Combined neural network/Phillips–Tikhonov approach to aerosol retrievals over land from the NASA Research Scanning Polarimeter

Noia, Antonio Di; Hasekamp, Otto; Wu, Lianghai; Diedenhoven, Bastiaan van; Cairns, Brian; Yorks, John E.

doi:10.5194/amt-10-4235-2017

Cited by 37 publications

(25 citation statements)

References 52 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…POLDER-3 data have been used for a wide range of applications, including aerosol retrievals (Dubovik et al, 2011;Hasekamp et al, 2011;Waquet et al, 2013), retrieval of cloud properties (Riedi et al, 2010;Parol et al, 2013;van Diedenhoven et al, 2014b;Desmons et al, 2017), water vapour retrievals , estimation of surface properties (Bréon and Maignan, 2017) and validation of surface reflection models (Kokhanovsky and Bréon, 2012).…”

Section: The Polder-3 Instrumentmentioning

confidence: 99%

Retrieval of liquid water cloud properties from POLDER-3 measurements using a neural network ensemble approach

et al. 2019

Self Cite

View full text Add to dashboard Cite

Abstract. This paper describes a neural network algorithm for the estimation of liquid water cloud optical properties from the Polarization and Directionality of Earth's Reflectances-3 (POLDER-3) instrument aboard the Polarization & Anisotropy of Reflectances for Atmospheric Sciences coupled with Observations from a Lidar (PARASOL) satellite. The algorithm has been trained on synthetic multi-angle, multi-wavelength measurements of reflectance and polarization and has been applied to the processing of 1 year of POLDER-3 data. Comparisons of the retrieved cloud properties with Moderate Resolution Imaging Spectroradiometer (MODIS) products show that the neural network algorithm has a low bias of around 2 in cloud optical thickness (COT) and between 1 and 2 µm in the cloud effective radius. Comparisons with existing POLDER-3 datasets suggest that the proposed scheme may have enhanced capabilities for cloud effective radius retrieval, at least over land. An additional feature of the presented algorithm is that it provides COT and effective radius retrievals at the native POLDER-3 Level 1B pixel level.

show abstract

Section: The Polder-3 Instrumentmentioning

confidence: 99%

Retrieval of liquid water cloud properties from POLDER-3 measurements using a neural network ensemble approach

et al. 2019

Self Cite

View full text Add to dashboard Cite

show abstract

“…Figure 2 shows a schematic of the used method and detailed descriptions of the method can be found in the textbooks of the subject (e.g., Buduma and Locascio, 2017;Duda et al, 2012). In general, the NNs are built by stacking interconnected layers of atomic units, or neurons.…”

Section: Discussionmentioning

confidence: 99%

“…There are as many learning methods as there are NN topologies; however, they can be sorted in a few categories. For instance, the learning method may require a labeled training set, falling into the supervised learning (SL) category (Duda et al, 2012;Amari, 1998), or it may learn by itself without an already classified set, in which case the method is referred to as unsupervised learning (UL) (Le, 2013;Radford et al, 2016). Another relevant classification of the learning methods is the depth of the NN.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Identification of new particle formation events with deep learning

et al. 2018

View full text Add to dashboard Cite

Abstract. New particle formation (NPF) in the atmosphere is globally an important source of climate relevant aerosol particles. Occurrence of NPF events is typically analyzed by researchers manually from particle size distribution data day by day, which is time consuming and the classification of event types may be inconsistent. To get more reliable and consistent results, the NPF event analysis should be automatized. We have developed an automatic analysis method based on deep learning, a subarea of machine learning, for NPF event identification. To our knowledge, this is the first time that a deep learning method, i.e., transfer learning of a convolutional neural network (CNN), has successfully been used to automatically classify NPF events into different classes directly from particle size distribution images, similarly to how the researchers carry out the manual classification. The developed method is based on image analysis of particle size distributions using a pretrained deep CNN, named AlexNet, which was transfer learned to recognize NPF event classes (six different types). In transfer learning, a partial set of particle size distribution images was used in the training stage of the CNN and the rest of the images for testing the success of the training. The method was utilized for a 15-yearlong dataset measured at San Pietro Capofiume (SPC) in Italy. We studied the performance of the training with different training and testing of image number ratios as well as with different regions of interest in the images. The results show that clear event (i.e., classes 1 and 2) and nonevent days can be identified with an accuracy of ca. 80 %, when the CNN classification is compared with that of an expert, which is a good first result for automatic NPF event analysis. In the event classification, the choice between different event classes is not an easy task even for trained researchers, and thus overlapping or confusion between different classes occurs. Hence, we cross-validated the learning results of CNN with the expert-made classification. The results show that the overlapping occurs, typically between the adjacent or similar type of classes, e.g., a manually classified Class 1 is categorized mainly into classes 1 and 2 by CNN, indicating that the manual and CNN classifications are very consistent for most of the days. The classification would be more consistent, by both human and CNN, if only two different classes are used for event days instead of three classes. Thus, we recommend that in the future analysis, event days should be categorized into classes of "quantifiable" (i.e., clear events, classes 1 and 2) and "nonquantifiable" (i.e., weak events, Class 3). This would better describe the difference of those classes: both formation and growth rates can be determined for quantifiable days but not both for nonquantifiable days. Furthermore, we investigated more deeply the days that are classified as clear events by experts and recognized as nonevents by the CNN and vice versa. Clear misclassifications seem to ...

show abstract

“…For example, Hasekamp et al (2011), while looking at multi-angular measurements, considered also the case of adding polarization to I-only retrievals and found improved agreement with ground-based (AERONET) data. Di Noia et al (2017) found that use of a neural network to provide an initial guess for an iterative algorithm led to a decrease in processing time and an increase in the number of converged retrievals. And neural networks have been used to directly retrieve products, e.g., ozone column amounts from ground-based irradiance measurements (Fan et al, 2014) or satellite water-leaving radiances (Fan et al, 2017), and have also been used with optimal estimation, e.g., retrieval of snow products from ground-based total radiance measurements (Tanikawa et al, 2015).…”

Section: Introductionmentioning

confidence: 99%

Advantages of Measuring the Q Stokes Parameter in Addition to the Total Radiance I in the Detection of Absorbing Aerosols

et al. 2018

Self Cite

View full text Add to dashboard Cite

A simple but novel study was conducted to investigate whether an imager-type spectroradiometer instrument like MODIS, currently flying on board the Aqua and Terra satellites, or MERIS, which flew on board Envisat, could detect absorbing aerosols if they could measure the Q Stokes parameter in addition to the total radiance I, that is if they could also measure the linear polarization of the light. Accurate radiative transfer calculations were used to train a fast neural network forward model, which together with a simple statistical optimal estimation scheme was used to retrieve three aerosol parameters: aerosol optical depth at 869 nm, optical depth fraction of fine mode (absorbing) aerosols at 869 nm, and aerosol vertical location. The aerosols were assumed to be bimodal, each with a lognormal size distribution, located either between 0 and 2 km or between 2 and 4 km in the Earth's atmosphere. From simulated data with 3% random Gaussian measurement noise added for each Stokes parameter, it was found that by itself the total radiance I at the nine MODIS VIS channels was generally insufficient to accurately retrieve all three aerosol parameters (∼15-37% successful), but that together with the Q Stokes component it was possible to retrieve values of aerosol optical depth at 869 nm to ± 0.03, single-scattering albedo at 869 nm to ± 0.04, and vertical location in ∼65% of the cases. This proof-of-concept retrieval algorithm uses neural networks to overcome the computational burdens of using vector radiative transfer to accurately simulate top-of-atmosphere (TOA) total and polarized radiances, enabling optimal estimation techniques to exploit information from multiple channels. Therefore such an algorithm could, in concept, be readily implemented for operational retrieval of aerosol and ocean products from moderate or hyperspectral spectroradiometers.

show abstract

Combined neural network/Phillips–Tikhonov approach to aerosol retrievals over land from the NASA Research Scanning Polarimeter

Cited by 37 publications

References 52 publications

Retrieval of liquid water cloud properties from POLDER-3 measurements using a neural network ensemble approach

Retrieval of liquid water cloud properties from POLDER-3 measurements using a neural network ensemble approach

Identification of new particle formation events with deep learning

Advantages of Measuring the Q Stokes Parameter in Addition to the Total Radiance I in the Detection of Absorbing Aerosols

Contact Info

Product

Resources

About