An Overview of Neural Network Methods for Predicting Uncertainty in Atmospheric Remote Sensing

In this paper, we present neural network methods for predicting uncertainty in atmospheric remote sensing. These include methods for solving the direct and the inverse problem in a Bayesian framework. In the first case, a method based on a neural network for simulating the radiative transfer model and a Bayesian approach for solving the inverse problem is proposed. In the second case, (i) a neural network, in which the output is the convolution of the output for a noise-free input with the input noise distribution; and (ii) a Bayesian deep learning framework that predicts input aleatoric and model uncertainties, are designed. In addition, a neural network that uses assumed density filtering and interval arithmetic to compute uncertainty is employed for testing purposes. The accuracy and the precision of the methods are analyzed by considering the retrieval of cloud parameters from radiances measured by the Earth Polychromatic Imaging Camera (EPIC) onboard the Deep Space Climate Observatory (DSCOVR).

Post-flight analysis of detailed size distributions of warm cloud droplets, as determined in situ by cloud and aerosol spectrometers

Warm clouds, consisting of liquid cloud droplets, play an important role in modulating the amount of incoming solar radiation to Earth's surface and thus the climate. The size and number concentration of these cloud droplets control the reflectance of the cloud, the formation of precipitation and ultimately the lifetime of the cloud. Therefore, in situ observations of the number and diameter of cloud droplets are frequently performed with cloud and aerosol spectrometers, which determine the optical diameters of cloud particles (in the range of up to a few tens of micrometers) by measuring their forward-scattering cross sections in visible light and comparing these values with Mie theoretical computations. The use of such instruments must rely on a fast working scheme consisting of a limited pre-defined uneven grid of cross section values that corresponds to a theoretically derived uneven set of size intervals (bins). However, as more detailed structural analyses of warm clouds are needed to improve future climate projects, we present a new numerical post-flight methodology using recorded particle-by-particle sample files. The Mie formalism produces a complicated relationship between a particle's diameter and its forward-scattering cross section. This relationship cannot be expressed in an analytically closed form, and it should be numerically computed point by point, over a certain grid of diameter values. The optimal resolution required for constructing the diagram of this relationship is therefore analyzed. Cloud particle statistics are further assessed using a fine grid of particle diameters in order to capture the finest details of the cloud particle size distributions. The possibility and the usefulness of using coarser size grids, with either uneven or equal intervals, is also discussed. For coarse equidistant size grids, the general expressions of cloud microphysical parameters are calculated and the ensuing relative errors are discussed in detail. The proposed methodology is further applied to a subset of measured data, and it is shown that the overall uncertainties in computing various cloud parameters are mainly driven by the measurement errors of the forward-scattering cross section for each particle. Finally, the influence of the relatively large imprecision in the real and imaginary parts of the refractive index of cloud droplets on the size distributions and on the ensuing cloud parameters is analyzed. It is concluded that, in the presence of high atmospheric loads of hydrophilic and light-absorbing aerosols, such imprecisions may drastically affect the reliability of the cloud data obtained with cloud and aerosol spectrometers. Some complementary measurements for improving the quality of the cloud droplet size distributions obtained in post-flight analyses are suggested.

A dataset of microphysical cloud parameters, retrieved from Emission-FTIR spectra measured in Arctic summer 2017

A dataset of microphysical cloud parameters from optically thin clouds, retrieved from infrared spectral radiances measured in summer 2017 in the Arctic, is presented. Measurements were conducted using a mobile Fourier-transform infrared (FTIR) spectrometer which was carried by the RV Polarstern. This dataset contains retrieved optical depths and effective radii of ice and water, from which the liquid water path and ice water path are calculated. These water paths and the effective radii are compared with derived quantities from a combined cloud radar, lidar and microwave radiometer measurement synergy retrieval, called Cloudnet. Comparing the liquid water paths from the infrared retrieval and Cloudnet shows significant correlations with a standard deviation of 8.60 g · m−2. Although liquid water path retrievals from microwave radiometer data come with a uncertainty of at least 20 g · m−2, a significant correlation and a standard deviation of 5.32 g · m−2 between the results of clouds with a liquid water path of at most 20 g · m−2 retrieved from infrared spectra and results from Cloudnet can be seen. Therefore, despite its large uncertainty, the comparison with data retrieved from infrared spectra shows that optically thin clouds of the measurement campaign in summer 2017 can be observed well using microwave radiometers within the Cloudnet framework. Apart from this, the dataset of microphysical cloud properties presented here allows to perform calculations of the cloud radiative effects, when the Cloudnet data from the campaign are not available, which was from the 22nd July 2017 until the 19th August 2017. The dataset is published at Pangaea (Richter et al., 2021).

Post-flight analysis of detailed size distributions of warm cloud droplets, as determined in situ by cloud and aerosol spectrometers

Warm clouds, consisting of liquid cloud droplets, play an important role in modulating the amount of incoming solar radiation to Earth’s surface and thus, the climate. The size and number concentration of these cloud droplets control the reflectance of the cloud, the formation of precipitation and ultimately, the lifetime of the cloud. Therefore, in situ observations of the number and diameter of cloud droplets are frequently performed with cloud and aerosol spectrometers, which determine the optical diameters of cloud particles (in the range of up to a few tens of microns) by measuring their forward scattering cross sections in visible light and comparing these values with Mie-theoretical computations. The use of such instruments must rely on a fast working scheme consisting of a limited pre-defined uneven grid of cross section values that corresponds to a theoretically derived uneven set of size intervals (bins). However, as more detailed structural analyses of warm clouds are needed to improve future climate projects, we present a new numerical post-flight methodology using recorded particle-by-particle sample files. The Mie formalism produces a complicated relationship between a particle’s diameter and its forward scattering cross section. This relationship cannot be expressed in an analytically closed form and it should be numerically computed point by point, over a certain grid of diameter values. The optimal resolution required for constructing the diagram of this relationship is therefore analysed. Cloud particle statistics are further assessed using a fine grid of particle diameters in order to capture the finest details of the cloud particle size distributions. The possibility and the usefulness of using coarser size grids, with either uneven or equal intervals is also discussed. For coarse equidistant size grids, the general expressions of cloud microphysical parameters are calculated and the ensuing relative errors are discussed in detail. The proposed methodology is further applied to a subset of measured data and it is shown that the overall uncertainties in computing various cloud parameters are mainly driven by the measurement errors of the forward scattering cross section for each particle. Finally, the influence of the relatively large imprecision in the real and imaginary parts of the refractive index of cloud droplets on the size distributions and on the ensuing cloud parameters is analysed. It is concluded that, in the presence of high atmospheric loads of hydrophilic and light absorbing aerosols, such imprecisions may drastically affect the reliability of the cloud data obtained with cloud and aerosol spectrometers. Some complementary measurements for improving the quality of the cloud droplet size distributions obtained in post-flight analyses are suggested.

Retrieving H₂O/HDO columns over cloudy and clear-sky scenes from the Tropospheric Monitoring Instrument (TROPOMI)

This paper presents an extension of the scientific HDO/H2O column data product from the Tropospheric Monitoring Instrument (TROPOMI) including clear-sky and cloudy scenes. The retrieval employs a forward model which accounts for scattering, and the algorithm infers the trace gas column information, surface properties and effective cloud parameters from the observations. The extension to cloudy scenes greatly enhances coverage, particularly enabling data over oceans. The data set is validated against co-located ground-based Fourier transform infrared (FTIR) observations by the Total Carbon Column Observing Network (TCCON). The median bias for clear-sky scenes is 1.4 × 1021 molec cm−2 (2.9 %) in H2O columns and 1.1 × 1017 molec cm−2 (−0.3 %) in HDO columns, which corresponds to −17 ‰ (9.9 %) in a posteriori δD. The bias for cloudy scenes is 4.9 × 1021 molec cm−2 (11 %) in H2O, 1.1 × 1017 molec cm−2 (7.9 %) in HDO, and −20 ‰ (9.7 %) in a posteriori δD. At low-altitude stations in low and middle latitudes the bias is small, and has a larger value at high latitude stations. At high altitude stations, an altitude correction is required to compensate for different partial columns seen by the station and the satellite. The bias in a posteriori δD after altitude correction depends on sensitivity due to shielding by clouds, and on realistic prior profile shapes for both isotopologues. Cloudy scenes generally involve low sensitivity below the clouds, and since the information is filled up by the prior, it plays an important role in these cases. Over oceans, aircraft measurements with the Water Isotope System for Precipitation and Entrainment Research (WISPER) instrument from a field campaign in 2018 are used for validation, yielding a bias of −3.9 % in H2O and −3 ‰ in δD over clouds. To demonstrate the added value of the new data set, a short case study of a cold air outbreak over the Atlantic Ocean in January 2020 is presented, showing the daily evolution of the event with single overpass results.

The Use of Weather Research and Forecasting Model to Predict Rainfall in Tropical Peatland: 1. Model Parameterization

Rainfall dynamics play a vital role in tropical peatland by providing sufficient water to keep peat moist throughout the year. Therefore, information of rainfall data either historical or forecasting data has risen in recent decades especially for an alert system of fire. Here the Weather and Research Forecasting (WRF) model may act as a tool to provide forecasting weather data. This study aims to do parameterization on WRF parameters for peatland in Sumatra, and to perform bias correction on the WRF’s rainfall output with observed data. We performed stepwise calibration to choose the best five physical schemes of WRF for use in the study area. The output WRF’s rainfall was bias corrected by spatially observed rainfall data for 2019 at day resolution. Our results showed the following schemes namely (i) Eta scheme for cloud microphysical parameters; (ii) GD scheme for cumulus cloud parameters, (iii) MYJ scheme for planetary boundary layer parameters; (iv) RRTM for longwave radiation; and (v) New Goddard schemes for shortwave radiation are best combination for being used to predict rainfall in maritime continent. The spatially interpolated observed rainfall with the Inverse Distance Weighting (IDW) was outperformed for calibration process of WRF’s rainfall as shown by statistical indicators used in this study. Further, the findings have contributed to advance knowledge of rainfall forecasting in maritime continent, particularly in providing data to support the development of fire danger rating system for Indonesian peatland.

Biomass Burning Aerosol Heating Rates from the ORACLES 2016 and 2017 Experiments

Aerosol heating due to shortwave absorption has implications for local atmospheric stability and regional dynamics. The derivation of heating rate profiles from space-based observations is challenging because it requires the vertical profile of relevant properties such as the aerosol extinction coefficient and single scattering albedo (SSA). In the southeast Atlantic, this challenge is amplified by the presence of stratocumulus clouds below the biomass burning plume advected from Africa, since the cloud properties affect the magnitude of the aerosol heating aloft, which may in turn lead to changes in the cloud properties and life cycle. The combination of spaceborne lidar data with passive imagers shows promise for future derivations of heating rate profiles and curtains, but new algorithms require careful testing with data from aircraft experiments where measurements of radiation, aerosol and cloud parameters are better collocated and readily available. In this study, we derive heating rate profiles and curtains from aircraft measurements during the NASA ObseRvations of CLouds above Aerosols and their intEractionS (ORACLES) project in the southeastern Atlantic. Spectrally resolved irradiance measurements and the derived column absorption allow for the separation of total heating rates into aerosol and gas (primarily water vapor) absorption. The nine cases we analyzed capture some of the co-variability of heating rate profiles and their primary drivers, leading to the development of a new concept: The Heating Rate Efficiency (HRE; the heating rate per unit aerosol extinction). The HRE, which accounts for the overall aerosol loading as well as vertical distribution of the aerosol layer, varies little with altitude as opposed to the standard heating rate. The large case-to-case variability for ORACLES is significantly reduced after converting from heating rate to HRE, allowing us to quantify its dependence on SSA, cloud albedo, and solar zenith angle.

LES study of precipitation/condensation dependance on cumulus clouds dynamics

Parameters affecting condensation/evaporation rates (CR/ER) in trade wind cumulus clouds were analyzed using LES model simulations. The model was initialized with data observed during the RICO field project, and simulated in a rather large 50.0×50.0×4 km3 domain. 2031 clouds were analyzed seeking relationships between CR/ER and thermo-dynamical cloud parameters. The condensation/evaporation rates were analyzed by stratifying the clouds by their size. The analyzed parameters included, among others, integral mass and buoyancy fluxes, as well as cloud and rain water and drop concentration. The results revealed rather remarkable relationship between integral condensation/evaporation rate and integral upward mass flux. Identified relathionship may be useful for parameterization of subgrid latent heat in meso and large-scale models.