Interval Predictions of Upper and Lower Bounds for Global 2019 HCHO Surface Concentration Distribution
This is the PyTorch implementation for our work--Mapping 2019 Global Surface HCHO Distribution and Confidential Interval by Satellite Observation of Sentinel-5P and Neural Network Model. With the usage of quality-driven interval estimation algorithm(High-Quality Prediction Intervals for Deep Learning), we manage to give the point and interval global HCHO surface concentration distribution in 2019.
Contact: Bohan Jin (2018200684@ruc.edu.cn); Yizhe Ding (1810015@mail.nankai.edu.cn)
Notice that ReLU activations in the last block are disabled.
Our project contains 4 code files:
- dataset.py
- function.py
- interval.ipynb
- point.ipynb
Hyperparameters are included in our files and you can run interval.ipynb and point.ipynb to reproduce our results.
Point Estimation of Global 2019 HCHO Surface Concentration Distribution
You can download the data from the website mentioned in our paper to train the network.
We have also provided our results in .tif format for downloading.