Predicting Rain Attenuation for Satellite Signals Based on Publicly Available Rainfall Data

Abstract

Predicting attenuation from hourly rainfall on satellite signals would improve response times and operator workload allocated to boost signal strength when affected by rain, improving user experience. This paper examines a composite dataset from the National Climactic Data Center representing 61 original variables engineered into 92 predictor variables supporting the precipitation parameter, spread across 8,877 observations. This paper examines 8 regression models: a baseline, 3 classical, and 4 neural network models. The training dataset had a precipitation mean of 0.1485 / . The Mean Squared Error (MSE) for the trivial baseline model was 0.8898 (/ )^2 . The best-performing model was a deep neural network model with 10,177 trainable parameters, one input layer, 3 hidden layers, including a dropout layer, and one output layer. It was also trained with early stopping. This models MSE was 7.4048 (/ )^2 . It showed the most consistent residual plots out of all examined models yet still failed to provide an operationally valuable model. Future efforts to provide a meaningful model include different data aggregation methods, consulting with a meteorologist, and switching from a regression approach predicting the amount of rainfall to a classification approach predicting whether or not rain is likely.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Mar 06, 2023
Accession Number
AD1201358

Entities

People

  • David M. Vermillion

Organizations

  • Air Force Institute of Technology

Tags

Communities of Interest

  • Energy and Power Technologies
  • Space

DTIC Thesaurus Topics

  • 5G Wireless Networks
  • Air Force Research Laboratories
  • Artificial Satellites
  • Communication Systems
  • Data Centers
  • Drops
  • Extremely High Frequency
  • Frequency
  • Geosynchronous Orbits
  • Ground Stations
  • Machine Learning
  • Measurement
  • Networks
  • Neural Networks
  • Precipitation
  • Radio Communications
  • Radio Frequency
  • Radio Waves
  • Satellite Communications
  • User Interface

Fields of Study

  • Environmental science

Readers

  • Adaptive Control and Estimation with Uncertainty in Dynamic Systems.
  • Atmospheric Science/Meteorology
  • Economics

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - Neural Networks
  • Space