On December 18, 2023, a M_S6.2 earthquake occurred in Jishishan county, Gansu Province, causing significant casualties and serious economic losses. This earthquake occurred in the northeastern margin of the Qinghai-Xizang Plateau, which is at the junction of the Loess Plateau and the Qinghai-Xizang Plateau. A series of studies have been conducted on this earthquake, primarily focusing on damage assessment or focal mechanism solution. However, from a broader perspective of earthquake precursor research, it has been documented that thermal radiation anomalies have been detected prior to many moderate or strong earthquakes. And, the detection of pre-earthquake anomalies for the 2023 Jishishan earthquake is still scarce. Whether there were thermal radiation anomalies prior to Jishishan earthquake is a scientific question worthy of exploring.

Currently, there are mainly two types of satellite remote sensing data sources utilized to detect pre-earthquake thermal radiation anomalies, one is the thermal infrared remote sensing data, the other is the passive microwave remote sensing data. However, due to the limited atmosphere penetration capability of thermal infrared band, data-continuity of satellite thermal infrared remote sensing is strongly affected by the variant meteorological conditions. In contrast, passive microwave remote sensing data offer unique advantages in terms of data continuity for the detection of pre-earthquake thermal radiation anomalies, thanks to its longer wavelengths and stronger atmospheric penetration capability. To answer the abovementioned scientific question, the microwave brightness temperature （MBT） data covering （34°N−42°N, 99°E−110°E） acquired by the Advanced Microwave Scanning Radiometer 2 （AMSR2） sensor is collected. This dataset is analyzed using a novel wavelet-based two-step difference （WTSD） method to detect the thermal radiation anomalies preceding the Jishishan earthquake.

In the WTSD method, the radiation received by microwave sensors is considered to comprise three different components, i.e., the low-frequency background field, high-frequency meteorological signals, and the seismic MBT anomalies. The low-frequency background field is considered as a stable contribution to the microwave radiation, which is mainly caused by topography, surface coverage and seasonal variations with strong regularity. The highfrequency meteorological signals are mainly caused by frequent variation of the meteorological conditions （e.g., precipitation and temperature change, etc）, which are considered as a random contribution. The seismic MBT anomalies can be extracted by sequentially subtracting the low-frequency background field and the high-frequency meteorological signals from the original MBT data. Specifically, the hierarchical clustering is first conducted to classify the ground covers into different types. Then, MBT background fields for different types of ground covers are established and removed separately by adopting wavelet decomposition on the multi-year MBT data （from 2003 to 2022）. After that, a second wavelet decomposition is performed to eliminate the high-frequency meteorological signals, thereby extracting the MBT anomalies associated with the Jishishan earthquake. The temporal, spatial, and intensity evolution characteristics of the pre-earthquake MBT anomalies for the Jishishan earthquake are summarized as follows:

1） Temporally, the MBT anomalies lasted for approximately 133 days （from August 7 to December 18, 2023） and experienced three different periods, i.e., the initial active period, the subsequent calm period and the final pre-earthquake rebound period.

2） Spatially, a “//”-shaped distribution was found preceding the earthquake, consisting of MBT increase stripes interspersed with a MBT decrease region. During the active period, two MBT increase stripes appeared, and then gradually extended in the northeastern margin of the Qinghai-Xizang Plateau, forming a “//”-shaped anomalies distribution pattern. However, there was a MBT decrease region between the two MBT increase stripes.

3） In terms of intensity, the maximum intensity within both MBT increase stripes exceeded 10 K, and the daily average intensity within them reached 3 K. It was worth noting that on 19 November （one month preceding the earthquake）, both MBT increase stripes showed significant MBT anomalies decrease of approximately 7 K. Meanwhile, the MBT in the decrease region had been consistently at a low level, with a maximum intensity of less than 2 K. On October 10 （approximately two months prior to the earthquake）, the MBT anomalies in the decreasing region also exhibited a significant decrease of around 12 K. Generally, the evolution trend of the MBT decrease region was similar to that of the two MBT increase areas, but in the MBT decrease region, the amplitude of the anomalies was smaller, the calm period was earlier and its decrease amplitude was larger.

Interestingly, the spatial distribution of the MBT anomalies preceding Jishishan earthquake is similar to the infrared anomalies observed during our rock-loading experiment. This indicates a large-scale development of microfractures within the crust preceding the earthquake. According to the remote sensing rock mechanics experiments and the meta-instability model, it can be concluded that the pre-earthquake MBT anomalies are mainly caused by variation of the crustal stress. The calm period shortly preceding the earthquake may imply that the crustal stress has entered the meta-instability stage, which can be regarded as a precursor to the Jishishan earthquake.