1 Introduction

Remotely sensed thermal infrared (TIR) images have been used to detect geothermal activity for over half a century. In 1961, a geothermal survey on Yellowstone National Park of USA initiated the application of TIR remote sensing in geothermal exploration, which the US Army Cold Regions Research and Engineering Laboratory incorporated with the University of Michigan successfully identified hot springs and other near-surface geothermal anomalies with thermal infrared scanning technique (Qin et al. 2011). Land surface temperature (LST) is required in the assessment of remote sensing to detect shallow thermal anomalies for geothermal exploration and field management. LST is also used as an indicator of the thermal information associated with faults or volcanic complexes (Gaudin et al. 2013; Gutiérrez et al. 2012; Wu et al. 2012) and as a key factor to obtain surface heat fluxes for a variety of investigations such as urban heat island (UHI) effect and land–air interactions (Chang et al. 2010; Chang and Liou 2005; Chien et al. 2008). Hence LST with relatively high accuracy is needed in the assessment of potential area for geothermal resource. This information can be retrieved from Landsat 7 ETM+ data provided by US Geological Survey (USGS) archives. A body of literature has shown that it is feasible to conduct the assessment of LST or geothermal heat flux (GHF) from high spatial resolution satellite data (Akhoondzadeh 2013; Davies et al. 2008; Gutiérrez et al. 2012; Kruse 2012; Kuenzer et al. 2007; Mia et al. 2012, 2013; Oguro et al. 2011; Qin et al. 2011; Reath and Ramsey 2013; Shaw et al. 2010; Vaughan et al. 2010, 2012; Wu et al. 2012). However, it appears that there is no previous study of geothermal exploration using satellite-based infrared data in Taiwan. This work aims to apply TIR remote sensing for the geothermal exploration in Ilan plain of Taiwan.

1.1 Geothermal Energy of Taiwan

Considering the geodynamic setting, geothermal energy holds promising for the Taiwan Island. Taiwan is located in the Pacific Rim of Fire and possesses rich geothermal resources because of volcanic activities originating from the plate convergence. Previous reports from Taiwan’s Bureau of Energy suggest that Taiwan has geothermal power potential in excess of 33.6 GWe (Tsanyao 2015), which would be sufficient for the island’s annual electricity demand when fully tapped. In addition, the ongoing debate on the safety of nuclear power, which currently accounts for around 20% of Taiwan’s power, has shifted focus on alternative renewable energies, especially after the 2011 Fukushima nuclear power plant accident in Japan. In this regard, geothermal energy appears as one of the best alternatives to replace nuclear and provide clean electricity for the island.

Ilan plain (around 330 km2; Fig. 1) in the northeastern Taiwan is a geologically active area affiliated with the Okinawa Trough. It is a deltaic plain with the flat topography which is surrounded by the Hsuehshan Range to the northwest and the Backbone Range to the Southeast. The study area is mainly covered by Oligocene to Miocene rocks (shale and sandstone in the main) and quaternary alluvial deposits, which is acquired from Ho’s 1:50,000 Geology Map of Taiwan as shown in Fig. 2 (Ho and Chen 2000). The shallow geothermal potential in Ilan plain was evaluated by thermal gradients (TGs) of 30 water wells (27–178 m in depth) from 2003 to 2005, and the probable heat source of the high geothermal is the volcanic intrusion beneath the plain (Chiang et al. 2007; Tong et al. 2008). The National Science and Technology Program-Energy (NSTPE) was established in December 2007 by the National Science Council (NSC; now renamed as Ministry of Science and Technology, MOST) for integrating resources and formulating an energy technology development strategy. This MOST’s master project of geothermal energy intends to drill a well of 3000 m to build up a 1 MW geothermal pilot power plant, by dint of developing the exploration and evaluation techniques of the deep enhanced geothermal system (EGS) (Song 2016; Tsanyao 2015).

Fig. 1
figure 1

Geographic location of Ilan plain in northeastern Taiwan

Full size image
Fig. 2
figure 2

a The topographic and b geologic map of the study area. Major faults (Kang et al. 2015; Lee 1999; Liu 1995; Tong et al. 2008) were illustrated on the geologic map

Full size image

Obviously, geothermal energy development in Taiwan presents opportunities both for the renewable energy and economic growth in the island. Various studies in Ilan plain have been conducted by existing geophysical methods, such as seismic tomography, gravity gradiometry, electrical resistivity tomography, magnetic prospecting, and well logging. However, Satellite remote sensing data such as Landsat is not yet integrated into this arena in Taiwan. Thus, this research proposes a top-down approach for detecting geothermal anomaly in Taiwan based on the largest free source of Landsat data provided by the USGS.

1.2 Geomagnetic, Gravity, and Magnetotelluric Survey in the Ilan Plain and Chingshui Area

The Chingshui geothermal field in the southwest of Ilan plain is the most productive geothermal area in Taiwan. Various geophysical methods have been performed for the exploration in the region. For instance, a geomagnetic survey in the Ilan plain was conducted in 1978. The acquired dataset of 425 stations has been reprocessed in 2008 for analyzing the possible heat sources and fluid channels of the geothermal field. As a result, the surface projection of the fault and dyke solutions in the Ilan plain is shown in Fig. 3. Three groups of dyke solutions were identified in the northern part of Ilan plain as shown in Fig. 3a, and three groups of fault solutions were observed as shown in Fig. 3b. In addition, a gravity survey in the Chingshui area was performed by the ITRI (Industrial Technology Research Institute of Taiwan) in 1976, and data of 636 stations were collected. This dataset has been reprocessed to analyze fault structures in the Chingshui area. Besides, the magnetotelluric (MT) survey method has been utilized to explore the structure of the geothermal reservoir in the Chingshui area in 2006, in which 33 broadband magnetotelluric data points were acquired by the ITRI (Tong et al. 2008).

Fig. 3
figure 3

The surface projection of the fault and dyke solutions in the Ilan plain. a Three groups of dykes (DA, DB and DC) were noticed. b Three groups of faults (FA, FB and FC) were observed. The solid yellow and green circles indicate the depth of fault and dyke solutions beneath the Ilan plain, respectively. The shaded area between DB and DC could be related to the WE high magnetic anomaly area associated with igneous intrusive rock (Tong et al. 2008)

Full size image

Clearly, the integration of the above-mentioned geophysical methods strives to delineate the insight of the geothermal structure in the Ilan plain. They have drawn the preliminary conclusion: (1) the presence of a magma chamber in the shallow crust and shallow intrusive igneous rock causes the high heat flow and geothermal gradient in the Ilan plain. (2) Geothermal fluid circulated within the fracture zone in the deep was heated by the hot rock in the Chingshui geothermal field, and the geothermal reservoir of the Chingshui geothermal field likely linked with the fracture zone of the Chingshuihsi fault (Tong et al. 2008).

2 Data and Method

This research uses Landsat 7 ETM+ LT1 data products. The Landsat scene covers northern part of Taiwan (WRS2 Path/Row 117/43; acquisition time: UTC 02:09:19) obtained on December 3, 2001 (Fig. 4). Each scene size by default is about 170 km north–south by 183 km east–west (106 mile by 114 mile). The image is in good quality of rank 9 (image quality ranges from 0 to 9 with 9 being the best). Clear weather condition was presented with average wind speed of 2.3 m/s and no precipitation according to the meteorological report of Central Weather Bureau (CWB). Landsat scenes are processed to Standard Terrain Correction (Level 1T—precision and terrain correction) which gives systematic radiometric and geometric accuracy through integrating ground control points and topographic accuracy by the digital elevation model (DEM) (NASA 2015).

Fig. 4
figure 4

Landsat 7 ETM+ thermal infrared image acquired on December 3, 2001

Full size image

2.1 Landsat 7 Products

The Landsat program is a civilian satellite program that was initiated in 1965 by the USGS. First satellite Landsat 1 was launched on July 23, 1972. Landsat 7 (launched in 1999) and Landsat 8 (launched in 2013) are currently active missions. Landsat 7 Enhanced Thematic Mapper Plus (ETM+) imagery contains eight spectral bands. The spatial resolution for Bands 1–5 and 7 is 30 ms, Band 6 is 60 m, and for Band 8 (panchromatic) is 15 m. All bands can acquire one of two gain settings (high or low) depending on radiometric sensitivity and dynamic range, while Band 6 (thermal band) collects both high and low gain for all scenes (NASA 2013). Table 1 provides selected features of the Band 6 (thermal band) of the instrument.

Table 1 Selected features of the thermal bands of Landsat 7 ETM+
Full size table

2.2 Data Processing

The physical basis of LST retrieval is the Blackbody Radiation and the Planck Function. Planck Function is employed for calculating the radiance emitted from a “Black Body”. Inverse of the Planck Function is to derive the “brightness temperature” of an object. Landsat 7 ETM+ sensor acquires radiance information and stores it to the format of digital number (DN) in the range between 0 and 255. So the LST can be retrieved by converting these DN values to degrees Kelvin or Celsius. In this study, main steps of data processing include radiometric calibration, atmospheric correction, topographic correction and emissivity calculation. Specific procedures are illustrated as shown in Fig. 5.

Fig. 5
figure 5

Flow chart of LST retrieval

Full size image

Radiometric calibration is applied as the first step to convert the DN value into the top of atmosphere (TOA) radiance according to formulas from the Landsat 7 Science Data Users Handbook (NASA 1998). Two Atmospheric Correction Modules are adopted to Landsat scene depending on the wavelengths. For wavelengths in the visible through near-infrared and shortwave infrared regions (up to 3 µm), fast line-of-sight atmospheric analysis of spectral hypercubes (FLAASH) is applied. FLAASH is a first-principles atmospheric correction tool incorporating the MODTRAN (MODerate resolution atmospheric TRANsmission) radiation transfer model to compensate for atmospheric effects, which is being established by the Air Force Phillips Laboratory, Hanscom AFB and Spectral Sciences, Inc. (Adler-Golden et al. 1998). For the thermal band, National Aeronautics and Space Administration (NASA) provide a web-based atmospheric correction tool “The Atmospheric Correction Parameter Calculator” for single thermal band sensors on the website (http://atmcorr.gsfc.nasa.gov/). It adopts atmospheric global profiles (modeled by the National Centers for Environmental Prediction; NCEP) for a specific date, time and location as input. It also uses MODTRAN model; the site-specific atmospheric transmission, and upwelling and downwelling radiances can be calculated from the calculator (Barsi et al. 2003).

In areas with rugged terrain, topographic correction can be carried out to remove the topographic effects. Both the optical and thermal bands are influenced by topographical variations in solar illumination. A number of methods for normalizing these effects in images have been developed. Amongst them the most used are empirical approaches which are based on regression and model-based methods which assume Lambertian reflectance (Warner and Chen 2001). In this research, Lambertian cosine correction (C-Correction) based on the 20-m-resolution DEM from the Department of Land Administration, Ministry of the Interior, Taiwan is applied for Topographic Correction. The Lambertian model defines that reflected radiance from the inclined surface (L T) is related to the horizontal surface (L H) by the following equation:

$$L_{\text{T}} = L_{\text{H}} \cdot \cos i/\cos z,$$
(1)

where i is the incidence angle (the angle between the illumination and the normal to the surface) and z is the sun zenith angle (the angle between the vertical and the sun).

Emissivity calculation uses the general method to incorporate emissivity as a function of the atmospherically corrected red and near-infrared bands. It is based on the work of Sobrino et al. (2001) and referred to as NDVI (Normal Differential Vegetation Index) threshold methods. For further elaboration, emissivity of each pixel in the imagery is derived from vegetation fraction (F r) from NDVI. NDVI is defined as a function of the surface derived reflectance in the red band (R red) and the near-infrared band (R nir): i.e.,

$${\text{NDVI}} = \frac{{R_{\text{nir}} - R_{\text{red}} }}{{R_{\text{nir}} + R_{\text{red}} }},$$
(2)

Vegetation fraction (F r) is defined as the ratio of the vertical projection over the vegetation canopy in each pixel: i.e.,

$$F_{\text{r}} = \left( {\frac{{{\text{NDVI}} - {\text{NDVI}}_{\text{s}} }}{{{\text{NDVI}}_{\text{v}} - {\text{NDVI}}_{\text{s}} }}} \right)^{2},$$
(3)

where NDVIs is the NDVI value corresponding to bare soil, and NDVIv is the value corresponding to full vegetation. NDVIv as 0.87 and NDVIs as 0.22 are proximately taken according to the NDVI histogram values of the study area. Considering the various landcover of the study area, emissivity values can be calculated in three cases:

  1. (a)

    NDVI < 0.22, the pixel is mainly bare soil (F r = 0) and the mean value of soil emissivity (ε s) is assumed as 0.97 (Sobrino et al. 2008).

  2. (b)

    NDVI > 0.87, the pixel is considered fully vegetated (F r = 1) with the mean value of vegetation emissivity (ε v) of 0.99 (Sobrino et al. 2008).

  3. (c)

    0.22 ≤ NDVI ≤ 0.87, pixels are consisted of bare soil and vegetation, and the pixel level emissivity is derived from the equation as follows:

    $$\varepsilon_{\text{i}} = F_{\text{r}} \cdot \varepsilon_{\text{v}} + (1 - F_{\text{r}} ) \cdot \varepsilon_{\text{s}} ,$$
    (4)

    where ε v is the vegetation emissivity, ε s is the soil emissivity, and ε i is the pixel’s effective emissivity.

Finally, the LST result is computed from a single-channel algorithm proposed by Artis and Carnahan (1982).

2.3 Validation of Landsat ETM+ Derived LSTs

Parameters retrieved from remotely sensed data can be used with confidence only after the proper accuracy assessment. Two approaches were adopted to assess the accuracy of the Landsat ETM+ derived LST in this study.

2.3.1 Ground-Truth Validation from Meteorological Temperature Data

The interpolated meteorological temperature data from Central Weather Bureau (CWB) of Taiwan at 02:09:00 UTC, 10:09:00 local time are employed to evaluate the reasonability of LST at 02:09:19 UTC, 10:09:19 local time. The 11 stations in the whole scene of imagery are selected to assess the accuracy by comparing with the LST as shown in Fig. 6 and Table 2. Figure 6 explains how the author obtains the specific points of meteorological stations for validation. The result in Table 2 shows that differences between LST and the meteorological air temperatures are in 0.53–2.27 °C in the Ilan plain (Suao station and Ilan station).

Fig. 6
figure 6

Geographical location of meteorological stations and the corresponding LST retrieval (scene-specific)

Full size image
Table 2 LST validation by the meteorological temperature data (scene-specific)
Full size table

The differences can be caused by various factors. LST is influenced by wind blowing over the surface and cooling the upper few microns, and soil moisture. The temperatures may rise rapidly as the time approaches toward the noon and the highest temperatures are present at the noon 12:00:00 local time (Fig. 7a). The measured air temperature usually depends on air conditions near ground. For slight vegetated areas the day LST is higher than air temperature. At night, LST is lower than air temperature. However, this relation varies with cover types, seasons and regions. Theoretically, the thermal responding of the land surface to the time-varying energy input is determined by the thermal inertia. Thermal inertia is the key property dominating the variations of diurnal surface temperature. It depends on the physical parameters of the top few centimeters of the land surface and represents the complicated combination of rock and soil properties (Mellon et al. 2000; Price 1977).

Fig. 7
figure 7

a CWB hourly measured air temperature variation for 2 sites in Ilan plain on December 3, 2001. b Scatter plot of the measured air temperature and retrieved LST with fitted linear curves

Full size image

The comparison between air temperature and LST retrieval is shown by the clustering points in Fig. 7b; the relationship is rather significant in the context of linear regression despite the limited stations (correlation coefficient 0.76). Major discrepancies can be caused by radical differences in the scales of resolution between the satellite and ground-based CWB meteorological stations (i.e., the scale mismatch). The stronger LST heterogeneity generally leads to the greater scale-mismatch effect. Therefore, the ground measurements may not be representative of the complex landform observed by the satellite sensor. This is especially obvious for the CWB stations viewed on non-vegetated area enclosing each tower while the major satellite footprint includes vegetated land (Quattrochi and Luvall 2004). However, in this study, despite the pros and cons of ground measurements, the only ground-truth data set available for validation is the air temperature of meteorological stations.

Furthermore, regression analysis between the measured air temperature and retrieved LST are conducted for accuracy assessment as shown in Table 3. Examinations of the results are represented by r, R-Squared (R 2), and standard error.

Table 3 Linear regression summary and parameter estimates of the measured air temperature and retrieved LST
Full size table

In summary, the error of LST retrieval can be attributed as the following sources: first, converting radiances to LST contains a number of assumptions and approximations such as sensor properties. Past studies indicate the calibration error of ETM+ data acquired after December 2000 is within ± 0.6 K. Second, the error stems from the accuracy of water vapor measurements using the atmospheric correction model. Taking MODTRAN atmospheric correction model for example, target temperature of 300 K could cause a LST error of about 0.5 K for ETM+ data (Schmugge et al. 1998). Third, the error may arise from the estimate of surface emissivity. The emissivity error can come from the NDVI (Normalized Difference Vegetation Index) estimation error; however, the emissivity error should be smaller than 0.005, which leads to a variance of 0.2 K (ETM+ data) in the LST providing the target temperature is 300 K (Li et al. 2004). Finally, the scale-mismatch effect is also considerably responsible for discrepancies between the point measurements and remote sensing methods.

2.3.2 Cross Validation from MODIS LST Products

Landsat 7 ETM+ dataset has the advantage of high spatial resolution (60 m) but the low temporal resolution (revisit in 16 days) is a disadvantage. Compared to Landsat 7, Terra MODIS (Moderate Resolution Imaging Spectroradiometer) provides daily imagery with spatial resolution of 1 km in the study area. According to the general accuracy statement from MODIS land team, The LST accuracy is better than 1 K (0.5 K in most cases), as expected pre-launch (Coll et al. 2009). Thus, MODIS LST product is applied for the cross validation of Landsat retrieved LST. MODIS/Terra Land Surface Temperature Daily L3 Global 1 km (Product ID: MOD11A1) daytime LSTs acquired by Terra MODIS at local time 10:30 AM on December 3, 2001 is displayed in Fig. 8. The comparison on the statistics of retrieved Landsat LST and MODIS LST product in Ilan plain is shown in Table 4.

Fig. 8
figure 8

Pattern of LST distribution in Ilan plain from MODIS LST product on December 3, 2001. The spatial resolution of MODIS LST products is 1 km

Full size image
Table 4 Comparison on the statistics of retrieved Landsat LST and MODIS LST product in Ilan plain
Full size table

3 Results and Discussion

3.1 Retrieved LSTs in Ilan Plain

Figure 9a shows the retrieved LST distribution of Ilan plain from Landsat 7 ETM+ band 6 (High Gain setting) data on December 3, 2001. Figure 9b displays the corresponding Land Use and Land Cover (LULC) classification map based on the NDVI value. The pattern indicates that the lowest temperature in the study area is about 9 °C and the highest is 27 °C with color spanned from blue to red. Six areas with distinct red color are selected and marked with A, B, C, D, E and F. Temperatures of A, B, C, D, E and F areas are overall 3–6 °C higher compared to the ambient background temperature. The pattern of thermal anomaly is consistent with those presented in the previous studies (Chiang et al. 1979, 2007), which focus on point measurements rather than remote sensing methods.

Fig. 9
figure 9

a Pattern of LST distribution in Ilan plain from Landsat 7 ETM+ on December 3, 2001. Major thermal anomalous areas are indicated by A, B, C, D, E and F. b LULC classification map of the study area

Full size image

Faults and earthquakes are closely related in the way that earthquakes occur when rocks slip along faults. Though many faults were identified in the surrounding mountain areas of Ilan plain, the features beneath the plain are not much investigated due to the thick sediment deposit. The seismicity of the Ilan plain is mainly distributed in the southern portion. Tectonically, the plain is usually divided into two entities: the Ilan basin in the north and its collision zone in the south. The Ilan basin related to Okinawa Trough is an extension setting dominated by normal faults and the collision zone is in the compressive condition caused by the arc-continental collision (Huang et al. 2012). Figure 10 shows the faults and subsurface structure beneath the Ilan plain through seismic proofing (Liu 1995).

Fig. 10
figure 10

LST compared with fault structure (dark umber bold line) and basement depth contour (blue line) in Ilan plain

Full size image

Comparing thermal anomaly areas in Fig. 9a with the fault structure and basement depth in Fig. 10, it shows that spatial correlations between thermal anomalies and faults for areas near mountains (i.e., area A, D, E and F in Fig. 9a) is more significant than areas near the plain center (i.e., area B and C in Fig. 9a). This pattern may be caused by the thickness of sediment deposits where shallow deposit area has the better heat transfer efficiency. The urban heat island (UHI) effect in the subsurface should also be accounted for thermal anomaly areas B (Ilan City) and C (Loudong Town), considering the fact that Ilan plain is basically a rural area in Taiwan and the pixel size of satellite imagery used is 30 m by 30 m.

3.2 Multi-temporal Brightness Temperature Imagery for the Verification of the LST Anomaly Results

LST anomaly maybe affected by the precipitation history and soil moisture in different seasons of the year. Thus, a single Landsat ETM+ scene on December 3, 2001 is may not be sufficient for the proof of LST anomaly results shown in Fig. 9a. For the verification of LST anomaly in Ilan plain. Additional 8 imageries of different year and season (Acquisition date: 1999-08-08, 2001-07-28, 2008-05-12, 2009-05-31, 2013-07-29, 2015-06-17, 2015-08-04, and 2016-08-22) which selected meticulously from the Landsat archive were processed to obtain the brightness temperature (BT) of the study area as shown in Fig. 11 and Table 5. The BT is the radiance measurement of the electromagnetic radiation traveling upward from the top of the atmosphere to the satellite, expressed in units of the temperature of an equivalent black body. The ratio between the BT and LST is known as surface emissivity. Since surface emissivity in the nature is smaller than one. Therefore, LST is always higher than BT.

Fig. 11
figure 11

Pattern of Brightness Temperature distribution in Ilan plain from Landsat 7 ETM+ on 8 August 1999, 28 July 2001, 12 May 2008, 31 May 2009, 29 July 2013, 17 June 2015, 4 August 2015, and 22 August 2016, respectively. Blank areas (white color) indicate “No Data” caused by either cloud coverage or image gaps. Thermal anomalous areas are illustrated by red color in the imagery

Full size image
Table 5 Statistic summary of Landsat ETM + multi-temporal brightness temperature imagery in the study area
Full size table

All the selected images suffered from the cloud coverage or image gap caused by the satellite’s mechanical malfunction. However, the study-interested triangle Ilan plain is mostly retained in the imagery. The imagery set from 1999 to 2016 shows clear consistency in the thermal anomaly area discussed above (A, B, C, D, E and F as shown in Fig. 9a.) which provides ample support for the LST anomaly results.

3.3 Spatial Correlation Between LST Anomaly and Geothermal Occurrences, Faults and Resistivity Profiles

The selected thermal anomaly areas can be validated by the field investigation as shown in Fig. 12. Results indicate that area A, E, and F are three sightseeing geothermal landscapes of tourist attractions. Hot springs and hydrothermal explosions are widely distributed in these geothermal areas. For example, the hot spring in area A, Jiaoxi hot spring is the most historic hot spring site, which is quite rare in Taiwan for its occurrence on a flatland. Meanwhile, Tangwei hot spring (Tangweigou Park) in area A has been famous since Qing dynasty (ruling from 1644 to 1912 AD in China) and placed on the list of “Eight scenes of Lanyang area”. In the mountain area near E, Chinhsui geothermal park is also a well-known geothermal spot located in the Chinhsui river valley, southwest to Ilan plain. The temperature of the spring water is as high as 95 °C, and its geothermal energy resources are located beneath the shallow riverbed. In area F, both hot and cold springs occur at Suao, which is one of the three rarest cold spring sites in the world. Area D features with the two drilling wells with the high geothermal gradient of 6.2 °C/100 m depth (Lize well) and 7.6 °C/100 m depth (Longde well), respectively. Finally, it should also be noted that in the area of Lanyang River, which is the main river of Ilan plain in between area B and C, thermal anomaly pattern obviously shows the continuous point-shaped distribution along the riverbank. Geothermal drillings such as Wujie well (4.3 °C/100 m), Dazhou well (5.3 °C/100 m) and Gengshen well (7.5 °C/100 m) are approximately distributed by way of the Zhuoshui fault in between area B and C. They might as well have certain correlations with the fault pattern in this area. Correspondingly, other faults (e.g., Niutou, Jiaoxi, Yilan, Chingshuihsi and Kulu faults) also display the similar pattern of correlation.

Fig. 12
figure 12

Geothermal drillings and hot springs in Ilan area. Purple squares in the plain indicate geothermal drillings and red circles annotate hot springs. The dashed box shows the Chingshui geothermal area in Fig. 13

Full size image

Magnetotelluric (MT) survey is applied for exploring structures of the geothermal reservoir. Thirty-three broadband magnetotelluric data by the ITRI were analyzed for the Chingshui geothermal area is shown in Fig. 13. Profile A (A–A′, NE–SW) and profile B (B–B′, NW–SE) are determined for the resistivity analysis. After electric fields and magnetic fields were calculated as a function of frequency, resistivity sections were then obtained for these two profiles by 2D data inversion (Tong et al. 2008).

Fig. 13
figure 13

MT stations in the Chingshui geothermal area. Most stations are confined in the valley because of the hilly topography. Profile A (A–A′, NE–SW) and profile B (B–B′, NW–SE) are constructed for further analysis (Tong et al. 2008)

Full size image

Figure 14c, d shows the 2-D resistivity sections of profiles A and B, respectively. Figure 14a, b are the corresponding LST profiles. Comparing resistivity sections with LST profiles (i.e., Fig. 14a, b vs. Fig. 14c, d), the horizontal axes are aligned to the same distance and locations. The vertical axes indicate variations on LST and resistivity. The significant low-resistivity zones observed in the resistivity sections (Fig. 14c, d) are echoed with the low-temperature zones of LST profiles (Fig. 14a, b). The subsurface resistivity is affected by the combined effects of temperature, pressure, mineral composition, geological structure, fluid, and partial melting. Generally, the decrease in pressure or the increase in the temperature, pore fluid, and the proportion of high conductive minerals (such as magnetite, sulfide minerals, etc.) in the rock will lead the decrease of resistivity. Previous MT surveys in the study area show that conductive structures with significant low-resistivity (less than 100 Ω-m) usually were identified as the fracture zone (Chang et al. 2014; Ho et al. 2014). The low-resistivity zones in the resistivity sections (Fig. 14c, d) contain rocks with higher porosity and permeability, and water saturation (amount of water), may subsequently lead to the lower temperature on LST.

Fig. 14
figure 14

LST profiles compared with MT resistivity section of profile A and profile B from Tong et al. 2008. The low-resistivity zones identified in the resistivity sections (c, d) are echoed in the LST profiles (a, b)

Full size image

Emphasis should be made that the analysis in this study concerns temperature anomalies rather than absolute temperatures. Temperature anomaly is defined as the departure from a reference value. Absolute temperatures may vary notably in short distances due to various factors described in Sect. 2.3, while temperature anomalies are representative of a much larger scale. They present a frame of reference that provides much meaningful comparisons among locations, which is deemed to be more suitable for geothermal prospecting.

In summary, LST is accountable for the localized temperature anomaly which primarily came from solar radiation and the Earth’s interior heat. Thus, comprehension of surface energy balance and underground heat transfer will facilitate the detection of geothermal regions. It is suggested that the volcanic intrusion inferred by magnetic anomaly could be the heat source of the high geothermal in the study area. A 3D geothermal conceptual model of Chingshui geothermal field was also proposed (Tong et al. 2008). However, the mechanism of geothermal anomaly for Ilan plain is needed for further verification.

4 Conclusions and Future Directions

This study aims to apply and integrate TIR remote sensing technology with existing geophysical methods for geothermal prospecting in Taiwan. Land surface temperature (LST) has been retrieved from the calibrated NASA Landsat 7 ETM+ data. Accuracy assessment of satellite-derived LST is conducted by comparing with the ground-truth air temperature data and the MODIS LST product. Landsat ETM+ derived LST anomaly results were verified by the multi-temporal brightness temperature images. Six thermal anomaly areas (A, B, C, D, E and F as shown in Fig. 9a) with overall 3–6 °C higher than the ambient background temperature are determined in the study area. Among them, selected geothermal anomaly areas (A, D, E and F) are validated in detail by the field investigation of hot springs and geothermal drillings. Results imply that occurrences of hot springs and geothermal drillings are in close agreement with anomaly areas. Thermal anomaly patterns also indicate that the distributions of geothermal areas appear correlating with the development of faulted structures in Ilan plain. The significant low-resistivity zones identified in the resistivity sections are echoed with the LST profiles when compared with the Chingshui area.

This work suggests that TIR remote sensing is a valuable tool for thermal anomaly detection and mapping with its high efficiency, cost-effective, and accuracy in temperature retrieval. TIR remote sensing provides a rapid way of mapping and quantifying surface features to facilitate the exploration and assessment of geothermal resources (Kuenzer and Dech 2013). Most significantly, compared with conventional in situ methods of point measurements, remote sensing methods are able to produce reliable results which are spatially representative at larger regional scales. However, it is limited to detecting the surficial and the shallow buried geothermal resources. Thus, TIR remote sensing cannot be expected to suffice as the sole tool for exploring and monitoring geothermal resources. Further geologic analysis and mechanisms of geothermal anomaly are needed to assist the identification of geothermal areas. The integration of geothermal mechanism with the structure geology analysis will greatly improve the applicability of geothermal detection using TIR remote sensing. Future research focus is to incorporate the heat conservation equation and the rich of previous geophysical research efforts to estimate spatially distributed surface fluxes for the establishment of a geothermal 3D model in Ilan plain.