Towards an operational CyGNSSnet - automated ocean wind speed prediction
- 1German Climate Computing Center (DKRZ) , Hamburg, Germany (grover@dkrz.de)
- 2Helmholtz AI, Germany
The Cyclone Global Navigation Satellite System (CyGNSS) is a constellation of eight microsatellites launched in 2016 with the goal of measuring global ocean wind speed. With four channels on each satellite, it produces up to 32 Delay Doppler maps (DDMs) per second. CyGNSSnet [1] is a machine learning algorithm developed to predict ocean surface wind speed directly from DDMs. Evaluated on an independent test set, CyGNSSnet achieved an RMSE of 1.36 m/s. It is however unknown whether the algorithm’s performance is stable, the further the evaluation date is from the training data range.
New DDMs are provided every day through the NASA EarthData cloud [2]. Here, we present an automatic machine learning pipeline to evaluate CyGNSSnet continuously using Prefect, a Python library for workflow orchestration. This allows us to schedule the pipeline daily and to handle the process smoothly in case of any failures.
The workflow of the pipeline is as follows: the CyGNSS data and the ERA5 windspeed labels are downloaded and pre-processed. Wind speed predictions are made with the pretrained CyGNSSnet. Metrics of the model, including root mean squared error and bias, as well as visualizations are stored in a MongoDB database. All saved data can be accessed through a website, where users can analyze the current performance of CyGNSSnet and access previous visualizations and results.
The pipeline can be set up system-independent via Docker compose. It can easily be adapted to other remote sensing data sources and machine learning algorithms, provides a valuable software tool to leverage big data in remote sensing, and enables continuous validation of machine learning algorithms. In our contribution, we will demonstrate the performance of CyGNSSnet for the months leading up to the conference.
[1] Asgarimehr, M., Arnold, C., Weigel, T., Ruf, C., Wickert, J. (2022): GNSS reflectometry global ocean wind speed using deep learning: Development and assessment of CyGNSSnet. - Remote Sensing of Environment, 269, 112801.
[2] CYGNSS. CYGNSS Level 2 Science Data Record Version 3.1. Ver. 3.1. PO.DAAC, CA, USA. accessed 2022/2023 at 10.5067/CYGNS-L2X31
How to cite: Grover, H., Albrecht, F., and Arnold, C.: Towards an operational CyGNSSnet - automated ocean wind speed prediction, EGU General Assembly 2023, Vienna, Austria, 24–28 Apr 2023, EGU23-1475, https://doi.org/10.5194/egusphere-egu23-1475, 2023.
