Bias correction, adjustment
See:
# Resources
# Courses
# References
- #PAPER #BSC Use of bias correction techniques to improve seasonal forecasts for reservoirs - A case-study in northwestern Mediterranean (Marcos 2018)
- #PAPER
Neural networks for post-processing ensemble weather forecasts (Rasp 2018)
- Extend to gridded data with CNNs?
- #PAPER Bias adjustment and ensemble recalibration methods for seasonal forecasting: a comprehensive intercomparison using the C3S dataset (Manzanas 2019)
- #PAPER Machine learning for observation bias correction with application to dust storm data assimilation (Jin 2019)
- #PAPER Calibration of ECMWF seasonal SEAS5 models monthly temperature re-forecasts over the southeast Asia region (Yun 2020)
- #PAPER Comparative Assessment of Various Machine Learning‐Based Bias Correction Methods for Numerical Weather Prediction Model Forecasts of Extreme Air Temperatures in Urban Areas (Cho 2020)
- #TALK Downscaling and bias correction of seasonal forecasts to support climate services for the Alpine regions (Crespi 2020)
- #PAPER
Deep learning for Post-Processing Ensemble Weather Forecasts (Gronquist, 2020) ^gronquist20
- https://royalsocietypublishing.org/doi/abs/10.1098/rsta.2020.0092?af=R
- #CODE https://github.com/spcl/deep-weather
- #TALK https://www.youtube.com/watch?v=5REg7_UtJcs
- Bias correction with AI/Deep learning/DL
- Simulations are started from GT (from data assimilation)
- Then get statistics of the ensemble. This is very costly. Can we learn mu and sigma from smaller ensembles?
- ENS10 and ERA5, selected variables
- Bias correction input (ERA5) model
- Uncertainty quantification (ENS10)
- ResNet, 3D UNET with added locally-connected network (LCN) as the last layer
- SSIM as training loss
- CRPS loss function
- Extreme events too
- #PAPER A new bias-correction method for precipitation over complex terrain suitable for different climate states: a case study using WRF (Velasquez 2020)
- #PAPER Deep learning for bias correction of MJO prediction (Kim 2021)
# MOS
- MOS is a multiple linear regression technique in which predictands, often near-surface quantities, such as 2-meter (AGL) air temperature, horizontal visibility, and wind direction, speed and gusts, are related statistically to one or more predictors
#PAPER Deep Learning for Climate Model Output Statistics (Steininger 2020)
#PAPER A Model Output Deep Learning Method for Grid Temperature Forecasts in Tianjin Area (Chen 2020)
- This paper proposes a model output deep learning (MODL) method for post-processing
- Samples are multi-variable and spatio-temporal (53 time steps as leadtime)
- The core of the MODL is 3D Fully Convolutional Neural Networks (3D FCNN)
- The 3D FCNN or MODL-PLAIN is composed of three convolutional layers. Compared with a UNET-style CNN architecture
- Each CNN block has CON3d -> activation -> BatchNorm
- Two DL models, one with 3 CNN layers, one with a UNET-like structure
- MSE loss function
- MODL method is better than the univariate linear MOS method, the MOML method based random forest, and linear regression with a running period, and it has the ability to improve grid temperature forecast results in Tianjin area
- Weather (few days) forecasting
- #CODE https://github.com/bastien-francois/MBC_CycleGAN
- multivariate bias correction (MBC) method
- adapted a computer vision technique used for Image-to-Image translation tasks (CycleGAN) for the adjustment of spatial dependence structures of climate model projections
- the method is applied to adjust maps of temperature and precipitation from climate simulations through two cross-validation approaches
- The first one is designed to assess two different post-processing schemes (Perfect Prognosis and Model Output Statistics)
- The second one assesses the influence of nonstationary properties of climate simulations on the performance of MBC-CycleGAN to adjust spatial dependences
- For the performance assessment of the CycleGAN model during training, the energy distance is used
- This metric, already used in the bias correction literature permits to measure the statistical discrepancy between two multivariate distributions that are potentially in high dimension
#PAPER Deep learning for bias correction of MJO prediction (Kim 2021)