Evaluating the Dark Sky Weather Forecast Data Over Florida and Georgia

Haimanote Kebede Bayabil, Fitsum Tilahun Teshome, Songzi Wu, Ziwen Yu

Research Square · 01 August 2022

In one sentence. Evaluates Dark Sky daily forecasts against 124 Florida and Georgia stations; finds good yes/no rain detection (POD 0.91–1.0) but poor rainfall amounts.

Abstract

Abstract The Dark Sky (DKS; © 2020 Apple Inc.) provides weather forecasts for most locations throughout the world. The objectives of this study were to 1) evaluate the accuracy of the DKS system in capturing selected weather parameters compared to observations at 124 weather stations in Florida and Georgia and 2) provide a strategy using machine learning to identify features that affect the accuracy of rainfall forecast. Daily forecast weather data were obtained using the DKS’ application programming interface (API). Data included daily total rainfall, minimum (Tmin) and maximum (Tmax) temperatures, daily average wind speed, dew point, and relative humidity (RH). The DKS system performed better in capturing forecasts of Tmin and Tmax, and dew point but its performances with rainfall, RH, and wind speed were poor. In addition, the DKS forecast system over-predicts rainfall events <5mm while under-predicting events >5mm. However, the DKS system was able to forecast the occurrence (yes/no) of rainfall events with a probability of detection (POD) values between 0.91 and 1.0. Machine learning model results showed that the accuracy of rainfall forecast was affected by changes in temperature after a rainfall event. The error of a next-day precipitation forecast was < 2.5 mm when post rainfall temperature drops by 5℃. Overall, results suggest that the DKS forecast system could be a reliable source of dichotomous (yes/no) rainfall forecasts, but its quantitative rainfall predictions were poor.

Labels

Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.