Exploring the accuracy of correlation coefficients representing the long-term meteorological data for projecting weather in Bahrain for sustainability

Abstract

This study aims at modelling long-term weather parameters (1955 to 2020), recorded in Bahrain, and at investigating which parameters have strong correlation coefficients with such data to account for mitigation and achieve sustainability. Eight forms of fitting methods were utilized to calculate the correlation coefficient between each parameter versus each given year. These parameters (Average and Anomalies) are the average temperature, maximum temperature, minimum temperature, humidity, wind speed, dust and precipitation. The forms of correlations used are as follows: Linear regression, Quadratic regression, Cubic regression, Power regression, ab-Exponential regression, Logarithmic regression, Hyperbolic regression and Exponential regression. Among all forms of regressions, the Exponential (cubic) regression is found to have the highest correlation coefficient for such data, both Average and Anomalies data; the highest is for humidity versus year (r = 0.900, strong relation), and the least is for precipitation (r = 0.1647, poor relation) for Average data. As for the Anomalies data, the highest is for humidity versus year (r = 0.9019) and the least is for precipitation versus (r = 0.1647, no relation). The novelty of this paper lies in concluding that the exponential (cubic) regression is the most accurate correlation (among 8 correlation coefficients) to predict all the recorded long-term (65 years) weather parameters in Bahrain eight correlation. This regression fit is the most useful one in projecting the weather trend by 2050 in Bahrain to account for the future built environment to become more resilient, sustainable and tolerant with the environment due to the anticipated damage resulting from climate change. More investigation is to be made for other data set in the Gulf Cooperation Countries, and worldwide, to explore whether this Exponential (cubic) regression will still have the highest correlation coefficient among the others.

Keywords:

• Bahrain
• Built Environment
• Correlation Coefficients
• Meteorological Parameters
• sustainability

1. Introduction

Long-term weather parameters (climate change) and the Built Environment nexus were classified into categories. These categories include building structure, which is affected by several factors (floods, landslides, storms and snow loads), building constructions (influenced by fastening systems and water supply), building materials (impacted by frost resistance, ultraviolet resistance and insulation properties) and the indoor climate (affected by humidity and temperature) (Andrić, Koc, & Al-Ghamdi, 2019).

Currently, with devastating weather disasters witnessed worldwide; in Asia (flooding of Pakistan and China), Europe (drought of rivers, wildfires, flooding, massive ice stones), North America and South America (drought of rivers, wildfires, flooding), scientists are obliged to conduct more advanced research in climatology, modelling, forecasting the future using Big Data analytics (Sebestyén, Czvetkó, & Abonyi, 2021). Now, there is a strong tie and Nexus between Weather (Climate), Prediction & Projection and Built Environment. The latter is defined as human made environment that provides proper setting of houses, buildings, zoning, streets, open spaces transportation options and more. According to the Environmental Protection Agency (EPA) (2022) the built environment touches all aspects of our lives, including the buildings and city habitats, the distribution systems that provide us with water and electricity, roads, bridges, and transportation systems used. It is the man-made or modified structures that facilitate the living, working, and recreational spaces. There is a strong relation between Built Environment and Health. The built environment influences our level of physical activity; for instance, the inaccessibility or nonexistence of sidewalks and bicycle or walking paths contributes to sedentary habits, leading to poor health outcomes such as obesity, cardiovascular disease, diabetes, and some types of cancer (Centre for Disease Control and Prevention (CDC), 2022). Additionally, there is a strong relationship between the built environment and weather parameters, and hence climate change has been characterized in order to propose systemic changes to improve the adaptation of cities to climate change (Zięba et al., 2020).

Among the useful research in using long-term meteorological data to create awareness and contribute to a growing body of climate change that is necessary for the development of effective mitigation and adaptation strategies is the recent published work of Kazemzadeh, Noori, Jamali, & Abdi (2022). Forty years of air temperature changeover was utilized and the polynomial fitting scheme was used—to both monthly and annual air temperature data—to detect trends and classify them into linear and nonlinear (quadratic and cubic) categories. Applying similar approach of utilizing long term data is the study on a new globally reconstructed sea surface temperature analysis dataset since 1900 developed by the China Meteorological Administration (Chen, Cao, Zhou, Zhang, & Liao, 2021).

Furthermore, building environment management systems should allow for minimizing the possible future negative impact of natural hazards caused by the behavior of long-term weather parameters. This can be achieved by looking into the correlation coefficients between these parameters with the passing and forth coming years (Świąder et al., 2018). Proper planning and configuring built environment disaster, economic, social, energy, water and comforting risk from monitoring and studying correlation between various weather parameters throughout the years, i.e. modelling, will lead to great reduction and comfort urban life (Mabon, Kondo, Kanekiyo, Hayabuchi, & Yamaguchi, 2019; Otto-Zimmermann, 2010; Sharifi & Yamagata, 2018).

Indeed, to have a sustainable building, one needs reliable information on the long-term performance and, more specifically, the expected service-life of building materials, components, assemblies and most importantly anticipated effects of climate change on the built environment (Lacasse, 2019).

Al-Humaiqani and Al-Ghamdi (2022) listed important indicators that play a crucial role in marking a resilient quality to build environment planning, analysis, and design where among them is weather modelling.

The variation of long-term meteorological parameters, past and future projections, is anticipated to have a significant impact on the built environment, especially, whereby the world faces climate change as one of the biggest environmental challenges ever (Dervis, 2007; European Environment Agency (EEA), 2021; Carpenter, Walker, Anderies, & Abel, 2001; Chen, Moufouma-Okia, Masson-Delmotte, Zhai, & Pirani, 2018; Solomon, Plattner, Knutti, & Friedlingstein, 2009). In 2015, the estimated climate change cost to private businesses was about USD 43 trillion during the 21^st century (UNFCCC, 2017). This explains why researchers do modelling and correlations to their weather data (Nabipour et al., 2020; Sharafati & Pezeshki, 2020; Zhao et al., 2021).

The officially accurate recorded data in Bahrain from 1955 to 2020 were used as an example to calculate the various correlation coefficients resulted from each form of statistical fitting between each meteorological parameter and the year. These weather parameters (Average and Anomalies) are the average temperature, maximum temperature, minimum temperature, humidity, wind speed, dust and precipitation. The forms of correlations used are eight: Linear regression, Quadratic regression, Cubic regression, Power regression, ab-Exponential regression, Logarithmic regression, Hyperbolic regression and Exponential regression.

The purpose of the paper is to explore whether there is a specific regression coefficient, among the known eight regression coefficients, which is more accurate than the others to represent all the long-term (65 years) recorded weather data parameters. It also aims to alert policy makers, architects, engineers, and environmentalists to project the temperature, humidity, precipitation, etc., in 2050 to allow for actions, mitigations and adaptation to climate change and be resilient towards it.

This study is one of the measures that policy makers and developers should conduct in order to forecast the required built environment to be of resilient quality. For this reason, we tested eight forms of fitting methods to calculate the long-term meteorological parameters’ correlation coefficient for future built environments. The outcome of this study can be tested when doing similar investigation on other weather data set in the Gulf Cooperation Council Countries (GCCC) and worldwide to explore which form of the aforementioned eight-correlation regression will have the highest correlation coefficient and be used for forecasting.

2. Methodology

Recorded data from 1955 to 2020 were made by Bahrain Meteorological Directorate, Ministry of Transport and Communication, Kingdom of Bahrain. The station is well calibrated and regularly updated. It has data since 1905 but the most reliable data are those from 1955. The station is located at Bahrain International Airport, Muharraq town, which was the first Capital of Bahrain.

The data is measured continuously and distributed at 10 min intervals, then at hour interval and per day to per month to per year. The station records various meteorological parameters; however, for this research we have selected the most important parameters for built environment. Although solar radiation data is important, it was only recorded accurately and systemically in Bahrain since 2016. However, the recorded temperature (minimum, average and maximum) can be considered to represent the solar radiation; it’s a result of absorbing and emitting solar energy. This is a valid point since we are not studying the reason for changing the recorded values of meteorological parameters, but rather we are exploring which correlation coefficient—among eight—is more accurate to represent the long-term meteorological data in Bahrain. Besides, Bahrain has solar radiation data for the past 7 years only while the other data were since 1955. These weather parameters (Average and Anomalies) are average temperature, maximum temperature, minimum temperature, humidity, wind speed, dust and precipitation. These data may represent the entire island because the area of the main island is about 604 km² (Length about 51 km and width 18 km) with latitude 26.03°N and longitude 50.55°E.

The Kingdom of Bahrain consists of 30 small islands, five of which are relatively large islands. Muharraq island is considered to be one of those large islands. The length of the coastline is 130 km and the highest elevation is in south middle of Bahrain—Aldokhan Mountain (134 m)—Figure 1. This means there is no much variation in the weather data; it does not exceed 5%.

Figure 1. Bahrain Map showing the location of the official weather station in Muharraq City, where Bahrain International Airport is located.

In order to produce regression equations for each meteorological parameter (y) versus year (x), we used several regression fits. These are available online, and they provide target function table data in the form of points [x, f(x)] to build several regression models listed below (PLANETCALC Online, 2022):

1. Linear regression: y = ax + b

2. Quadratic regression: y = ax²+bx + c

3. Cubic regression: y = ax³+bx²+cx + d

4. Power regression: y = a x^b

5. ab-Exponential regression: y = a⋅b^x

6. Logarithmic regression: y = a+b ln(x)

7. Hyperbolic regression: y = a−b/x

8. Exponential regression: y = e (a + bx)

The correlation between each meteorological parameter (y) and year (x) for each type of regression fit can be compared using the following:

Correlation coefficient (r): It is called also Pearson’s correlation. It is used to measure how strong a relationship is between two variables and is commonly used in linear regression to find how strong a relationship is between data. The formulas return a value between −1 and 1, where 1 indicates a strong positive relationship, −1 indicates a strong negative relationship and zero indicates no relationship at all.

Coefficient of determination (R²): It is a number between 0 and 1 that measures how well a statistical model predicts an outcome. R² can be interrupted as the proportion of variation in the dependent variable that is predicted by the statistical model.

Average relative error (ARE): It is used as a measure of precision; it is the ratio of the absolute error of a measurement to the measurement being taken. This type of error is relative to the average size of the item being measured. RE is expressed as a percentage and has no units.

Such statistical data allow decision makers to prioritize a particular meteorological parameter that most likely will affect the future built environment.

Table 1 lists all regression fits for the average long-term weather data in Bahrain from 1955 to 2022 while Table 2 lists all regression fits for the anomalies average long-term weather data from 1955 to 2022. The Regression fit that has the highest correlation coefficient for each weather set parameter (in Bold blue color) was marked in bold black color.

Table 1. Regression fits for the average long-term weather data in the Kingdom of Bahrain from 1955 to 2020.

Type of regression	Linear correlation coefficient	Coefficient of determination	Average relative error %
1-Annual Mean Temperature (°C)
1.1 Linear regression y = 0.02316021 x + 25.99731469	0.6058	0.3670	1.7462
1.2 Quadratic regression: y = 0.00087273 x^2−0.03531283 x + 26.66000918	0.7198	0.5182	1.4271
1.3 Cubic regression y = 0.00002255 x^3−0.00139315x^2+0.02587062x + 26.3056245	0.7392	0.5464	1.4288
1.4 Power regression y = 25.62615617 x^0.01342003	0.4504	0.2029	1.9911
1.5 ab-Exponential regression y = 26.00603783 ⋅ 1.00085718^x	0.6085	0.3703	1.7408
1.6 Logarithmic regression y = 25.60021975 + 0.36254711⋅ln(x)	0.4478	0.2005	1.9932
1.7 Hyperbolic regression y = 26.85418631 − 1.11977765/x	0.2145	0.0460	2.1749
1.8 Exponential regression y = e^3.25832874 + 0.00085681x	0.6085	0.3703	1.7408
2-Annual Mean Maximum Temperature (°C)
2.1 Linear regression y = 0.04120760x + 29.17424242	0.7828	0.6127	1.6292
2.2 Quadratic regression y = 0.00027096 x^2+0.02305329 x + 29.37999126	0.7877	0.7877	1.6311
2.3 Cubic regression y= −0.00000217x^3+0.00048919x^2+0.01716064x + 29.414122	0.7877	0.6206	1.6273
2.4 Power regression y = 28.15025934 x^0.02516629	0.6919	0.4788	2.0047
2.5 ab-Exponential regression y = 29.18944314 ⋅ 1.00134929^x	0.7837	0.6142	1.6275
1.6 Logarithmic regression y = 28.07658748 + 0.76595097 ⋅ ln(x)	0.6870	0.4720	2.0169
3-Annual Mean Minimum Temperature (°C)
3.1 Linear regression y = 0.02236030 x + 22.58956643	0.5380	0.2894	2.4553
3.2 Quadratic regression y = 0.00153250 x^2− 0.08031697 x + 23.75324213	0.8270	0.6839	1.5111
3.3 Cubic regression y = 0.00002351x^3−0.00083028x^2−0.01651730x + 23.3837038	0.8425	0.7098	1.4492
3.4 Power regression y = 22.52616601 x ^0.01077662	0.2978	0.0886	2.6565
3.5 ab-Exponential regression y = 22.60555510 ⋅1.00093620^x	0.5425	0.2943	2.4402
3.6 Logarithmic regression y = 22.49732383 + 0.26003861⋅ln(x)	0.2954	0.0872	2.6707
3.7 Hyperbolic regression y = 23.36384184 − 0.34843168/x	0.0613	0.0037	2.6283
3.8 Exponential regression y = e^3.11819568 + 0.00093576^x	0.5425	0.2943	2.4402
4-Annual Mean Relative Humidity (%)
4.1 Linear regression y = −0.20684688x + 69.97482517	0.8633	0.7453	2.9850
4.2 Quadratic regression y= −0.00352317x^2+0.02920567x + 67.29956294	0.8989	0.8081	2.6438
4.3 Cubic regression y = 0.00003617x^3−0.00715872x^2+0.12737265x + 66.73096348	0.8999	0.8099	2.6165
4.4 Power regression y = 75.47062669x^−0.05645229	0.6694	0.4481	4.3273
4.5 ab-Exponential regression y = 70.37072584⋅0.99664219^x	0.8552	0.7315	3.0915
4.6 Logarithmic regression y = 74.36023165 − 3.49724839⋅ln(x)	0.6892	0.4751	4.1890
4.7 Hyperbolic regression y = 62.24201345 + 11.10648712/x	0.3394	0.1152	5.6320
4.8 Exponential regression y = e^4.25377735 − 0.00336346^x	0.8552	0.7315	3.0915
5-Annual Mean Wind Speed (in Knots)
5.1 Linear regression y = − 0.02468010 x + 10.04648019	0.6664	0.4442	4.0062
5.2 Quadratic regression y = 0.00023195x^2 − 0.04022090 x + 10.22260927	0.6749	0.4555	3.9981
5.3 Cubic regression y= −0.00000295x^3+0.00052846x^2−0.04822720x + 10.2689831	0.6753	0.4561	3.9941
5.4 Power regression y = 10.92352954x^−0.05331573	0.6376	0.4066	4.3441
5.5 ab-Exponential regression y = 10.03428690⋅0.99738899^x	0.6680	0.4463	4.0267
5.6 Logarithmic regression y = 10.85768324 − 0.50627996⋅ln(x)	0.6456	0.4168	4.2730
5.7 Hyperbolic regression y = 9.07921748 + 1.94193906/x	0.3840	0.1474	5.6475
5.8Exponential regression y = e^2.30600792 − 0.00261442^x	0.6680	0.4463	4.0267
6-Annual Total Number Of Days With Dust, Dust storm or Sandstorm (Visibility 1 km or Less)
6.1 Linear regression y = − 0.02903664 x + 6.01818182	0.1344	0.0180	Infinity
6.2 Quadratic regression y = 0.00126917x^2− 0.11407109x + 6.98190559	0.1676	0.0281	Infinity
6.3 Cubic regression y = 0.00003336 x^3 − 0.00208314x^2 − 0.02355215x + 6.457604	0.1733	0.0300	Infinity
6.4 Power regression N/A	N/A	N/A	N/A
6.5 ab-Exponential regression N/A	N/A	N/A	N/A
6.6 Logarithmic regression y = 6.63600122 − 0.49161700⋅ln(x)	0.1075	0.0115	Infinity
6.7 Hyperbolic regression y = 5.09051282 − 0.62286977/x	0.0211	0.0004	Infinity
6.8 Exponential regression N/A	N/A	N/A	N/A
7-Annual Total Precipitation (mm)
7.1 Linear regression y = 0.01811711x + 78.51883450	0.0063	0.0000	108.0079
7.2 Quadratic regression y = −0.01072109x^2+0.73643004x + 70.37795455	0.0640	0.0041	107.6366
7.3 Cubic regression y = −0.00152660x^3+0.14270187x^2 −3.40629514x + 94.3733051	0.1646	0.0271	104.7709
7.4 Power regression y = 50.27720544x^0.05854884	N/A	N/A	81.0899
7.5 ab-Exponential regression y = 51.36907021⋅1.00502574 ^x	N/A	N/A	79.5556
7.6 Logarithmic regression y = 84.06077113 − 1.52534761⋅ln(x)	0.0251	0.0006	108.6017
7.7 Hyperbolic regression y = 75.63063353 + 48.31536548/x	0.12343	0.0152	107.7999
7.8 Exponential regression y = e ^3.93903624 + 0.00501316 ^x	N/A	N/A	79.5556

Table 2. Regression fits for the anomalies average long-term weather data in the Kingdom Bahrain from 1955 to 2020.

1.  Annual Mean Temperature Anomalies
8.1 Linear regression y = 0.02316021x − 0.34268531	0.6058	0.3670	Infinity
8.2 Quadratic regression y = 0.00087273x^2 −0.03531283x + 0.32000918	0.7198	0.5182	Infinity
8.3 Cubic regression y = 0.00002255x^3−0.00139315x^2+0.02587062x − 0.0343754	0.73925	0.5464	Infinity
8.4 Power regression N/A	N/A	N/A	N/A
8.5 ab-Exponential regression N/A	N/A	N/A	N/A
8.6 Logarithmic regression y= −0.73978025 + 0.36254711⋅ln(x)	0.4478	0.2005	Infinity
8.7 Hyperbolic regression y = 0.51418631 − 1.11977765/x	0.2145	0.0460	Infinity
8.8 Exponential regression N/A	N/A	N/A	N/A
9- Annual Mean Maximum Temperature Anomalies
9.1 Linear regression y = 0.04121991x − 0.93632168	0.7829	0.6130	260.4867
9.2 Quadratic regression y = 0.00027023×2 + 0.02311452x − 0.73112719	0.7878	0.6206	238.5421
9.3 Cubic regression y= −0.00000214x^3+0.00048514x^2+0.01731140x − 0.6975145	0.7878	0.6207	241.6524
9.4 Power regression N/A	N/A	N/A	N/A
9.5 ab-Exponential regression N/A	N/A	N/A	N/A
9.6 Logarithmic regression y = −2.03468412 + 0.76629717⋅ln(x)	0.6873	0.4723	304.9447
9.7 Hyperbolic regression y = 0.64953202 − 2.83366221/x	0.3941	0.155	282.6200
9.8 Exponential regression N/A	N/A	N/A	N/A
10- Annual Mean Minimum Temperature Anomalies
10.1 Linear regression y = 0.02235299x − 0.24034033	0.5377	0.2892	263.8650
10.2 Quadratic regression y = 0.00153258x^2−0.08032974x + 0.92339729	0.8268	0.6831	105.7010
10.3 Cubic regression y = 0.00002354x^3−0.00083334x^2−0.01644526x + 0.5533678	0.8424	0.7096	120.2438
10.4 Power regression N/A	N/A	N/A	N/A
10.5 ab-Exponential regression N/A	N/A	N/A	N/A
10.6 Logarithmic regression y= −0.33240578 + 0.25990821⋅ln(x)	0.29527924	0.0871	332.7731
10.7 Hyperbolic regression y = 0.53366063 − 0.34802116/x	0.0613	0.0037	343.4627
10.8 Exponential regression N/A	N/A	N/A	N/A
11- Annual Mean Relative Humidity Anomalies
11.1 Linear regression y= −0.20757750x + 3.98414918	−0.8654	0.7489	Infinity
11.2 Quadratic regression y= −0.00351502x^2+0.02792872x + 1.31507867	0.9007	0.8113	Infinity
11.3 Cubic regression y = 0.00003930x^3−0.00746464x^2+0.13457631x + 0.69735820	0.9019	0.8134	Infinity
11.4 Power regression N/A	N/A	N/A	N/A
11.5 ab-Exponential regression N/A	N/A	N/A	N/A
11.6 Logarithmic regression y = 8.38727083 − 3.51028898⋅ln(x)	0.6910	0.4776	Infinity
11.7 Hyperbolic regression y = −3.77610776 + 11.14753914/x	0.3403	0.1158	Infinity
11.8 Exponential regression N/A	N/A	N/A	N/A
12- Annual Mean Wind Speed Anomalies (in Knots)
12.1 Linear regression y = −0.02482413x + 0.54675991	−0.6667	0.4445	Infinity
12.2 Quadratic regression y = 0.00024087 × 2 − 0.04096268x + 0.72966346	0.6757	0.4566	Infinity
12.3 Cubic regression y= −0.00000215 × 3 + 0.00045725 × 2 − 0.04680536x + 0.763505	0.6759	0.4569	Infinity
12-Annual Mean Wind Speed Anomalies (in Knots)
12.4 Power regression N/A	N/A	N/A	N/A
12.5 ab-Exponential regression N/A	N/A	N/A	N/A
12.6 Logarithmic regression y = 1.36341176 − 0.50945551⋅ln(x)	0.6461	0.4174	Infinity
11.7 Hyperbolic regression y = −0.42617018 + 1.95358145/x	0.3842	0.1476	Infinity
12.8 Exponential regression N/A	N/A	N/A	N/A
13- Annual Total Number of Days With Dust, Dust storm Or Sandstorm Anomalies (Visibility 1 km or Less)
13.1 Linear regression y = − 0.02903664x + 0.01818182	−0.13443	0.0180	Infinity
13.2 Quadratic regression y = 0.00126917x^2−0.11407109x + 0.98190559	0.1676	0.0281	Infinity
13.3 Cubic regression y = 0.00003336x^3−0.00208314x^2 − 0.02355215x + 0.457604	0.1733	0.0300	Infinity
13.4 Power regression N/A	N/A	N/A	N/A
13.5 ab-Exponential regression N/A	N/A	N/A	N/A
13.6 Logarithmic regression y = 0.63600122 − 0.49161700⋅ln(x)	0.1075	0.0115	Infinity
13.7 Hyperbolic regression y = − 0.90948718 − 0.62286977/x	0.0211	0.0004	Infinity
13.8 Exponential regression N/A	N/A	N/A	N/A
14 -Annual Total Precipitation Anomalies9mm)
14.1 Linear regression y = 0.01811711x + 7.51883450	0.00639	0.0000	114.6974
14.2 Quadratic regression y = −0.01072109 x^2+0.73643004 x − 0.62204545	0.0640	0.0041	113.5036
14.3 Cubic regression y= −0.00152660x^3+0.14270187x^2−3.40629514x + 23.373305	0.1646	0.0271	118.1365
14.4 Power regression N/A	N/A	N/A	N/A
14.5 ab-Exponential regression N/A	N/A	N/A	N/A
14.6 Logarithmic regression y = 13.06077113 − 1.52534761⋅ln(x)	0.0251	0.0006	112.5132
14.7 Hyperbolic regression y = 4.63063353 + 48.31536548/x	0.1234	0.0152	110.3017
14.8 Exponential regression N/A	N/A	N/A	N/A

3. Results

Table 3 summarizes the correlation coefficients (r) for the Average Data of the long-term Meteorological parameters; average temperature T, maximum temperature T_max, minimum temperature T_min, average humidity H, average wind speed V, dust D and precipitation Pr . It clearly shows that the highest correlation coefficient (R) is when we use the cubic regression type equation. These results are derived from the following regression fits for the Average data (Table 1, sections 1.1 to 7.8).

Table 3. Summary of the correlation coefficients (r) for the different quantities fitted to the different forms for Average Data.

	T	Tmax	Tmin	H	V	D	Pr
Linear	0.6058	0.7828	0.5380	−0.8634	−0.6665	−0.1345	0.0063
Quadratic	0.7199	0.7877	0.8269	0.8990	0.6750	0.1676	0.0641
Cubic	0.7392	0.7878	0.8425	0.9000	0.6754	0.1733	0.1647
Power	0.4505	0.6920	0.2978	0.6694	0.6377	N/A	N/A
ab- Exponential	0.6085	0.7837	0.5426	0.8553	0.6681	N/A	N/A
Logarithmic	0.4478	0.6871	0.2954	0.6893	0.6456	0.1075	0.0251
Hyperbolic	0.2145	0.3940	0.0614	0.3395	0.3840	0.0211	0.1235
Exponential	0.6085	0.7837	0.5426	0.8553	0.6681	N/A	N/A

Table 4 summarizes the correlation coefficients (r) for the Anomalies Data of the long-term meteoritical parameters; average temperature T, maximum temperature Tmax, minimum temperature Tmin, average humidity H, average wind speed V, dust D and precipitation Pr. Herein, Anomalies are the deviation of a measurable unit (e.g. temperature or precipitation) over a period in a given location from the long-term average, often the 65-year mean, for that location. Similarly, it clearly shows that the highest correlation coefficient (R) is evident when the cubic regression type equation is used. These results are derived from the following regressions fit for the Anomalies data (Table 2, sections 8.1 to 14.8).

Table 4. Summary of the correlation coefficients (r) for the different quantities fitted to the different forms for Anomalies Data.

	T	Tmax	Tmin	Humidity	Wind Speed	Dust	Precip.
Linear	0.6058	0.7830	0.5378	−0.8654	−0.6667	−0.1345	0.0063
Quadratic	0.7199	0.7878	0.8269	0.9007	0.6758	0.1676	0.0641
Cubic	0.7392	0.7879	0.8424	0.9019	0.6760	0.1733	0.1647
Power	NA	NA	NA	NA	NA	N/A	N/A
ab- Exponential	NA	NA	NA	NA	NA	N/A	N/A
Logarithmic	0.4478	0.6873	0.2953	0.6911	0.6461	0.1075	0.0251
Hyperbolic	0.2145	0.3942	0.0613	0.3404	0.3842	0.0211	0.1235
Exponential	NA	NA	NA	NA	NA	N/A	N/A

The data reveal that the Exponential (cubic) regression has the highest correlation coefficient for both the Average and Anomalies data (fitting) for temperature, maximum temperature, minimum temperature, humidity, wind speed, dust and precipitation against the year, using eight forms of correlations. The highest correlation is for humidity versus year (r = 0.900, strong relation) and the least is for precipitation (r = 0.1647, no relation) for Average data fitting. These are presented in sections Table 1, sections 1.3, 2.3, 3.3, 4.3, 5.3, 6.3, 7.3 and 8.3. This exponential (cube) regression fit has also the highest Coefficient of determination (R²)—compared to others for the Average data, with highest R² of 0.8100 for humidity. Similar findings for Anomalies data fitting are reported (Table 2, sections 8.3, 9.3, 10.3, 11.3, 12.3, 13.3 and 14.3), where Exponential (cubic) regression has the highest correlation coefficient; the highest is for humidity versus year (r = 0.9019) and the least is for precipitation (r = 0.1647). This Exponential (cube) fit has the highest Coefficient of determination (R²)—compared to others, with the highest correlation for humidity (R² = 0.8135) and least for precipitation (R² = 0.0271).

Therefore, the best fitting regression equation to predict or foresee a particular meteorological parameter is the exponential cube regression fit, for both Average data and Anomalies data. These can be used to plan for future built environment. This is crucial due to concentration of greenhouse gases (GHG), led by carbon dioxide (CO₂), in our atmosphere, along with all negative man-made destruction to the environment which has resulted in climate change and subsequently to the evolution or creation of extremities.

4. Discussion

Figure 2 shows the annual variation of the recorded average long-term temperature throughout the years from 1955 to 2022. The figure shows that the temperature is expected to increase exponentially in Bahrain. It might reach an average of 36.0 °C by 2050. Considering the temperature, the built environment should undergo substantial changes. It seems that, taking into account the house walls’ thickness and the use of shade forecasting on the houses, is not enough. It is notable, too, that some serious consideration should also be given to the exposure of the roofs or facades to sun light in order to harvest solar electricity to avoid CO₂ emission. It is indeed a big challenge to architects, developers and environmentalists.

Figure 2. The annual variation of the recorded average long-term temperature throughout the years from 1955 up to 2022. The curve is represented by the Following equation: $y=0.00002255x^3-0.00139315x^2+0.02587062x+26.30562451,\text{ r }=0.7392$.

Figure 3 shows the annual variation of the recorded maximum long-term temperature throughout the years from 1955 to 2020. The figure shows that the temperature is expected to increase exponentially in Bahrain. It might reach an average of 33.7 °C by 2050. This seems to be an unusual result as it should be higher than the expected average temperature in 2050 (36.0 °C).

Figure 3. The annual variation of the recorded average long-term maximum temperature throughout the years from 1955 up to 2022. The curve is represented by the following equation: $y=-0.00000217x^3+0.00048919x^2+0.01716064x+29.41412247,\text{ r }=0.7878$.

Figure 4 shows the annual variation of the recorded minimum long-term minimum temperature throughout the years from 1955 to 2022. The figure shows that the temperature is expected to increase exponentially in Bahrain. It might reach an average of 35.0 °C by 2050. This result is consistent with the expected average temperature in 2050 (36.0 °C), where both are subjected to nearly a “worrying” exponential increase. The result means that no significant winter is expected in the future. Accordingly, the built environment should be thought off very well.

Figure 4. The annual variation of the recorded average long-term minimum temperature throughout the years from 1955 up to 2022. The curve is represented by the following equation: $y=0.00002351x^3-0.00083028x^2-0.01651730x+23.38370387,\text{ r }=0.84254175$.

Figure 5 shows the annual variation of the long-term average temperature anomalies of the years from 1955 up to 2022. The curve (exponential cube regression fit) shows that the temperature in 2050 will be larger by 9.5 °C than the long-term average temperature (which was normalized to 0 °C). This is a substantial increase for a span of 30 years.

Figure 5. The annual variation of the long-term average temperature anomalies of the years from 1955 up to 2022. The curve is represented by the following equation: $y=0.00002255x^3-0.00139315x^2+0.02587062x-0.03437549,\text{ r }=0.7879$.

The annual variation of the long-term maximum temperature anomalies of the years from 1955 up to 2022 (Figure 6) shows that by 2050, the rise of the maximum temperature will be 9.5 °C compared to long-term maximum average (65 years) which is alerting especially because the correlation coefficient r is relatively high (0.7879). However, the situation seems to be worse when looking at the annual variation of the long-term minimum temperature anomalies of the years from 1955 to 2022 (Figure 7). It shows that in about 30 years the increase in the long-term average minimum temperature will be 12 °C, given the fact that the correlation coefficient r is relatively high (0.8424).

Figure 6. The annual variation of the long-term average maximum temperature anomalies of the years from 1955 up to 2022. The curve is represented by the following equation: $y=-0.00000214x^3+0.00048514x^2+0.01731140x-0.69751457,\text{ r }=0.7392$.

Figure 7. The annual variation of the long-term average minimum temperature anomalies of the years from 1955 up to 2022. The curve is represented by the following equation: $y=0.00002351x^3-0.00083028x^2-0.01651730x+23.38370387,\text{ r }=0.84243154$.

The most surprising result is the variation of annual mean relative humidity from 1955 to 2020, which shows that humidity is expected to decrease from 66% (the long-term average) to 46% by 2050 (Figure 8). This is an advantage for people living on islands as the cooling system may perform better, particularly for air conditioners working in water due to the evaporation mechanism.

Figure 8. The annual variation of the long-term average relative humidity of the years from 1955 up to 2022. The curve is represented by the following equation: $y=0.00003617x^3-0.00715872x^2+0.12737265x+66.73096348,\text{ r }=0.9000$.

The highest correlation coefficient, among all the examined meteorological parameters, belongs to humidity (r = 0.9000). The same observation is applied to the annual variation of relative humidity anomalies (Figure 9). The relative humidity will be less than the long-term average humidity by 22 unit percentage. This is the highest correlation coefficient calculated in this study (r = 0.9019).

Figure 9. The annual variation of the long-term average anomalies relative humidity of the years from 1955 up to 2022. The curve is represented by the following equation: $y=0.00003930x^3-0.00746464x^2+0.13457631x+0.69735820,\text{ r }=0.9019$.

Figure 10 illustrates the long-term average annual wind speed variation (from 1955 to 2022). The results indicate that the wind speed has a tendency to reduce annually reaching an average wind speed of 8.2 knots (4.1 m/s) by 2050. This means that the potential of using wind energy for city and homes electrification is less optimistic unless micro turbines are used where they have a low cut-in speed value (1 m/s). It is interesting to note that the correlation coefficient is not high (r = 0.6754). For the annual variation of the wind speed anomalies, the result shows that the wind speed drops by 1.4 knots (0.7 m/s) by 2050 compared to the long-term average (Figure 11).

Figure 10. The annual variation of the long-term average wind of the years from 1955 up to 2022. The curve is represented by the following equation: $y=-0.00000295x3+0.00052846x^2-0.04822720x+10.26898310,\text{ r }=0.6754$.

Figure 11. The annual variation of the long-term average wind anomalies of the years from 1955 up to 2022. The curve is represented by the following equation: $y=-0.00000215x^3+0.00045725x^2-0.04680536x+0.76350524,\text{ r }=0.6760$.

Figure 12 illustrates the long-term average annual variation from 1955 to 2022, taking into consideration the annual total number of days with dust, dust storms or sandstorms (visibility 1000 meters or less). It seems that there is a tendency that storm or sand storm events will increase to 15 by 2050 (Figure 12). This means that the performance of PV panels and, to a larger degree, the solar concentrators parabolic (CSP) will be hugely affected as they are very sensitive to sand and dust. If this is the trend, then future renewable energy option for the built environment should exclude CSP, otherwise, the consumption of water will be high which will offset the advantage of solar benefit. For the annual variation of dust, dust storm or sandstorm anomalies, the result shows that 9 more dust storms or sandstorms will occur by 2050 compared to the long-term average (Figure 13). Fortunately, the correlation coefficient between dust events (both average and anomalies) are low, i.e. r = 0.1733 and 0.1733, respectively.

Figure 12. The annual variation of the long-term average annual total number of days with dust, dust storm or sandstorm (visibility 1000 meters or less) from 1955 up to 2022. The curve is represented by the following equation: $y=0.00003336x^3-0.00208314x^2-0.02355215x+6.45760490,\mathrm{r}=0.1733$.

Figure 13. The annual variation of the long-term average annual total number of days with dust, dust storm or sandstorm anomalies (visibility 1000 meters or less) from 1955 up to 2022. The curve is represented by the following equation: $y=-0.00152660x^3+0.14270187x^2-3.40629514x+23.37330517,\text{ r }=0.17331857$.

Figure 14 shows the results for the long-term average annual precipitation variation from 1955 to 2022. The results indicate that there is a great tendency to reduce annually reaching deficit value (−270 mm) by 2050 (Figure 14), while for the anomalies, the annual precipitation is less than the long-term average (71 mm) by 340 mm (huge deficit) by 2050 compared to the long-term average (Figure 15). Due to the very low correlation coefficient between precipitation versus year—which is 0.1647 for both average and anomalies data, no consideration is given for this analysis.

Figure 14. The annual variation of the long-term average annual precipitation from 1955 up to 2022. The curve is represented by the following equation: $y=-0.00152660x^3+0.14270187x^2-3.40629514x+94.37330517,\text{ r }=0.1647$.

Figure 15. The annual variation of the long-term average annual precipitation anomalies from 1955 up to 2022. The curve is represented by the following equation: $y=-0.00152660x^3+0.14270187x^2-3.40629514x+23.37330517,\text{ r }=0.1647$.

In order to mitigate the impacts of random, short-term fluctuations on the weather over a specified time frame, the moving average was calculated and illustrated for the annual mean temperature average (Figure 16) and annual mean temperature anomalies average (Figure 17). This moving average offer an analysis of average temperature data points by creating a series of averages of different subsets of the full data set, herein, every five years. We had selected the temperature (both average and anomalies) because of its significance in built environment and both have relatively high correlation coefficient (r = 0.73922495, R² = 0.54645353 and r = 0.73922495, R² = 0.54645353, respectively) as this weather parameter is significant in the building design, either to minimize heat input (in high solar radiation countries), or maximize heat input (in low radiation countries) (Build, 2021; El Bakkush, Bondinuba, Bondinuba, & Harris, 2015). Figures 16 and 17 ascertained that temperature increase is expected to be substantial and a lot of mitigation should be made to offset such expected increase.

Figure 16. The annual variation of the recorded average long-term temperature, along with using moving average per 5 years subset, throughout the years from 1955 up to 2022.

Figure 17. The annual variation of the recorded mean long-term temperature anomalies, along with using moving average per 5 years subset, throughout the years from 1955 up to 2022.

It is worth mentioning that the occurrence of construction disasters in Poland, during a period of 12 years, were due to random extreme weather events (73%) such as strong wind or strong wind with heavy precipitation causing floods and landslides (Zięba et al., 2020). Therefore, studying correlation between time scale (years) and different weather parameters is significant and could save lives, economy, safety and security.

Finally, the outcome of this study can be tested by doing similar investigation on other weather data set in the Gulf Cooperation Countries (GCCC) and worldwide to explore whether still the Exponential (cubic) regression will have the highest correlation coefficient or other may other regression from the other seven regression coefficients tested in this study. This is important because most of these data set of weather parameter is used for estimating the potential of renewable energy (Alnaser, 1993; Alnaser, Awadalla, & Almoataz, 1992; Alnaser & Aldudiafa, 1990; Alnaser & Awadalla, 1990; Danny, Li, & Lam, 2012) as well as in built environment & sustainable cities (Ceccato, Ramirez, Manyangadze, Gwakisa, & Thomson, 2018; Hui & Tsang, 2005) and sustainability & Sustainable Development Goals (SDGs) (Griggs et al., 2021).

As mentioned earlier, the purpose of this paper is not to monitor the weather parameters and then make projection; if this were the case, then one can use artificial intelligence techniques (Duhoon & Bhardwaj, 2021; Hung, Babel, Weesakul, & Tripathi, 2009; Litta, Idicula, & Mohanty, 2013; Sfetsos & Coonick, 2000). However, the purpose of the paper is to explore whether there is a specific regression coefficient that is more accurate than the other eight regressions coefficient (the exponential cubic regression) to represent all the long-term recorded weather data parameters in Bahrain, or if there may be different correlation coefficients are suitable for each different long-term recorded weather data parameter. The novel finding, herein, is that the exponential cubic correlation is found to be more accurate to represent all the long-term meteorological recorded data (65 years) in Bahrain. In future research paper, we will attempt to use Artificial intelligent technique for projecting the weather parameters using Sequential Minimal Optimization (SMO), Radial Basis Function (RBF) and Multilayer perceptron (MLP).

5. Conclusion

The exponential (cubic) regression is the most accurate correlation (among 8 correlation coefficients) to predict all the recorded long-term (65 years) weather parameters in Bahrain; there is no multi correlations between different parameters.

Weather parameters can cause hazards for communities and may cause huge disturbances in achieving suitability. It is important to anticipate the variation of weather parameters (extremes) in order to achieve mitigation and establish resilience strategies and technologies to minimize such hazards, i.e. convert to sustainable cities and make buildings more resilient to changes in weather such as energy, water, food and transport. Therefore, this study focused on modelling the weather parameters and exploring which weather parameter has a strong correlation coefficient throughout the years.

The analysis in this study showed that the two weather parameters that have high correlation coefficient in modelling their annual changes (past and future) are temperature and humidity, while the other parameters give relatively lower value of correlation coefficient (r). The weather parameter that gives the lowest r is precipitation and dust storms or sand storms.

The investigation and modeling of weather parameters throughout the years in this study will help to reduce the risk and mitigate the effect of construction disasters if they occur.

This study is a milestone in modelling the variation of meteorological parameters to improve the built environment resilience qualities and probably help in setting mitigation and resilience strategies based on the projection of future weather parameters. However, further research to develop plans and policies for effective and timely responses to climate change effects is also required. The limitation of this study lies on using one country weather data set (Bahrain) and not testing for other GCC countries, including cities like Dharan City in KSA, Doha City, Qatar, Kuwait City, Kuwait or other worldwide countries. This limitation is mainly due to the inability to obtain officially recorded data. This limitation, however, will not demote the significance of the outcome and concept of this paper.

Acknowledgements

The authors thank Arabian Gulf University (AGU) and University of Bahrain (UoB) for their support to publish this paper. Thanks, are also due to Mr Nader Ahmed Abdulla, Chief of Climate & Seismology Meteorological Directorate, Ministry of Transport and Communication, Kingdom of Bahrain, for providing accurate long-term meteorological data. The authors are also grateful to Dr. Ghada Ahmed, Head of the Department of English Language and Literature at UoB, and Mrs. Luma Al Salah, Language Instructor at AGU for editing this paper.

Disclosure statement

No potential conflict of interest was reported by the authors.