1 Introduction

Satellite infrared remote sensing observation has many advantages including its large field of view, high-precision, short cycle, and not restricted by ground conditions, so it becomes the promising means to study the infrared seismic anomalies. Several studies have been carried out to analyze thermal infrared (TIR) anomalies appearing in the area of earthquake occurrence a few days before the seismic shock. These studies analytically compare images of pre (vs. post) earthquake satellite TIR imagery (Tronin et al. 2002; Shen 2013; Xie et al. 2013); study outgoing longwave radiation (OLR) data and identify anomalous variations prior to a number of medium to large earthquakes (Defu et al. 1997; Ouzounov et al. 2007; Xiong et al. 2010a, b; Jing et al. 2013; Xiong et al. 2013); perform a multispectral thermal infrared component analysis during earthquakes using land surface temperature (LST) (Ouzounov and Freund 2004; Tronin et al. 2004); or assess the anomalous surface latent heat flux (SLHF) peaks a few days prior to the earthquake event(Cervone et al. 2004, 2005; Qin 2013).

At the same time, the study of multi-angle infrared sensing mechanisms, modeling, and multi-angle remote sensing technology has made great advances, multi-angle infrared remote sensing technology and its application has promoted to develop mutually, which has become a frontier research field, but the applied research of earthquake science using multi-angle infrared remote sensing is still blank. Therefore, the development of seismic infrared anomalies identification and extraction using multi-angle remote sensing technology, and exploring the applicability of the study of multi-angle seismic infrared radiation anomalies will be a new technology and method to study earthquake infrared remote sensing.

Actually, the main problems associated with seismic anomalies identification are lack of rigorous definition for TIR anomalies in the current study. A possible solution to these problems is to employ a new approach and an advanced data analysis technique to detect earthquake anomalies embedded in TIR remote sensing satellite data. Recently, a data analysis technique named the robust satellite data analysis technique (RST) (Tramutoli 2007) has been proposed for natural and environmental hazards monitoring and mitigation. This technique has already been validated in the case of many earthquakes (Tramutoli et al. 2001; Filizzola et al. 2004; Corrado et al. 2005; Tramutoli 2005; Aliano 2007).

Based on the characteristics of the Envisat bi-angular advanced along-track scanning radiometer (AATSR) infrared data, this paper has introduced and developed the method of the RST to detect seismic anomalies, and formed a set of data processing procedure. On this basis, this paper has carried out the case study of the Yushu earthquake occurred on 14th April 2010, combined with the tectonic explanation of the spatial and temporal continuity of the abnormal phenomena, the analyzed results have indicated a number of anomalies associated with the possible seismic characteristics of Yushu earthquake, which also has explored the applicability of multi-angle infrared remote sensing in the study of seismic infrared radiation anomalies.

2 The bi-angular AATSR data

The AATSR onboard ESA’s Envisat spacecraft has three visible/near-infrared channels at 0.55, 0.67, and 0.87 micron, designed specifically for remote sensing applications over land and a visible channel at 1.6 micron which is used primarily for cloud clearing, three channels at the thermal infrared wavelengths of 3.7, 11, and 12 µm, from which surface temperatures are derived over both sea and land surfaces (Llewellyn-Jones et al. 2001). The nominal spatial resolution of AATSR is 1 km × 1 km in the nadir view and 1.5 km × 2 km in the forward view, with a swath width of about 500 km. Equator crossing time is 10:00 a.m. local time (descending node) and revisit time is about 3 days (Coll et al. 2012).

The AATSR has two onboard calibration mechanisms, low-noise detectors, and mechanical coolers that provide high radiometric accuracy and stability to the TIR data (better than 0.05 K for the 11 and 12 μm bands). A special feature of the AATSR is its conical scanning mechanism to give a dual view of the Earth’s surface, first in the forward view at an angle of around 55° and 150 s later in the nadir view at an angle close to vertical (from 0° to 21.7°). From two views with different atmospheric path lengths, it is possible to obtain independent information about the atmospheric contributions to the signal, and perform an accurate atmospheric correction and thus enable more precise data (ESA 2007).

The AATSR data used in this paper are usually provided in so-called Gridded Brightness Temperature/Reflectance (GBTR) product, which is the single Level 1B product from AATSR. It contains cloud-free top-of-atmosphere (TOA) brightness temperature (BT) values for the three infrared channels and reflectance values for the 1.6 µm and visible channels, and all pixels are co-located and resampled to a 1 km latitude-longitude grid(ESA 2007). In this paper, we propose to utilize the bi-angular 12-μm brightness temperature data to detect seismic anomalies using the Robust Satellite Technique.

3 Yushu earthquake

Yushu earthquake occurred on 14th April 2010, with its magnitude of 7.1, the location of the epicenter is at 33.2 north latitude, 96.6 east longitude, which is located in the Ganzi-Yushu fault zone (Fig. 1). The fault was a large strike slip fault zone of the interior of the Qinghai Tibet plateau. It was also an important boundary fault of the Sichuan-Yunnan rhombic block body (Rongjun et al. 1997). The fracture was formed in the early West HOLLEY period, and moved strongly during the Indo-Chinese epoch. Subsequently, thrust activities were the main activities of this fracture, and the fault has been a long-term active fault since the Quaternary period. The fracture was extending northwest for part of the Xianshuihe fault zone and the northern border of the Sichuan Yunnan rhombic block of Qinghai Tibet Plateau with eastward extrusion. The structure of this fault was complex (Wen et al. 1985; Wang et al. 2008; Ren et al. 2010).

Fig. 1
figure 1

The investigated area of Yushu earthquake. The epicenter of Yushu earthquake occurred on 14th April 2010 is marked with a yellow cride, the main faults are marked with pink line

Full size image

4 Methodology

There are several methods to choose for the analysis of TIR data, mainly including the Eddy field calculation mean (Defu et al. 1997; Ouzounov et al. 2007), the wavelet time–frequency analysis (Cervone et al. 2005; Xiong et al. 2009a, b, 2010a, b; Xiong 2011), and the Robust Satellite data analysis Technique (Tramutoli 1998). The RST method is based on a preliminary multi-temporal analysis on several years of satellite TIR records, which is devoted to characterize the TIR signal for each pixel of the satellite image processed. This previous study concluded that IR anomalies could be identified within TIR data related to an earthquake (Tramutoli et al. 2001, 2005, 2013; Tramutoli 2005, 2007, Temimi et al. 2011). Each step of the RST method for detecting seismic anomalies using bi-angular AATSR data will be explained and detailed in the following.

4.1 Caculating the differential variable ΔT

Calculating differential variable ΔΤ is expected to reduce possible contributions due to day-to-day and/or year-to-year climatological changes and/or season time-drifts, the differential variable ΔΤ is the value of the difference between the punctual value of AATSR BT and its spatial average computed on the investigated area by taking consideration of only cloud-free locations, all belonging to the same land or sea.

The first step in calculating the differential variable ΔΤ is to build a cloud mask: hereinafter, we use the nadir-forward tests (Závody et al. 2000). For each pixel, the difference Τ11 – Τ12 is determined, and used to predict the view difference using this relationship. The actual view difference is compared with the prediction, and if they differ by more than a pre-defined threshold, the pixel is assumed to be cloudy (ESA 2007); after performing the cloud detection, one can build an image R CM(x,y) for each pixel x,y of the scene R CM = 1 or R CM = 0 if the location has been recognized as clear or cloudy, respectively.

Given latitude and longitude of the image pixel, the surface type for each image pixel is derived from land/sea mask information obtained by an external AATSR auxiliary product; after performing the land/sea identification, one can build an image R LS(x,y) having for each pixel x,y of the scene with R LS = 1 or R LS = 0 if the location has been recognized as sea or land.

Computing the average T(t) for the sea using only pixels over the scene which have R CM = 1 and R LS = 1 and for the land using only pixels over the scene which have R CM = 1 and R LS = 0.

The last step is to calculate differential variable ΔΤ using the formula (1)

$$ \Delta \,T = T\left( {r,t} \right) - T\left( t \right), $$
(1)

where, r ≡ (x, y) represents location coordinates of the pixel center on a satellite image; t is the time of image acquisition.

4.2 Calculating the reference field

In order to calculate the reference field for this study, we use a binning algorithm to preprocess the selected data, in which binning refers to as the process of distributing the contributions of pixels to be processed in satellite coordinates to a fixed grid using a geographic reference system. In most cases, a sinusoidal projection is used to generate a grid comprising a fixed number of equal area bins with global coverage (Campbell et al. 1995).

We use cloud-free records from the selected data-set (t ∈ τ) to compute the reference fields for mean \( \mu_{{\Delta \,T}} \left( r \right) \) and standard deviation \( \sigma_{{\Delta \,T}} \left( r \right) \) at the location r ≡ (x, y) using the following equations:

$$ \mu_{{\Delta \,T}} \left( r \right) = \frac{1}{N}\sum\limits_{t \in \tau }^{N} {\Delta \,T(r,t)} $$
(2)
$$ \sigma_{{\Delta \,T}} \left( r \right) = \sqrt {\frac{1}{N}\sum\limits_{t \in \tau }^{N} {\left( {\Delta \,T(r,t) - \mu_{{\Delta \,T}} (r)} \right)^{2} } }. $$
(3)

A data-set used here is the bi-angular AATSR gridded BT, a coverage of longitude from 91 east longitude to 101 east longitude and latitude from 28 north latitude to 38 north latitude, The duration is from 2003 to 2010 during the months of March, April, and May has been selected and processed for computing \( \mu_{{\Delta \,T}} \left( r \right) \) and \( \sigma_{{\Delta \,T}} \left( r \right) \) reference fields shown in Fig. 2.

Fig. 2
figure 2

The reference fields for the investigated area of forward view and nadir view computed from bi-angular AATSR data from 2003 to 2010 during the months of March (left panel) and April (middle panel) and May (right panel)

Full size image

4.3 Anomalous detection step

Anomalous TIR of bi-angular AATSR data are detected by an index as used in (Tramutoli 2007), robust estimator of TIR anomalies (RETIRA), which can be defined as follows:

$$ \otimes_{{\Delta \,T}} (r,t) = \frac{{\Delta \,T(r,t) - \mu_{{\Delta \,T}} (r)}}{{\sigma_{{\Delta \,T}} (r)}}. $$
(4)

The difference ΔT(r, t) − μ ΔT (r) then represents the signal (S) to be investigated for its possible relation with seismic activities. It is always evaluated by comparison with the corresponding natural/observational Noise (N), represented by \( \sigma_{{\Delta \,T}} \left( r \right) \) which describes the overall (local) variability of S including all (natural and observational, known, and unknown) sources of its variability, as historically observed at the same site in similar observational conditions (sensor, time of day, month, etc.). This way the relative importance of the measured TIR signal (or the intensity of anomalous TIR transients) can naturally be evaluated in terms of S/N ratio by the RETIRA index. Moreover, the larger is \( \sigma_{{\Delta \,T}} \left( r \right) \), the lower will be \( \otimes_{{\Delta \,T}} \left( {r,\,t} \right) \), so that the RETIRA index results intrinsically protects against false alarm proliferation (Tramutoli 2005).

4.4 The bi-angle brightness temperature synthetic approach

This paper proposes the multi-angular BT synthetic approach. In this approach, the RETIRA (Robust Estimator of TIR Anomalies) index \( \otimes_{{\Delta \,T}} \left( {r,\,t} \right) \) of bi-angular AATSR BT data are processed by normalization. The Normalized Robust Estimator of TIR Anomalies index NRTIRnadir(r, t) can be defined as follows:

$$ {\rm NRTIR}_{\text{nadir}} (r,t) = \frac{{ \otimes_{{\Delta \,T,{\text{nadir}}}} (r,t) - { \hbox{min} }\left( { \otimes_{{\Delta \,T,{\text{nadir}}}} (r,t)} \right)}}{{{ \hbox{max} }\left( { \otimes_{{\Delta \,T,{\text{nadir}}}} (r,t)} \right) - { \hbox{min} }\left( { \otimes_{{\Delta \,T,{\text{nadir}}}} (r,t)} \right)}}. $$
(5)

Secondly, the paper carries out the synthetic analysis of the Normalized Robust Estimator of TIR Anomalies index NRTIRnadir(r, t) and NRTIRforward(r, t) using the following equation:

$$ \oplus (r,t) = 1 - \left( {1 - {\rm NRTIR}_{\text{nadir}} (r,t)} \right) \times \left( {1 - {\rm NRTIR}_{\text{forward}} (r,t)} \right). $$
(6)

Finally, the index ⊕ (r, t) is the multi-angular BT synthetic index within AATSR BT data, which represented the multi-angular infrared radiation characteristics.

5 Results and discussion

Using the above method and the bi-angular AATSR top-of-atmosphere BT data from 11 March, 2010 to 25 May, 2010, we get the curves of results of the RETIRA index computation on the investigated area before and after the Yushu earthquake using AATSR nadir view and forward view data. Pixels with \( \otimes_{{\Delta \,T}} \left( {r,t} \right) > 4.0 \) are depicted in mars red, pixels with \( 3.5 < \otimes_{{\Delta \,T}} \left( {r,t} \right) < 4 \) are depicted in seville orange, pixels with \( 3.0 < \otimes_{{\Delta \,T}} \left( {r,t} \right) < 3.5 \) are depicted in solar yellow, pixels with \( 2.5 < \otimes_{{\Delta \,T}} \left( {r,t} \right) < 3.0 \) are depicted in autunite yellow, pixels with \( 2.0 < \otimes_{{\Delta \,T}} \left( {r,t} \right) < 2.5 \) are depicted in yucca yellow. Epicenter is marked with red beryl green cride, the investigated area is with longitude from 91°E to 101°E and latitude from 28°N to 38°N.

5.1 Nadir view

From the results of the robust Estimator of infrared radiation Anomalies index computed on the investigated area before and after the Yushu earthquake using the AATSR BT data (Nadir view) from 11 March, 2010 to 25 May, 2010 (Fig. 3), the feature of the anomalies could be summarized as follows:

Fig. 3
figure 3

Results of the RETIRA index computed on the investigated area before and after the Yushu earthquake using AATSR BT data (Nadir view)

Full size image

(1) About 33 days before the earthquake, on March 11th, the infrared radiation anomalies appear and enhance gradually in the course of earthquake development and occurrence along the Ganzi-Yushu fault, Nu River fault, and Jiali-Chayu fault.

(2) From March 16th to 26th, the abnormal strength rises to a maximum value, then decrease, and the infrared radiation anomalies along the Nu River fault disappear, infrared radiation anomalies on the other two faults decrease but continually develop.

On April 15th, one day after the earthquake, with the crustal stress release, the infrared radiation anomalies intensity weaken sharply along the Ganzi-Yushu fault and Jiali-Chayu fault; until the end of April, infrared radiation information becomes very weak on the two faults. On May 5th, the infrared radiation anomalies disappear along Jiali-Chayu fault. Then infrared radiation anomalies on Ganzi-Yushu fault present discrete, diffuse distribution, and disappears on May 25th at last.

In order to examine the reliability of the analyzed results for the Yushu earthquake above, this paper also carries out further analysis of infrared anomalies along faults of Yushu earthquake, this paper reduces the investigated area and selects the Ganzi-Yushu fault zone as the research area (Fig. 4).

Fig. 4
figure 4

The investigated area of the Ganzi-Yushu seismic fault

Full size image

From the results (Fig. 5) of the RETIRA index computed on the investigated area before and after the Ganzi-Yushu seismic fault using the AATSR BT data (Nadir view) from 11 March, 2010 to 25 May, 2010, we could conclude the follwing features: from March 11 to 31, the infrared anomalies appear in the north and south along Ganzi-Yushu fault, then gradually expand the scope, and there is an increasing trend before the earthquake; from April 5 to 15, near the earthquake time, the infrared radiation anomalies of the fault belt reach maximum, which is mainly distributed in the north and southwest side along the fault. About 7–16 days after the earthquake, many obvious anomalies still appear near the earthquake fault zone.

Fig. 5
figure 5

Results of the RETIRA index computed on the investigated area before and after the Ganzi-Yushu seismic fault using AATSR BT data (Nadir view)

Full size image

5.2 Forward view

From Fig. 6, we could conclude that the characteristics of infrared anomalies from forward view and Nadir view are similar, but not the same. The similar feature is that obvious earthquake infrared radiation anomalies appear along the Ganzi-Yushu fault. However, from the results of the RETIRA index using AATSR Forward view BT data, the feature of anomalies in the area is large but the strength of anomalies decreased, the direction of the anomalies was not obvious, the process of anomalies preparation appear diffuse type, and there were not tight relationship between structure and the anomalies.

Fig. 6
figure 6

Results of the RETIRA index computed on the investigated area before and after the Yushu earthquake using AATSR BT data (Forward view)

Full size image

The results of the RETIRA index computed on the investigated area between nadir view (Fig. 5) and forward view (Fig. 6) data are greatly different. The possible reasons could be concluded that the local changes in development of seismic activity fault are small, but infrared anisotropy information of seismic activity is very obvious. The AATSR onboard ESA’s Envisat spacecraft has a viewing geometry where each terrestrial scene is viewed at two angles, at nadir and at a forward angle of 55°. From the dual views, we get different patterns of ground, especially in complicated terrain area. And these dual views can be used to detect seismic anomalies from bi-directional observation based on infrared anisotropy of seismic activities.

5.3 Bi-angular synthetic analysis

This paper carries out the bi-angular synthetic analysis on the investigated area before and after the Yushu earthquake using the AATSR BT data (bi-angular view) from 11 March, 2010 to 25 May, 2010. From the Results (Fig. 7), the feature of the anomalies is included below:

Fig. 7
figure 7

Results of the RETIRA index computed on the investigated area before and after the Yushu earthquake using AATSR BT data (bi-angle synthetic analysis)

Full size image

(1) The infrared radiation anomalies appear and enhance gradually in the course of earthquake development and occurrence along the Ganzi-Yushu fault, Nu River fault and Jiali-Chayu fault, Lancang fault, and NingGashi-deqing fault from March 11th, which detect two more fault anomalies (Lancang fault and NingGashi-deqing fault) compared with the anomalies detected using the nair view data.

(2) From March 16th to March 26th, the strength of the anomalies rises to the maximum value, then decrease. Infrared radiation anomalies along the Nu River fault reach maximum on March 21st and then disappear, infrared radiation anomalies on other several faults decrease but continually develop, and the anomalies phenomenon are obvious.

(3) On April 15th, one day after the earthquake, with the crustal strain release, infrared radiation intensity of the whole investigated area weakens sharply, and anomalies disappear on the Nanjing Temple Deqin fault. Then until the end of April, the infrared radiation feature of the earthquake on the investigated area shows discrete, diffuse distribution, no trend here.

(4) The earthquake infrared radiation on May is still present large area coverage, and the direction of sesmic anomalies is not obvious, the earthquake infrared radiation anomalies disappear on May 25th.

5.4 Confutation analysis

In order to examine the reliability of the analyzed results for the Yushu earthquake above, we use the same procedure to perform a confutation analysis on a relatively seismically unperturbed period: the same period (from 11 March to 25 May) but in a different year (2009) on the same investigated area (Fig. 8).

Fig. 8
figure 8

Results of the RETIRA index computed on the investigated area during 2009 using AATSR BT nadir data

Full size image

The anomalies from Fig. 8 are not obvious and in discontinuity. Only isolated anomalous pixels are detected. the quick comparison between images in Fig. 3 and confutation phases (Fig. 8) allows us to realize that a substantial difference exists in the occurrence of TIR anomalies. In fact, it is evident, from the results shown in Fig. 8, that only isolated (05 April, 10 April, 15 April, 20 April, 25 April and 30 April) and not time persistent anomalous pixels are detected. From Fig. 8, it also shows a closer view of these sporadic anomalies.

6 Conclusions

The paper has developed the methods of robust satellite data analysis techniques to detect seismic information, and formed a set of data processing procedure. On this basis, this paper has carried out the case study of Yushu (Qinghai, China) earthquake in 2010 using Envisat-AATSR bi-angular infrared data, which also has explored the applicability of multi-angle infrared remote sensing in the study of seismic infrared radiation anomalies. The main conclusions of this paper could be drawn as follows:

(1) The paper has developed the RST method to extract seismic information, which takes into account the influences of different radiation characteristics and climate, seasonal changes between cloud, land, and water, and it could eliminate the impacts on identification and extraction of seismic anomalies because of ground temperature increase effect caused by seasons changing and abrupt climate variations; and based on the multi-angular background reference field for many years, the paper has used the RETIRA index to detect the seismic infrared radiation anomalies, and the detected infrared radiation anomalies can be characterized by the signal-to-noise ratio, the greater earthquake-related noise information, and the smaller the RETIRA index, and vice versa. Therefore, the RETIRA index could avoid the infrared anomalies caused by non-shock, and has excellent robustness in the detected seismic infrared anomalies.

(2) The seismic anomalies characteristic detected by the AATSR nadir BT data are better than the forward BT data, which show up in that there are tighter relationships between structure and the anomalies, the spatial and temporal development of seismic infrared radiation characteristics is more obvious, and is more close to the fault activity from the results of the RETIRA index using the nadir view data.

(3) In the case study of Yushu earthquake in 2010, the paper proposes a synthesis analysis method of the AATSR bi-angular infrared data, which efficiently analyze and synthesize the seismic infrared radiation anomalies from two different views. The result of the synthesis of seismic infrared radiation anomalies using the AATSR nadir view and forward view data has provided more information and detected more infrared radiation anomalies along the earthquake fault zone (Lancangjiang fracture and Ning Karma Monastery-Deqin fracture), which can characterize the activity of seismic faults more precise.

Therefore our studies provide a finding that anomalies discovered within Envisat-AATSR bi-angular infrared data. This finding will be further validated using more earthquake data in the future.