Split-Window Algorithm for Land Surface Temperature Retrieval from Joint Polar-Orbiting Satellite System JPSS-2/NOAA-21 ^†

Abstract

Land surface temperature (LST) plays a pivotal role in the dynamic exchange of energy between the Earth’s surface and the atmosphere. This research centers on the assessment of LST from satellite data acquired by the Joint Polar-orbiting Satellite System (JPSS), specifically JPSS-2/NOAA-21, employing an innovative split-window algorithm (SWA). Atmospheric water vapor content (WVC) and surface emissivity are the two main input variables in the split-window technique. Therefore, the moderate resolution transmittance code, version 4.0 (MODTRAN 4.0), was used to simulate WVC and atmospheric transmittance. The performance of the SWA was rigorously assessed against standard atmospheric conditions, revealing its capacity to achieve an LST retrieval accuracy of 1.4 Kelvin (K), even in the presence of various errors. Moreover, the LST retrieval algorithm was validated using ground truth data sets from two Australian sites, and the RMSE value was 1.71 K. The achieved results demonstrate the algorithm’s capability to provide accurate LST estimation for NOAA-21 satellite data.

1. Introduction

Land surface temperature (LST) is a necessary parameter with a profound impact on the physical processes of land surfaces, influencing a range of phenomena from local to global scales. It drives the outgoing longwave radiation and turbulent heat fluxes at the interface between the Earth’s surface and the atmosphere. Consequently, LST is routinely applied in various fields such as evapotranspiration [1,2,3,4], the estimation of soil moisture [5,6,7,8], and environmental studies [9,10,11,12]. Furthermore, the International Geosphere and Biosphere Program (IGBP) [13] considers LST as one of the high-priority parameters.

According to one of the sustainable development goals (SDGs) promoted by the United Nations, the increase in the earth’s surface temperature is regarded as a major phenomenon [14]. Hence, it is crucial to monitor this dilemma in order to evaluate the rapid variations in LST spatially and temporally in the globe for vast geographic areas. The only way to measure LST on a worldwide scale is through remote sensing satellite data, which makes this conceivable [15].

The estimation of LSTs from TIR satellite data requires two primary parameters: emissivity and atmospheric corrections [16,17]. Over the course of several decades, researchers have dedicated their efforts to refining algorithms for deriving LST from TIR remote sensing data, using a range of approaches to deal with emissivity and atmospheric effects. Among these algorithms, the split-window (SW) technique stands out, as it directly mitigates atmospheric distortions by leveraging the brightness temperature (BT) from two adjacent TIR channels at the top of the atmosphere. This method is frequently employed for producing operational LST products [18,19,20,21]

The satellite NOAA-21, designated Joint Polar Satellite System JPSS-2 prior to launch [22], was launched on 10 November 2022 [22] by the National Oceanic and Atmospheric Administration (NOAA). Its primary objective is to furnish comprehensive global environmental data, encompassing insights into weather patterns, atmospheric dynamics, and various environmental indicators. A scanning radiometer sensor onboard JPSS-2/NOAA-21 called VIIRS gathers visible and infrared imagery, as well as radiometric measurements of the land, atmosphere, and oceans. Interestingly, two of the 22 spectral bands on VIIRS, which range in wavelength from 0.4 to 12.5 m, are thermal infrared channels that will be used for LST retrieval.

In this study, an SW algorithm was developed for JPSS-2; validation and comparison with ground-based measurements verified the algorithm’s efficacy in providing accurate and reliable land surface temperature estimates over diverse landscapes and climatic conditions.

2. Methodology

2.1. Split-Window Algorithm for LST Retrieval

The SW technique, which is based on the differential absorption in two neighboring infrared channels, was initially developed for calculating sea surface temperature (SST) from satellite observations. Then, it was expanded to estimating land surface temperature. In this study, emissivity and water vapor effects have been taken into consideration by using the SW-LST algorithm structure described by Sobrino and Raissouni [23] to retrieve LST from VIIRS NOAA-21 data. This algorithm is written as follows:

$$T_S=T_{15}+c_1\left(T_{15}-T_{16}\right)+c_2{\left(T_{15}-T_{16}\right)}^2+c_0+\left(c_3+c_{4}W\right)\left(1-\epsilon \right)+\left(c_5+c_{6}W\right)\Delta \epsilon$$

(1)

where T_S is the earth surface temperature (in K), T₁₅ and T₁₆ are the at-sensor brightness temperatures (in K) of VIIRS NOAA-21, ε = (ε₁₅ + ε₁₆)/2 presents the mean effective emissivity, ∆ε = (ε₁₅ − ε₁₆) is the difference between the emissivities of VIIRS channels M15 and M16, W (g·cm⁻²) is the total atmospheric water vapor column, and c_k (k = 0, 1 … 6) are the SW algorithm coefficients.

2.2. VIIRS Sensor Characteristics

The Visible Infrared Imaging Radiometer Suite (VIIRS) is a whiskbroom radiometer designed for use onboard S-NPP, NOAA-20, NOAA-21, and future JPSS series satellites. The characteristics of the thermal infrared M15 and M16 bands that have been used in the SWA for LST retrieval are presented in

Table 1. The characteristics of the JPSS-2/NOAA-21 VIIRS M15 and M16 bands.

JPSS-VIIRS Band	Wavelength (µm)	Bandwidth (µm)	Spatial Resolution (m)
M15	10.763	10.26–11.26	750
M16	12.013	11.54–12.49	750

and Figure 1.

2.3. MODTRAN 4.0 Simulations

In order to determine the atmospheric parameters (downwelling and upwelling atmospheric radiances and atmospheric transmittance), the radiative transfer model (RTM) MODTRAN was used. The atmospheric profiles were extracted from the Thermodynamic Initial Guess Retrieval (TIGR) database [24]. To better characterize surface variations, the calculations were performed over a vast gradient of temperatures, T − 5, T, T + 5, T + 10, and T + 20 (taking into account the fact that T is the initial boundary layer temperature of the profiles). A total of 100 emissivities of various types of surfaces were taken from the Advanced Spaceborne Thermal Emission Reflection Radiometer (ASTER) spectral [25]. Furthermore, five different view angles (0°, 10°, 20°, 30° and 40°) and 54 atmospheric water vapor (W) values at nadir (varying between 0.15 g·cm⁻² and 4.65 g·cm⁻²) were used in the simulation in order to consider the viewing angle and atmospheric water vapor effects. Therefore, 135,000 simulation data were composed.

2.4. Numerical Coefficients and Sensitivity Analysis

We conducted a sensitivity analysis based on the error theory to assess the performance of the SW algorithm and the impact of the possible errors in LST estimation. The sensitivity analysis is given by the following equation:

$${\delta }_{Total}(T_S)=\sqrt{{\delta }_{a\lg }^2+{\delta }_{NE\Delta T}^2+{\delta }_{\epsilon }^2+{\delta }_W^2}$$

(2)

where δ_alg is the algorithm’s standard deviation, and δ_NE∆T, δ_W, and δ_ε are the impacts on total error due to uncertainties of sensor temperatures, atmospheric water vapor, and land surface emissivity, respectively. δ_NE∆T, δ_W, and δ_ε are expressed by the following equations:

$${\delta }_{\mathrm{NE}\Delta \mathrm{T}}=\sqrt{\left(\frac{\partial T_S}{\partial T_{15}}\right)e^2(T_{15})+\left(\frac{\partial T_S}{\partial T_{16}}\right)e^2(T_{16})}$$

(3)

$${\delta }_W=\left(\frac{\partial T_S}{\partial W}\right)e(W)$$

(4)

$${\delta }_{\epsilon }=\sqrt{\left(\frac{\partial T_S}{\partial T_{15}}\right)e^2({\epsilon }_{15})+\left(\frac{\partial T_S}{\partial T_{16}}\right)e^2({\epsilon }_{16})}$$

(5)

Thus, we assume that both the brightness temperature errors of the M15 and M16 channels e(T₁₅) = e(T₁₆) = 0.05 K or 0.01 K, the emissivity errors in the VIIRS channels M15 and M16 e(ε₁₅) = e(ε₁₆) are 0.01 or 0.005 [26], and the atmospheric water vapor content can be considered as e(W) = 0.5 g·cm⁻² [27].

3. Results and Discussion

3.1. Sensitivity Analysis

The SW coefficients (C₀ to C₆) of the developed SWA for LST estimation from the NOAA-21 satellite are presented in

Table 2. Split-window algorithm coefficients (C₀ to C₆) for JPSS-2/NOAA-21.

Satellite	λieff	λjeff	C0	C1	C2	C3	C4	C5	C6
JPSS-2/NOAA-21	10.763	12.013	−0.16	1.330	0.230	58.1	−0.57	−112	8.84

.

The results of the sensitivity analysis are shown in

Table 3. The sensitivity analysis of the parameters influencing JPSS-2/NOAA-21 LST estimation.

Satellite	λeff (µm)	λeff (µm)	R	δalg (K)	δNE∆T (K)	δε (1%)	δε (0.5%)	δW (K)	δTotal(Ts) (1%)	δTotal(Ts) (0.5%)
JPSS-2/NOAA-21	10.654	11.934	0.93	1.07	0.22	1.26	0.63	0.02	1.67	1.26

. Emissivity uncertainty is about 1.26 K and 0.63 K for e(ε₁₅) = e(ε₁₆) = 1% and e(ε₁₅) = e(ε₁₆) = 0.5%, successively. The total LST uncertainty, δ_Total(Ts), is about 1.67 K, considering e(ε₁₅) = e(ε₁₆) = 1%, and it is less than 1.26 K for e(ε₁₅) = e(ε₁₆) = 0.5%. Therefore, the uncertainty in emissivity has an insignificant effect on the LST estimation. Thus, an accurate knowledge of the surface is required.

3.2. LST Validation

The accuracy of the proposed SWA for LST retrieval from VIIRS/NOAA-21 is also evaluated using ground truth data sets from the Hay and Walpeup sites [28].

Figure 2 presents the LST retrieved from the NOAA-21 satellite using the developed split-window algorithm and the in situ LST ground data from two Australian sites (Hay and Walpeup), as well as the correlation coefficient R, bias, standard deviation differences (SDV), and root-mean-square error (RMSE). The results show that the split-window algorithm onboard JPSS-2/NOAA-21 can estimate LST with a bias of 0.97 K, a standard deviation difference of 1.31 K, and an RMSE of less than 1.71 K for the Hay and Walpeup site measurements, confirming the algorithm’s accuracy in LST retrieval.

4. Conclusions

An alternate split-window technique for LST estimation from NOAA-21 satellite data was proposed in this study. The algorithm coefficients were obtained from the simulation dataset of atmospheric profiles. To assess the performance of the SW-LST method, a sensitivity analysis was performed.

The LST’s accuracy derived was validated using ground truth data sets from two Australian sites. The recovered LST shows a good fitting with the in situ LST at both sites. The bias and RMSE are, respectively, 0.97 and 1.71 K. This indicates that this algorithm offers an alternate and feasible method for retrieving LST using NOAA-21 satellite data. Nonetheless, more LST validation under various atmospheric conditions and surface types is required to adequately evaluate the efficacy of this approach.