skip to main content


Title: VDAM: VAE based domain adaptation for cloud property retrieval from multi-satellite data
Domain adaptation techniques using deep neural networks have been mainly used to solve the distribution shift problem in homogeneous domains where data usually share similar feature spaces and have the same dimensionalities. Nevertheless, real world applications often deal with heterogeneous domains that come from completely different feature spaces with different dimensionalities. In our remote sensing application, two remote sensing datasets collected by an active sensor and a passive one are heterogeneous. In particular, CALIOP actively measures each atmospheric column. In this study, 25 measured variables/features that are sensitive to cloud phase are used and they are fully labeled. VIIRS is an imaging radiometer, which collects radiometric measurements of the surface and atmosphere in the visible and infrared bands. Recent studies have shown that passive sensors may have difficulties in prediction cloud/aerosol types in complicated atmospheres (e.g., overlapping cloud and aerosol layers, cloud over snow/ice surface, etc.). To overcome the challenge of the cloud property retrieval in passive sensor, we develop a novel VAE based approach to learn domain invariant representation that capture the spatial pattern from multiple satellite remote sensing data (VDAM), to build a domain invariant cloud property retrieval method to accurately classify different cloud types (labels) in the passive sensing dataset. We further exploit the weight based alignment method on the label space to learn a powerful domain adaptation technique that is pertinent to the remote sensing application. Experiments demonstrate our method outperforms other state-of-the-art machine learning methods and achieves higher accuracy in cloud property retrieval in the passive satellite dataset.  more » « less
Award ID(s):
1942714 1730250 1948399
NSF-PAR ID:
10400896
Author(s) / Creator(s):
; ; ;
Date Published:
Journal Name:
Proceedings of the 30th International Conference on Advances in Geographic Information Systems (SIGSPATIAL 22)
Page Range / eLocation ID:
1 to 10
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Abstract. As wildfires intensify and fire seasons lengthen across the western US, the development of models that can predict smoke plume concentrations and track wildfire-induced air pollution exposures has become critical. Wildfire smoke plume height is a key indicator of the vertical placement of plume mass emitted from wildfire-related aerosol sources in climate and air quality models. With advancements in Earth observation (EO) satellites, spaceborne products for aerosol layer height or plume injection height have recently emerged with increased global-scale spatiotemporal resolution. However, to evaluate column radiative effects and refine satellite algorithms, vertical profiles of regionally representative aerosol properties from wildfires need to be measured directly. In this study, we conducted the first comprehensive evaluation of four passive satellite remote-sensing techniques specifically designed for retrieving plume height. We compared these satellite products with the airborne Wyoming Cloud Lidar (WCL) measurements during the 2018 Biomass Burning Flux Measurements of Trace Gases and Aerosols (BB-FLUX) field campaign in the western US. Two definitions, namely, “plume top” and “extinction-weighted mean plume height”, were used to derive the representative heights of wildfire smoke plumes, based on the WCL-derived vertical aerosol extinction coefficient profiles. Using these two definitions, we performed a comparative analysis of multisource satellite-derived plume height products for wildfire smoke. We provide a discussion related to which satellite product is most appropriate for determining plume height characteristics near a fire event or estimating downwind plume rise equivalent height, under multiple aerosol loadings. Our findings highlight the importance of understanding the sensitivity of different passive remote-sensing techniques on space-based wildfire smoke plume height observations, in order to resolve ambiguity surrounding the concept of “effective smoke plume height”. As additional aerosol-observing satellites are planned in the coming years, our results will inform future remote-sensing missions and EO satellite algorithm development. This bridges the gap between satellite observations and plume rise modeling to further investigate the vertical distribution of wildfire smoke aerosols. 
    more » « less
  2. Due to the ability of deep neural nets to learn rich representations, recent advances in unsupervised domain adaptation have focused on learning domain-invariant features that achieve a small error on the source domain. The hope is that the learnt representation, together with the hypothesis learnt from the source domain, can generalize to the target domain. In this paper, we first construct a simple counterexample showing that, contrary to common belief, the above conditions are not sufficient to guarantee successful domain adaptation. In particular, the counterexample exhibits conditional shift: the class-conditional distributions of input features change between source and target domains. To give a sufficient condition for domain adaptation, we propose a natural and interpretable generalization upper bound that explicitly takes into account the aforementioned shift.Moreover, we shed new light on the problem by proving an information-theoretic lower bound on the joint error of any domain adaptation method that attempts to learn invariant representations.Our result characterizes a fundamental tradeoff between learning invariant representations and achieving small joint error on both domains when the marginal label distributions differ from source to target. Finally, we conduct experiments on real-world datasets that corroborate our theoretical findings. We believe these insights are helpful in guiding the future design of domain adaptation and representation learning algorithms. 
    more » « less
  3. null (Ed.)
    Human activity recognition (HAR) from wearable sensors data has become ubiquitous due to the widespread proliferation of IoT and wearable devices. However, recognizing human activity in heterogeneous environments, for example, with sensors of different models and make, across different persons and their on-body sensor placements introduces wide range discrepancies in the data distributions, and therefore, leads to an increased error margin. Transductive transfer learning techniques such as domain adaptation have been quite successful in mitigating the domain discrepancies between the source and target domain distributions without the costly target domain data annotations. However, little exploration has been done when multiple distinct source domains are present, and the optimum mapping to the target domain from each source is not apparent. In this paper, we propose a deep Multi-Source Adversarial Domain Adaptation (MSADA) framework that opportunistically helps select the most relevant feature representations from multiple source domains and establish such mappings to the target domain by learning the perplexity scores. We showcase that the learned mappings can actually reflect our prior knowledge on the semantic relationships between the domains, indicating that MSADA can be employed as a powerful tool for exploratory activity data analysis. We empirically demonstrate that our proposed multi-source domain adaptation approach achieves 2% improvement with OPPORTUNITY dataset (cross-person heterogeneity, 4 ADLs), whereas 13% improvement on DSADS dataset (cross-position heterogeneity, 10 ADLs and sports activities). 
    more » « less
  4. Abstract. Many passive remote-sensing techniques have beendeveloped to retrieve cloud microphysical properties from satellite-basedsensors, with the most common approaches being the bispectral andpolarimetric techniques. These two vastly different retrieval techniqueshave been implemented for a variety of polar-orbiting and geostationarysatellite platforms, providing global climatological data sets. Priorinstrument comparison studies have shown that there are systematicdifferences between the droplet size retrieval products (effective radius)of bispectral (e.g., MODIS, Moderate Resolution Imaging Spectroradiometer)and polarimetric (e.g., POLDER, Polarization and Directionality of Earth'sReflectances) instruments. However, intercomparisons of airborne bispectraland polarimetric instruments have yielded results that do not appear to besystematically biased relative to one another. Diagnosing this discrepancyis complicated, because it is often difficult for instrument intercomparisonstudies to isolate differences between retrieval technique sensitivities andspecific instrumental differences such as calibration and atmosphericcorrection. In addition to these technical differences the polarimetricretrieval is also sensitive to the dispersion of the droplet sizedistribution (effective variance), which could influence the interpretationof the droplet size retrieval. To avoid these instrument-dependentcomplications, this study makes use of a cloud remote-sensing retrievalsimulator. Created by coupling a large-eddy simulation (LES) cloud modelwith a 1-D radiative transfer model, the simulator serves as a test bed forunderstanding differences between bispectral and polarimetric retrievals.With the help of this simulator we can not only compare the two techniquesto one another (retrieval intercomparison) but also validate retrievalsdirectly against the LES cloud properties. Using the satellite retrievalsimulator, we are able to verify that at high spatial resolution (50m) thebispectral and polarimetric retrievals are highly correlated with oneanother within expected observational uncertainties. The relatively smallsystematic biases at high spatial resolution can be attributed to differentsensitivity limitations of the two retrievals. In contrast, a systematicdifference between the two retrievals emerges at coarser resolution. Thisbias largely stems from differences related to sensitivity of the tworetrievals to unresolved inhomogeneities in effective variance and opticalthickness. The influence of coarse angular resolution is found to increaseuncertainty in the polarimetric retrieval but generally maintains aconstant mean value.

     
    more » « less
  5. Abstract. During the Chequamegon Heterogeneous Ecosystem Energy-balance Study Enabled by a High-density Extensive Array of Detectors 2019 (CHEESEHEAD19) field campaign, held in the summer of 2019 in northern Wisconsin, USA, active and passive ground-based remote sensing instruments were deployed to understand the response of the planetary boundary layer to heterogeneous land surface forcing. These instruments include radar wind profilers, microwave radiometers, atmospheric emitted radiance interferometers, ceilometers, high spectral resolution lidars, Doppler lidars, and collaborative lower-atmospheric mobile profiling systems that combine several of these instruments. In this study, these ground-based remote sensing instruments are used to estimate the height of the daytime planetary boundary layer, and their performance is compared against independent boundary layer depth estimates obtained from radiosondes launched as part of the field campaign. The impact of clouds (in particular boundary layer clouds) on boundary layer depth estimations is also investigated. We found that while all instruments are overall able to provide reasonable boundary layer depth estimates, each of them shows strengths and weaknesses under certain conditions. For example, radar wind profilers perform well during cloud-free conditions, and microwave radiometers and atmospheric emitted radiance interferometers have a very good agreement during all conditions but are limited by the smoothness of the retrieved thermodynamic profiles. The estimates from ceilometers and high spectral resolution lidars can be hindered by the presence of elevated aerosol layers or clouds, and the multi-instrument retrieval from the collaborative lower atmospheric mobile profiling systems can be constricted to a limited height range in low-aerosol conditions. 
    more » « less