Pooling Data Improves Multimodel IDF Estimates over Median-Based IDF Estimates: Analysis over the Susquehanna and Florida

Abstract

Traditional multimodel methods for estimating future changes in precipitation intensity, duration, frequency (IDF) curves rely on mean or median of models’ IDF estimates. Such multimodel estimates are impaired by large estimation uncertainty, shadowing their efficacy in planning efforts. Here, assuming that each climate model is one representation of the underlying data generating process – i.e. the Earth system, we propose a novel extension of current methods through pooling model data that follows: (i) evaluate performance of climate models in simulating the spatial and temporal variability of the observed annual maximum precipitation (AMP), (ii) bias-correct and pool historical and future AMP data of reasonably performing models, and (iii) compute IDF estimates in a non-stationary framework from pooled historical and future model data. Pooling enhances fitting of the extreme value distribution to the data and assumes that data from reasonably performing models represent samples from the “true” underlying data generating distribution. Through Monte Carlo simulations with synthetic data, we show that return periods derived from pooled data have smaller biases and lesser uncertainty than those derived from ensembles of individual model data. We apply this method to NA-CORDEX models to estimate changes in 24-hr precipitation intensity-frequency (PIF) estimates over the Susquehanna watershed and Florida peninsula. Our approach identifies significant future changes at more stations compared to median-based PIF estimates. The analysis suggests that almost all stations over the Susquehanna and at least two-thirds of the stations over the Florida peninsula will observe significant increases in 24-hr precipitation for 2-100 year return periods.

Evaluation of precipitation indices in suites of dynamically and statistically downscaled regional climate models over Florida

“…While these studies highlight important aspects of biases in RCMs, they are of limited utility to stakeholders concerned with smaller subregions (e.g., Kissimmee-Southern Florida water managers). Srivastava et al (2021) show that some of the NA-CORDEX models analyzed in this manuscript do not simulate Rx1day well over Florida peninsula, though they perform well over the Susquehanna watershed in the Northeastern US. Also, biases in GCMs and RCMs lead to biases in hydrologic models (Ashfaq et al 2010;Li et al 2014) and uncertainties in future climate projections (Ashfaq et al 2010).…”

Projected U.S. drought extremes through the twenty-first century with vapor pressure deficit

“…The Central Valley has shown consistent VPD increases over time throughout this study. Overall, ensemble model uncertainty increases with time, and while we have 30 years of simulated data, pooled from three timeframes 54 , larger uncertainty in the 50-year return levels is likely due to the length of the data set (see “ Methods ” section).

Figure 6 Ensemble return periods for the Northwest, Midwest, South and Los Angeles locations.

…”

“…We fit a GEV distribution to a 30-year sample by combining the three different timeframes to characterized extreme VPD and assess future drought intervals 54 . There are two extreme data analysis techniques used in extreme value analysis: block-maxima 65 on which GEV are then fit, and the peak-over threshold 66 for which Generalized Pareto distributions are appropriate.…”

Metrics as tools for bridging climate science and applications

“…Stakeholders showed concern about the large estimation uncertainty in multimodel IDF estimates. Consequently, a novel methodology was developed for reducing uncertainty in the multimodel IDF estimates, which consists of three steps: (i) historical evaluation of climate models, (ii) bias‐correct the reasonably performing models, and (iii) pooling the bias‐corrected model data (A. K. Srivastava et al, 2021). Since climate model performance is highly dependent on region (A. Srivastava et al, 2020; A. K. Srivastava et al, 2021), models should be assessed separately for each region so that underperforming models can be excluded for IDF estimation in that region.…”

“…Consequently, a novel methodology was developed for reducing uncertainty in the multimodel IDF estimates, which consists of three steps: (i) historical evaluation of climate models, (ii) bias‐correct the reasonably performing models, and (iii) pooling the bias‐corrected model data (A. K. Srivastava et al, 2021). Since climate model performance is highly dependent on region (A. Srivastava et al, 2020; A. K. Srivastava et al, 2021), models should be assessed separately for each region so that underperforming models can be excluded for IDF estimation in that region. Using Monte Carlo simulations, it was shown that multimodel IDF estimates based upon pooling of model data have smaller biases (difference between model and observation‐based IDF estimates) and uncertainty (confidence interval) than traditional median‐based multimodel IDF estimates.…”