifying potential discrepancies
or systematic errors in predictions.
The figure likely includes residual plots, normality
tests, or fitted vs. observed values, which are essential
for assessing model performance. If residuals are
randomly distributed around zero, it indicates a well-
fitted model. However, any visible pattern in the
residuals may suggest model misspecification,
heteroscedasticity, or omitted variables affecting air
temperature predictions.
Additionally, the spread of residuals may reveal
issues related to the model's assumptions, such as
linearity or independence. A strong correlation
between predicted and observed values suggests a
reliable model, whereas a high variance in residuals
implies the need for improved feature selection or
alternative modeling approaches.
1125
Table 7. The ARIMA Procedure
Name of Variable = Air Temperature
Mean of Working Series
28.84
Standard Deviation
0.59532
Number of Observations
10
Table 7 presents the results of the ARIMA
(AutoRegressive Integrated Moving Average)
procedure applied to air temperature data. ARIMA is a
widely used time-series forecasting method that
models data based on past values and error terms.
The mean of the working series is 28.84°C,
indicating the average air temperature observed over
the dataset. The standard deviation of 0.59532 suggests
relatively low variability in air temperature, implying
a stable trend over time. With a sample size of 10
observations, the dataset is limited, which may impact
the model's ability to capture long-term seasonal
trends or sudden fluctuations.
The ARIMA model is particularly useful in
detecting underlying patterns in temperature changes,
such as seasonal variations, trends, and short-term
fluctuations. A proper ARIMA configuration (p, d, q)
would be needed to ensure accurate predictions. The
small dataset size may require additional data points
or alternative models, such as exponential smoothing
or machine learning-based forecasting, for better
precision.
Figure 2. Residual regressprs for air temperature
Figure 2 illustrates the residual regression analysis
for air temperature, providing critical insights into the
accuracy and effectiveness of the predictive model.
Residuals represent the difference between the
observed and predicted values, and analyzing them
helps determine whether the regression model meets
its assumptions.
A well-performing regression model should exhibit
randomly distributed residuals around zero,
indicating that no systematic bias exists in the
predictions. However, if patterns emerge in the
residual plot—such as clustering or a clear trend—it
may suggest model misspecification,
heteroscedasticity (variance inconsistencies), or
missing predictor variables affecting air temperature
forecasts.
Additionally, the spread of residuals can highlight
potential weaknesses in the model. If the variance of
residuals increases or decreases with the predicted
values, it indicates a non-constant variance issue,
requiring adjustments like data transformation or
weighted regression techniques.
This residual analysis is essential for validating the
regression model’s reliability. If errors show significant
deviations from normality, incorporating non-linear
models, additional environmental variables, or
machine learning techniques may improve air
temperature predictions, enhancing the model’s
overall forecasting accuracy for maritime applications
Figure 3. Trend and Correlation Analysis for Air
Temperature
Figure 3 presents an analysis of trends and
correlations in air temperature data, helping to
understand its variations over time and relationships
with other environmental factors. The trend analysis
examines whether air temperature exhibits an
increasing, decreasing, or stable pattern over the
observed period. A consistent upward or downward
trend may indicate long-term climate shifts or seasonal
effects.
The correlation analysis investigates the
relationship between air temperature and other
meteorological variables such as rainfall, wind speed,
air pressure, and sea temperature. A strong positive or
negative correlation with these factors suggests their
influence on air temperature fluctuations. For example,
if air temperature shows a high correlation with sea
temperature, it may indicate that oceanic conditions
significantly impact atmospheric temperatures in the
study area.
Understanding these trends and correlations is
crucial for improving weather forecasting models and
climate studies. If significant correlations exist, they
can be integrated into predictive models, such as
regression analysis or machine learning algorithms, to
enhance the accuracy of air temperature forecasting.
This analysis helps in making informed decisions for
maritime navigation, agriculture, and climate
adaptation strategies.
Table 8. Conditional Least Squares Estimation
Conditional Least Squares Estimation
Parameters
Estimate
Standard
Error
t Value
Approx
Pr > |t|
Lag
MU
28.76863
0.25771
111.63
<.0001
0
MA1,1
-0.98667
0.18618
-5.3
0.0011
1
AR1.1
-0.49958
0.43416
-1.15
0.2877
1
1126
Table 8 presents the Conditional Least Squares
Estimation results for air temperature modeling. This
method is commonly used in time series analysis,
particularly in ARIMA (AutoRegressive Integrated
Moving Average) models, to estimate parameters by
minimizing the sum of squared residuals.
The MU parameter represents the mean of the air
temperature data, estimated at 28.76863°C with a low
standard error (0.25771), indicating a stable mean
value. The high t-value (111.63, p < 0.0001) suggests
that this estimate is statistically significant.
The MA (Moving Average) coefficient MA1,1 is -
0.98667, with a significant p-value of 0.0011, indicating
that past error terms have a strong influence on the
current air temperature value. This suggests that short-
term fluctuations in temperature are largely dependent
on previous deviations from the predicted values.
The AR (AutoRegressive) coefficient AR1,1 is -
0.49958, with a p-value of 0.2877, meaning it is not
statistically significant. This suggests that past air
temperature values do not strongly influence current
temperatures in this dataset.
Table 9. The ARIMA model's constant estimate
Constant Estimate
43.1408
Variance Estimate
0.39026
Std Error Estimate
0.6247
AIC
21.4025
SBC
22.3102
Number of Residuals
10
Table 9 presents the constant estimate and statistical
parameters of the ARIMA model used for air
temperature forecasting. The constant estimate is
43.1408, which represents the baseline value of the
model. This value plays a crucial role in defining the
overall level of the predicted air temperature trend.
The variance estimate is 0.39026, indicating the
degree of dispersion in the data. A lower variance
suggests that the model’s predictions remain relatively
stable. The standard error estimate of 0.6247 quantifies
the uncertainty in the model’s predictions; a smaller
value generally indicates a more reliable model.
Two important model selection criteria are
included: Akaike Information Criterion (AIC) = 21.4025
and Schwarz Bayesian Criterion (SBC) = 22.3102. These
values help in comparing different ARIMA models—
lower values generally indicate a better-fitting model.
Lastly, the number of residuals is 10, referring to the
number of observations used in the model evaluation.
This small sample size may limit the robustness of the
model, potentially affecting generalization to larger
datasets.
Table 10. AIC and SBC do not include log determinant
Correlations of Parameter Estimates
Parameters
MU
MA1,1
AR1.1
MU
1
0.099
-0.112
MA1,1
0.099
1
0.632
AR1.1
-0.112
0.632
1
Table 10 presents the correlation coefficients among
the estimated parameters of the ARIMA model for air
temperature forecasting. The table includes three key
parameters: MU (constant term), MA1,1 (Moving
Average parameter), and AR1.1 (Autoregressive
parameter).
The correlation between MU and MA1,1 is 0.099,
indicating a weak positive relationship. This suggests
that variations in the constant estimate (MU) have
minimal impact on the moving average component.
The correlation between MU and AR1.1 is -0.112,
showing a weak negative relationship. This implies
that changes in the constant estimate have a slightly
inverse effect on the autoregressive component,
though the effect is not strong.
The highest correlation is between MA1,1 and
AR1.1, with a value of 0.632. This moderate positive
correlation suggests a notable relationship between the
moving average and autoregressive components,
meaning that changes in one of these parameters could
significantly influence the other.
Table 11. Autocorrelation Check of Residuals
Autocorrelation Check of Residuals
To
Lag
Chi-
Square
DF
Pr >
ChiSq
Autocorrelations
6
1.98
4
0.7404
-0.08
-0.25
-0.112
0.14
-0.043
-0.098
Table 11 presents the autocorrelation check of
residuals to assess whether the ARIMA model's
residuals exhibit randomness. This is a crucial step in
validating the model’s adequacy for forecasting air
temperature trends.
The Chi-Square test statistic for up to lag 6 is 1.98
with 4 degrees of freedom (DF) and a p-value of 0.7404.
Since the p-value is much greater than 0.05, it indicates
that there is no significant autocorrelation in the
residuals, meaning the model sufficiently captures the
time-dependent patterns in the data.
The autocorrelation values for lags 1 to 6 are as
follows:
− Lag 1: -0.08 (weak negative autocorrelation)
− Lag 2: -0.25 (moderate negative autocorrelation)
− Lag 3: -0.112 (slight negative autocorrelation)
− Lag 4: 0.14 (weak positive autocorrelation)
− Lag 5: -0.043 (almost negligible correlation)
− Lag 6: -0.098 (minor negative autocorrelation)
Table 12. Model for variable TemperatureAir
Model for variable Temperature Air
Estimated Mean
28.76863
Table 12 presents the estimated mean for the
variable Air Temperature, which is calculated as
28.76863°C. This estimated mean represents the
expected value of air temperature based on the applied
statistical model.
The model used for this estimation likely
incorporates historical data trends and predictive
analysis techniques, such as ARIMA (AutoRegressive
Integrated Moving Average) or regression analysis, to
generate a reliable forecast. The estimated mean
provides a central tendency measure, which is crucial
for understanding long-term patterns in temperature
variations.
This value is particularly useful in maritime and
meteorological applications, where predicting
atmospheric conditions can enhance navigation safety,
route planning, and operational efficiency. By
comparing the estimated mean with actual
observations, researchers can assess the model's
accuracy and reliability. If discrepancies exist, further
refinement of the model, such as incorporating
1127
additional predictors (e.g., humidity, wind patterns),
may be required.
In conclusion, the estimated mean of 28.76863°C
offers a valuable reference for monitoring air
temperature trends, aiding in climate studies and
forecasting applications.
Figure 4. Residual Correlation Diagnostics For Air
Temperature
Figure 4 illustrates the residual correlation
diagnostics for the air temperature forecasting model,
providing a detailed assessment of the model’s
accuracy and reliability. This diagnostic analysis helps
determine whether the residuals exhibit any patterns
that could indicate model misspecification or the
presence of autocorrelation.
The residual correlation plot typically includes
autocorrelation function (ACF) and partial
autocorrelation function (PACF) charts, which display
the correlation between residuals at different time lags.
If the model is correctly specified, these correlations
should remain within the confidence bounds,
indicating that no systematic pattern remains
unaccounted for.
In this case, most residual correlations are close to
zero, with no significant spikes beyond the critical
threshold. This suggests that the residuals behave
randomly, confirming that the ARIMA model has
effectively captured the key patterns in air temperature
variations.
Additionally, the normality of residuals can be
assessed through a histogram or Q-Q plot, ensuring
that they follow a Gaussian distribution. If deviations
from normality exist, further model refinement may be
necessary. However, based on the observed
diagnostics, the model appears to be statistically robust
and suitable for temperature forecasting
Figure 5. Residual Normality Diagnostics For Air
Temperature
Figure 5 presents the residual normality diagnostics
for air temperature, an essential step in validating the
assumptions of the ARIMA model. The purpose of this
analysis is to assess whether the residuals follow a
normal distribution, which is a key requirement for
ensuring the accuracy and reliability of the model’s
predictions.
A histogram of residuals is typically included in this
diagnostic, showing the distribution of error terms. If
the histogram forms a bell-shaped curve, it indicates
that the residuals are normally distributed.
Additionally, a Q-Q plot (quantile-quantile plot)
compares the residuals against a theoretical normal
distribution. If the points lie along the 45-degree line, it
confirms normality.
Another critical test is the Shapiro-Wilk or
Kolmogorov-Smirnov test, which provides a statistical
measure of normality. If the p-value is greater than
0.05, the null hypothesis of normality is not rejected,
confirming that the residuals exhibit a normal
distribution.
In this case, the residuals appear to follow a near-
normal distribution, with slight deviations. If
significant skewness or kurtosis is observed,
transformations or modifications to the model might
be necessary to improve its predictive performance
Table 13. Autoregressive Factors
Autoregressive Factors
Factor 1:
1 + 0.49958 B**(1)
Table 13 presents the autoregressive factor for the
air temperature model, expressed as 1 + 0.49958 B^(1).
This mathematical representation indicates that the
model incorporates a first-order autoregressive (AR)
component, where B represents the backshift operator,
meaning that past values of air temperature influence
the current observation.
The coefficient 0.49958 suggests a moderate positive
relationship between previous and current air
temperature values. This means that past temperature
fluctuations significantly impact the current value, but
with some degree of dissipation over time. In time
series forecasting, an autoregressive factor like this is
essential in capturing temporal dependencies and
improving predictive accuracy.
The presence of an AR(1) process implies that the
temperature data follows a pattern where each
observation is influenced by its immediate past value.
This is particularly useful in meteorology and climate
research, where temperature trends often exhibit serial
correlation due to gradual atmospheric changes.
In conclusion, the autoregressive factor 1 + 0.49958
B^(1) highlights the importance of past temperature
records in forecasting future trends, making it a crucial
component in time series analysis for climate
monitoring and prediction.
Table 14. Forecasts for variable TemperatureAir
Forecasts for variable TemperatureAir
Obs
Forecast
Std Error
95% Confidence Limits
11
28.7466
0.6247
27.522
29.971
12
28.7796
0.6949
27.418
30.142
13
28.7631
0.7113
27.369
30.157
14
28.7714
0.7153
27.369
30.173
15
28.7673
0.7164
27.363
30.171
1128
Table 14 presents the forecasted values for air
temperature (TemperatureAir) along with their
standard errors and 95% confidence limits for five
future observations (Obs 11 to 15). These predictions
are generated using an ARIMA-based forecasting
model, which incorporates past temperature trends to
estimate future values.
The forecasted air temperatures remain relatively
stable, ranging between 28.7466°C and 28.7796°C,
indicating minimal fluctuation in the predicted trend.
The standard error varies between 0.6247 and 0.7164,
reflecting the degree of uncertainty associated with
each prediction. The 95% confidence limits provide a
range within which the true air temperature is
expected to fall, showing a lower bound between
27.363°C and 27.522°C and an upper bound between
29.971°C and 30.173°C.
These forecasts suggest a consistent temperature
pattern with minor variations, which could be
attributed to climatic stability in the observed region.
The narrow confidence intervals indicate a high level
of reliability in the predictions. Such forecasts are
crucial for maritime navigation, climate studies, and
operational planning, enabling informed decision-
making regarding weather conditions and safety
measures.
Figure 6. Forecast for Air Temperature
Figure 6 illustrates the forecasted trend of air
temperature over a specified period, providing
insights into expected temperature variations. The
graph presents a smooth projection of future air
temperature values, highlighting the model’s ability to
predict short-term trends with reasonable accuracy.
The forecasted values remain relatively stable,
indicating minimal fluctuations in air temperature.
This suggests a consistent climatic pattern, with only
slight variations over time. The presence of confidence
intervals around the forecasted line further
demonstrates the level of uncertainty associated with
each prediction. A narrow confidence band implies a
highly reliable model, while a wider band may indicate
potential variations due to external factors such as
weather disturbances or seasonal influences.
This forecast is particularly valuable for maritime
navigation, climate monitoring, and operational
planning. Accurate temperature predictions enable
ship operators, meteorologists, and researchers to
anticipate environmental conditions, ensuring
improved safety, efficiency, and preparedness in
maritime activities. Additionally, understanding
temperature trends contributes to better decision-
making in sectors reliant on climatic stability, such as
shipping, fishing, and offshore operations
5 CONCLUSION AND RECOMMENDATIONS
The analysis of weather data using Big Data Analytics,
ARIMA, and regression models highlights significant
insights into the relationship between climatic
variables and maritime safety. The results indicate that
wind speed, wave height, and sudden drops in
atmospheric pressure are critical factors influencing
risks for fishermen and vessel operations. The ARIMA
model demonstrated a high prediction accuracy, with
an estimated mean temperature of 28.76863°C and a
95% confidence range between 27.363°C and 30.173°C,
ensuring reliable temperature forecasts. The regression
analysis, however, showed a moderate correlation
between weather parameters and temperature
variations, with an R-squared value of 0.632,
suggesting that other environmental factors might
contribute to temperature fluctuations.
Recommendations To enhance maritime safety and
operational efficiency, the following measures are
suggested Integration of AI-Driven Early Warning
Systems – Utilizing real-time data from IoT sensors and
meteorological satellites to provide accurate forecasts
and risk assessments for fishermen. Enhanced Weather
Prediction Models – Combining ARIMA with Machine
Learning techniques to improve the accuracy of
extreme weather event forecasting. Policy and Training
for Fishermen – Conducting awareness programs on
interpreting weather predictions and using
technology-driven navigation systems to mitigate
risks. Development of Smart Maritime Decision
Support Systems – Implementing automated alerts
based on predictive analytics to assist vessel operators
in making informed decisions.
REFERENCES
[1] Priyadharshini, S., Vadivazhagan, K. (2024) Enhanced
Vessel Detection In Maritime Surveillance Using Multi-
modal Data Integration And Deep Learning 2024 8th
International Conference On I-SMAC (Iot In Social,
Mobile, Analytics And Cloud) (I-SMAC), 1090-1099
[2] Jang, H., Yang, W., Kim, H., Lee, D., Kim, Y., Park, J., Jeon,
M., Koh, J., Kang, Y., Jung, M., Jung, S., Hao, C., Z., Hin,
W., Y., Yihang, C., Kim, A. (2024) MOANA: Multi-Radar
Dataset For Maritime Odometry And Autonomous
Navigation Application Arxiv Abs/2412.03887
[3] Kalliovaara, J., Jokela, T., Asadi, M., Majd, A., Hallio, J.,
Auranen, J., Seppänen, M., Putkonen, A., Koskinen, J.,
Tuomola, T., Moghaddam, R., M., Paavola, J. (2024) Deep
Learning Test Platform For Maritime Applications:
Development Of The Em/S Salama Unmanned Surface
Vessel And Its Remote Operations Center For Sensor Data
Collection And Algorithm Development Remote.
Unrated 1545
[4] Otto ‐ Bliesner, B. (1999) El Niño/La Niña And Sahel
Precipitation During The Middle Holocene Geophysical
Research Letters 26
[5] Guan, C., Hu, S., Mcphaden, M., Wang, F., Gao, S., Hou,
Y. (2019) Dipole Structure Of Mixed Layer Salinity In
Response To El Niño ‐ La Niña Asymmetry In The
1129
Tropical Pacific Geophysical Research Letters 46, 12165-
12172
[6] Yuniasih, B., Harahap, W., N., Wardana, D., A., S. (2023)
El Nino and La Nina Climate Anomalies in Indonesia in
2013-2022 AGROISTA : Journal of Agrotechnology
[7] Hyvärinen, M. (2012) ANALYSIS OF SHIP CASUALTIES
IN THE BALTIC, GULF OF FINLAND AND GULF OF
BOTNNJA IN 197-972
[8] Eickschen, S. (2000) Wind Speed And SWH Calibration
For Radar Altimetry In The North Sea
[9] Petrucci, O., Pasqua, A. (2012) Damaging Events Along
Roads During Bad Weather Periods: A Case Study In
Calabria (Italy) Natural Hazards And Earth System
Sciences 12, 365-378
[10] Tolani, H., Neogi, S., Gupta, S., D., Mishra, S., S.,
Samtani, R. (2024) Analyzing Dynamics Of Extreme
Weather Events (EWE) In India: Unfolding Trends
Through Statistical Assessment Of 50 Years Data (1970–
2019) BMC Environmental Science
[11] Huang, W., Zheng, S., Du, Z. (2024) Research On
Insurance Decision-Making Model For Extreme Weather
Based On ARIMA Algorithms 2024 International
Conference On Power, Electrical Engineering, Electronics
And Control (PEEEC), 1154-1157
[12] Noh, S., Lee, S. (2024) Forecasting Meteorological
Drought Conditions In South Korea Using A Data-Driven
Model With Lagged Global Climate Variability
Sustainability
[13] Brandt, P., Munim, Z., H., Chaal, M., Kang, H. (2024)
Maritime Accident Risk Prediction Integrating Weather
Data Using Machine Learning Transportation Research
Part D: Transport And Environment
[14] Panda, S., Ray, P. (2023) A Survey On Weather Prediction
Using Big Data And Machine Learning Techniques 2023
5th International Conference On Energy, Power And
Environment: Towards Flexible Green Energy
Technologies (ICEPE), 1-6
[15] Vaishnavi, J., Minmini, V., Panda, M. (2024) Weather
And Emission Data Analysis And Prediction Using
Machine Learning On A Big Data Platform 2024 15th
International Conference On Computing Communication
And Networking Technologies (ICCCNT), 1-7
[16] Mehta, S., Manisha, E. (2023) Preventive And Predictive
CNN Based Solution For Pipeline Leak, Blockage And
Corrosion Detection International Journal Of Scientific
Research In Computer Science, Engineering And
Information Technology
[17] Sharma, E., Deo, R., Davey, C., P., Carter, B. (2024)
Artificial Intelligence-Empowered Doppler Weather
Profile For Low-Earth-Orbit Satellites Sensors (Basel,
Switzerland) 24
[18] Hasan, M., M. (2024) Regional Analysis Of Extreme
Weather Events Using Deep Learning Innovatech
Engineering Journal
[19] Li, Y., Goda, K. (2022) Hazard And Risk-Based Tsunami
Early Warning Algorithms For Ocean Bottom Sensor S-
Net System In Tohoku, Japan, Using Sequential Multiple
Linear Regression Geosciences
[20] Li, Y., Tong, D., Makkaroon, P., Delsole, T., Tang, Y.,
Campbell, P., Baker, B., Cohen, M., Darmenov, A.,
Ahmadov, R., James, E., Hyer, E., Xian, P. (2024) Multi-
Agency Ensemble Forecast Of Wildfire Air Quality In The
United States: Toward Community Consensus Of Early
Warning Bulletin Of The American Meteorological
Society
[21] Jaya, I., Handoko, B., Andriyana, Y., Chadidjah, A.,
Kristiani, F., Antikasari, M. (2023) Multivariate Bayesian
Semiparametric Regression Model For Forecasting And
Mapping HIV And TB Risks In West Java, Indonesia
Mathematics
[22] Røstad, J., Aarset, M., F. (2024) Boosting Offshore
Uptime: Accurate Real-Time Sea State ADIPEC
[23] Zahorodnia, Y., Maksymov, S. (2021) COMMERCIAL
RISKS IN THE SEA TRANSPORTATION SYSTEM ON
THE EXAMPLE OF THE «EVER GIVEN» CONTAINER
CARRIER Development Of Management And
Entrepreneurship Methods On Transport (ONMU)
[24] Maulida, Z., Hafidzah, N., Purba, D., Kusumawati, E.
(2024) Identification of Passage Plan Process with Risk
Assessment Analysis Globe: Publication of Engineering,
Earth Technology, Marine Science
[25] Kogawa, T., Takayabu, Y. (2013) Environmental
Conditions On The Selection Of MJO And Moist Kelvin
Waves
[26] Bhatla, R., Bhattacharyya, S., Verma, S., Mall, R., Singh,
R., S. (2021) El Nino/La Nina And IOD Impact On Kharif
Season Crops Over Western Agro-Climatic Zones Of
India Theoretical And Applied Climatology 151, 1355-
1368
[27] Aji, T., Pranowo, W., S., Asmoro, N., W., Agustinus, A.,
Kurniawan, M., A., Rahmatullah, A. (2023) The
Characteristics Of The Mixed Layer Depth During La
Niña, El Niño, And Normal Years In The North Natuna
Sea Omni-Aquatics
[28] Yuniasih, B., Harahap, W., N., Wardana, D., A., S. (2023)
El Nino and La Nina Climate Anomalies in Indonesia in
2013-2022 AGROISTA : Journal of Agrotechnology
[29] (2022) Predictive Maintenance Beyond Prediction Of
Failures
[30] Margaretha, R., Syuzairi, M., Mahadiansar, M. (2024)
Digital Transformation In The Maritime Industry;
Opportunities And Challenges For Indonesia Journal Of
Maritime Policy Science
[31] Du, Y., Li, C., Wang, T., Xu, Y. (2023) Special Issue On
"Smart Port And Shipping Operations" In Maritime
Policy & Management Maritime Policy & Management
50, 413-414
[32] Nie, W., Chen, J., Song, D., Dong, L., Liu, X., Wang, E.
(2024) Three-Dimensional Intelligent Monitoring And
Early Warning Technology For Tailings Ponds Based On
Spatiotemporal Fusion Of Multisource Big Data.
Environmental Monitoring And Assessment 196 11, 1081
[33] Salim, S., Hussain, I., Kaur, J., Morita, P. (2023) An Early
Warning System For Air Pollution Surveillance: A Big
Data Framework To Monitoring Risks Associated With
Air Pollution 2023 IEEE International Conference On Big
Data (Bigdata), 3371-3374
[34] Liu, X., Member, J., L., Zhao, Y., Ding, T., Member, X., L.,
S., Liu, J. (2024) A Bayesian Deep Learning-Based
Probabilistic Risk Assessment And Early-Warning Model
For Power Systems Considering Meteorological
Conditions IEEE Transactions On Industrial Informatics
20, 1516-1527
[35] Jin, W., Liu, Y., Fang, Y., Wang, P., Liu, L. (2023)
Construction And Application Of National Urban
Waterlogging Risk Assessment System Based On Big
Data 2023 IEEE 14th International Conference On
Software Engineering And Service Science (ICSESS), 290-
296
[36] Huang, X., Wu, Y., Zhao, X., Zhang, J., Li, J. (2024) Using
Big Data Mining Algorithm To Improve The Accuracy Of
Risk Assessment Of Key Operating Vehicles 2024 3rd
International Conference On Data Analytics, Computing
And Artificial Intelligence (ICDACAI), 583-587
[37] Du, Z., Zhu, Y., Li, D. (2024) A Risk Assessment Model
For Navigation Safety Of Maritime Aquaculture Platform
Based On AIS Ship Trajectory Journal Of Electrical
Systems