Search for: All records

TRISH (Tree-Ring Integrated System for Hydrology), a new web-based tool for reconstruction of water-balance variables from tree-ring proxies is described. The tool makes use of a mapping application, a global water balance model and R-based reconstruction software. Long time series of water balance variables can be reconstructed by regression or analog statistical methods from tree-ring data uploaded by the user or available in TRISH as previously uploaded public datasets. A predictand hydroclimatic time series averaged or summed over a river basin or arbitrary polygon can be generated interactively by clicking on the map. Control over reconstruction modeling includes optional lagging of predictors, transformation of predictand, and reduction of predictors by principal component analysis. Output includes displayed and downloadable graphics, statistics, and time series. The two-stage reconstruction approach in TRISH allows assessment of the strength of the hydroclimatic signal in individual chronologies in addition to providing a reconstruction based on the tree-ring network. TRISH facilitates the testing of sensitivity of reconstructions to modeling choices and allows a user to explore hydrologic reconstruction in ungauged basins. The R software for reconstruction is available for running offline in the RStudio development environment. TRISH is an open-science resource designed to be shared broadly across the Earth Science research community and to engage water resource management. 

ABSTRACT Fast radio bursts (FRBs) are short-duration radio pulses of cosmological origin. Among the most common sources predicted to explain this phenomenon are bright pulses from a class of extremely highly magnetized neutron stars known as magnetars. Motivated by the discovery of an FRB-like pulse from the Galactic magnetar SGR 1935+2154, we searched for similar events in Messier 82 (M82). With a star formation rate 40 times that of the Milky Way, one might expect that the implied rate of events similar to that seen from SGR 1935+2154 from M82 should be 40 times higher than that of the Milky Way. We observed M82 at 1.4 GHz with the 20-m telescope at the Green Bank Observatory for 34.8 d. While we found many candidate events, none had a signal-to-noise ratio greater than 8. We also show that there are insufficient numbers of repeating low-significance events at similar dispersion measures to constitute a statistically significant detection. From these results, we place an upper bound for the rate of radio pulses from M82 to be 30 yr−1 above a fluence limit of 8.5 Jy ms. While this is less than nine times the rate of radio bursts from magnetars in the Milky Way inferred from the previous radio detections of SGR 1935+2154, it is possible that propagation effects from interstellar scattering are currently limiting our ability to detect sources in M82. Further searches of M82 and other nearby galaxies are encouraged to probe this putative FRB population. 

Designing deep neural networks is an art that often involves an expensive search over candidate architectures. To overcome this for recurrent neural nets (RNNs), we establish a connection between the hidden state dynamics in an RNN and gradient descent (GD). We then integrate momentum into this framework and propose a new family of RNNs, called MomentumRNNs. We theoretically prove and numerically demonstrate that MomentumRNNs alleviate the vanishing gradient issue in training RNNs. We study the momentum long-short term memory (MomentumLSTM) and verify its advantages in convergence speed and accuracy over its LSTM counterpart across a variety of benchmarks. We also demonstrate that MomentumRNN is applicable to many types of recurrent cells, including those in the state-of-the-art orthogonal RNNs. Finally, we show that other advanced momentum-based optimization methods, such as Adam and Nesterov accelerated gradients with a restart, can be easily incorporated into the MomentumRNN framework for designing new recurrent cells with even better performance.