Predicting Hurricane Trajectories Using a Recurrent Neural Network

Authors: Sheila Alemany, Jonathan Beltran, Adrian Perez, Sam Ganzfried468-475

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Results show that this proposed technique is competitive to methods currently employed by the NHC and can predict up to approximately 120 hours of hurricane path.
Researcher Affiliation Collaboration 1School of Computing and Information Sciences, Florida International University, Miami, FL 2Ganzfried Research, Miami, FL {salem010, jbelt021, apere946}@fiu.edu, sam.ganzfried@gmail.com
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any concrete access information (e.g., specific repository link, explicit code release statement) for the source code of the methodology described in this paper.
Open Datasets Yes The raw Atlantic hurricane/tropical storm data used in the study were extracted from the NOAA database.2 The data contains all hurricanes and tropical storms from 1920 to 2012. Footnote 2 points to: http://weather.unisys.com/hurricane/atlantic/
Dataset Splits Yes The data provided by Unisys Weather data was divided where 85% of the total hurricanes were used for training, and 15% were used for testing the accuracy of our model. In other words, the training set and testing contained 27,477 and 4,850 individual data tuples, respectively. As a result, validation of the training set was completed on 10% of the 85% training set, or 2,747 data tuples.
Hardware Specification Yes The model was trained on an NVIDIA Ge Force GTX 1060 with 6GB of RAM which allowed the model to complete training in 200 seconds.
Software Dependencies No The paper mentions using 'Keras' and 'TensorFlow' but does not provide specific version numbers for these software dependencies.
Experiment Setup Yes The number of grid blocks in our model could be tuned depending on the amount of hurricane data available. ... we utilized a total of 7,256 grid blocks. The grid blocks were of size 1x1 degrees latitude by longitude. A regularization hyperparameter, known as the dropout value, randomly ignores a percentage of the input to prevent the model from co-adapting to the training set, or overfitting, of hurricane trajectories (Srivastava et al. 2014). This value was set to 0.1 and was tuned and selected using cross-validation... Therefore, we employed three hidden layers each with a long short-term memory cell... Keras provides a default learning rate hyperparameter of the value 0.001.