Search

Neurodegenerative disease often affects speech. Speech acoustics can be used as objective clinical markers of pathology. Previous investigations of pathological speech have primarily compared controls with one specific condition and excluded comorbidities. We broaden the utility of speech markers by examining how multiple acoustic features can delineate diseases. We used supervised machine learning with gradient boosting (CatBoost) to delineate healthy speech from speech of people with multiple sclerosis or Friedreich ataxia. Participants performed a diadochokinetic task where they repeated alternating syllables. We subjected 74 spectral and temporal prosodic features from the speech recordings to machine learning. Results showed that Friedreich ataxia, multiple sclerosis and healthy controls were all identified with high accuracy (over 82%). Twenty-one acoustic features were strong markers of neurodegenerative diseases, falling under the categories of spectral qualia, spectral power, and speech rate. We demonstrated that speech markers can delineate neurodegenerative diseases and distinguish healthy speech from pathological speech with high accuracy. Findings emphasize the importance of examining speech outcomes when assessing indicators of neurodegenerative disease. We propose large-scale initiatives to broaden the scope for differentiating other neurological diseases and affective disorders.

Accurate and timely forecasting of forest fuel moisture is critical for decision making in the context of bushfire risk and prescribed burning. The moisture content in forest fuels is a driver of ignition probability and contributes to the success of fuel hazard reduction burns. Forecasting capacity is extremely limited because traditional modelling approaches have not kept pace with rapid technological developments of field sensors, weather forecasting and data-driven modelling approaches. This research aims to develop and test a 7-day-ahead forecasting system for forest fuel dryness that integrates an automated fuel sensor network, gridded weather, landscape attributes and machine learning models. The integrated system was established across a diverse range of 30 sites in south-eastern Australia. Fuel moisture was measured hourly using 10-hour automated fuel sticks. A subset of long-term sites (5 years of data) was used to evaluate the relative performance of a selection of machine learning (Light Gradient Boosting Machine (LightGBM) and Recurrent Neural Network (RNN) based Long-Short Term Memory (LSTM)), statistical (VARMAX) and process-based models. The best performing models were evaluated at all 30 sites where data availability was more limited, demonstrating the models' performance in a real-world scenario on operational sites prone to data limitations. The models were driven by daily 7-day continent-scale gridded weather forecasts, in-situ fuel moisture observation and site variables. The model performance was evaluated based on the capacity to successfully predict minimum daily fuel dryness within the burnable range for fuel reduction (11 – 16%) and bushfire risk (

The world's most complex climate models are currently running a range of experiments as part of the Sixth Coupled Model Intercomparison Project (CMIP6). Added to the output from the Fifth Coupled Model Intercomparison Project (CMIP5), the total data volume will be in the order of 20PB. Here, we present a dataset of annual, monthly, global, hemispheric and land/ocean means derived from a selection of experiments of key interest to climate data analysts and reduced complexity climate modellers. The derived dataset is a key part of validating, calibrating and developing reduced complexity climate models against the behaviour of more physically complete models. In addition to its use for reduced complexity climate modellers, we aim to make our data accessible to other research communities. We facilitate this in a number of ways. Firstly, given the focus on annual, monthly, global, hemispheric and land/ocean mean quantities, our dataset is orders of magnitude smaller than the source data and hence does not require specialized ‘big data’ expertise. Secondly, again because of its smaller size, we are able to offer our dataset in a text‐based format, greatly reducing the computational expertise required to work with CMIP output. Thirdly, we enable data provenance and integrity control by tracking all source metadata and providing tools which check whether a dataset has been retracted, that is identified as erroneous. The resulting dataset is updated as new CMIP6 results become available and we provide a stable access point to allow automated downloads. Along with our accompanying website (cmip6.science.unimelb.edu.au), we believe this dataset provides a unique community resource, as well as allowing non‐specialists to access CMIP data in a new, user‐friendly way.