Search for: All records

Supraglacial lakes on the Greenland Ice Sheet (GrIS) can impact both the ice sheet surface mass balance and ice dynamics. Thus, understanding the evolution and dynamics of supraglacial lakes is important to provide improved parameterizations for ice sheet models to enable better projections of future GrIS changes. In this study, we utilize the growing inventory of optical and microwave satellite imagery to automatically determine the fate of Greenland-wide supraglacial lakes during 2018 and 2019; cool and warm melt seasons respectively. We develop a novel time series classification method to categorize lakes into four classes: 1) refreezing, 2) rapidly draining, 3) slowly draining, and 4) buried. Our findings reveal significant interannual variability between the two melt seasons, with a notable increase in the proportion of draining lakes in 2019. We also find that as mean lake depth increases, so does the percentage of lakes that drain, indicating that lake depth may influence hydrofracture potential. However, we also observe that non-draining lakes are deeper during the cooler 2018 melt season, suggesting that additional factors may predispose lakes to drain earlier in a warmer year. Our automatic classification approach and the resulting two-year ice-sheet-wide dataset provide unprecedented insights into GrIS supraglacial lake dynamics and evolution, offering a valuable resource for future research. 

Supraglacial lakes form on the surface of the Greenland Ice Sheet during the summer months and can directly impact ice sheet mass balance by removing mass via drainage and runoff or indirectly impact mass balance by influencing ice sheet dynamics. Here, we utilize the growing inventory of optical and microwave satellite imagery to automatically determine the fate of Greenland-wide supraglacial lakes during 2018 and 2019, a cool and warm melt season respectively. We use a machine learning time series classification approach to categorize lakes into four different categories: lakes that 1) refreeze, 2) rapidly drain, 3) slowly drain, and 4) become buried lakes at the end of the melt season. We find that during the warmer 2019, not only was the number of lake drainage events higher than in 2018, but also the proportion of lakes that drained was greater. By investigating mean lake depths for these four categories, we show that drained lakes were, on average, 22% deeper than lakes that refroze or became buried lakes. Interestingly, drained lakes had approximately the same maximum depth in 2018 and 2019; however, lakes that did not drain were 29% deeper in 2018, a cooler year. Our unique two-year dataset describing the fate of every Greenland supraglacial lake provides novel insight into lake drainage and refreeze in a relatively warm and cool year, which may be increasingly relevant in a warming climate. 

Causal discovery is an important problem in many sciences that enables us to estimate causal relationships from observational data. Particularly, in the healthcare domain, it can guide practitioners in making informed clinical decisions. Several causal discovery approaches have been developed over the last few decades. The success of these approaches mostly relies on a large number of data samples. In practice, however, an infinite amount of data is never available. Fortunately, often we have some prior knowledge available from the problem domain. Particularly, in healthcare settings, we often have some prior knowledge such as expert opinions, prior RCTs, literature evidence, and systematic reviews about the clinical problem. This prior information can be utilized in a systematic way to address the data scarcity problem. However, most of the existing causal discovery approaches lack a systematic way to incorporate prior knowledge during the search process. Recent advances in reinforcement learning techniques can be explored to use prior knowledge as constraints by penalizing the agent for their violations. Therefore, in this work, we propose a framework KCRL that utilizes the existing knowledge as a constraint to penalize the search process during causal discovery. This utilization of existing information during causal discovery reduces the graph search space and enables a faster convergence to the optimal causal mechanism. We evaluated our framework on benchmark synthetic and real datasets as well as on a real-life healthcare application. We also compared its performance with several baseline causal discovery methods. The experimental findings show that penalizing the search process for constraint violation yields better performance compared to existing approaches that do not include prior knowledge.