Recalibrating Extremes for Decadal Predictions

<p>Decadal predictions have become essential for near-term decision making and adaptation strategies. In parallel, interest in weather and climate extremes has increased strongly in the past. Thus, a combination of decadal predictions and extreme value theory is reasonable and necessary. Since decadal predictions suffer from typical discrepancies, such as start- and lead-year dependent conditional and unconditional biases, many ways for their recalibration have been proposed (Eade et al., 2014; Fučkar et al.,2014; Fyfe et al., 2011; Kharin et al., 2012; Kruschke et al., 2016; Raftery et al., 2005; Sansom et al., 2016; Sloughter et al., 2007). However, in previous studies, extremes have not been considered. Therefore, the aim of this study is to investigate how extremes from decadal predictions can be adequately recalibrated and how this affects forecasting skill. Pasternack et al. (2018) introduced a parametric Decadal Climate Forecast Recalibration Strategy (DeFoReSt 1.0), based on estimating polynomial adjustment terms (Gangstø et al., 2013). DeFoReSt assumes normality for the probability distribution (PDF) to be recalibrated and optimizes the cross-validated continuous ranked probability score (CRPS) with this assumption build in Gneiting et al. (2005). For a proof of concept, Pasternack et al. (2018) introduced a toy model for generating pseudo decadal forecast-observation pairs. For toy model data and surface temperatures from MiKlip hindcasts, improvement of forecast quality over a simple calibration from Kruschke et al. (2016) has been found. We extend these methods to extreme values with two modifications: (1) Follow DeFoReSt, but assume general extreme value (GEV) distributed forecasts. Again the CRPS is optimized but with the GEV build into the score (Friederichs and Thorarinsdottir, 2012). Both DeFoReSt strategies (DeFoReSt-normaland DeFoReSt-GEV) and the calibration from Kruschke et al. (2016) are compared to a forecast based on climatology. (2) The toy model is modified to generate pseudo decadal forecast-observation pairs with GEV distributed observations. For validation, a bootstrapping scheme is applied to temperature maxima hindcasts from MiKlip verified with HadEX2 observations. After recalibration, both DeFoReSt strategies perform similar for the toy model and MiKlip hindcasts, none significantly outperforms the other. However, they consistently show considerable improvements over the climatological forecast for the lower and upper quartiles in the toy model data. For the recalibrated MiKlip hindcasts, the findings are in accordance, but not as considerable, presumably due to their very small ensemble size (Sienz et al., 2016). This suggests that extremes may be directly recalibrated with the assumption of a Normal distribution, as long as this represents the characteristics of the decadal forecast ensemble. Thus, the forecasting skill of recalibrations appears to be unaffected by the underlying distribution of the observations.</p>

Can a model weighting scheme be used to obtain skillful, reliable and seamless climate information for the next 1-40 years?

<p>Skillful, reliable and seamless climate information for the next 1-40 years is crucial for policy- and other decision makers to develop suitable planning strategies. This poses a challenge for the scientific community, which is split up into the prediction community (developing initialized predictions up to multi-annual time scales, e.g. 10 years), and the climate projection community (providing long-term projections). As predictions are initialized with the observed climate state at the start of the integration, they are often more skillful for lead times of a few years (depending on variable and region) compared to uninitialized climate projections, which can provide information beyond 10 years. Thus, most useful climate information for the next 1-40 years would likely need to draw upon information from both sources. However, temporal merging from different sources is challenging, e.g., it can lead to discontinuities in the central estimates at the respective transition points, which pose problems for interpretation and communication alike.</p><p>The aim of this study is to explore if skillful and seamless climate information can be provided by applying a model weighting scheme to initialized decadal predictions and projections. The model specific weights are based on the respective past model performance compared to observations. Whereas for climate projections each model is assigned a single weight, for initialized decadal predictions these weights are calculated for each forecast year separately. Here, we apply the weighting technique to CMIP6 decadal predictions and climate projections from 8 different models. </p><p><br><br></p>

The DWD climate prediction website

<p>DWD provides operational seasonal and decadal predictions of the German climate prediction system since 2016 and 2020, respectively. We plan to present these predictions together with post-processed ECMWF sub-seasonal forecast products on the DWD climate prediction website www.dwd.de/climatepredictions. In March 2020, this climate service was published with decadal predictions for the coming years; sub-seasonal and seasonal predictions for the coming weeks and months will follow.</p><p>The user-oriented evaluation and design of this climate service has been developed in close cooperation with users from various sectors at workshops of the German MiKlip project and will be consistent across all time scales. The website offers maps, time series and tables of ensemble mean and probabilistic predictions in combination with the prediction skill for 1-year and 5-year means/ sums of temperature and precipitation for different regions (World, Europe, Germany, German regions).</p><p>For Germany, the statistical downscaling EPISODES was applied to reach high spatial resolution required by several climate data users. Decadal predictions were statistically recalibrated in order to adjust bias, drift and standard deviation and optimize ensemble spread. We used the MSESS and RPSS to evaluate the skill of climate predictions in comparison to reference predictions, e.g. ‘observed climatology’ or ‘uninitialized climate projections’ (which are both applied by users until now as an alternative to climate predictions). The significance was tested via bootstraps.</p><p>Within the ‘basic climate predictions’ section, a user-oriented traffic light indicates whether regional-mean climate predictions are significantly better (green), not significantly different (yellow) or significantly worse (red) than reference predictions. Within the ‘expert climate predictions’ section, prediction maps show per grid box the prediction itself (via the color of dots) and its skill (via the size of dots representing the skill categories of the traffic light). The co-development of this climate prediction application with users from different sectors strongly improves the comprehensibility and applicability by users in their daily work.</p><p>In addition to sub-seasonal and seasonal predictions, plans for future extensions of this climate service include multi-year seasonal predictions, e.g. 5-year summer or winter means, combined products for climate predictions and climate projections, further user-oriented, extreme or large-scale variables, e.g. ENSO, or high-resolution applications for German cities based on statistically downscaled predictions.</p>

Assessing short-term climate change impacts on water supply at the Wupper catchment area, Germany

The occurrence of dry periods on the Wupper catchment has increased in the last decades in conjunction with the shifting of the precipitation regime. In the frame of the Horizon 2020 project BINGO (Bringing INnovation to onGOing water management), the effects of climate change scenarios on the water cycle in the Wupper catchment area were investigated. To quantify these effects, a set of hydrological models (NASIM and SWAT) has been set-up, calibrated, and validated for the upper part of the Dhünn River catchment area – Wupper River's main tributary. This sub-catchment corresponds to one of the inflows to the Große Dhünn Reservoir (GDR), the second largest drinking water reservoir in Germany. Both models were driven with climate data from decadal predictions, which have been selected instead of IPCC-RCP scenarios, as they provide a more realistic assumption of climate variability for the next 10 years. Ten decadal members based on the MiKlip (Mittelfristige Klimaprognose – medium-term climate prediction) framework have been prepared for the time span of 2015 to 2024. Additionally, a simulation with TALSIM-NG (a reservoir-oriented hydrological model) was carried out to obtain future reservoir storage. Special focus was given to identify observed trends and compare them to future trends. Past hydro-meteorological extreme dry periods were evaluated based on observed data. Standardized Precipitation Index (SPI), Standardized Precipitation-Evapotranspiration Index (SPEI), and Standardized Runoff Index (SRI) were estimated for different seasons to determine if they were abnormally dry or wet. SPI, SPEI, and SRI were also calculated with decadal predictions to evaluate future extreme dry periods. Uncertainties in climate data predictions are one of the greatest challenges. Observed and forecast time series were compared by means of statistical tests in order to assess uncertainties in climate data predictions. Also, the application of two hydrological models aims to determine potential uncertainties, so that predictions are more reliable. Results indicate that SRI might be more appropriate to estimate drought periods for the study area in the frame of reservoir management – where inflow rates are of crucial importance – as this index quantifies losses in runoff formation processes. In terms of inflow rates to GDR, future changes indicate a reduction in runoff for the spring season, while an increment during winter. On the other hand, a clear change in pattern for fall and summer seasons remains uncertain. Simulations of GDR reservoir volume with different climate scenarios show that water stress by the end of 2024 is not unlikely, so sustainable adaptation measures should be further considered. Effectively managing the GDR will become consequently more complex.

Evaluation of land water storage prediction skill in CMIP5 decadal hindcasts by means of a GRACE-based data set

<p>Reliable predictions of terrestrial water storage (TWS) changes for the next couple of years would be extremely valuable for, e.g., agriculture and water management. In contrast to long-term projections of future climate conditions, so-called decadal predictions do not depend on prescribed CO<sub>2 </sub>scenarios but provide unconditional forecasts similar to numerical weather models. Therefore, opposed to climate projections, decadal predictions (or hindcasts, if run for the past) can directly be compared to observations. Here, we evaluate decadal hindcasts of TWS related variables from an ensemble of 5 coupled CMIP5 climate models against a TWS data set based on GRACE satellite observations.</p> <p>Since data from the CMIP5 models and GRACE is jointly available in only 9 years, we access a GRACE-like reconstruction of TWS derived from precipitation and temperature data sets (Humphrey and Gudmundsson, 2019), which expands the analysis time-frame to 41 years. The skill of the decadal hindcasts is assessed by means of anomaly correlations and root-mean-square deviations (RMSD) for the yearly global average and aggregated over different climate zones. Furthermore, we compute global maps of correlation and RMSD.</p> <p>We find that at least for the first two prediction years the decadal model experiments clearly outperform the classical climate projections, regionally even for the third year. We can thereby demonstrate that the observation type “terrestrial water storage” as available from the GRACE and GRACE-FO missions is suitable as additional data set in the validation and/or calibration of climate model experiments.</p>