Fast and robust quantification of uncertainty in non-linear diffusion MRI models.

Diffusion MRI (dMRI) allows for non-invasive investigation of brain tissue microstructure. By fitting a model to the dMRI signal, various quantitative measures can be derived from the data, such as fractional anisotropy, neurite density and axonal radii maps. We investigate the Fisher Information Matrix (FIM) and uncertainty propagation as a generally applicable method for quantifying the parameter uncertainties in linear and non-linear diffusion MRI models. In direct comparison with Markov Chain Monte Carlo (MCMC) sampling, the FIM produces similar uncertainty estimates at much lower computational cost. Using acquired and simulated data, we then list several characteristics that influence the parameter variances, including data complexity and signal-to-noise ratio. For practical purposes we investigate a possible use of uncertainty estimates in decreasing intra-group variance in group statistics by uncertainty-weighted group estimates. This has potential use cases for detection and suppression of imaging artifacts.

Prediction of diffuse solar radiation by integrating radiative transfer model and machine-learning techniques.

Diffuse radiation is a major component of solar radiation that is important in carbon exchanges and material, energy, and information flows in agricultural ecosystems; however, measuring diffuse radiation is difficult and expensive, leaving only few stations in China that can record diffuse radiation. Therefore, five high-speed and highly accurate hybrid models were developed and compared to simulate diffuse radiation based on the aerosol optical properties and radiation parameters provided by the Aerosol Robotic Network (AERONET), Baseline Surface Radiation Network (BSRN), Wuhan University, Chinese Ecosystem Research Network (CERN), GLASS surface albedo data, and combined radiative transfer model (RTM) with machine learning (ML) models that include random forest (RF), extreme gradient boosting (XGBoost), multi-layer perceptron (MLP), deep neural networks (DNN), and convolutional neural network (CNN). Furthermore, the uncertainty in the simulated diffuse radiation due to the measurement uncertainties of aerosol optical properties and land surface albedo was quantified, and the relative contributions of multiple variables to diffuse radiation were analyzed. The results showed that RTM-RF was the most successful, with determination coefficients (R2) of 0.95, 0.94, and 0.98, and minimum root mean square errors (RMSE) of 9.56, 10.05, and 13.27 W m-2 at the Lulin, Wuhan, and Xianghe sites, respectively. The largest measurement uncertainty in the aerosol optical depth (AOD) was found at the Lulin site, while that of the single-scattering albedo led to the largest errors in Wuhan and Xianghe. AOD, solar zenith angle (SZA), and single-scattering albedo contributed significantly more than the asymmetry factor, land surface albedo, precipitable water vapor, and ozone. This was especially true for AOD, which was higher than 28 % at all sites. Overall, the proposed RTM-RF method exhibited superior performance, therefore we recommend it for estimating diffuse radiation in China.

Gaussian processes retrieval of crop traits in Google Earth Engine based on Sentinel-2 top-of-atmosphere data.

The unprecedented availability of optical satellite data in cloud-based computing platforms, such as Google Earth Engine (GEE), opens new possibilities to develop crop trait retrieval models from the local to the planetary scale. Hybrid retrieval models are of interest to run in these platforms as they combine the advantages of physically- based radiative transfer models (RTM) with the flexibility of machine learning regression algorithms. Previous research with GEE primarily relied on processing bottom-of-atmosphere (BOA) reflectance data, which requires atmospheric correction. In the present study, we implemented hybrid models directly into GEE for processing Sentinel-2 (S2) Level-1C (L1C) top-of-atmosphere (TOA) reflectance data into crop traits. To achieve this, a training dataset was generated using the leaf-canopy RTM PROSAIL in combination with the atmospheric model 6SV. Gaussian process regression (GPR) retrieval models were then established for eight essential crop traits namely leaf chlorophyll content, leaf water content, leaf dry matter content, fractional vegetation cover, leaf area index (LAI), and upscaled leaf variables (i.e., canopy chlorophyll content, canopy water content and canopy dry matter content). An important pre-requisite for implementation into GEE is that the models are sufficiently light in order to facilitate efficient and fast processing. Successful reduction of the training dataset by 78% was achieved using the active learning technique Euclidean distance-based diversity (EBD). With the EBD-GPR models, highly accurate validation results of LAI and upscaled leaf variables were obtained against in situ field data from the validation study site Munich-North-Isar (MNI), with normalized root mean square errors (NRMSE) from 6% to 13%. Using an independent validation dataset of similar crop types (Italian Grosseto test site), the retrieval models showed moderate to good performances for canopy-level variables, with NRMSE ranging from 14% to 50%, but failed for the leaf-level estimates. Obtained maps over the MNI site were further compared against Sentinel-2 Level 2 Prototype Processor (SL2P) vegetation estimates generated from the ESA Sentinels' Application Platform (SNAP) Biophysical Processor, proving high consistency of both retrievals (R 2 from 0.80 to 0.94). Finally, thanks to the seamless GEE processing capability, the TOA-based mapping was applied over the entirety of Germany at 20 m spatial resolution including information about prediction uncertainty. The obtained maps provided confidence of the developed EBD-GPR retrieval models for integration in the GEE framework and national scale mapping from S2-L1C imagery. In summary, the proposed retrieval workflow demonstrates the possibility of routine processing of S2 TOA data into crop traits maps at any place on Earth as required for operational agricultural applications.

Neural Stochastic Differential Equations with Neural Processes Family Members for Uncertainty Estimation in Deep Learning.

Existing neural stochastic differential equation models, such as SDE-Net, can quantify the uncertainties of deep neural networks (DNNs) from a dynamical system perspective. SDE-Net is either dominated by its drift net with in-distribution (ID) data to achieve good predictive accuracy, or dominated by its diffusion net with out-of-distribution (OOD) data to generate high diffusion for characterizing model uncertainty. However, it does not consider the general situation in a wider field, such as ID data with noise or high missing rates in practice. In order to effectively deal with noisy ID data for credible uncertainty estimation, we propose a vNPs-SDE model, which firstly applies variants of neural processes (NPs) to deal with the noisy ID data, following which the completed ID data can be processed more effectively by SDE-Net. Experimental results show that the proposed vNPs-SDE model can be implemented with convolutional conditional neural processes (ConvCNPs), which have the property of translation equivariance, and can effectively handle the ID data with missing rates for one-dimensional (1D) regression and two-dimensional (2D) image classification tasks. Alternatively, vNPs-SDE can be implemented with conditional neural processes (CNPs) or attentive neural processes (ANPs), which have the property of permutation invariance, and exceeds vanilla SDE-Net in multidimensional regression tasks.

Environmental impact and risk assessments and key factors contributing to the overall uncertainties.

There is a significant number of nuclear and radiological sources that have contributed, are still contributing, or have the potential to contribute to radioactive contamination of the environment in the future. To protect the environment from radioactive contamination, impact and risk assessments are performed prior to or during a release event, short or long term after deposition or prior and after implementation of countermeasures. When environmental impact and risks are assessed, however, a series of factors will contribute to the overall uncertainties. To provide environmental impact and risk assessments, information on processes, kinetics and a series of input variables is needed. Adding problems such as variability, questionable assumptions, gaps in knowledge, extrapolations and poor conceptual model structures, a series of factors are contributing to large and often unacceptable uncertainties in impact and risk assessments. Information on the source term and the release scenario is an essential starting point in impact and risk models; the source determines activity concentrations and atom ratios of radionuclides released, while the release scenario determine the physico-chemical forms of released radionuclides such as particle size distribution, structure and density. Releases will most often contain other contaminants such as metals, and due to interactions, contaminated sites should be assessed as a multiple stressor scenario. Following deposition, a series of stressors, interactions and processes will influence the ecosystem transfer of radionuclide species and thereby influence biological uptake (toxicokinetics) and responses (toxicodynamics) in exposed organisms. Due to the variety of biological species, extrapolation is frequently needed to fill gaps in knowledge e.g., from effects to no effects, from effects in one organism to others, from one stressor to mixtures. Most toxtests are, however, performed as short term exposure of adult organisms, ignoring sensitive history life stages of organisms and transgenerational effects. To link sources, ecosystem transfer and biological effects to future impact and risks, a series of models are usually interfaced, while uncertainty estimates are seldom given. The model predictions are, however, only valid within the boundaries of the overall uncertainties. Furthermore, the model predictions are only useful and relevant when uncertainties are estimated, communicated and understood. Among key factors contributing most to uncertainties, the present paper focuses especially on structure uncertainties (model bias or discrepancies) as aspects such as particle releases, ecosystem dynamics, mixed exposure, sensitive life history stages and transgenerational effects, are usually ignored in assessment models. Research focus on these aspects should significantly reduce the overall uncertainties in the impact and risk assessment of radioactive contaminated ecosystems.

A general purpose feature extractor for light detection and ranging data.

Feature extraction is a central step of processing Light Detection and Ranging (LIDAR) data. Existing detectors tend to exploit characteristics of specific environments: corners and lines from indoor (rectilinear) environments, and trees from outdoor environments. While these detectors work well in their intended environments, their performance in different environments can be poor. We describe a general purpose feature detector for both 2D and 3D LIDAR data that is applicable to virtually any environment. Our method adapts classic feature detection methods from the image processing literature, specifically the multi-scale Kanade-Tomasi corner detector. The resulting method is capable of identifying highly stable and repeatable features at a variety of spatial scales without knowledge of environment, and produces principled uncertainty estimates and corner descriptors at same time. We present results on both software simulation and standard datasets, including the 2D Victoria Park and Intel Research Center datasets, and the 3D MIT DARPA Urban Challenge dataset.