2020/08/17 by David B. Huberman, Huberman, David B., Brian J. Reich +3 · 1 citation
Earth and Planetary Sciences · Environmental Science · #Applications (stat.AP) #FOS: Computer and information sciences #Flood Risk Assessment and Management #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Tropical and Extratropical Cyclones Research
paper · pdf · doi:10.48550/arxiv.2008.07653
openalex publication_date 2020/08/17 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
Short-term forecasting is an important tool in understanding environmental\nprocesses. In this paper, we incorporate machine learning algorithms into a\nconditional distribution estimator for the purposes of forecasting tropical\ncyclone intensity. Many machine learning techniques give a single-point\nprediction of the conditional distribution of the target variable, which does\nnot give a full accounting of the prediction variability. Conditional\ndistribution estimation can provide extra insight on predicted response\nbehavior, which could influence decision-making and policy. We propose a\ntechnique that simultaneously estimates the entire conditional distribution and\nflexibly allows for machine learning techniques to be incorporated. A smooth\nmodel is fit over both the target variable and covariates, and a logistic\ntransformation is applied on the model output layer to produce an expression of\nthe conditional density function. We provide two examples of machine learning\nmodels that can be used, polynomial regression and deep learning models. To\nachieve computational efficiency we propose a case-control sampling\napproximation to the conditional distribution. A simulation study for four\ndifferent data distributions highlights the effectiveness of our method\ncompared to other machine learning-based conditional distribution estimation\ntechniques. We then demonstrate the utility of our approach for forecasting\npurposes using tropical cyclone data from the Atlantic Seaboard. This paper\ngives a proof of concept for the promise of our method, further computational\ndevelopments can fully unlock its insights in more complex forecasting and\nother applications.\n