@article{SteiningerAbelZiegleretal.2023,
  author    = {Steininger, Michael and Abel, Daniel and Ziegler, Katrin and Krause, Anna and Paeth, Heiko and Hotho, Andreas},
  title     = {ConvMOS: climate model output statistics with deep learning},
  series = {Data Mining and Knowledge Discovery},
  volume    = {37},
  journal   = {Data Mining and Knowledge Discovery},
  number    = {1},
  issn      = {1384-5810},
  doi       = {10.1007/s10618-022-00877-6},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-324213},
  pages     = {136-166},
  year      = {2023},
  abstract  = {Climate models are the tool of choice for scientists researching climate change. Like all models they suffer from errors, particularly systematic and location-specific representation errors. One way to reduce these errors is model output statistics (MOS) where the model output is fitted to observational data with machine learning. In this work, we assess the use of convolutional Deep Learning climate MOS approaches and present the ConvMOS architecture which is specifically designed based on the observation that there are systematic and location-specific errors in the precipitation estimates of climate models. We apply ConvMOS models to the simulated precipitation of the regional climate model REMO, showing that a combination of per-location model parameters for reducing location-specific errors and global model parameters for reducing systematic errors is indeed beneficial for MOS performance. We find that ConvMOS models can reduce errors considerably and perform significantly better than three commonly used MOS approaches and plain ResNet and U-Net models in most cases. Our results show that non-linear MOS models underestimate the number of extreme precipitation events, which we alleviate by training models specialized towards extreme precipitation events with the imbalanced regression method DenseLoss. While we consider climate MOS, we argue that aspects of ConvMOS may also be beneficial in other domains with geospatial data, such as air pollution modeling or weather forecasts.},
  subject      = {Klima},
  language  = {en}
}
@article{ZieglerPollingerBoelletal.2020,
  author    = {Ziegler, Katrin and Pollinger, Felix and B{\"o}ll, Susanne and Paeth, Heiko},
  title     = {Statistical modeling of phenology in Bavaria based on past and future meteorological information},
  series = {Theoretical and Applied Climatology},
  volume    = {140},
  journal   = {Theoretical and Applied Climatology},
  issn      = {0177-798X},
  doi       = {10.1007/s00704-020-03178-4},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-232717},
  pages     = {1467-1481},
  year      = {2020},
  abstract  = {Plant phenology is well known to be affected by meteorology. Observed changes in the occurrence of phenological phases arecommonly considered some of the most obvious effects of climate change. However, current climate models lack a representationof vegetation suitable for studying future changes in phenology itself. This study presents a statistical-dynamical modelingapproach for Bavaria in southern Germany, using over 13,000 paired samples of phenological and meteorological data foranalyses and climate change scenarios provided by a state-of-the-art regional climate model (RCM). Anomalies of severalmeteorological variables were used as predictors and phenological anomalies of the flowering date of the test plantForsythiasuspensaas predictand. Several cross-validated prediction models using various numbers and differently constructed predictorswere developed, compared, and evaluated via bootstrapping. As our approach needs a small set of meteorological observationsper phenological station, it allows for reliable parameter estimation and an easy transfer to other regions. The most robust andsuccessful model comprises predictors based on mean temperature, precipitation, wind velocity, and snow depth. Its averagecoefficient of determination and root mean square error (RMSE) per station are 60\% and ± 8.6 days, respectively. However, theprediction error strongly differs among stations. When transferred to other indicator plants, this method achieves a comparablelevel of predictive accuracy. Its application to two climate change scenarios reveals distinct changes for various plants andregions. The flowering date is simulated to occur between 5 and 25 days earlier at the end of the twenty-first century comparedto the phenology of the reference period (1961-1990).},
  language  = {en}
}