Web-based macroseismic intensity study in Turkey: entries in Ek¸si Sözlük

. Ek¸si Sözlük is one of the most visited websites in Turkey. Registered users of the website share their knowledge about any topic. In this study, we collect the user entries on the topics of 20 earthquakes in Turkey and the surrounding area. Entries with city and district level information are converted to intensity values. Shake maps of the earthquakes are created by using a ground motion to intensity conversion equation. User entries and created shake maps are compared. It is found that entries correlate with the predicted intensities. It is also found that local soil conditions and building types have an ampliﬁer 5 effect on entries in the web site. Several entries in the earthquake topics have magnitude estimations. Difference between predicted and observed intensities also vary with distance. Users are able to predict the magnitudes of the earthquakes with ± 0 . 54 misﬁt. This study shows that Ek¸si Sözlük has a potential to be a reliable source of macroseismic intensity for the earthquakes in Turkey, if the felt reports are collected with a predetermined format.

The answers are then converted to macroseismic intensity scales, and felt maps are created as an end product of the earthquake.
Users of the website create topics for earthquakes that they have felt and also for major earthquakes occurring around the world.
In this study, we have collected entries from various major earthquakes in western Turkey.Entries which provide city and district information are analyzed.Macroseismic intensity maps are created for the earthquakes by using the magnitude information.Analyzed entries are compared with predicted intensities.Furthermore we collect the magnitude guesses of the users and compare them with the real results.
To analyze the entries in the earthquake topics, 20 titles with the highest number of entries for earthquakes that have occurred in western Turkey and the surrounding area are selected (Table 1).Western Turkey is chosen due to higher access to the internet (Turkish Statistical Institute, 2019).Entries that provide location information in district levels are filtered.These entries are labelled according to the modified Mercalli-Cancani-Sieberg (MCS) scale (Sieberg, 1930).Several entries also provide a guess of the magnitude of the earthquake.
Topics are created almost immediately after the arrival of S waves.Several topics are created for some earthquakes that are widely felt in many regions.Moderators of the website combine the topics into the proper topic title which is in general the date of the earthquake and the nearest major city or the name of the sea to the epicenter.Topics are created with the same format, which is provided below: 10-aralik-2019-balikesir-depremi-6277350 day-month-year-location-earthquake-topicID When an aftershock is felt, the entries indicate the existence of an aftershock.In these cases, we subdivided the earthquake to multiple earthquakes by adding a number to the end of the topicID.Several MCS labels, along with the entry, can be seen below: Topic: 20-aralik-2018-yalova-depremi-5881852 Entry: "I felt it in Gebze (Kocaeli) but I was the only one.
Lights are not swinging, none of my family members have felt it."

MCS:3
Topic: 26-eylul-2019-istanbul-depremi-6191375 Entry: "We saw the rattling windows not only in our office building but also on the neighboring building.I tried to walk forward but instead I staggered backwards . . ."

MCS:6
Hereafter, earthquakes are represented by their topicID.In Table 2, we present information about the entries.5392358 is excluded from further analysis due to lack of entries.

Results
All topics are analyzed with the methods explained in Section 2 and Section 3. 4 of them are represented in this section.Maps of all the earthquakes can be found at Github repository of the study.We compare the computed intensities with the entries; and we also compare them with the EMSC felt report maps.Furthermore, magnitude guesses are compared with the measured magnitudes.İzmir and Yunusemre district (≈85 km) of Manisa (Fig. 1).In Karaburun district, user overbalanced during the earthquake with many others inside the government office.Cracks on walls and columns of the buildings are reported in two entries.

12th
Intensity map has strong correlation with the labelled maximum MCS scale.Minimum MCSs are hard to interpret since some of the entries contain only words such as "strongly felt", which is labelled as MCS = 2.Such entries lowered the average MCSs.
Hence we preferred to rely on maximum MCS labels.
There are 635 felt reports at AFAD for the earthquake.It is hard to interpret the data due to lack of location information of the earthquake.There are intensity values of 11 in several reports which is highly unlikely for magnitude 6.2 earthquake.
EMSC have 755 felt reports for the earthquake (Fig. 3 : 1).Intensities in EMSC are larger than the ones in Ekşi Sözlük.There are intensity 5 reports in far away cities such as Istanbul ::::::: İstanbul and Sofia which is unlikely.Furthermore, there are various intensity 10+ in EMSC in Izmir :::: İzmir.However, there is no report that supports such destruction in Izmir :::: İzmir.Various masonry buildings have cracks on their walls but none of them have collapsed (9 Eylul :::: Eylül, 2017).

25th ::
25 August 2019 Ankara Earthquake 25th :: 25 : August 2018 Ankara earthquake, M l = 3.5, (topicID: 6155192) had occurred south of Kecioren district of Ankara and was felt locally in the city (Fig. 3).There are 129 entries with city and district information.Depth of the earthquake is measured as 5 km.Even though there are large number of magnitude 3.5 earthquakes in Turkey, this was felt by many inhabitants since hypocenter was located beneath the city of Ankara and the earthquake had a shallow hypocentral depth.Due to the shallow depth, almost all districts provided relatively higher maximum MCS values.
There is no felt report in AFAD for the earthquake.There are 205 felt reports in EMSC (Fig. 5 : 3).Intensity measures of Ekşi Sözlük and EMSC are highly correlated for the earthquake.In Besiktas ::::::: Beşiktaş : district, maximum MCS of 5 is given in 3 entries.All of them are due to the evacuation of the buildings.
Two of these entries are from high-rise office buildings, which probably caused extra panic due to the swing of the tall buildings.
The evacuation also influenced by panic is also one of the reasons for the maximum MCS of 5 that is given to the entries from Fatih, Beyoglu and Kadikoy :::::: Beyoglu :::: and ::: Kadı ::: köy districts.In Kartal district, MCS = 5 is due to the fallen objects from shelves.
There are 70 felt reports in AFAD for the earthquake with maximum intensity of 6, which is also the case in Ekşi Sözlük.
In EMSC (Fig. 7 : 4) there are 2027 felt reports.In Istanbul :::::: İstanbul, intensities are reported slightly higher with respect to Ekşi Sözlük.However, it is important to keep in mind that felt report in EMSC is designed for this purpose, whereas in Ekşi Sözlük, entries are written in free format.In general, EMSC and Ekşi Sözlük are correlated in Buyukcekmece and Avcilar İstanbul : and the surrounding cities (Fig. 5).There are 68 entries with city and district information.Most of the entries are coming from Kartal district of Istanbul :::::: İstanbul : and Gebze district of Kocaeli.In Gebze, maximum MCS is 5, which is due to the evacuation of a building with panic.The rest of the entries in Gebze are claiming 2 and 3. MCS scale is highly subjective on an individual's feelings on earthquakes.Thus, users in Ekşi Sözlük are writing their feelings without using any guidelines.
As in this example, unexpectedly high MCSs may occur.
There are more datapoints in EMSC than in Ekşi Sözlük, which provides more information in different regions of the area, especially in the city of Yalova.There are several unexpectedly high intensity values for Istanbul :::::: İstanbul : in EMSC.

Magnitude Guesses
Users in Ekşi Sözlük also provide magnitude information depending on their feelings.We include entries without location information.In various entries, the magnitude is guessed with a semi-infinite range, such as ". . . it was at least 4.8" in 20th :: 20 February 2019 Dardanelles earthquake (topicID: 5947040), or it is guessed within a full range, such as ". . . it is between 2-3 . . ." in 10th :: 10 December 2019 Balikesir ::: Balı :::: kesir earthquake (topicID: 6277350).In semi-definite guesses, given edge value is considered.The average of the range is used when it is provided.
Measured magnitudes along with the user guesses can be seen in Fig. 6.The average misfit between guessed magnitudes and measured magnitudes is 0.54.We also examine the earthquakes with at least two guesses.Misfit is calculated, again, 0.54.
However, it is important to keep in mind that, guesses with range are averaged and various guesses are semi-definite.

Discussion and Conclusion
In this study, we gather entries from earthquake topics in Ekşi Sözlük.In the topics, users discuss the earthquake and their experiences.We filter the entries with city and district level information that can be converted to MCS scale.When there is an aftershock, it is discussed in the same topic on the website.In such incidences, we divide the topic into sub topics.
In total 27 earthquakes are chosen for the analysis.Intensity maps are created for the earthquakes and correlation between the predicted intensities and the entries which were converted to intensity values are roughly interpreted.The values are mapped to the districts of Turkey due to lack of precise location information.Interpretation is done over maximum MCSs.Uncertainty of the data points in terms of location varies depending on the epicenter of the earthquake and the positioning of the governmental district.
To have an insight of the relation between the predicted MCS and observed MCS, a rough relation between the two parameters and the distance is analysed.Observation points are labelled with their districts.However, the district is most likely to have more than one intensity inside its border depending on the epicentral distance from the earthquake and the border of the district (Gebze district in Fig. 5 has 3 different MCS values inside its border).To overcome this problem, we have calculated the centroid point of each district and treated the district as a point.
We have calculated the residuals of MSC differences between the predicted and observed values (Fig 7).To do that, we binned the distance between the epicenter and the centroid points with 10 km intervals from 0 km-100 km.We combined more distant points with +100 km label.Weighted average of average MSCs are used for the residual calculations.Asymmetric errors are calculated by using the minimum and maximum MSC value for each bin and the predicted MSC.If all data points have the same MSCs, then errors are not calculated.A line is fitted to residuals when more than one data is associated in different bins.
Furthermore, population distribution of districts are mostly heterogeneous.An example can be seen in Section 4.3.Silivri district of Istanbul ::::::: İstanbul has all of its EMSC data from the coastline.However, its centroid position is located in northern part of the district, for which MSC value is one integer lower with respect to the highly populated coastal area.
City of Istanbul ::::::: İstanbul and Kocaeli have many reports with MSC = 2.However, all the centroid points of the districts is in MSC = 1 area.
Entries that are written from high-rise buildings were affected more due to these buildings' tendency to amplify the motions of longer periods (eg.Besiktas ::::::: Beşiktaş district in the Istanbul :::::: İstanbul : earthquake).Another reason for higher intensities with respect to the predicted ones, is the evacuation of buildings, even if it is not necessary.There are unexpectedly high intensity values in various districts both in Istanbul :::::: İstanbul : and Yalova earthquakes.
In 20th :: 20 : August 2019 Ankara earthquake (6148327), there is a large gap between the predicted and observed MSCs.It is due to the fact that people of city of Ankara are not used to feel an earthquake with respect to seismically active cities such as Izmir, Istanbul ::::: İzmir, ::::::: İstanbul and Yalova.Moreover, the epicenter of the earthquake is close to the city center.
Effect of epicentral depth in small earthquakes can be seen in Ankara earthquake (Section 4.2).Districts of Sincan, Yeni Mahalle, Mamak and Cankaya ::::::: Çankaya : have at least one degree higher maximum MCS than the predicted ones.
In conclusion, entries in Ekşi Sözlük can provide intensity distribution of earthquakes with limits.Entries are written in free form which creates uncertainties in the MSC labeling process.Entries do not reveal the exact position of the data provider which makes it hard to analyze the differences between observed and predicted MSC values.Despite the limitations, gathered data have similarities with the predictions.The website can provide near real-time intensity information after an earthquake.
Intensity maps and user feedback about the earthquake can be used for the creation of real time intensity maps and can be published for each topic.Felt reports of 26th September 2019 Istanbul earthquake submitted to EMSC on top of the labelled entries.

Figure 7 .
Figure 7. Residuals of predicted and observed MSCs.Red dashed line represents the baseline.Black circles are the residual of weighted average of the bin.Vertical black lines are the residuals from predicted MSCs and minimum and maximum MSCs that are observed in the bin.Black dotted line is the fitted line to the residual of weighted averages.Number of data points inside bins are provided beneath each bin point with data :::::: (created :::: using :::::::: Matplotlib :::::: library :: of :::::: Python).

Table 1 .
Topics of various earthquakes in Ekşi Sözlük.First entry, additionally, gives the information of the creation time of the topic.∆t is the time between the origin time of the earthquake and the creation of topic of the earthquake.Entries do not have the 'second' information for the time.Magnitudes are moment magnitude unless otherwise stated.

Table 2 .
Table of extracted information from entries of each topicID.Number of entries with city and district information along with the MCS value is represented by No of Entry.Unique number of cities and districts of these cities are given by No of City and No of District, respectively.Maximum MCS, minimum MCS and average MCS of these entries are represented by Max MCS, Min MCS and Av MCS, respectively.Entries with magnitude guesses regardless of location information are given in No of Mag.topicID No of Entry No of City No of District Max MCS Min MCS Av MCS No of Mag