Abstract
Regional rainfall forecasting is an important issue in hydrology and meteorology. Machine learning algorithms especially deep learning methods have emerged as a part of prediction tools for regional rainfall forecasting. This paper aims to design and implement a generic computing framework that can assemble a variety of machine learning algorithms as computational engines for regional rainfall forecasting in Upstate New York. The algorithms that have been bagged in the computing framework include the classical algorithms and the state-of-the-art deep learning algorithms, such as K-Nearest Neighbors, Support Vector Machine, Deep Neural Network, Wide Neural Network, Deep and Wide Neural Network, Reservoir Computing, and Long Short Term Memory methods. Through the experimental results and the performance comparisons of these various engines, we have observed that the SVM- and KNN-based method are outstanding models over other models in classification while DWNN- and KNN-based methods outstrip other models in regression, particularly those prevailing deep-learning-based methods, for handling uncertain and complex climatic data for precipitation forecasting. Meanwhile, the normalization methods such as Z-score and Minmax are also integrated into the generic computing framework for the investigation and evaluation of their impacts on machine learning models.
Highlights
The global climate changes and the uneven weather conditions in different spatialtemporal scales are causes for severe problems like droughts and floods [1]
The KNearest Neighbors (KNN), Support Vector Machine (SVM), Linear, and Reservoir Computing (RC) models for rainfall forecasting were implemented based on SKLearn libraries, running on M1 Mac (Apple M1 chip, 8-core CPU, 8-core GPU, 16-core Neural Engine, 16.0 GB RAM, 1TB SSD HD)
For models whose parameter had a virtually unlimited set of possible values, like the number of layers for Deep Neural Network (DNN) or Deep and Wide Neural Network (DWNN), we tried multiple values spaced apart in such a way to try and capture the best that could be done in a reasonable amount of time
Summary
The global climate changes and the uneven weather conditions in different spatialtemporal scales are causes for severe problems like droughts and floods [1]. As droughts and floods become more frequent in the changing climate, accurate rainfall forecasting becomes more important for planning in agriculture and other relevant activities. Regional rainfall forecasting constrains the spatial variable to local or a particular region, making the prediction processing relatively controllable comparing with the globe weather prediction. The problem of regional rainfall forecasting can be briefly described as follows: Given a number of historical weather data including rainfall information in a particular place or a region, one tries to devise a computational model that can predict and tell the rainfall status either categorically or quantitatively in the period of time in the future.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
