FY-4A/AGRI has three flexible scanning modes: full disk (images of the whole Earth as seen from satellite) every 15 min, China area (3–55 °N, 70∼140 °E) every 5 min and target area (1000 km × 1000 km) every 1 min. GOES-R’s Advanced Baseline Imager (GOES-R /ABI) has similar scanning modes: full disk every 15 min and the continental U.S. every 5 min. When storm activity is present, GOES-R/ABI has two smaller scan areas every 1 min (or one every 30 s). In comparison, Himawari 8-9/AHI has more complicated scanning modes including full disk every 10 min, the Japan area every 2.5 min, target area every 2.5 min, and two landmark areas every 30 s (Bessho et al., 2016).

AGRI, ABI, and AHI have extremely flexible scanning modes. For full disk observation, AHI can scan up to every 10 min while AGRI and ABI scan every 15 min. For native continent observation, AGRI and ABI can scan every 5 min while AHI can scan every 2.5 min. In addition, AHI, ABI, and AGRI have similar scanning modes (from every 0.5 to every 1 min) for target area observation. However, at present only full disk AHI data are available to users worldwide, which limits their application in the Asia and Pacific area.

In this paper, FY-4A/AGRI is further compared with two LEO instruments: FY-3/MERSI and NPP/VIIRS. The Medium Resolution Spectral Imager (MERSI) is an important instrument on board the Chinese new-generation LEO satellite FY-3 series. The number of spectral bands, the spatial resolution and the radiometric accuracy of MERSI increase from FY-3A to FY-3D (for FY-3D: 25 channels; spatial resolution 250–1000 m; swath width 2916 km). The Visible Infrared Imaging Radiometer Suite (VIIRS) is the main instrument on Suomi NPP and JPSS (22 channels; spatial resolution 370–1000 m; swath width 3060 km; Cao et al., 2013). As the U.S. and China’s second-generation medium-resolution imagers, VIIRS and MERSI are responsible for monitoring the reflective and radiative properties of land, atmosphere, ice, and ocean. Their high spatial and multi-spectral properties give MERSI and VIIRS traditional advantages in detecting land-surface change such as fire dynamics, vegetation change, and ocean color.

As two typical LEO instruments, MERSI and VIIRS scan the Earth twice a day. In comparison, AGRI has more flexible scan modes. Therefore, AGRI has obvious advantages over MERSI and VIIRS for temporal resolution. Through further multi-temporal synthesis, clearer sky images can be acquired by AGRI, which are particularly suited to monitoring rapid changes in the land surface and atmosphere.

We compare the spectral channels and spatial resolutions of AGRI, MERSI, and VIIRS so as to make best use of these different sensors in operational work (Table 2). FY-4A/AGRI has 11 spectral channels while FY-3D/MERSI has 25 and VIIRS has 22. AGRI has 3 spectral channels in the visible and near-infrared, whereas MERSI has up to 15 and VIIRS up to 10. MERSI and VIIRS are more suitable for ocean color monitoring due to the extra channels. The spatial resolution of the AGRI 0.65 μm channel is 500 m, approaching that of the similar channel of MERSI (250 m) and VIIRS (370 m). In the shortwave infrared region, AGRI has similar channels to MERSI and VIIRS, but MERSI and VIIRS have better spatial resolution (2 km for AGRI, 1 km for MERSI and VIIRS). MERSI and VIIRS, but not AGRI, have a special channel (1.24 μm) that is sensitive to cloud particle size. AGRI has two channels in the mid-infrared (“3.75 μm high” and “3.75 μm low”) whereas MERSI and VIIRS have only one channel (3.84 μm for MERSI and 3.74 μm for VIRRS). AGRI has two water-vapor channels in the 6–8 μm range, whereas MERSI and VIIRS do not, demonstrating that AGRI still has advantages in atmospheric monitoring. In the thermal infrared region, AGRI, MERIS, and VIIRS have an 8.5 μm channel and two “split-window” channels, which are helpful for dust, volcanic ash detection, and retrieving cloud top height as well as cloud micro-physical parameters. However, the spatial resolution of AGRI is only 4 km in the thermal infrared region, which is just one fourth that of MERSI and VIIRS. The spatial resolution in the thermal infrared region is the main limitation of AGRI for non-meteorological applications.

Aerosol is an important component of the atmosphere that significantly influences global climate by affecting the radiative energy balance and hydrological cycle (e.g., IPCC, 2007). Aerosol is also the main component of smog and air pollution. The efficient removal of the contribution of the land surface is the first step in aerosol retrieval. The standard methods for this include the Dark Target (DT) algorithm, the multi-angle polarization algorithm (Kahn et al., 1998), and the Deep Blue (DB) algorithm (Remer et al., 2005). The DT algorithm uses the linear relationship between the blue and red channels to effectively remove the contribution of the land surface and has been used in developing the Moderate Resolution Imaging Spectroradiometer (MODIS) aerosol products (Levy et al., 2010). Since FY-4A/AGRI also has the corresponding channels for DT algorithm retrieval, an aerosol optical depth (AOD) product has been developed based on the DT algorithm and FY-4A/AGRI data (the FY-4 aerosol product). Under the assumption of a plane-parallel atmosphere and Lambertian land surface conditions, the FY-4 aerosol product uses the traditional DT algorithm to retrieve clear sky AOD over a dark land surface (e.g., vegetation).

Figure 1 shows a typical AOD retrieval result from FY-4A/AGRI. High AOD values are located mainly in central and eastern China, which agrees well with the distribution of haze on that day. This result is also consistent with the results derived from Himawari/AHI and MODIS with the same algorithm (not shown). The scatter plots in Figure 2 demonstrate that FY-4A/AGRI AOD and AHI AOD are closely correlated (r = 0.925). Similarly, the relationship between FY-4A/AGRI AOD and MODIS AOD is strong, with r up to 0.89. This case study shows that FY-4A/AGRI has comparable accuracy with the widely used GEO and LEO satellites. Compared with LEO instruments, FY-4A/AGRI has better temporal resolution; therefore, FY-4A/AGRI has great potential in detecting hourly variations in aerosol.

The FY-4A/AGRI dust detection product combines 12 widely used dust detection indexes with the radiative properties measured by FY-4A/AGRI (Zhang et al., 2006; Chen et al., 2016, 2017). The probability density function (PDF) and cumulative probability density function (CDF) are calculated for different backgrounds including cloud, dust, vegetation, and desert. A reliability index is attached to each dust index according to its threshold. Finally, the FY-4A/AGRI dust detection product is the combined result of the 12 dust detection indexes and a corresponding reliability index, so as to avoid missed detection and false detection. FY-4A/AGRI dust detection products include dust detection and the Infrared Difference Dust Index (IDDI). Figure 3 shows a dust event in north China on 4 May 2017. Ground-based dust observation data are used to validate the satellite retrieval of IDDI. The result shows that the FY-4A/AGRI dust product can detect 88% of the ground-based dust observations. This case study shows that the FY-4A/AGRI dust product can effectively detect dust areas, especially under cloud-free conditions. However, when the dust layer is under or mixed with meteorological cloud, the false detection rate is higher.

Volcanic ash has serious impacts on aviation. As volcanic ash disperses, it also has a long-term influence on climate and the environment (Zhu et al., 2011). FY-4A/AGRI offers a dynamic method of observing visible to infrared signals of volcanic ash from space with high spectral and spatial resolution. A new algorithm for retrieving volcanic ash cloud height from FY-4A/AGRI thermal infrared channels has been developed from the simulation of volcanic ash microphysical properties and statistical optimal estimation theory (Zhu et al., 2017). Unlike previous methods, the new FY-4 ash height algorithm does not require simultaneous atmospheric profiles, providing a flexible way to estimate volcanic ash height using passive satellite infrared measurements (Zhu et al., 2017).

A case study was conducted during the eruption of Indonesia’s Agung volcano (8.34°S, 115.51°E) on 27 November 2017. Large amounts of volcanic ash were being released by Agung volcano at 00:00 UTC. The main body of ash was spreading to the east and had already affected the adjacent Lombok island, 110 km from the crater of the volcano. FY-4A/AGRI’s volcanic ash height retrieval result shows that the volcanic ash was at a height of 1.86–8.38 km, with 96% of the total pixels in the range 2–6 km (Figure 4). The FY-4 ash height algorithm was also applied to Himawari/AHI data, as shown in Figure 5. The height of volcanic ash derived from Himawari/AHI is in the range 2.52–8.40 km, with 81% of the total pixels in the range 2–6 km. The two results display similar spatial variations and have comparable accuracy, with a correlation coefficient of 0.58 and a standard deviation of 0.57 (Figures 4–6). Positive bias is observed for FY-4A/AGRI at lower heights, and FY-4A/AGRI tends to under-estimate the ash height at greater altitudes. One possible reason for the difference between the FY-4A/AGRI and Himawari/AHI ash heights is that we did not account for the spectral response differences between the two instruments. Another reason may be related to the spatial resolution in the thermal infrared channels, which is 2 km for Himawari/AHI and 4 km for FY-4A/AGRI. In addition, positioning deviation between FY-4A/AGRI and Himawari/AHI can give rise to differences in final retrievals, especially at the edge of the volcanic ash region.

According to weekly reports from both the Global Volcanism Program¹ and the Darwin VAAC (Volcanic Ash Advisory Center), ash plumes rose over 6 km above the crater rim on 27 November. The ash height derived from FY-4A/AGRI generally agrees well with these reports.

For the mid-infrared bands, FY-4A/AGRI uses a linear conversion between Digital Number (DN) and a quantitative physical value (such as radiance), which cannot strike a balance for the detection of both high- and low-temperature objects (Lu et al., 2017). Therefore, two mid-infrared bands were designed for FY-4A/AGRI: channel 7 (3.72 μm, high) and channel 8 (3.72 μm, low). The spatial resolution of channel 7 is 2 km. Its dynamic range mainly covers the high-temperature range (>290 K) so as to enhance the ability to detect fire and strong reflection from the sun. In contrast, the spatial resolution of channel 8 is 4 km and its dynamic range covers low temperatures (Lu et al., 2017). Channel 8 is more suitable for relatively low-temperature detection such as fog, low cloud and fire with a very low temperature (<290 K).

According to Wien’s displacement law, there is an inverse relationship between the peak wavelength of black body emission and its temperature. FY-4A/AGRI’s mid-infrared sensors with a central wavelength at 3.72 μm can respond strongly to thermal emission from the surface at magmatic temperatures, even though those surface areas are very small. The FY-4A/AGRI fire product uses three main criteria to detect fire: (1) when there is fire, the radiation in the mid-infrared channel will increase and (2) the radiation in thermal infrared channels will also increase, but not as much as that in the mid-infrared channel (Dozier, 1981); and (3) the increase between thermal infrared channels will also differ. A weighted mean is therefore calculated for the radiation from the thermal and mid-infrared channels. When there is fire, the increase of the weighted mean for a fire pixel will be more significant than for a no-fire pixel. In this way, fires can be detected from FY-4A/AGRI multi-channel radiative information.

A case study was conducted for 25 September 2017, when an area of anomalously high land-surface temperature was detected east of Inner Mongolia (50.36°N, 129.56°E) at 15:00 Beijing time (07:00 UTC). At 15:30 Beijing time (07:30 UTC), the hot spot (red spot in Figure 7) was approaching the border. Using the sub-pixel area calculation method, it is estimated that the total area of the anomalously high temperature is 1.76 hm². The FY-4A/AGRI fire detection result was further confirmed by the local fire prevention department as a grass fire at that location. Then the fire developed very quickly. At 16:00 Beijing time (08:00 UTC), the fire density reached its maximum with a total burned area of 2.32 hm².

The performance of AGRI for fire detection was compared with Himawari/AHI using the same fire detection algorithm and at the same time (Figure 8). The fire location derived from FY-4A/AGRI agrees well with that from Himawari/AHI, indicating FY-4A/AGRI’s great potential for automatically detecting the location of fire. However, there is still an obvious difference between the two instruments in the estimate of total burned area (Table 3), possibly caused by different dynamic response ranges between FY-4A/AGRI channel 7 and Himawari/AHI channel 7, and different satellite viewing angles.

The reflectivity of water in the visible and near-infrared spectral region is very different from that of vegetation. The reflectivity of water is generally very low (4–5%) and decreases with increasing wavelength. In the red to near-infrared region, especially the latter, water absorbs nearly all the light (Sun et al., 2016). Relatively high reflectivity is mainly concentrated in the blue and green regions. FY-4A/AGRI uses the red and near-infrared bands to extract water information. Under clear sky conditions, an empirical threshold is set to detect a water area. When thin cloud is present, the ratio of the near-infrared to red bands (500 m resolution) is used to detect water (Sheng et al., 1998).

Figure 9 is a case study of the detection of Poyang and Dongting lakes in China on 30 October 2017 using both FY-4A/AGRI data and FY-3B/MERSI data. The spatial distribution of water area from FY-4A/AGRI (500 m resolution) agrees well with that from FY-3B/MERSI (250 m resolution), indicating that FY-4A/AGRI performs well in water detection. The area of Poyang Lake extracted from FY-4A/AGRI is 2626 km², 3.9% less than that extracted from FY-3B/MERSI (Table 4). The Dongting Lake area is 1167 km², 10% less than that extracted from FY-3B/MERSI. The mean deviation between the FY-4A/AGRI and FY-3B/MERSI detected water area is 6.9%, due to the different spatial resolutions of the two instruments.

Cyanobacteria (also known as blue-green algae) are a type of primitive photosynthetic aquatic organism. Their toxicity and rapid growth properties, have made them a common ecological problem for shallow freshwater water bodies such as lakes. Cyanobacterial blooms have been reported frequently for Taihu, Caohu, and Dianchi lakes in China (Shi et al., 2017). Under low wind conditions intense cyanobacterial blooms often form floating mats at the water surface that cover large areas and can change rapidly, making it difficult to monitor the dynamic change from LEO satellites alone (Hu, 2009). FY-4A/AGRI, with its greater temporal resolution, offers a new data source for dynamic monitoring of floating algae.

Floating algae have similar spectral reflectance properties to vegetation when they accumulate on the water surface demonstrating high reflectivity in the near-infrared channel and relatively low reflectivity in the visible channels. Therefore, a vegetation index such as the Normalized Difference Vegetation Index (NDVI) can also be used in floating algae detection. The relationships between NDVI and floating algae are determined by fitting for pure water conditions and for a complicated background water condition. A self-adapted model is developed based on the fitted relationships using FY-4A/AGRI data to dynamically detect floating algae.

Figure 10 shows a case study of detecting the whole life cycle of a floating algal bloom on Taihu Lake based on FY-4A/AGRI 1 km spatial resolution and high temporal resolution data. The floating algae appeared at 08:00 (Beijing time) 06 May 2017 and developed very quickly. Four hours later, the density and area of floating algae reached their peak. After that, the area of floating algae gradually reduced and the density also weakened. The relationships between NDVI and floating algae developed from FY-4A/AGRI data were also applied to Himawari/AHI data for a preliminary comparison of the performance of FY-4A/AGRI and Himawari/AHI in floating algae detection. The location and area of floating algae detected by Himawari/AHI agree well with those detected by FY-4A/AHI (Figure 11). Area affected by floating algae estimated from FY-4A/AGRI and from Himawari/AHI display a similar trend (Table 5). Compared with Himawari/AHI, FY-4A/AGRI seems to overestimate the total area affected in the early stage of the floating algal bloom while it underestimates the area affected by floating algae during the weakening stage. At the peak of the floating algal bloom, there is only a very slight difference between Himawari/AHI and FY-4A/AGRI, indicating a good capability of FY-4A/AGRI for detecting floating algal blooms over a large area. Since the FY-4A/AGRI floating algae detection algorithm was directly applied to Himawari/AHI data, the differences in spectral response function and satellite view angle between the two instruments were not taken into account in the floating algae detection algorithm. This was probably the main cause of the final difference in floating algae location detection, area estimation, and density grading between the two instruments.