Advancing Hyperspectral LWIR Imaging of Soils with a Controlled Laboratory Setup

Highlights

What are the main findings?

• Imaging laboratory spectroscopy in the longwave infrared (LWIR) range allows reliable and repeatable measurements of soil emissivity.
• Tests on different soils showed that the LWIR laboratory results agree well with outdoor measurements and mineral analyses.

What are the implications of the main findings?

• Our approach provides a weather-independent way to measure soils in the LWIR range and semi-quantify their minerology.
• It establishes a solid basis for calibrating satellite and airborne LWIR data, improving soil and environmental monitoring with the potential of becoming a standard protocol for these types of campaigns.

Abstract

This study introduces a controlled laboratory setup for hyperspectral longwave infrared (LWIR) imaging of soils, designed to bridge the gap between laboratory measurements and remote sensing observations. A Fourier-transform hyperspectral LWIR imaging spectrometer (Telops Hyper-Cam LW) was employed, together with a specialized heating plate, rigorous calibration protocols, and a Spatial Averaging Before Blackbody Fitting (SABBF) method to enable accurate LWIR indoor measurements. Unlike established laboratory techniques that measure reflectance and calculate emissivity indirectly, this setup enables direct passive measurement of soil emissivity, replicating airborne and spaceborne LWIR measurements of the surface. The emissivity spectra of 12 variable soil samples obtained with the lab setup were compared and validated based on LWIR Hyper-Cam LW spectra acquired under outdoor conditions, then were subsequently analyzed to determine the mineral composition of each sample. Spectral features and indices were used to estimate the relative content of quartz, clay minerals, and carbonates, from the most to least abundant. The results demonstrate that the laboratory-based setup preserves spectral fidelity while offering improved repeatability, scheduling flexibility, and reduced dependence on weather. Beyond replicating outdoor measurements, this controlled setup is easy to install and provides a reproducible framework for LWIR soil spectroscopy that could be considered for standard laboratory protocols, enabling reliable mineral identification, calibration/validation of airborne and satellite LWIR data, and broader applications in soil monitoring and environmental remote sensing.

1. Introduction

Soil is a complex medium with highly variable physical and chemical properties. It originates from weathered rocks of the Earth’s crust and undergoes continuous change. Soils contain plant debris, water, roots, small and large fauna, and minerals derived in situ or transported by runoff, dust storms, and human activity [1,2,3,4,5,6]. Several remote sensing techniques have demonstrated the ability to obtain information about soil surface composition. In particular, hyperspectral remote sensing, with high spectral resolution, in the visible and short-wavelength infrared ranges between 0.4 and 2.5 μm, has proven to be a valuable method of identifying and mapping soil characteristics, using sensors in laboratories, air-, and spacecrafts [1,2,3,4,5,6].

Recent research highlighted the additional value of the thermal infrared region, including the mid-wavelength infrared (MWIR, 3–5 μm) and long-wavelength infrared (LWIR, 8–12 μm) domains, for retrieving soil properties such as texture, carbon and nitrogen content, pH, and mineral composition when combined with statistical modeling [4,7,8,9,10,11]. In particular, Eisele et al. (2012, 2015) were able to quantitatively model Al₂O₃ and SiO₂ content based on FTIR and MIDAC laboratory spectroscopy, using a Illuminator M4401 and a Perkin & Elmer Spectrum GX [8,9]. Notesco et al. (2019) demonstrated that outdoor ground-based hyperspectral LWIR measurements, based on the Telops Hyper-Cam LW, were used to derive semi-quantitative estimates of quartz, clay minerals, and carbonates from soil emissivity spectra, with results in good agreement with XRD mineralogy [4]. Note that quantitative mineral estimates provide precise, reproducible concentrations of minerals (usually expressed in weight %, molar ratios, or absolute values) based on calibrated analytical techniques, and semi-quantitative estimates give approximate relative abundances or ranges, derived from indirect proxies, and are useful for rapid screening but not for definitive concentration calculations [12].

Despite these advances, a methodological gap remains between laboratory measurements and remote sensing observations. Conventional laboratory approaches typically rely on reflectance-based measurements with indirect emissivity estimation, while outdoor LWIR measurements are weather-dependent and logistically demanding. A standardized protocol for controlled laboratory acquisition of direct emissivity spectra using hyperspectral LWIR imaging is currently lacking. Addressing this gap is essential for producing reproducible reference spectra that are directly comparable to airborne and satellite LWIR observations.

This study introduces a laboratory-based protocol for measuring soil emissivity spectra using hyperspectral LWIR imaging. The method employs the Telops Hyper-Cam LW sensor under controlled indoor conditions, enabling direct emissivity measurements comparable to outdoor acquisitions. Validation was performed by comparing the laboratory-based emissivity spectra obtained with emissivity spectra of the same soil samples acquired under a different protocol in outdoor conditions. Furthermore, mineralogical indices developed by Notesco et al. (2019) were used to validate the mineralogical identification of the soil samples [4]. The proposed setup allows the generation of reliable ground-truth spectra, independent of weather constraints, using the same instrument applied in airborne campaigns, providing a reproducible and flexible workflow for soil compositional analysis.

2. Materials and Methods

2.1. Soil Samples and Mineral Abundance

Twelve soil samples (see

Table 1. Soils and their mineral abundance.

Soil (Symbol, USDA Great Group Name)	Mineral Abundance (%) a
	Quartz	Clay Minerals b	Carbonates c	Organic Carbon d
E2, Rhodoxeralf	90	5	0	0.17
E7, Rhodoxeralf	75	15	1	0.16
C4, Haploxeroll	55	30	12	0.73
B8, Haploxeroll	30	40	31	1.95
A3, Rhodoxeralf	35	58	2	2.64
H2, Xerert	4	67	21	1.85
K2, Calciorthid	15	0	60	1.27
O3, Torriorthent	25	10	59	0.30
P3, Torriorthent	25	7	68	0.19
H11, Haplargid	36	10	54	0.56
H14, Xerert	25	45	28	1.22
S19, Torriorthent	47	10	43	0.77

^a Based on X-ray powder diffraction (XRD). ^b Clay minerals refer to smectite, illite, and kaolinite. ^c Carbonates refer to calcite (mainly) and dolomite. ^d Organic Carbone on Walkley and Black.

) from the legacy soil spectral library of the Tel-Aviv University (Israel) were selected for this study [13,14]. The samples represent diverse formation conditions, including differences in climate, parent material, and topography. All samples were collected from the surface layer (0–5 cm) at multiple sites across Israel, air-dried and crushed to a grain size of ≤2 mm. These soils have previously been characterized in the context of outdoor hyperspectral LWIR measurements by Notesco et al. (2019) [4], providing a basis for direct comparison between laboratory-based and outdoor-derived emissivity spectra.

2.2. Laboratory Setup

Laboratory measurements were carried out using the Hyper-Cam LW imaging spectrometer. The Hyper-Cam LW is a lightweight and compact imaging radiometric Fourier-transform spectrometer (FTIR) built by Telops, Quebec, QC Canada, and covers a range from 7.6 to 11.7 μm (see

Table 2. Telops Hyper-Cam LW sensor specifications [15,16,17].

Sensor	Telops Hyper-Cam LW (FTIR)
Spectral Range (μm)	7.7–11.8
Spectral Resolution (cm−1)	Up to 0.25
Spectral Resolution used (cm−1)	4
Spatial Resolution	320 × 256 (Full Frame)
FOV (Degrees)	6.4 × 5.1
Typical NESR (NW/CM2 SR CM−1)	20
Radiometric Accuracy (K)	<1.0
Calibration	2 Internal Blackbodies

) [15,16]. It is based on a Michelson interferometer coupled to a 320 × 256 longwave infrared photovoltaic MCT (mercury-cadmium-telluride) focal plane array detector, with a user-adjustable field of view (FOW) [15,16,17]. The instrument offers a selectable resolution ranging from 0.25 to 150 cm⁻¹. Each pixel of the images contain the complete spectrum and have an instantaneous field of view of 0.35 mrad [15,16,17]. The Hyper-Cam LW features two calibration blackbodies mounted in front of the instrument to perform a complete end-to-end radiometric calibration of the infrared measurements [15,16,17]. Besides acquisition electronics, it also includes processing capabilities, and is able to convert the raw interferograms into radiometrically calibrated spectra using a real-time discrete-Fourier transform (DFT) technique [15,16,17]. Its weatherproof enclosure provides operability in harsh environments from −10 °C to +45 °C [16].

For the laboratory measurements, the Hyper-Cam LW was mounted on a stable and adjustable tripod designed for the instrument. Around 10 g of dried soil sample material was placed in each aluminum sample holder. The soil samples were then placed on a custom laboratory heating plate set to 80 °C for continuous heating prior and during measurement, while the camera was equipped with a wide-angle ×0.25 telescope and gold mirror form Telops, Quebec, QC, Canada to provide an adequate nadir field of view (FOV) (see Figure 1 and Figure A5). The aluminum sample holders here have the purpose of transferring the heat from the plate to the soil sample material. Heating prior to measurement was applied until thermal equilibrium was reached, monitored by a thermocouple placed at the soil surface (see Figure 2). The 10 g of soil sample material usually reached their maximum temperature in under 15 min of heating. An amount of 10 g was chosen as this was enough soil for all samples to cover the sample holder surfaces while at the same time reaching stable temperatures in a sensible amount of time. Heating of the samples creates a thermal contrast between the samples and the laboratory surrounding. This improves the signal-to-noise ratio of the measurements as the downwelling radiance from the air-conditioned (20 °C) environment is smaller in relation to the signal from the samples. The higher the temperature difference, the lower the relative influence of the background radiation of the surroundings on the at-sensor radiance signal of the samples.

The 80 °C plate temperature was selected in order not to alter soil structure while producing soil surface temperatures of approximately 60–70 °C and creating a sufficient temperature contrast between the samples and the laboratory. Such soil surface temperatures are also reached, or even exceeded, under natural conditions in Mediterranean and Middle Eastern climates during sunny summer periods [18,19,20]. Laboratory-based LWIR images were subsequently acquired with the Telops Hyper-Cam LW, covering the 7.9–11.5 µm range with 124 spectral bands at a resolution of 4 cm⁻¹. This resolution was chosen to increase the signal-to-noise ratio while keeping a good enough spectral resolution to resolve the mineralogical features. Furthermore, it is stated to be the optimal resolution for the system and is often used in remote sensing applications during, e.g., airborne acquisitions and/or their ground/sample validation [21,22].

2.3. Data Pre-Processing

The processing of the Hyper-Cam data involved blackbody calibration using Reveal Calibrate 5.3.1, wavenumber to wavelength conversion, temperature emissivity separation (TES), and downwelling radiance correction using a Python 3.12 algorithm described in the next sections.

The wavenumber (wn) to wavelength $\left(\lambda \right)$ conversion is performed via a Python algorithm that converts the spectral radiance via pixel-by-pixel iterations by applying Equation (1).

$${\mathrm{L}}_{\mathrm{o}\mathrm{b}\mathrm{s}}\left(\lambda \right)[\mathrm{W}/{\mathrm{m}}^2/\mathrm{s}\mathrm{r}/\mathrm{\mu }\mathrm{m}]={\mathrm{L}}_{\mathrm{o}\mathrm{b}\mathrm{s}}\left(wn\right)[\mathrm{W}/{\mathrm{m}}^2/\mathrm{s}\mathrm{r}/{\mathrm{c}\mathrm{m}}^{-1}]\times {10}^4/{\lambda }^2 ,$$

(1)

where $\lambda \left[\mathrm{\mu }\mathrm{m}\right]={10}^4/wn \left[{\mathrm{c}\mathrm{m}}^{-1}\right]$.

2.3.1. Laboratory TES Using a Blackbody Fitting Approach

The at-sensor radiance measured for each soil pixel in the LWIR image contains contributions from both temperature and emissivity. The emissivity spectrum was derived by applying a self-developed Temperature Emissivity Separation (TES) procedure implemented in Python, as described below:

The emissivity spectrum of the surface, reflecting its chemical and physical properties, can be calculated using Equation (2), where ${\mathrm{L}}_{\mathrm{o}\mathrm{b}\mathrm{s}}\left(\lambda \right)$ is the observed at-sensor radiance, $\mathrm{B}(\mathrm{T},\lambda )$ is the blackbody radiance, $\epsilon \left(\lambda \right)$ is the emissivity, $\tau \left(\lambda \right)$ is the atmospheric transmittance, ${\mathrm{L}}^{\downarrow }\left(\lambda \right)$ is the downwelling radiance, and ${\mathrm{L}}^{\uparrow }\left(\lambda \right)$ is the upwelling radiance, with all factors at temperature $\mathrm{T}$ and wavelength $\lambda$.

$${\mathrm{L}}_{\mathrm{o}\mathrm{b}\mathrm{s}}\left(\lambda \right)=\mathrm{B}(\mathrm{T},\lambda )\epsilon \left(\lambda \right)\tau \left(\lambda \right)+[1-\epsilon (\lambda \left)\right]{\mathrm{L}}^{\downarrow }\left(\lambda \right)\tau \left(\lambda \right)+{\mathrm{L}}^{\uparrow }\left(\lambda \right)$$

(2)

Under laboratory conditions, the atmospheric transmittance $\tau \left(\lambda \right)$ is equal to 1 and the upwelling radiance ${\mathrm{L}}^{\uparrow }\left(\lambda \right)$ is negligible due to the very short sensor-target distance, simplifying Equations (2) to (3).

$${\mathrm{L}}_{\mathrm{o}\mathrm{b}\mathrm{s}}\left(\lambda \right)=\mathrm{B}(\mathrm{T},\lambda )\epsilon \left(\lambda \right)+[1-\epsilon (\lambda \left)\right]{\mathrm{L}}^{\downarrow }\left(\lambda \right)$$

(3)

Equation (3) can be rearranged for emissivity calculations, accounting for downwelling radiance, as shown in Equation (4) [23].

$$\epsilon \left(\lambda \right)={\displaystyle \frac{{\mathrm{L}}_{\mathrm{o}\mathrm{b}\mathrm{s}}\left(\lambda \right)-{\mathrm{L}}^{\downarrow }\left(\lambda \right)}{\mathrm{B}\left(\mathrm{T},\lambda \right)-{\mathrm{L}}^{\downarrow }\left(\lambda \right)}}$$

(4)

Equations (3) and (4) are used as part of our Python processor for hyperspectral radiance images. The program calculates the temperature and emissivity for each pixel as follows.

First, constants (Planck’s constant, Boltzmann’s constant, and the speed of light) are defined for blackbody radiation calculations. An initial temperature T, based on the heating plate, is used to generate the first blackbody curve. This curve is iteratively fitted by adjusting temperature to match the measured radiance. The wavelength $\left(\lambda \right)$ array, defined from the image header, provides the basis for modeling the Planck function and performing TES.

The Planck function calculates blackbody spectral radiance for temperature T over the defined wavelength range, while its inverse is used to estimate brightness temperature from ${\mathrm{L}}_{\mathrm{o}\mathrm{b}\mathrm{s}}\left(\lambda \right)$. For each pixel, the brightness temperature is derived using the inverse Planck function, ensuring that blackbody radiance at this temperature is greater than or equal to the pixel radiance. The temperature inversion is performed by minimizing the residual between the observed radiance spectrum and the Planck function using the “fsolve” function provided by the Python package SciPy [24]. The SciPy “fsolve” function is a Python wrapper for MINPACK’S “hybrd” and “hybrdj” algorithms, which are based on Powell’s hybrid method, an iterative optimization approach for non- linear functions and a variant of gradient descent [25,26,27]. Specifically, we define the cost function (5) as follows:

$$J\left(T\right)={\sum }_{\lambda }{\left[L_{obs}\left(\lambda \right)-B(T,\lambda )\right]}^2,$$

(5)

and solve for the temperature $T$ that minimizes $J\left(T\right)$. This corresponds to the brightness temperature, i.e., the blackbody temperature that best fits the observed spectrum. Because the observed radiance is a product of soil temperature and emissivity, the brightness temperature is not yet the true soil temperature. However, it provides a physically consistent upper bound: emissivity depresses the spectrum below the blackbody curve, so the temperature that minimizes residuals represents the soil temperature prior to emissivity correction.

The measured downwelling radiance ${\mathrm{L}}^{\downarrow }\left(\lambda \right)$ is taken from a gold plate in the scene, identified either by thresholding or predefined coordinates. This represents the radiation incident on the sample from laboratory surroundings and is essential for accurate emissivity retrieval. The emissivity ε(λ) for each pixel is calculated as the ratio of measured radiance to the blackbody radiance at the estimated temperature, corrected for L↓(λ) according to Equation (4). Then the algorithm iteratively adjusts the fitted blackbody curve in 0.05 K steps until it exceeds the observed radiance at all wavelengths, following the approach of Notesco et al. (2014), outputting the adjusted temperature and emissivity images [28].

In summary, this implementation estimates temperature and emissivity from hyperspectral LWIR radiance images involving the use of Planck’s law, the concept of blackbody radiation, and the Temperature Emissivity Separation (TES) by blackbody fitting. The brightness temperature is calculated using the inverse Planck function, and the emissivity is estimated by fitting the observed radiance to the blackbody radiance at the estimated temperature. The code also includes corrections for downwelling radiance. The output consists of temperature and spectral emissivity images (see Figure 3).

2.3.2. Automated ROI Extraction

The calculated emissivity data from the measurements of the samples were processed using an automated Region of Interest (ROI) Python algorithm that detects the samples in the emissivity images using object detection, based on OpenCV edge detection (see [29]). When provided with a sample names text file, it extracts 70 × 40-pixel ROIs from the center of each sample tray (see Figure 4). The extracted sample ROIs are then stored as individual hyperspectral emissivity data cubes containing 70 × 40 pure sample pixels without sample tray edges. Another Python script then averages the 70 × 40-pixel image cubes to obtain a single emissivity curve for each sample. The resulting emissivity curve of each sample is hence a 70 × 40-pixel average.

2.3.3. Spatial Averaging Before Blackbody Fitting (SABBF)

In addition to the ROI-based workflow described above, a modified processing chain was implemented to further improve the signal-to-noise ratio (SNR) of the radiance data prior to emissivity retrieval. This method, termed Spatial Averaging Before Blackbody Fitting (SABBF), applies spatial filtering before the TES step. After ROI detection, the radiance data of each sample was averaged using a 10 × 10 pixel running average filter. This procedure reduces high-frequency noise in the radiance domain by combining the hyperspectral radiance of adjacent pixels. Although the image dimension per ROI remains at 70 × 40 pixels, the effective spatial resolution is reduced to 7 × 4 pixels (Figure 5). The noise-suppressed radiance spectra were then processed using the blackbody fitting algorithm described in Section 2.3.1. A comparison of emissivity retrieval from standard pixel-level fitting versus SABBF-processed data is shown in Figure 6, demonstrating the improvement in spectral smoothness and stability.

2.4. Validation with Outdoor Lab Measurements

For the validation of our method we chose samples that have been measured outdoors with the same instrument (Telops Hyper-Cam LW) before by Notesco et al. (2019) (see Section 2.1) [4]. The data collected by Notesco et al. (2019) has been processed by a slightly different algorithm using Interactive Data Language (IDL), described in detail in Notesco et al. (2014) [4,28]. The differences are considered minor because both the blackbody fitting and TES algorithms use a generally similar approach, differing primarily in terms of mathematical processing details. The main distinction occurs in the outdoor measurements, where atmospheric downwelling correction is required to account for environmental influences. The general physical foundation of the blackbody fitting and the related mathematics of the formulas in Section 2.3 are given in both approaches and we consider the outdoor data measured by Notesco et al. (2019) to be similar enough to validate our indoor laboratory method [4].

2.5. Mineralogical Identification

The mineralogical characterization of the laboratory-measured soil samples was conducted using indices derived from known emissivity absorption features of specific minerals as also described in Notesco et al. (2019) [4]. A triplet-like absorption feature between 8.00 µm and 10.40 µm, with minima at 8.25 µm, 8.79 µm, and 9.35 µm, indicates the presence of both quartz and clay minerals. A decrease in the amount of quartz with an increase in the amount of clay minerals shifts the main absorption to 9.00–10.40 µm, with a minimum at 9.58 µm. On the other hand, a minimum in the 8.00–8.18 µm range and/or a noticeable absorption feature between 10.11 µm and 11.38 µm indicates the presence of carbonates in the soil. These mineral-related emissivity features were used to create spectral indicants, described in

Table 3. Spectral indicants of minerals.

Most Abundant Mineral	Spectral Indicant
Clay minerals	Nελ=9.58 µm a < Nελ=8.25 µm and Nελ=8.25 µm > 0.98
Carbonates	ελ=8.00–8.18 µm b < ελ=8.25 µm and/or Nελ=11.22 µm < 0.995 with Nελ=8.25 µm > 0.98
Quartz	Excluding the above

^a Nε is the normalized emissivity value at the indicated wavelength. ^b ε is the emissivity value in the indicated wavelength range.

, to identify the most abundant mineral(s) in each soil sample.

Next, two indices (Formulas (6) and (7)) were created to identify the less abundant minerals in each soil sample:

SQCMI (Soil Quartz Clay Mineral Index) = Nε_{λ=9.58 µm}/(Nε_{λ=8.25 µm} × Nε_{λ=8.79 µm})

(6)

SCI (Soil Carbonate Index) = Nε_{λ=11.22 µm} × Nε_{λ=10.56 µm}/Nε_{λ=8.79 µm}

(7)

where Nε is the normalized emissivity value at the indicated wavelength.

A higher SQCMI value corresponds to a greater proportion of quartz compared to clay minerals, while a lower SCI value reflects a higher carbonate content within the soil sample [4]. The indicants based on these indices are described in

Table 4. Spectral indicants of relative amounts of minerals.

Soil Type a	Indicant	Relative Amount of Mineral(s)
Q	SCI < 1.010	C > CM
	1.010 ≤ SCI < 1.020 and SQCMI > 1.020	C > CM
	SCI > 1.020 SCI > 1.050, SQCMI > 1.200	CM > C no C, no CM
CM	Absorption at 8.16 µm and/or SCI < 1.005	C > Q
C	SQCMI > 1.010 with Nελ=8.25 µm < 0.990	Q > CM

^a Soil type refers to the most abundant mineral(s)—quartz (Q), clay minerals (CM, smectite, illite, and kaolinite), or carbonates (C, calcite, mainly, and dolomite)—in the soil.

.

2.6. Statisical Analysis

Further statistical analysis has been performed by calculating statistical indicators between the outdoor and SABBF measurements. The $r^2$, the $\mathrm{RMSE}$, and the $\mathrm{MAE}$ have been calculated for the laboratory-acquired SABBF data and the outdoor measurements according to Formulas (8)–(10):

$$r^2={\left({\displaystyle \frac{cov\left(x,y\right)}{{\mathsf{\sigma }}_x\hspace{0.17em}{\mathsf{\sigma }}_y}}\right)}^2,$$

(8)

where $cov\left(x,y\right)$ is the covariance between the two spectral datasets, and ${\mathsf{\sigma }}_x$ and ${\mathsf{\sigma }}_y$ are their standard deviations.

$$\mathrm{RMSE}=\sqrt{{\displaystyle \frac{1}{n}}{\sum }_{i=1}^n{\left(x_i-y_i\right)}^2}$$

(9)

$$\mathrm{MAE}={\displaystyle \frac{1}{n}}{\sum }_{i=1}^n\left|x_i-y_i\right|,$$

(10)

where $x_i$ and $y_i$ are corresponding values of the two spectral measurements, and $n$ is the number of spectral bands.

3. Results

3.1. Emissivity Spectra

In order to provide a proof of concept of the laboratory measurement setup with the thermal hyperspectral imaging system and heating plate, 12 soil samples, also present in the legacy soil spectral library of Israel and measured in an outdoor setup by Notesco et al. (2019) (see Section 2 and [4]), were measured and compared.

Figure 7 shows the emissivity spectra of selected soil samples, highlighting mineral-related features. As descried above in Section 2.5, the triplet-like absorption features between 8.00 µm and 10.40 µm indicates the presence of both quartz and clay minerals, which can be seen in soil E7. In soil H2, a shift in the main absorption feature to 9.00–10.40 µm correlates with a decrease in quartz towards an increase in clay minerals. In sample K2, the presence of carbonates is indicated by the positive slope between 8.00 µm and 8.40 µm, with absorption between 10.20 µm and 11.40 µm.

The samples E2, E7, C4, B8, A3, H2 represent soils that are quartz-rich to clay mineral-rich. Comparing the spectra from these soils show that the lab measurements can reproduce the outdoor measurements very well (Figure 8). There are little to no noticeable absorption position differences. The most quartz rich samples E2 and E7 both show the typical triplet-like absorption feature with minima at 8.21 µm, 8.85 µm, and 9.33 µm, associated with both quartz and clay minerals. The overview of all samples measured in the lab and under outdoor conditions shows that the general absorption features and characteristics of the soils can be reproduced via the laboratory measurements (see Figure 8 and Figure A1).

3.2. Impact of Spatial Averaging Before Blackbody Fitting (SABBF)

The Spatial Averaging method (SABBF) is used to extract the data prior to blackbody fitting and the application of a running filter before the data is processed by the blackbody fitting algorithm, and has some advantages and disadvantages, depending on the sample nature and specific use case. The running filter reduces the spatial resolution information, resulting in spatially blurred images of the samples, making spatial surface features and textures of the individual samples less resolved and harder to detect. When spatial surface features or structures of samples are of interest, this is a major disadvantage of the method. This would be the case if, for example, rock samples or structurally preserved soil samples are measured when spatial surface structures are of interest.

In our case, the spectral information is more important and the resulting radiance data per pixel is smoother and less noisy after the averaging filter is applied. This enables the simulated blackbody, as described in Section 2.3.3., to be fitted closer to the radiance curve of each pixel of the sample, allowing a more precise emissivity estimation by the algorithm. For the purpose of building a spectral emissivity soil library where the results are only spectra without spatial information, the averaging method presents a way to further improve the blackbody fitting, resulting in a closer estimation of the true emissivity of the soil samples. The comparison of all samples that have been measured in the laboratory setting and then processed by the standard blackbody fitting approach and the SABBF method show that the overall calculated emissivity is generally slightly higher (see Figure 9, Figure 10, and Figure A2).

3.3. Mineralogical Identification and Semi-Quantification

The laboratory-based emissivity spectra processed with the SABBF method resembles the outdoor-based spectra regarding the mineral-related features. These emissivity features were used to calculate indicants and indices created by Notesco et al. (2019) and are described in Table 3 and Table 4, and were also used to determine the content of quartz, clay minerals, and carbonates in each soil in a semi-quantitative manner, from more to less abundant minerals (Table 3) [4]. Laboratory-based classification regarding the most abundant mineral fit the outdoor-based classification, as well as the XRD analysis results, for all soil samples. Regarding the less abundant minerals, the laboratory-based classification fit the outdoor-based results of most (75%) soil samples (

Table 5. Content of quartz (Q), clay minerals (CMs), and carbonates (Cs) in soil samples. Letters in bold represent the most abundant mineral.

Soil	Mineralogy (More to Less Abundant)
	Laboratory-Based	Outdoor-Based	XRD Analysis
E2	Q CM C	Q CM C	Q CM
E7	Q CM C	Q CM C	Q CM C
C4	Q C CM	Q CM C	Q CM C
B8	CM C Q	CM C Q	CM C Q
A3	CM C Q	CM Q C	CM Q C
H2	CM C Q	CM C Q	CM C Q
K2	C CM Q	C CM Q	C Q
O3	C CM Q	C CM Q	C Q CM
P3	C Q CM	C Q CM	C Q CM
H11	C CM Q	C Q CM	C Q CM
H14	CM C Q	CM C Q	CM C Q
S19	Q C CM	Q C CM	Q C CM

). It should be noted that soil moisture may affect the results, as well as organic tissues such as litter, and particularly lignin (e.g., Elvidge 1988) [30]. Nonetheless, all soils were air-dried and exposed to heat conditions that minimize any moisture effects. The organic carbon content of these arid and semi-arid soils was very low (Table 1), and the existing organic matter was already decomposed into humus, which is not expected to significantly influence the LWIR spectrum.

3.4. Statistical Agreement Between SABBF and Outdoor Measurements

The comparison between the SABBF laboratory emissivity measurements and the outdoor spectra shows a consistently high agreement across most of the LWIR atmospheric window (Figure 11). Band-wise correlation coefficients remain above 0.95 for the majority of channels between 8 and 10.5 µm, with corresponding RMSE values typically below 0.006 (Figure 12), indicating that both systems capture the same spectral structure and sample-to-sample variability with high fidelity. This behavior is also reflected in the overall scatter analysis, where emissivity values from both setups align closely along the 1:1 line, yielding an overall coefficient of determination of r² = 0.91 and an RMSE of 0.0079 (

Table 6. Overall coefficient of determination, RMSE and MAE for the SABBF, and outdoor measurements.

n Emissivity Bands	r2	RMSE [Emissivity %]	MAE [Emissivity %]
1368	0.91	0.79%	0.58%

and Figure 13). Systematic deviations emerge toward the long-wavelength end of the spectrum, and from approximately 10.6 µm onward, correlations decrease markedly and RMSE increases beyond the 1% emissivity threshold (Figure 12), with the strongest discrepancies occurring between 11.0 and 11.5 µm. The detailed per-band scatter plots provided in Appendix A, Figure A6, Figure A7, Figure A8 and Figure A9, confirm these patterns, showing tight agreement in the 8–10.5 µm range and increasing departure from the 1:1 line at longer wavelengths. A small but consistent negative bias of –0.004 indicates that the outdoor measurements tend to underestimate emissivity relative to the SABBF measurements in this region. Overall, the two systems demonstrate excellent agreement across the diagnostically important part of the LWIR window, with notable wavelength-dependent degradation at the spectral longwave edge.

4. Discussion

The strong agreement across the 8–10.5 µm spectral range suggests that differences between the indoor laboratory and outdoor measurement setups have minimal impact on the retrieval of key silicate-related emissivity features that underpin most LWIR-based soil mineralogical applications. The deviations observed beyond approximately 10.6 µm (Figure 11 and Figure 12) can likely be attributed to a combination of physical and instrumental constraints inherent to open-sky emissivity measurements. Increased atmospheric absorption and emission toward the long-wavelength end, combined with the lower signal-to-noise ratio of FTIR-based sensors in this region, make downwelling radiance correction particularly sensitive to small temperature inaccuracies or calibration residuals. Differences in calibration workflows may further amplify these effects, leading to the systematic underestimation of emissivity in the outdoor spectra documented in Appendix A, Figure A6, Figure A7, Figure A8 and Figure A9, and summarized in the overall scatter plot (Figure 13). For downstream soil property estimation, these wavelength-dependent biases imply that spectral features in the 11–11.5 µm region should be used with caution or excluded when merging datasets from different measurement setups. Nevertheless, because the most diagnostic absorption features for quartz, feldspar, and clay minerals lie within the high-agreement region, the overall implications for soil mineral, texture, and related property estimation are expected to be moderate, provided that modeling approaches down-weight or mask affected long-wavelength bands. The primary difference between indoor and outdoor measurements lies in the consistency and control of the experimental conditions. Indoor measurements offer several advantages over outdoor and existing laboratory protocols for LWIR emissivity measurements. Firstly, the protocol is both cheap and easy to implement if a hyperspectral thermal imaging system is available, making it accessible for various laboratories. Secondly, it is easy to replicate, ensuring that different labs can achieve consistent results. Unlike FTIR spectroscopy, which analyzes soil samples in a different manner, this method allows soil samples to be analyzed in their own volume. This approach ensures that the emissivity measurements are more representative of real-world conditions, replicating air/spaceborne measurements. Additionally, the use of standard TES and adapted spatial averaging techniques allows for the derivation of high-quality emissivity data.

Despite the advantages, comparing measurements taken under such different conditions as indoor and outdoor remains challenging. The differences in emissivity, absorption depth, and mineral indices could be due to slight differences in the processing algorithm, and/or due to the fact that the sample mixtures surface exposure was composed slightly differently. A rather unlikely explanation would be chemical changes as a result of storage time, as the long-term stability of soil sample characteristics has been demonstrated in a recent study by Shepard et al. (2024) [31]. The re-arrangement of particles at the soil surface and varying atmospheric conditions can lead to discrepancies in the data. Considering the comparison of the standard blackbody fitting approach, the SABBF method, and the outdoor measurements, it can be shown that the SABBF method brings the laboratory measurements closer to the outdoor measurements in terms of the general shape of the emissivity (see Figure 10, Figure A3 and Figure A4). The results of this study indicate a very good agreement between the two types of measurements, validating both methods. This agreement suggests that while indoor measurements provide a controlled and reproducible environment, they can still produce results that are consistent with outdoor measurements, thereby bridging the gap between laboratory and field data.

5. Conclusions

A laboratory setup protocol is presented that utilizes a heating plate, calibration and temperature measurements, blackbody fitting, and a TES processor with spatial averaging to improve signal-to-noise ratio. This approach is used to obtain LWIR spectral emissivity data for 12 soil samples with varying mineralogical compositions. Data acquisition and processing are validated by comparing them with outdoor measurements of the same samples (see [4]). This replication is crucial for ensuring the reliability and consistency of data collected under different environmental conditions. The Spatial Averaging (SABBF) method further aligns the emissivity of lab measurements with those of outdoor soils. By leveraging this method, we can ensure that the emissivity values obtained in a controlled indoor environment are well concordant with those measured in natural outdoor settings. Furthermore, mineralogical identification and semi-quantification was derived and compared with XRD analyses. A good concordance was obtained, due to the well resolved mineralogical features in the spectra.

Producing ground-truth spectra indoors under any weather conditions with the same instrument used for aerial acquisition enhances our workflow in remote soil compositional analysis. This capability allows for more flexible scheduling and reduces dependency on favorable weather conditions that are necessary for outdoor measurements. Furthermore, it ensures that the data collected are consistent and comparable across various study sites and temporal frames. The ability to conduct thorough analyses indoors also provides opportunities to refine methodologies and validate hypotheses before conducting extensive fieldwork, thereby optimizing resources and time. The protocol developed is suggested as a foundation for a standard laboratory protocol for LWIR measurements and could be easily followed in different laboratories.

The integration of controlled lab measurements with aerial acquisition instruments bridges the gap between laboratory research and practical field applications. This approach promotes the development of more robust soil classification models, which can be applied to monitor and manage agricultural lands, assess soil health, and track environmental changes over time. By combining precise laboratory techniques with advanced remote sensing technologies, we can achieve a deeper understanding of soil properties and dynamics across diverse landscapes.