Page 33 - NGTU_paper_withoutVideo

Modern Geomatics Technologies and Applications

3.6 GWR regression model evaluation

Under conditions of non-stationary in OLS modeling, GWR method was applied to potentially improve the results by
2
examining local spatial diversities. The value of adjusted R slightly increased from 0.55 (obtained by OLS) to 0.62, indicating
that the GWR model could explain 62% of the entire variations of COVID-19 incidence rates in Iran. Furthermore, the value of
corrected AIC decreased from 44.273 to 40.352, which demonstrates that GWR was the most parsimonious model. In this
study, as expected, differences between the results obtained by OLS and GWR regression methods, showed that the GWR had
the ability of better explaining the spatial context of COVID-19 incidence rates across Iran.
One of the restrictions of this research was lack of data availability. In order to develop a model, which predicts the COVID-19
incidence rates accurately, using other influential explanatory variables is of the essence. However, due to lack of accessibility
to other behavioural and social variables, we did not insert them into the current model. Another limitation was the difference
between lockdown rules enforced by the local government, also the way people behave, which was completely different among
provinces. For instance, Gilan and Mazandaran were the first provinces announced that travelling into them banned. However,
Qom, Tehran, and Alborz, three of the most infected provinces of the country, did not have any strict policies in terms of
lockdown. These types of distinctions among provinces could have adverse impacts on COVID-19 incidence rates.
To sum up, incidence rate of COVID-19 has a strong relationship with the age, especially those who are older than 60.
Moreover, provinces with higher rates of urban population could be more infectious.
Well-detailed results obtained by this study, could become helpful for health officials and local leaders in terms of adjusting
rules and policies, resulting in better management of COVID-19 incidence, especially in high-risk areas.

4. Conclusion
Regarding data collected over 8 months from February 18 to October 21, 2020, Qom could be considered as the center of
the COVID-19 disease, having the highest incidence rate among other provinces in Iran.
Spatial distribution of the COVID-19 disease represented a clustered pattern in the study area. Other most infected
provinces were located not too far away from Qom, including Tehran, Alborz, Semnan, Markazi, and Zanjan.
Hot spot analysis also demonstrated that provinces located in north and western parts of Iran, are the most at-risk provinces
with the confidence level of 99%. However, southern east provinces of Iran have had the minimum rates of COVID-19, such
as Sistan and Baluchistan.
Two of the most uncorrelated explanatory variables, namely the percentage of elderly population and the percentage of urban
population had significant relationships with the COVID-19 incidence rates. Noteworthy, mentioned explanatory variables
were used in both OLS and GWR model. After applying both local and global approaches, the GWR method provided an
improvement in explaining the spatial distribution of COVID-19 incidence rate across Iran.
In conclusion, spatial analysis and modeling of COVID-19 incidence rates could be considered as beneficial tools in order
to help principals take necessary actions.

5. Acknowledgments
We are very grateful to Iran’s Minister of Health and Statistical Center of Iran for sharing their valuable data.

8