**1. Introduction**

At 3:22 am UTC on 22 December 2016, two wide-swath pushbroom hyperspectral imaging microsatellites, SPARK-01 and -02, which were manufactured by the Shanghai Engineering Center for Microsatellites, were successfully launched at the Jiuquan satellite launch center by the CZ-2D rocket. The spectrometers on the satellites were developed by the Academy of Opto-electronics, Chinese Academy of Sciences, less than one year previously. SPARK-01 and -02 have spectral ranges of 400–1000 nm, a swath of ~100 km, a spatial resolution of 50 m and 2048 pixels along the cross-track direction. The spectrometers use prisms to split the beam into different bands, and thus, the spectral resolution (or full width at half maximum, FWHM) varies from 1 to 10 nm. Figure 1 shows a schematic of the satellite; major satellite characteristics are described in Table 1.

**Figure 1.** Diagram of the SPARK satellite.


**Table 1.** Main characteristics of SPARK satellite and imaging sensor.

The SPARK satellites are lightweight and inexpensive. They provide the advantages of fine spectral resolution and large swath. These two hyperspectral satellites can be used for applications such as environmental and disaster monitoring, target detection, and precise classification. They provide basic information to support quantitative applications, resource exploration, and business applications [1]. However, due to size, weight, and cost limitations, SPARK-01 and -02 do not have

on-board calibration systems. Also, complete preflight radiometric calibrations were not performed in the laboratory due to the short manufacture time and prioritization of more urgen<sup>t</sup> tasks before the satellite launch. Only the spectral calibration for each detector was conducted by a monochromator in the laboratory. The spectral response curves followed the Gauss function quite well after data processing, and thus, the central wavelengths and the full-width at half-maximum (FWHM) of the SPARK satellites were determined. The averaged central spectral wavelengths of these two satellites are slightly different (Figure 2). Moreover, the spectral smile effect is minor for SPARK-01 but is evident in SPARK-02 (Figure 3). This aspect should be considered in the data processing flow. Therefore, in-orbit vicarious calibration must be used to transform the satellite data into meaningful physical information. Previous studies used reflectance-, irradiance-, and radiance-based techniques [2,3] to successfully calibrate satellites such as the SPOT HRV [4], Landsat TM/ETM [3,5,6], Airborne Visible and Infrared Spectrometer [7], EO-1 Hyperion [8,9], and FY [10,11], MISR [12], Landsat OLI [13], CBERS-4 [14], and many other optical remote sensors [15]. The reflectance- and irradiance-based methods have been compared with cross-calibration methods to derive the calibration coefficients for the BJ-1 microsatellite [16]. The results showed the irradiance-based method to be superior to the reflectance- based method, especially under low-visibility atmosphere conditions. In reality, vicarious calibration methods have always been used in combination with the pre-launch calibration and on-board calibration to determine calibration accuracy and monitor the sensor's radiometric stability [17–19]. Apart from the vicarious calibration methods frequently applied to multispectral remote sensing satellites, some novel methods for hyperspectral sensors have also been proposed in recent years, such as the improved irradiance-based method [20] and supervised vicarious calibration method [21,22]. A distinguishing characteristic of a hyperspectral sensor is its high spectral resolution, and spectral smile effect and spectral shift may greatly affect the radiometric accuracy near the atmospheric absorption wavelength regions [7,23,24]. Due to the lack of on-board calibrator and pre-launch radiometric calibration, the in-orbit calibration of the SPARK-01 and -02 satellites was achieved via a calibration experiment performed at the dry Dunhuang site in the Gobi Desert in western China from 28 February to 10 March 2017. In-situ measurements, including both ground reflectance and atmospheric parameters, were also acquired during this calibration period. Two vicarious calibration methods (i.e., reflectance-based and irradiance-based) were used independently to predict the top-of-atmosphere (TOA) radiance (*LTOA*) using MODTRAN® 5 software. The vicarious method results were then used to obtain the final SPARK-01 and -02 calibration coefficients.

**Figure 2.** Central spectral wavelengths of various SPARK-01 and -02 bands.

**Figure 3.** Cross-track central spectral wavelengths for channels centered near 760 nm for SPARK-01 (**a**) and -02 (**b**), respectively.

#### **2. Calibration Site and Measurements**

Three simultaneous measurement datasets from the Dunhuang calibration site were required for the SPARK radiometric calibrations: raw data from the SPARK satellites, surface reflectance measurements, and atmospheric measurements. Also, in order to correct for non-uniform phenomenon detection due to differing detector responses, two more observations were performed around the time when the calibration experiment occurred: a 90◦ yaw observation or slide slither over the bright desert region during daytime and a dark current observation over the open ocean during nighttime. Use of the 90◦ yaw observation is efficient for correcting the non-uniform radiometric response among different detectors. This technique has been utilized for Hyperion [9], Quickbird [25], RapidEye [26], and Landsat 8 [27]. Using this technique, all the pixels along the cross-track direction would observe nearly the same scene. Owing to the wide swath (~100 km) of the SPARK satellites, it is difficult to find a uniform ground site wider than 100 km to permit normalization of the different responses among pixels in the cross-track direction. Thus, 90◦ yaw observation is necessary to perform the relative radiometric calibration. The surface reflectance measurements were conducted by a spectroradiometer (FieldSpec-4, ASD Inc., Longmont, CO, USA) one hour before and after the SPARK satellite overpass. The atmospheric measurements were acquired by a CE318 sunphotometer, a Microtops II sunphotometer (Solar Light Company, Inc., Glenside, PA, USA), an irradiance sphere combined with an SVC GER1500 spectrograph and radiosonde balloons. The details are illustrated as follows.

## *2.1. Calibration Site*

The Dunhuang calibration site (40◦5 32.80"N, 94◦23 35.78"E) is located on the eastern edge of the Kumutage Penniform Desert, which is in the Gobi Desert in northwestern China, about 35 km west of the city of Dunhuang, Gansu Province. The calibration area is approximately 1.2 km above sea level. The entire vicarious calibration target area (30 km × 30 km) is situated on a stabilized alluvial fan (see Figure 4). The area used for the vicarious calibration measurements for the high- and medium-spatial resolution sensors is approximately 400 m × 400 m and is located in the center of the alluvial fan; the surface is covered by cemented gravels. Several years ago, this calibration site was protected by the addition of protective fens along the edges to form a 500 m × 500 m square region.

(**b**) 

**Figure 4.** Dunhuang calibration site for medium-high resolution satellites, as illustrated in a Landsat 8/OLI image acquired on 2 February 2017. (**a**) Scaled subset image; (**b**) 5× magnification of a portion of the original image, where the red rectangle is the outline of the calibration site.

Figure 4 shows a Landsat/OLI image of the Dunhuang calibration site in which the surrounding fens can just be discerned. The local atmosphere is dry with low aerosol loading, which is beneficial for the calibration experiments. The atmospheric aerosol characteristics at the site are typical of a rural continental location, although some larger particles have been observed, possibly originating from sand dunes located to the northwest [10,11,28,29].

#### *2.2. SPARK Satellite Observations*

SPARK-01 and -02 data were acquired over the Dunhuang calibration site at 06:48:30 UTC on 7 March 2017 and at 06:52:32 UTC 28 February 2017, respectively. The dark current data and 90◦ yaw data for the relative calibration were acquired on 13 March and 11 March 2017, respectively, for the SPARK-01 satellite and on 27 February and 28 February 2017 for the SPARK-2 satellite. Figure 5 shows the SPARK-01 and -02 raw data; a number of vertical strips are evident. Clouds are evident in the SPARK-2 image over the southern and eastern areas of the calibration site. Although these atmospheric conditions are not ideal for SPARK-02 calibration, the observations over the calibration site were not affected by either clouds or shadows (Figure 5b), and, thus, the calibration results are expected to be comparable. Detailed imaging information for the calibration site and the relative radiometric calibration is listed in Tables 2 and 3, respectively.

**Figure 5.** *Cont.*

(**b**)

**Figure 5.** Subsets of SPARK-01 and -02 data acquired over the Dunhuang calibration site featuring pseudo color composited from the 141 (856.60 nm), 111 (648.60 nm) and 84 (550.30 nm) bands. (**a**) SPARK-01 data acquired on 7 March 2017 at 06:48:30 UTC; (**b**) SPARK-02 data acquired on 28 February 2017 at 06:52:32 UTC. These images were 180◦ rotated from the original raw data to maintain the northern and eastern directions on the top and the right hand, respectively.





#### *2.3. Ground Reflectance Measurements*

In-situ ground surface reflectance was measured over a 400 × 400 m square region one hour before and after the SPARK satellite overpass. The surface consists of cemented gravels of different colors and sizes (from mm to cm), as well as sand just beneath the gravel (Figure 6a). The measurements were taken by an ASD, Inc. spectroradiometer along a fixed route as shown in Figure 6b. Adjacent measurement points were ~40 m apart, and 10 measurements were taken around each measurement point. As a result, a total of nearly 1000 surface measurements were acquired from the Dunhuang calibration site. Thorough site measurements were repeated several times during the experiment in order to verify the stability of the surface reflectance. Measurements taken under clear atmospheric conditions were examined carefully and, after the exclusion of any erroneous measurements, averaged to produce the average reflectance of the calibration site. The ground reflectance measured on different dates during the experimental period is shown in Figure 7. The ground reflectance is relatively stable on different dates, with differences of less than 2%. The desert reflectance during the experimental period was also measured on 7 March 2017 as shown in Figure 8. These data were used to verify the radiometric calibration results.

**Figure 6.** (**a**) Photo of the calibration site; (**b**) Schematic of the surface measurement route.

**Figure 7.** Ground reflectance at the Dunhuang calibration site on different dates. Measurements on 7 March and 28 February 2017 correspond to SPARK-01 and SPARK-02, respectively.

**Figure 8.** Ground reflectance measured on 7 March 2017 over the desert area south of the Dunhuang calibration site.
