Earth’s Outgoing Longwave Radiation Variability Prior to M ≥6.0 Earthquakes in the Taiwan Area During 2009–2019

This paper proposes an analysis method, using the National Oceanic and Atmospheric Administration satellite data, to trace variations in outgoing longwave radiation (OLR) for finding the precursors of earthquakes. The significance of these observations is investigated using data sets of recent M ≥6.0 earthquakes around the Taiwan area from 2009 to 2019. We suggest that the precursory signal could be an EIndex anomaly (EA) in the form of substantial thermal releases distributed near the epicenter. The consecutive appearances of OLR EAs are observed as precursors 2–15 days before significant earthquakes, and we refer to this as a pre-earthquake OLR EIndex anomaly (POEA). We interpret these thermal sources as possibly originating from electromagnetics together with gas emissions associated with pre-seismic processes. This study highlights the potential of OLR anomalous changes in earthquake precursor studies, at least in the Taiwan region.


INTRODUCTION
Outgoing long-wave radiation (OLR) is energy emitted from Earth in the form of electromagnetic radiation, which passes out of the atmosphere and into space in the form of thermal radiation. The flux of energy transport by OLR can be measured by a satellite. OLR is primarily sensitive to nearsurface and atmospheric temperatures, the humidity of the air, and the presence of clouds, which are related to the intensity of convective activity as well as the latitude and altitude dependence of the variability (Kane and Buriti, 1997;Ouzounov et al., 2007;Graham et al., 2011).
Taiwan is located on the Pacific seismic belt, where earthquake activity is intense. Many large earthquakes have occurred in Taiwan and caused seismic disasters in the past hundred years as a result of the continuous collision between the Philippine Sea Plate and the Eurasian Plate at a rate of 82 mm/yr to the northwest (Yu et al., 1997). For example, the M L 7.3 Chi-Chi earthquake in 1999 killed more than 2400 people and had a substantial impact on Taiwan.
Previous studies have applied remote-sensing techniques to survey thermal anomalies in Taiwan Liu et al., 2007;Pulinets and Ouzounov, 2011;Genzano et al., 2015;Chan et al., , 2020. It is worth mentioning that a significant sequence of TIR anomalies was observed 2 weeks before the Chi-Chi earthquake and close to its epicenter. Previous studies have also proposed that the observed TIR anomalies were related to earthquakes with M >4 that occurred around Taiwan from 1995 to 2002 .
These studies provide the opportunity to explore OLR/TIR variations related to energy radiating from underground toward the atmosphere as a result of rock deformation and geodynamic processes in the Taiwan area. However, most of the abovereferenced studies have been limited by a lack of sufficient earthquake events with OLR results (M ≥ 6) to further investigate the potential of detecting significant earthquakes in Taiwan. Nevertheless, these potential precursory phenomena for Taiwan may provide an opportunity to better understand the physical and chemical processes before earthquakes, which would be crucial for exploring the theoretical mechanism of lithosphere-atmosphere-ionosphere coupling (LAIC) (Pulinets and Ouzounov, 2011;Kuo et al., 2011Kuo et al., , 2014. This paper aims to describe the characteristics of the OLR monitoring system, process the data, and systematically assess the feasibility of using the OLR method to find precursors prior to the M ≥6 earthquakes that occurred over 2009-2019 around Taiwan (Figure 1). We also interpret the observed precursors of OLR and a possible pre-earthquake generation process by comparing the results with multiple observations.

MATERIALS AND METHODS
The OLR data used in this study are daily measurements from the Advanced Very-High-Resolution Radiometer (AVHRR) polar orbiters NOAA-18, obtained from the NOAA Climate Prediction Center website. 1 To compute the daily zonal OLR anomaly over the Taiwan area, we use a daily mean with a 1 • × 1 • spatial resolution in both longitude and latitude, as shown in Figure 2.
Taiwan is located in the Pacific Ocean and has a subtropical climate, and it is frequently impacted by typhoons. Increasing sea surface temperatures can effectively increase the surface latent heat flux during the typhoon season, which means that the heavy precipitation and strong convective flow of a typhoon can induce a region of extreme negative OLR anomalies (Liebmann and Smith, 1996;Lee et al., 2006;Susskind et al., 2012;Hatchett, 2018). The latitude and altitude effects on OLR variations around the study area can be ignored, owing to the narrow latitude range and smooth topography, except in the Central Range of Taiwan. To acquire non-cloud satellite data at the time of a typhoon is difficult because of bad weather condition. In other words, the effect of the continuous presence of clouds during typhoons is strong, but the sporadic presence of clouds may be slight for data analysis. Therefore, we directly removed the OLR raw data during typhoons to diminish the contributions by typhoons to OLR variations.
It is important to note that OLR raw data were acquired twice for each location and at the same time of day (daytime and nighttime) under the similar background conditions to reduce the variability of signals due to the daily variation of temperature and humidity and the vegetation and cloud coverage. The local grid value of the OLR is subtracted from the spatial variation of the 1-year mean value as the daily eddy. The current day value (the current daily eddy) is deducted from the average value of 16 days preceding the observation day as the OLR anomaly. A standardized anomaly, E Index , can be determined, computed from the OLR anomaly divided by a standard deviation τ i,j . In this study, when E Index is observed to have consecutive abnormal changes in some specific area in a short period, it may then be considered as a possible precursor for identifying seismic events. Seismic data are acquired from the earthquake catalogs of the Central Weather Bureau and United States Geological Survey. There are 35 seismic events with magnitudes 1 http://www.cdc.noaa.gov/ FIGURE 1 | A regional map and the principal tectonic structures around Taiwan showing the Philippine Sea Plate moving northwest at ∼82 mm/yr toward the Eurasian Plate (Yu et al., 1997). The circles represent M ≥6 earthquake epicenters from 2009 to 2019 with the colors corresponding to different focal depths. These earthquakes are labeled and listed in Table 1. The red lines indicate the active faults of Taiwan. greater than 6.0 during 2009-2019 around the Taiwan area ( Figure 1) and the variation in E Index on the nearest grid-point adjacent to the epicenter of the earthquake was selected for further investigation.
The data processing method for extracting E Index from daytime and nighttime OLR data connected with the earthquakes is summarized in Figure 2. The procedure of data analysis uses the following equations: where S i,j is the OLR value of the target grid-point, as defined in Figure 2A, W i,j is the number of the surrounding grid points around the target point, and S i,j is the yearly average. By subtracting the weighted yearly average from the four surrounding grid points for removing the regional mean, S * X i,j , t can be obtained as the daily eddy (shown in Figure 2B) where S * x i,j , t is the current daily eddy andS * x i,j , t is the 16day average daily eddy fields excluding the current day. OLR (t) A indicates a residual value of the eddy as the OLR anomaly ( Figure 2C) that can be determined as the difference between the current eddy and the average eddy of the 16 days preceding the observation day: where E Index is defined by the OLR anomaly divided by the standard deviation of τ i,j . E Index can then represent the maximum change in OLR values for specific spatial locations Frontiers in Earth Science | www.frontiersin.org and predefined times as an indicator of abnormality (shown in Figure 2D). This methodology follows a similar analysis process to that described in detail by Ouzounov et al. (2007) and Pulinets and Ouzounov (2011). Figure 3 shows the daytime and nighttime variations of E Index for each significant earthquake. The threshold values were determined to be when the E Index value exceeded 2, which is called the E Index anomaly (EA) to identify each seismic event's precursor (Figure 1 and Table 1). The earthquake preparation is considered to be a continuous process, as shown in Figure 2D, and we propose that the consecutive appearances of EAs around the epicentral area are a precursor to an earthquake. Based on this scenario, the EQ-1 event in Figure 3 shows an example of the temporal variations of precursory EAs. Here, red and blue dots are the daily variations in the daytime and nighttime EAs, respectively. In this study, we propose that at least three appearances of EAs found within 4 days over the epicentral area before significant seismic events (M ≥ 6) be the criteria for earthquake precursory anomalies, and we refer to this as a preearthquake OLR E Index anomaly (POEA). The EAs are observed on July 5, July 7, and July 8, 2009, in the daytime and nighttime results, respectively, which means there are a total of six EAs within 4 days (6, 7, and 9 days in the daytime and 6, 7, and 9 days in the nighttime before the 2009/07/14 M6.0 earthquake, respectively). Similarly, precursors in the OLR usually indicate significant and anomalous variability associated with significant earthquakes lasting more than 1 day and repeating at the same local time (Jing et al., 2013;Tramutoli et al., 2013;Genzano et al., 2015;Ouzounov et al., 2015Ouzounov et al., , 2018). Figure 3 presents the temporal evolution of the propagation characteristics of the EA to the earthquake, which may be related to the POEA. Comparing the rates of anomaly occurrences in the daytime and nighttime and 35 earthquakes with M ≥6 during 2009-2019 shows that the POEA quite often occurred ∼2-15 days before the earthquakes, except for 11 of the 35 selected seismic events (EQ-2, EQ-3, EQ-6, EQ-8, EQ-9, EQ-19, EQ-21, EQ-25, EQ-28, EQ-31, and EQ-33 in Figure 3 and Table 1). In other words, the progressive POEA accompanying the relevant earthquake can be termed as the precursory event (PE). It is worth noting that, for all of the cases, significant EAs are much weaker and fewer after the earthquake occurrence (Figure 3).

The Influence of Typhoons
The disturbance to OLR E Index variations seems to be caused by the occurrence of a typhoon before but not after the earthquake. There is no frequent appearance of EA before the EQ-2, EQ-3, EQ-9, and EQ-19 events during the occurrence of a typhoon before the earthquake. Similarly, Liu et al. (2007) pointed out that high-temperature anomalies from a satellite observation could not be identified because of the influence of the typhoons. In contrast, a continuous anomalous E Index is observed as the POEA before the EQ-1 event; in this case, a typhoon occurs later than the earthquake. This is because the OLR raw data for the typhoon period were already removed before the data analysis so that the anomaly cannot be recognized without the complete results. The anomalous E Index signals might be present in the OLR raw data; however, the influence of the typhoon cannot entirely be eliminated by using the data processing method of this study. This example illustrates the potential of this methodology for finding a POEA during the occurrence of a typhoon over the earthquake period, as shown in Figure 3. When excluding four events during the occurrence of a typhoon before an earthquake, 24 out of 31 such earthquakes show the POEA (an ∼77% success rate).

Influence of the Focal Depth of the Earthquake
An earthquake's focal depth appears to be a significant parameter that affects E Index variations. Earthquakes with a hypocenter depth greater than 70 km, such as EQ-6, EQ-8, EQ-19, EQ-21, EQ-25, and EQ-28, have no precursors of the abnormal E Index anomaly, except for the EQ-11 event. It is worth noting that the POEA can be observed for EQ-11, but not for EQ-21, EQ-25, and EQ-28 in similar tectonic settings (Figure 1), which might be attributed to the different focal mechanism. However, no further discussion can be made as only one case of this has been observed. Positive hole charge carriers are proposed as a physical mechanism for the buildup of surface charges over stressed rock to transport energy from the Earth's crust to the surface. This hypothesis is used to explain the pre-earthquake thermal anomalies, by generating electric fields strong enough to ionize the air (Ouzounov and Freund, 2004;Freund et al., 2007) and drive current flow upward in the atmosphere and into the ionosphere (Freund, 2007a,b;Kuo et al., 2011). Alternatively, gases (e.g., radon, carbon dioxide, and methane) emitted along the fault zones over the epicenter area is another hypothesized mechanism for the ionization of the near-surface air for increased air temperature and latent heat flux (Tronin, 2002;Pulinets et al., 2006;Ouzounov et al., 2007;Tramutoli et al., 2009Tramutoli et al., , 2013Pulinets and Ouzounov, 2011). Changes in gas emissions may be attributed to near-or far-field earthquakes and strain-induced changes in permeability within the earthquake's preparation zone (Walia et al., , 2013Fu et al., 2015Fu et al., , 2017bFu et al., , 2019. Earthquakes occurring at a focal depth of greater than 70 km are classified as intermediate-focus earthquakes, suggesting that they are located lower than the crust-mantle boundary of Taiwan (Ustaszewski et al., 2012). The strength of the rock in the upper mantle is ductile and decreases with depth as the temperature increases, and has no substantial brittle strengths to cause microfractures and activate p-hole charge carriers. Furthermore, above 700 • C, the activated charge carriers may be consumed by a reaction with CH 4 from olivine crystal through oxidative coupling in the deep region (Chen and Molnar, 1983;Hadi et al., 2012). That might be why the POEA is rarely observed before earthquakes with large focal depths. Similarly, FIGURE 3 | Time-series of daily OLR E Index values associated with M ≥6 earthquakes over the epicenter area from 2009 to 2019 around Taiwan. The red and blue lines indicate the daily E Index variations in the daytime and nighttime, respectively, within 25 days before and after the earthquake. The earthquake's anomalous days are considered to be when the increasing E Index value is greater than the threshold value (E Index ≥ 2), as given by the red and blue dots. The occurrence of an earthquake is marked with the green dashed line. The shaded region indicates typhoon period. The relevant earthquakes are marked by the numbers shown in Figure 1 and listed in Table 1 Figures 1, 3, 5. b An E Index value higher than 2 can be called the E Index anomaly (EA), and only EAs observed before the EQ are listed. C "P" denotes "Precursor." d The number of the relevant earthquake denotes a typhoon occurring during 25 days before and after the EQ. (2017) pointed out that TIR anomalies are not easily observed during the earthquakes with depth larger than 60 km and Ho et al. (2018) reported that the precursory ion density and velocity anomalies associated with surface charges are small before and after earthquakes with a depth greater than 70 km. These observations correspond to the results of previous studies, indicating that the depth of earthquakes can play a significant role in earthquake-related effects (Rodger et al., 1999;Fu et al., 2017b). Significant changes in soil radon concentrations along the fault zones over the epicenter were observed ∼14 days before EQ-7, followed by the appearance of the POEA from satellite data and the crust with surface movement based on GPS residual displacement nearly 10 days before the event (Chen et al., 2013c;Fu and Lee, 2018). Similarly, the reoccurrence of a significant increase in soil radon concentrations was found ∼2 weeks before the 2020/02/06 M6.6 Meinong earthquake (EQ-26) in southern Taiwan, and a substantial POEA occurred following the radon anomalies 9-12 days before the EQ-26 event (Fu et al., 2017a). Furthermore, Chen et al. (2015a) reported that the high-conductivity materials were enhanced ∼12-13 days before the M6.2 2013/03/27 Nantou earthquake (EQ-14) and

Relationship Between OLR E Index Variations and Seismicity
were followed by significant changes in the POEA ∼4-10 days before the same event. In contrast to the anomalies observed in the lithosphere and atmosphere, the substantial precursory changes in the ionosphere often appear 1 day and 1-5 days before M ≥6.0 and M ≥5.0 earthquakes, respectively (Liu et al., 2000. Those observations would support a constant and positive relationship with the pre-earthquake generation process and be associated with the model for the LAIC (Pulinets and Ouzounov, 2011;Kuo et al., 2014). Most precursory phenomena in previous studies of earthquake precursors in Taiwan have been fragmentary, especially for atmospheric OLR investigations using satellite techniques. The mechanism of the earthquake generation process and the reasons for the OLR anomalies are not well-understood. According to the mechanisms and previous results described above, we propose a possible mechanism based on the previous observations together with the results of this study (e.g., multiprecursors before EQ-7, EQ-14, and EQ-26). Gases spread out along the fault zones together with the positive hole charge carriers, generated by the amount of gases degassed from compressed rock and the magnetic enhancements of highconductivity materials around the earthquake preparation zone emanating to the surface, which is caused by changes in rock permeability and conductivities due to pre-earthquake stress accumulation. The ionization of the near-surface air and latent heat exchange is related to gas emissions and the upward flow of current with the positive hole charge carriers, respectively, leading to an increase in air temperature and intensive outgoing longwave radiation. The time sequence of anomalous changes between the lithospheric and atmospheric observations explain FIGURE 4 | Temporal profile of daily OLR E Index anomalies (EA) associated with each precursory event within 25 days before and after the earthquake. The black and gray bars denote the EAs in the daytime and the nighttime, respectively. the appearances of the precursory OLR anomalies in the epicentral areas before earthquakes with M ≥6 around the Taiwan area. Then, the influence on the total electron content (TEC) leads to substantial ionospheric perturbations, followed by the contributions of the lithosphere-atmosphere before the impending earthquake. Hence, the significant relationship between different types of observations between the lithosphere, atmosphere, and ionosphere can be validated in Taiwan. Similar models have also been proposed by Freund (2007a;2007b), Ouzounov (2011), andTramutoli et al. (2013). Multi-parameter monitoring data could reveal multiple pre-earthquake signals that could provide a possible explanation of the origin of thermal fluctuations before earthquakes and support the LAIC's theoretical mechanism.
When the EA occurrence is present within a specific day in the daytime and the nighttime before and after an earthquake, we count one for a certain day to the relevant PE. Figure 4 shows all PEs to display the proportion of the number of total FIGURE 6 | Time-series of daily OLR E Index anomalies (EA) associated with the M ≥6 earthquakes from 2009 to 2019 around Taiwan for a confutation test over the same area. The red and blue squares indicate the daily EAs in the daytime and the nighttime, respectively, within 90 days before and 25 days after the earthquake. When at least three EAs are found within 4 days, this is referred to as a POEA (marked by the yellow highlighting). The occurrence of the strong earthquakes (M ≥ 6) and the 2013/03/07 M5.9 earthquake are marked with the green and gray dashed lines, respectively. The occurrence of the earthquakes with the magnitude between 5-6 and 4-5 are denoted by light-blue line and light-green line, respectively. The unperturbed period without the strong earthquakes during 90-25 days before the event is marked by the gray dashed rectangle.
EAs during the observation period. It can be seen that the occurrence of EAs mainly appears from 5 to 10 days before the earthquake, regardless of being in the daytime or the nighttime. This provides a useful precursory index for OLR monitoring of the impending significant seismic activity, at least in the Taiwan area. The dependent relationship between the number of total precursory EAs and the magnitude of each PE is shown in Figure 5. A weak correlation (R 2 = 0.05) is obtained between these variables, suggesting that the number of precursory EAs does not reflect the magnitude of the earthquake. The amounts of EAs may not be used to estimate the magnitude of an earthquake for neither inland nor offshore earthquakes.
A confutation test was performed to verify the appearance or absence of the POEA before or after significant earthquakes over the same area (Di Bello et al., 2004;Genzano et al., 2009;Tramutoli et al., 2013;Xiong et al., 2013;Lisi et al., 2015;Khalili et al., 2019). As shown in Figure 6, most of the POEAs were only observed once within 90 days before and 25 days after the related earthquake, except for eight of the 35 selected seismic events (EQ-5, EQ-7, EQ-14, EQ-16, EQ-17, EQ-18, EQ-24, and EQ-30). However, only a few sporadic EAs appeared only in 1 day, which did not approach the anomalous criteria of this study. It is important to note that a POEA observed during the 2013/02/23-2/27 event may relate the 2013/03/07 M5.9 earthquake (marked by a black arrow) as a true positive occurring in the same grid of the EQ-14. The POEA occurring 27-31 days before EQ-3 may be considered as a false positive as opposed to most POEAs that appear 2-15 days prior to the significant earthquakes. Therefore, excluding a POEA related to the M5.9 event in the time series of the EQ-14, only a few false positives (∼23%) appear in the absence of the strong earthquake (M ≥ 6), by considering the unperturbed period within 90-25 days before the event as shown by the dashed rectangle in Figure 6.
Based on the observations from the present study, the consecutive POEAs generally appear several days over the epicenter before a strong earthquake. This can be utilized to trace the appearance of EAs around Taiwan using the spatial-temporal monitoring of OLR observations. Currently, an automaticprocess monitoring system has been established in Taiwan to routinely monitor variations in OLR. As mentioned above, substantial POEAs can be observed, followed by an increase in radon concentrations, before significant seismic activities in Taiwan. The appearance of the significant POEAs could be further compared with soil radon observations along fault zones around Taiwan (Walia et al., 2013;Fu et al., 2015Fu et al., , 2017aFu et al., , 2019, which can be validated and used to evaluate the potential mechanism of the coupling processes between the lithosphere and atmosphere before significant earthquakes.

CONCLUSION
In the present study, we applied a satellite technique to outgoing longwave radiation (OLR) variations in order to search for the precursors of M ≥6.0 earthquakes over 2009-2019. The major findings of this study are listed below: (1) The results showed that most precursors, as shown by the pre-earthquake OLR E Index anomaly, mainly appeared 5-10 days before significant earthquakes. (2) There is a weak correlation between the number of E Index anomalies and the earthquake's magnitude, indicating that the increasing intensity of OLR anomalies is not associated with increased magnitude. (3) The effects of the focal depth and typhoon events are proposed as critical factors that affect OLR variations for earthquake precursor studies, as when an earthquake has a hypocenter depth greater than 70 km and a typhoon occurs before the earthquake, consecutive E Index anomalies are rarely observed. (4) A high percentage of earthquakes (∼77%) with preearthquake OLR anomalies are consistently observed when the possible factor of a typhoon's occurrence is removed. (5) We proposed a potential mechanism to interpret the preearthquake generation process, suggesting the existence of multiple responses triggered by coupling processes between the lithosphere, atmosphere, and ionosphere through OLR anomalies by combining different types of observations. (6) Therefore, our studies provide a useful indicator for exploring earthquake precursors, especially in the Taiwan region.

DATA AVAILABILITY STATEMENT
The datasets generated for this study are available on request from the corresponding author and from the NOAA Climate Prediction Center website (http://www.cdc.noaa.gov/).

AUTHOR CONTRIBUTIONS
C-CF organized and wrote the manuscript. L-CL provided the concepts of the manuscript. DO and J-CJ assisted in improving the data processing. All authors contributed to the article and approved the submitted version.

ACKNOWLEDGMENTS
We thank Mr. Peng-Kang Wang and Kuo-Hang Chen for help with data collection and preliminary analysis. We would like to thank Uni-edit (www.uni-edit.net) for editing and proofreading this manuscript. The TEC contribution number for this article is 00163. We are also thankful to the editor and reviewers and the referee for their valuable comments and suggestions, which helped us to improve the manuscript.