Spatial-temporal distribution of atmospheric temperature anomalies connected with seismic activity in Tien-Shan

In this paper we analyzed spatial-temporal temperature changes in the upper troposphere/lower stratosphere (UTLS) above the Northern and Central Tien-Shan detected by satellite remote sensing which have been compared against seismic activity in (1992-2015). These anomalous changes in temperature time series were used as preseismic indicators. Anomalous variation parameters were calculated accounting for amplitude and phase of short-time temperature variations at UTLS isobaric levels separated by the tropopause. The results show that the spatial structure and dynamics of temperature anomalies in the area of UTLS have a sufficiently stable relation to seismic activity. We estimated the spatial and time variability of anomalous temperature perturbations on the basis of 12 strongest earthquakes with magnitudes Mï5.0. The temperature anomalies were observed in all considered cases from ~3 to 72 hours before the main seismic event.


Introduction
Study of strong earthquakes impacting the atmospheric parameters have a long history with active period during the last decades due to substantial progress in the development and improvement of satellite technologies, as well as accessibility of a great number of dedicated services and databases (Tronin, 2010;Pulinets et al., 2014;Prakash and Srivastava, 2015;Yadav and Pathak, 2018;Jiao et al., 2018;Ouzounov et al., 2019).
Temperature measurements are of great importance for studying the atmospheric effects of earthquakes. It is the key parameter, which defines dynamic processes and structural changes of atmosphere. The atmospheric thermal stratification has a strongly pronounced layered (by the rate of temperature change or gradient ) nature. The tropopause inversion layer (TIL) separating the convectively mixed troposphere (0) from the more stable and stratified stratosphere (usually >0) is characterized by great dynamic variability and sensitivity to various perturbations and atmospheric wave activity (Pilch et al., 2017). Circulation processes of planetary and synoptic scale, primarily related to cyclone and anticyclone passing (Randel et al., 2007), change in the solar activity and stratospheric ozone content (Morozova et al., 2017), radiation processes and mass exchange between the troposphere and the stratosphere (Birner, 2006;Manney et al., 2017) play an important role in UTLS temperature variations modulations. In addition, retrospective analysis of satellite during the catastrophic earthquake in Fukushima, Japan, in 2011 showed that the strongest negative correlation between temperature changes at isobaric levels separated by the tropopause coincided with a period of high seismic activity (Kashkin et al., 2012;Kashkin, 2013). However, complex interaction of different-scale processes in UTLS makes it difficult to study the atmospheric response to seismic activity and requires new methods for the processing of experimental data. RST (Robust Satellite Techniques) method was widely applied to detect and localize deviations of parameters from their typical behavior (Tramutoli et al., 2001;Pergola et al., 2010;Zhang and Meng, 2019;Tramutoli et al., 2019). Basic principles of RST method combined with spectral and correlation analysis provide the basis of our algorithm, which, in contrast to conventional methods, is supplemented by a special module for the detection of short-time anomalies in temperature time series. The retrospective analysis of satellite data using the algorithm shows correlation between seismic activity and anomalous temperature variations in UTLS preceding strong seismic events of M>6.0 in the territory of European countries  and in various Asian regions Kashkin et al., 2020). This can indicate a probable relation between strong earthquakes and the observed temperature variations in UTLS.
In this paper we present a modified version of the previously developed  algorithm, which allows the detection of short-time anomalous variations in the spatial-temporal distribution of temperature. To evaluate our algorithm, we applied it to the analysis of atmospheric effects of earthquakes with magnitude of M≥5.0, registered in the territory of Kyrgyzstan and near its boundaries.

Initial seismic and satellite data
To study temperature variability in pre-seismic, coseismic and post-seismic periods we chose 12 earthquakes with magnitudes from 5.0 to 7.4, which took place in Kyrgyzstan and nearby territories in 1992-2015. The main characteristics of seismic events in the Northern and Central Tien-Shan presented in the Table 1. We used data from Institute Seismology of Kyrgyzstan and KNET seismology network of Research Station of the Russian Academy of Sciences in Bishkek. Magnitude estimation for the earthquakes is performed according to the following formula : M= [lg(E)4.8] /1.5 (E  seismic wave energy in joules). To obtain data for the seismic events outside of KNET network we used online version of the USGS global catalog (https://earthquake.usgs.gov/ earthquakes/search/). Locations of the earthquake epicenters are presented in Fig. 1. To assess the sensitivity of the developed algorithm we also analyzed seismic events with magnitudes 4.0<M<5.0. The epicenters of these 58 events are also presented in the map.
To study pre-seismic atmospheric perturbations we used temperature data from the MERRA-2 reanalysis system [https://disc.gsfc.nasa.gov/datasets]. Comparison of reanalysis satellite data of MERRA-2 with remote sensing data demonstrated good consistency and quality in spatial structure and UTLS dynamics reconstruction (Manney et al., 2017). This interactive service provides open access to M2I3NPASM Version V5.12.4 data files in the net CDF format. The data represent arrays of  Table) and 4.0< M < 5.0 synthesized temperature values for each atmospheric level (ASM data type). As an input data for the algorithms, we used atmospheric temperature values at standard isobaric levels from 450 to 70 hPa. The area of interest was 37 -46° N and 65 -85° E with the size of the grid 0.5° × 0.625°. Resolution of temperature time series T(t) was t = 3h, providing a sufficiently good detailing of the anomaly formation process.

Description of satellite data processing algorithm
Establishing the relation between dynamics of processes in the atmosphere and the lithosphere was based on the assumption that variations of parameters caused by seismic activity differ significantly from background fluctuations that occur during the periods without strong earthquakes. This implies the necessity to identify preseismic indicators of abnormal behavior in time series of temperature data. For this purpose, we used integral indexes, which were calculated with regard to variations in temperature amplitude and phase at UTLS levels separated by the tropopause . The new algorithm make it possible to determine not only temporal, but also spatial distribution of short-time anomalies in temperature; it includes the following stages:

Preprocessing satellite data
At the first stage, we performed a preliminary processing of satellite data fragments. We prepared time series of T(t) with resolution of t = 3h, containing temperature values for atmospheric levels (p k ) from 450 to 70 hPa for each seismic event. Length of the analyzed series was 90 days (45 days before and after the earthquake date).

Spectral analysis of atmospheric temperature time series
Temperature variations observed in UTLS have components with varying periodicity and amplitude. We used continuous wavelet transform to identify the stationary and non-stationary components in our data. The linear trend and low frequency periodic (seasonal) constituents were excluded from the initial time series.

Filtration of short-time temperature variations
To retrieve short-time variations we applied nonlinear filtering based on discrete wavelet transform, which has several advantages comparing to moving average or high-order polynomials (Donoho and Johnstone, 1994;Gadre et al., 2014). We consider anomalous changein quasiperiodic components with the period of 4-6 days as the main features characterizing atmospheric temperature prior to strong earthquakes (Sanchez-Dulcet et al., 2015;Sverdlik et al., 2019).

Calculation of atmospheric temperature anomalies
We converted short-time temperature variations at each isobaric level to dimensionless quantity. For this purpose, we retrieve dynamics of temperature anomalies (ΘТ) as deviation of the current temperature value from the average monthly level, normalized to standard deviation : At this isobaric leve used for pa