Evaluation of Cirrus Cloud Detection Accuracy of GOSAT/CAI and Landsat-8 with Laser Radar: Lidar and Confirmation with Calipso Data

Cirrus cloud detection accuracy of GOSAT/CAI and Landsat-8 is evaluated with a ground based Laser Radar: Lidar data and sky view camera data. Also, the evaluation results are confirmed with Calipso data together with a topographic representation of vertical profile of cloud structure. Furthermore, origin of cirrus clouds is estimated with forward trajectory analysis. The results show that GOSAT/CAI de4rived cirrus clouds is not accurately enough due to missing of cirrus cloud detection spectral channel while Landsat-8 derived cirrus cloud. Keywords—Cirrus cloud; GOSAT/CAI; Landsat; LiDAR; Sky view camera; Calipso; topogramphic representation of 3D clouds


INTRODUCTION
Cloud detection is one of though issues in satellite remote sensing in particular for cirrus clouds [1]- [16]. It is not so easy to detect cirrus clouds in particular for remote sensing satellite onboard instruments. In order to detect cirrus clouds, 1.38 micrometer wavelength channel is adopted for Moderate resolution of Imaging Spectrometer: MODIS1 and Landsat-8 Operational Land Imager: OLI 2 , etc. Green house gasses Observation Satellite / Cloud and Aerosol Imager: GOSAT 3 /CAI 4 is dedicated sensor for cloud and aerosol retrievals. Because that GOSAT/FTS (Fourier Transform Spectrometer 5 ) data is affected by clouds and aerosols, GOSAT/CAI is carried on the same platform of GOSAT satellite. Therefore, cloud flag and its confidence level are evaluated from the GOSAT/CAI and provide as Level 2 of GOSAT products together with Level 1B product as a source of Level 2 product. As mentioned above, it is not so easy to detect cirrus clouds. Although cirrus detection wavelength channel (1.38  Light Detection and Ranging、Laser Imaging Detection and Ranging: LiDAR data which allows measurement of back scattering ratio and depolarization ratio is used [17]- [29]. The ground based LiDAR is equipped at one of the GOSAT validation sites which is situated at Saga University, Japan. Therefore, vertical profile of aerosol particles as well as cloud particles are detected which results in detection of aerosols and clouds including cirrus clouds. Meantime, sky view camera observes hemispherical cloud conditions. Although it is possible to detect thick clouds, it is not easy to detect cirrus clouds with sky view camera. Vertical cloud structure can be retrieved with Cloud Aerosol Lidar and Infrared Pathfinder Satellite Observations: Calipso 6 data. Therefore, detected cirrus clouds can be validated with Calipso data. In this paper, a specific representation of vertical cloud structure is proposed. That is to representation of the retrieved structure on the topographic map which is projected on the globe. Forward trajectory analysis is also made for retrievals of the original source areas of the cirrus in concern through consideration of atmospheric conditions. In the next section, the proposed method for evaluation of cirrus detection accuracy of GOSAT/CAI and Landsat-8 is described followed by experiments (method and procedure as well as the results from the experiments). Then validation of the evaluation results with sky view camera data and Calipso data is described followed by the specific representation of vertical cloud structure on the earth. Finally, conclusion and some discussion are followed.

A. GOSAT/FTS and CAI
GOSAT satellite is operating since January 23 2009 as the joint project among Ministry of Environment, JAXA and National Institute Environmental Science: NIES. GOSAT carries FTS and CAI as mission instruments as shown in Fig.1.
Major mission of GOSAT is to measure total column of carbon dioxide and methane which can be done with FTS instrument. In order to avoid influence due to aerosols and clouds, TANSO/CAI is also carried on GOSAT. www.ijarai.thesai.org

B. GOSAT Validation Site
There are TCCON validation sites in the world. One of these is Saga University site in Japan. The location is shown in Fig.2. Fig.3 shows the LiDAR site (Laser light and the container in which LiDAR is equipped). Examples of the LiDAR data are shown in Fig.4 (a) together with PM2.5 data, CAI imagery data, the time series of PM2.5 data, and the sky view camera image. The right bottom graphs are the LiDAR data which is acquired at 14:00 Japan Standard time on May 29 2014. On the left, there is back scattering ratio data is situated while depolarization ratio is shown on the right. From these back scattering ratio and depolarization ratio, aerosol distribution and cloud vertical profile can be retrieved. Therefore, LiDAR data derived cloud vertical profile can be used for validation of CAI data derived clouds and Landsat-8 data derived clouds in particular, cirrus clouds. Meanwhile, Fig.4 (b) shows the LiDAR data which is acquired on April 26 2015. There is a peak of back scattered photon counts at around 10km of elevation (altitude above sea level). It is cirrus clouds. There is the ground based FTS for GOSAT validation which is situated just beside the LiDAR as shown in Fig.5. Other than these, there are sky view camera, sky radiometer which allows estimation of aerosol particle size distribution and refractive index retrievals for GOSAT validation. Examples of sky camera imagery data are shown in where F i denotes statistical test results which are shown in Table 1.  Fig.8 shows examples of the acquired color images of GOSAT/CAI imagery data together with sky view camera images which area acquired at the Saga University validation site at the same time as satellite over pass time. Although Landsat-8 OLI image shows the cirrus clouds pixels in Fig.10 (b), Fig.10 (c) does not indicate any cirrus cloud at all. On the other hand, LiDAR data shows evidence of cirrus cloud existing as shown in Fig.10 (e). Therefore, it may say that Landsat-8 band 9 of cirrus channel does work to detect cirrus cloud while GOSAT/CAI does not work for detection of cirrus cloud due to missing cirrus channel.

B. Adjectment of Acquisition Time Difference Between
Landsat-8 and GOSAT Local mean times of the orbits of Landsat-8 and GOSAT are different each other for 30 minutes. Therefore, some adjustment of the time difference between both is required. By using forward trajectory analysis software tool provided by NOAA, original positions of cirrus cloud (30 minutes before the acquisition time) are estimated. The results from the forward trajectory analysis are shown in Fig.11. For January 20 2015, there is North-West wind while there is North-East wind for April 26 2015.
Therefore, the cirrus cloud locations are shifted for the distance which is shown in Fig.11 within 30 minutes in those directions. Thus the cirrus cloud detection accuracy can be done through comparisons between LiDAR data and Landsat-8 OLI data which is acquired at 30 minutes apart from the LiDAR acquisition.

C. Summary of the Experimental Results
LiDAR data are acquired for 173 days within 518 days from April 1 2014 to August 31 2015 (Revisit cycle of the GOSAT satellite is 3 days). Within 173 days, LiDAR data are acquired 48 days (Acquisition ratio of LiDAR data to the total available days is just 33.01%). Cirrus clouds are observed for 11 days out of 48 days. Meanwhile, cirrus clouds are detected with CAI for just 8 days out of 11 days. On the other hand, cloud free situations are found with CAI for 18 days out of 37 days which is confirmed with LiDAR data. Due to the fact that the revisit cycle of Landsat-8 satellite is 16 days, just two match-up data between LiDAR and Landsat-8 OLI are collected for check cirrus detection accuracy. Two of match-up data show good coincidence between Landsat-8 OLI data utilized cirrus detection (no cirrus cloud and cirrus cloud existing situations).

D. Another Comparison Between Landsat-8 OLI data and GOSAT/CAI Imagery Data
Another match-up data between Landsat-8 OLI and GOSAT/CAI imagery data is found for April 7 2014 (Unfortunately LiDAR data is not acquired on that day). Fig.12 (a) shows GOSAT/CAI imagery data, (b) shows Landat-8 OLI imagery data on that day. Meanwhile, Fig.12 (c) shows the sky view camera image while (d) shows the results from the forward trajectory analysis for adjustment of the data acquisition time difference of 30 minutes between GOSAT/CAI and Landsat-8 OLI data.  Fig.12 (c), there are thick clouds in the sky above the test site at the GOSAT satellite over pass time. These, however, are not cirrus clouds at all. Forward trajectory analysis result shows that there is West wind at that time.

E. Confirmation of Cirrus Cloud Detection Capability with Calipso Data
Cirrus clouds can be confirmed with Calipso data. As shown in Fig.13, vertical profile of the existing clouds are investigated with Calipso data. The horizontal axis of the Fig.13 is sub-satellite track while vertical axis shows back scattered phone count from the cloud particles.

IV. CONCLUSION
Cirrus cloud detection accuracy of GOSAT/CAI and Landsat-8 is evaluated with a ground based Laser Radar: Lidar data and sky view camera data. Also, the evaluation results are confirmed with Calipso data together with a topographic representation of vertical profile of cloud structure. Furthermore, origin of cirrus clouds is estimated with forward trajectory analysis. The results show that GOSAT/CAI de4rived cirrus clouds is not accurately enough due to missing of cirrus cloud detection spectral channel while Landsat-8 derived cirrus cloud.