Stephanie R Clark

Dr. Stephanie R Clark

CSIRO Land and Water, Canberra, Australian Capital Territory, Australia
orcid

Research Focus & Metrics

Academic Career Length

13Years (2014-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 September 2020 in Advances in Water Resources
Verified
Journal Article
Closed access
SCStephanie R Clark
Stephanie R Clark
SSScott.A. Sisson
Scott.A. Sisson
Ashish SharmaAshish Sharma
Ashish Sharma

13 March 2024 in Hydrology and Earth System Sciences
Verified
Journal Article
Open access
PDF file
SCStephanie R Clark
Stephanie R Clark
JLJulien Lerat
Julien Lerat
JPJean-Michel Perraud
Jean-Michel Perraud
PFPeter Fitch
Peter Fitch
A deep learning model designed for time series predictions, the long short-term memory (LSTM) architecture, is regularly producing reliable results in local and regional rainfall–runoff applications around the world. Recent large-sample hydrology studies in North America and Europe have shown the LSTM model to successfully match conceptual model performance at a daily time step over hundreds of catchments. Here we investigate how these models perform in producing monthly runoff predictions in the relatively dry and variable conditions of the Australian continent. The monthly time step matches historic data availability and is also important for future water resources planning; however, it provides significantly smaller training datasets than daily time series. In this study, a continental-scale comparison of monthly deep learning (LSTM) predictions to conceptual rainfall–runoff (WAPABA model) predictions is performed on almost 500 catchments across Australia with performance results aggregated over a variety of catchment sizes, flow conditions, and hydrological record lengths. The study period covers a wet phase followed by a prolonged drought, introducing challenges for making predictions outside of known conditions – challenges that will intensify as climate change progresses. The results show that LSTM models matched or exceeded WAPABA prediction performance for more than two-thirds of the study catchments, the largest performance gains of LSTM versus WAPABA occurred in large catchments, the LSTMs struggled less to generalise than the WAPABA models (e.g. making predictions under new conditions), and catchments with few training observations due to the monthly time step did not demonstrate a clear benefit with either WAPABA or LSTM.

23 July 2019 in Clinical Journal of the American Society of Nephrology
Verified
Journal Article
Closed access
PDF file
SCStephanie R Clark
Stephanie R Clark
MKMatthew Kadatz
Matthew Kadatz
JGJagbir Gill
Jagbir Gill
JGJohn S. Gill
John S. Gill
Background and objectives: Patients who have failed a transplant are at increased risk of repeat transplant failure. We determined access to transplantation and transplant outcomes in patients with and without a history of transplant failure. Design, setting, participants, & measurements: In this observational study of national data, the proportion of waitlisted patients and deceased donor transplant recipients with transplant failure was determined before and after the new kidney allocation system. Among patients initiating maintenance dialysis between May 1995 and December 2014, the likelihood of deceased donor transplantation was determined in patients with (n=27,459) and without (n=1,426,677) a history of transplant failure. Among transplant recipients, allograft survival, the duration of additional kidney replacement therapy required within 10 years of transplantation, and the association of transplantation versus dialysis with mortality was determined in patients with and without a history of transplant failure. Results: The proportion of waitlist candidates (mean 14%) and transplant recipients (mean 12%) with transplant failure did not increase after the new kidney allocation system. Among patients initiating maintenance dialysis, transplant-failure patients had a higher likelihood of transplantation (hazard ratio [HR], 1.16; 95% confidence interval [95% CI], 1.12 to 1.20; P<0.001). Among transplant recipients, transplant-failure patients had a higher likelihood of death-censored transplant failure (HR, 1.44; 95% CI, 1.34 to 1.54; P<0.001) and a greater need for additional kidney replacement therapy required within 10 years after transplantation (mean, 9.0; 95% CI, 5.4 to 12.6 versus mean, 2.1; 95% CI, 1.5 to 2.7 months). The association of transplantation versus dialysis with mortality was clinically similar in waitlisted patients with (HR, 0.32; 95% CI, 0.29 to 0.35; P<0.001) and without transplant failure (HR, 0.40; 95% CI, 0.39 to 0.41; P<0.001). Conclusions: Transplant-failure patients initiating maintenance dialysis have a higher likelihood of transplantation than transplant-naïve patients. Despite inferior death-censored transplant survival, transplantation was associated with a similar reduction in the risk of death compared with treatment with dialysis in patients with and without a prior history of transplant failure.

SCStephanie R Clark
Stephanie R Clark
DPDaniel E. Pagendam
Daniel E. Pagendam
Louise RyanLouise Ryan
Louise Ryan
Time series data from environmental monitoring stations are often analysed with machine learning methods on an individual basis, however recent advances in the machine learning field point to the advantages of incorporating multiple related time series from the same monitoring network within a ‘global’ model. This approach provides the opportunity for larger training data sets, allows information to be shared across the network, leading to greater generalisability, and can overcome issues encountered in the individual time series, such as small datasets or missing data. We present a case study involving the analysis of 165 time series from groundwater monitoring wells in the Namoi region of Australia. Analyses of the multiple time series using a variety of different aggregations are compared and contrasted (with single time series, subsets, and all of the time series together), using variations of the multilayer perceptron (MLP), self-organizing map (SOM), long short-term memory (LSTM), and a recently developed LSTM extension (DeepAR) that incorporates autoregressive terms and handles multiple time series. The benefits, in terms of prediction performance, of these various approaches are investigated, and challenges such as differing measurement frequencies and variations in temporal patterns between the time series are discussed. We conclude with some discussion regarding recommendations and opportunities associated with using networks of environmental data to help inform future resource-related decision making.

7 June 2014 in Ecological Informatics
Verified
Journal Article
Closed access
SCStephanie R Clark
Stephanie R Clark
PSPeter Sarlin
Peter Sarlin
Ashish SharmaAshish Sharma
Ashish Sharma
SSScott A. Sisson
Scott A. Sisson

5 January 2022 in Environmental Modelling & Software
Verified
Journal Article
Closed access
SCStephanie R Clark
Stephanie R Clark