Layer detection algorithm for CALIPSO observation based on automatic segmentation with a minimum cost function

Highlights

• A layer detection algorithm is proposed based on automatic segmentation.

• We determine most of the cloud and aerosol layers at 1 and 5 km resolutions.

• The layer we detected is 170 m thicker than the CALIPSO product.

Abstract

CALIPSO (cloud-aerosol lidar and infrared pathfinder satellite observation) provides unique opportunities for profiling global cloud and aerosol. It is crucial to accurately detect the boundaries of cloud and aerosol layers from CALIPSO observation because the detecting error will be passed to further retrieval. Considered superior to other layer detection methods, the threshold method is the core of the selective iterated boundary location (SIBYL) algorithm developed for producing the CALIPSO official products. However, the threshold method can miss many tenuous layers, and the use of the slope method to refine the layer base in SIBYL leads to considerable uncertainty due to its high sensitivity to noise. This study proposed a new layer detection algorithm based on an automatic segmentation method with a minimum cost function. Results show that the new algorithm determines 21% and 13% more layers than SIBYL at 1 km and 1-5 km resolution, respectively, which indicates that the new algorithm has higher detection efficiency. Moreover, the layers detected by the new algorithm are 170 m thicker than that detected by SIBYL on average, which indicates that the SIBYL misses layer edges where the signal to noise ratio is low. The new algorithm can improve the accuracy and resolution of the layer products of CALIPSO as well as other space-based lidars.

Introduction

Cloud and aerosol layers play an essential role in local weather and global climate change, but the accurate estimation of their impacts is still challenging [1]. Accurate 3D observation of clouds and aerosols is critical to assess their meteorological and climatic impact and feedback [2,3]. Passive satellites are commonly used to capture global cloud and aerosol information. However, thin optical layers often failed to be detected [4], and the cloud base and top heights are difficult to be accurately measured [5]. The Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) sensor onboard the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) satellite is a unique active remote sensor that provides global clouds and aerosols profiles with high vertical resolutions [6], [7], [8]. However, the signal-to-noise ratio (SNR) of CALIPSO is very low due to the effects of the long detecting range and intense solar light in the daytime. Thus, it is still a challenge to accurately detect cloud and aerosol boundaries from CALIPSO observation. If this problem remains unresolved, the detecting error will be passed to further retrieval [9].

At present, many methods have been developed for layer detection, including the slope method [10], the wavelet method [11], the simple multi-scale method [12], and the threshold method [13,14], etc. The slope method identifies the layer based on the signal slope, which can be directly applied to un-calibrated raw data. However, when the slope oscillates to zero, layer detection would be difficult. The wavelet method identifies discontinuous signal points by continuous wavelet transform, which is insensitive to noise. However, the local modulus maxima of the wavelet correspond to the fastest-changing bins, which are not precise layer boundaries because changes near the boundaries are usually slow. The simple multi-scale method identifies the layer boundaries by determining the trend index of the signal [12]. The method is relatively insensitive to noise but cause bias due to missing attenuation losses within a layer. The threshold algorithm determines a layer top when the signal is continually larger than the threshold, which is considered relatively superior to the others, thus used as the core method of the selective iterated boundary location (SIBYL) algorithm developed for producing the CALIPSO official products [15].

However, many studies reported that the CALIPSO AOD was substantially underestimated due to missing layers compared with various observations [16], [17], [18], [19], [20]. For instance, Rogers et al. compared CALIPSO AOD with NASA's airborne high spectral resolution lidar and found that the undetected aerosols in the free troposphere introduced a mean underestimate of 0.02 in the CALIPSO AOD of the nighttime dataset examined [21]. Omar et al. found that CALIPSO AOD was smaller than AERONET AOD over ocean and land [22]. Kim retrieved aerosol extinction for the undetected aerosol layers and found a global mean undetected layer AOD of 0.031 [23]. Toth et al. found that the mean AOD at 550 nm of MODIS and AERONET were approximately 0.06 and 0.08, respectively, for the CALIOP profiles with no aerosols detected [24]. Thus, a more accurate and robust layer detection algorithm should be developed to improve CALIPSO retrievals and serve scientific applications.

In this study, a signal segmentation method based on a minimum cost function is proposed to determine layer boundaries. This algorithm can adapt to the general variation trend of signals, instead of being limited only to local variation. As a result, the accuracy and the resolution of the layer products of CALIPSO are effectively improved.