Search Results
How the Selection of Training Data and Modeling Approach Affects the Estimation of Ammonia Emissions from a Naturally Ventilated Dairy Barn—Classical Statistics versus Machine Learning
Environmental protection efforts can only be effective in the long term with a reliable quantification of pollutant gas emissions as a first step to mitigation. Measurement and analysis strategies must permit the accurate extrapolation of emission values. We systematically analyzed the added value of applying modern machine learning methods in the process of monitoring emissions from naturally ventilated livestock buildings to the atmosphere. We considered almost 40 weeks of hourly emission values from a naturally ventilated dairy cattle barn in Northern Germany. We compared model predictions using 27 different scenarios of temporal sampling, multiple measures of model accuracy, and eight different regression approaches. The error of the predicted emission values with the tested measurement protocols was, on average, well below 20%. The sensitivity of the prediction to the selected training dataset was worse for the ordinary multilinear regression. Gradient boosting and random forests provided the most accurate and robust emission value predictions, accompanied by the second-smallest model errors. Most of the highly ranked scenarios involved six measurement periods, while the scenario with the best overall performance was: One measurement period in summer and three in the transition periods, each lasting for 14 days.
Identifying cloud droplets beyond lidar attenuation from vertically pointing cloud radar observations using artificial neural networks
In mixed-phase clouds, the variable mass ratio between liquid water and ice as well as the spatial distribution within the cloud plays an important role in cloud lifetime, precipitation processes, and the radiation budget. Data sets of vertically pointing Doppler cloud radars and lidars provide insights into cloud properties at high temporal and spatial resolution. Cloud radars are able to penetrate multiple liquid layers and can potentially be used to expand the identification of cloud phase to the entire vertical column beyond the lidar signal attenuation height, by exploiting morphological features in cloud radar Doppler spectra that relate to the existence of supercooled liquid. We present VOODOO (reVealing supercOOled liquiD beyOnd lidar attenuatiOn), a retrieval based on deep convolutional neural networks (CNNs) mapping radar Doppler spectra to the probability of the presence of cloud droplets (CD). The training of the CNN was realized using the Cloudnet processing suite as supervisor. Once trained, VOODOO yields the probability for CD directly at Cloudnet grid resolution. Long-term predictions of 18 months in total from two mid-latitudinal locations, i.e., Punta Arenas, Chile (53.1 S, 70.9 W), in the Southern Hemisphere and Leipzig, Germany (51.3 N, 12.4 E), in the Northern Hemisphere, are evaluated. Temporal and spatial agreement in cloud-droplet-bearing pixels is found for the Cloudnet classification to the VOODOO prediction. Two suitable case studies were selected, where stratiform, multi-layer, and deep mixed-phase clouds were observed. Performance analysis of VOODOO via classification-evaluating metrics reveals precision > 0.7, recall ≈ 0.7, and accuracy ≈ 0.8. Additionally, independent measurements of liquid water path (LWP) retrieved by a collocated microwave radiometer (MWR) are correlated to the adiabatic LWP, which is estimated using the temporal and spatial locations of cloud droplets from VOODOO and Cloudnet in connection with a cloud parcel model. This comparison resulted in stronger correlation for VOODOO (≈ 0.45) compared to Cloudnet (≈ 0.22) and indicates the availability of VOODOO to identify CD beyond lidar attenuation. Furthermore, the long-term statistics for 18 months of observations are presented, analyzing the performance as a function of MWR-LWP and confirming VOODOO's ability to identify cloud droplets reliably for clouds with LWP > 100 g m-2. The influence of turbulence on the predictive performance of VOODOO was also analyzed and found to be minor. A synergy of the novel approach VOODOO and Cloudnet would complement each other perfectly and is planned to be incorporated into the Cloudnet algorithm chain in the near future.
Analogue pattern recognition with stochastic switching binary CMOS-integrated memristive devices
Biological neural networks outperform current computer technology in terms of power consumption and computing speed while performing associative tasks, such as pattern recognition. The analogue and massive parallel in-memory computing in biology differs strongly from conventional transistor electronics that rely on the von Neumann architecture. Therefore, novel bio-inspired computing architectures have been attracting a lot of attention in the field of neuromorphic computing. Here, memristive devices, which serve as non-volatile resistive memory, are employed to emulate the plastic behaviour of biological synapses. In particular, CMOS integrated resistive random access memory (RRAM) devices are promising candidates to extend conventional CMOS technology to neuromorphic systems. However, dealing with the inherent stochasticity of resistive switching can be challenging for network performance. In this work, the probabilistic switching is exploited to emulate stochastic plasticity with fully CMOS integrated binary RRAM devices. Two different RRAM technologies with different device variabilities are investigated in detail, and their potential applications in stochastic artificial neural networks (StochANNs) capable of solving MNIST pattern recognition tasks is examined. A mixed-signal implementation with hardware synapses and software neurons combined with numerical simulations shows that the proposed concept of stochastic computing is able to process analogue data with binary memory cells. © 2020, The Author(s).
Seasonal prediction of Indian summer monsoon onset with echo state networks
Although the prediction of the Indian Summer Monsoon (ISM) onset is of crucial importance for water-resource management and agricultural planning on the Indian sub-continent, the long-term predictability—especially at seasonal time scales—is little explored and remains challenging. We propose a method based on artificial neural networks that provides skilful long-term forecasts (beyond 3 months) of the ISM onset, although only trained on short and noisy data. It is shown that the meridional tropospheric temperature gradient in the boreal winter season already contains the signals needed for predicting the ISM onset in the subsequent summer season. Our study demonstrates that machine-learning-based approaches can be simultaneously helpful for both data-driven prediction and enhancing the process understanding of climate phenomena.
Optimization with learning-informed differential equation constraints and its applications
Inspired by applications in optimal control of semilinear elliptic partial differential equations and physics-integrated imaging, differential equation constrained optimization problems with constituents that are only accessible through data-driven techniques are studied. A particular focus is on the analysis and on numerical methods for problems with machine-learned components. For a rather general context, an error analysis is provided, and particular properties resulting from artificial neural network based approximations are addressed. Moreover, for each of the two inspiring applications analytical details are presented and numerical results are provided.
Toward Reliable Compact Modeling of Multilevel 1T-1R RRAM Devices for Neuromorphic Systems
In this work, three different RRAM compact models implemented in Verilog-A are analyzed and evaluated in order to reproduce the multilevel approach based on the switching capability of experimental devices. These models are integrated in 1T-1R cells to control their analog behavior by means of the compliance current imposed by the NMOS select transistor. Four different resistance levels are simulated and assessed with experimental verification to account for their multilevel capability. Further, an Artificial Neural Network study is carried out to evaluate in a real scenario the viability of the multilevel approach under study.