An official website of the United States government
Abstract Global climate models (GCMs) and Earth system models (ESMs) exhibit biases, with resolutions too coarse to capture local variability for fine-scale, reliable drought and climate impact assessment. However, conventional bias correction approaches may cause implausible climate change signals due to unrealistic representations of spatial and intervariable dependences. While purely data-driven deep learning has achieved significant progress in improving climate and earth system simulations and predictions, they cannot reliably learn the circumstances (e.g., extremes) that are largely unseen in historical climate but likely becoming more frequent in the future climate (i.e., climate non-stationarity). This study shows an integrated trend-preserving deep learning approach that can address the spatial and intervariable dependences and climate non-stationarity issues for downscaling and bias correcting GCMs/ESMs. Here we combine the super-resolution deep residual network (SRDRN) with the trend-preserving quantile delta mapping (QDM) to downscale and bias correct six primary climate variables at once (including daily precipitation, maximum temperature, minimum temperature, relative humidity, solar radiation, and wind speed) from five state-of-the-art GCMs/ESMs in the Coupled Model Intercomparison Project Phase 6 (CMIP6). We found that the SRDRN-QDM approach greatly reduced GCMs/ESMs biases in spatial and intervariable dependences while significantly better-reducing biases in extremes compared to deep learning. The estimated drought based on the six bias-corrected and downscaled variables captured the observed drought intensity and frequency, which outperformed state-of-the-art multivariate bias correction approaches, demonstrating its capability for correcting GCMs/ESMs biases in spatial and multivariable dependences and extremes.
more »
« less
Mesoscale Temporal Wind Variability Biases Global Air–Sea Gas Transfer Velocity of CO2 and Other Slightly Soluble Gases
The significance of the water-side gas transfer velocity for air–sea CO2 gas exchange (k) and its non-linear dependence on wind speed (U) is well accepted. What remains a subject of inquiry are biases associated with the form of the non-linear relation linking k to U (hereafter labeled as f(U), where f(.) stands for an arbitrary function of U), the distributional properties of U (treated as a random variable) along with other external factors influencing k, and the time-averaging period used to determine k from U. To address the latter issue, a Taylor series expansion is applied to separate f(U) into a term derived from time-averaging wind speed (labeled as ⟨U⟩, where ⟨.⟩ indicates averaging over a monthly time scale) as currently employed in climate models and additive bias corrections that vary with the statistics of U. The method was explored for nine widely used f(U) parameterizations based on remotely-sensed 6-hourly global wind products at 10 m above the sea-surface. The bias in k of monthly estimates compared to the reference 6-hourly product was shown to be mainly associated with wind variability captured by the standard deviation σσU around ⟨U⟩ or, more preferably, a dimensionless coefficient of variation Iu= σσU/⟨U⟩. The proposed correction outperforms previous methodologies that adjusted k when using ⟨U⟩ only. An unexpected outcome was that upon setting Iu2 = 0.15 to correct biases when using monthly wind speed averages, the new model produced superior results at the global and regional scale compared to prior correction methodologies. Finally, an equation relating Iu2 to the time-averaging interval (spanning from 6 h to a month) is presented to enable other sub-monthly averaging periods to be used. While the focus here is on CO2, the theoretical tactic employed can be applied to other slightly soluble gases. As monthly and climatological wind data are often used in climate models for gas transfer estimates, the proposed approach provides a robust scheme that can be readily implemented in current climate models.
more »
« less
- PAR ID:
- 10287498
- Date Published:
- Journal Name:
- Remote Sensing
- Volume:
- 13
- Issue:
- 7
- ISSN:
- 2072-4292
- Page Range / eLocation ID:
- 1328
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract. Systematic biases and coarse resolutions are major limitations ofcurrent precipitation datasets. Many deep learning (DL)-based studies havebeen conducted for precipitation bias correction and downscaling. However,it is still challenging for the current approaches to handle complexfeatures of hourly precipitation, resulting in the incapability ofreproducing small-scale features, such as extreme events. This studydeveloped a customized DL model by incorporating customized loss functions,multitask learning and physically relevant covariates to bias correct anddownscale hourly precipitation data. We designed six scenarios tosystematically evaluate the added values of weighted loss functions,multitask learning, and atmospheric covariates compared to the regular DLand statistical approaches. The models were trained and tested using theModern-era Retrospective Analysis for Research and Applications version 2(MERRA2) reanalysis and the Stage IV radar observations over the northerncoastal region of the Gulf of Mexico on an hourly time scale. We found thatall the scenarios with weighted loss functions performed notably better thanthe other scenarios with conventional loss functions and a quantilemapping-based approach at hourly, daily, and monthly time scales as well asextremes. Multitask learning showed improved performance on capturing finefeatures of extreme events and accounting for atmospheric covariates highlyimproved model performance at hourly and aggregated time scales, while theimprovement is not as large as from weighted loss functions. We show thatthe customized DL model can better downscale and bias correct hourlyprecipitation datasets and provide improved precipitation estimates at finespatial and temporal resolutions where regular DL and statistical methodsexperience challenges.more » « less
-
Abstract This study presents an evaluation of the skill of 12 global climate models from phase 6 of the Coupled Model Intercomparison Project (CMIP6) archive in capturing convective storm parameters over the United States. For the historical reference period 1979–2014, we compare the model-simulated 6-hourly convective available potential energy (CAPE), convective inhibition (CIN), 0–1-km wind shear (S01), and 0–6-km wind shear (S06) to those from two independent reanalysis datasets: ERA5 and Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA2). To obtain a comprehensive picture, we analyze the parameter distribution, climatological mean, extreme, and thresholded frequency of convective parameters. The analysis reveals significant bias in capturing both magnitude and spatial patterns, which also vary across the seasons. The spatial distribution of means and extremes of the parameters indicates that most models tend to overestimate CAPE, whereas S01 and S06 are underrepresented to varying extents. Additionally, models tend to underestimate extremes in CIN. Comparing the model profiles with rawinsonde profiles indicates that most of the high CAPE models have a warm and moist bias. We also find that the near-surface wind speed is generally underestimated by the models. The intermodel spread is larger for thermodynamic parameters as compared to kinematic parameters. The models generally have a significant positive bias in CAPE over western and eastern regions of the continental United States. More importantly, the bias in the thresholded frequency of all four variables is considerably larger than the bias in the mean, suggesting a nonuniform bias across the distribution. This likely leads to an underrepresentation of favorable severe thunderstorm environments and has the potential to influence dynamical downscaling simulations via initial and boundary conditions. Significance StatementGlobal climate model projections are often used to explore future changes in severe thunderstorm activity. However, climate model outputs often have significant biases, and they can strongly impact the results. In this study, we thoroughly examined biases in convective parameters in 12 models from phase 6 of the Coupled Model Intercomparison Project with respect to two reanalysis datasets. The analysis is performed for North America, covering the period 1979–2014. The study reveals significant biases in convective parameters that differ between models and are tied to the biases in temperature, humidity, and wind profiles. These results provide valuable insight into selecting the right set of models to analyze future changes in severe thunderstorm activity across the North American continent.more » « less
-
Measurements of turbulence, as rate of dissipation of turbulent kinetic energy (ε), adjacent to the air-water interface are rare but essential for understanding of gas transfer velocities (k) used to compute fluxes of greenhouse gases. Variability in ε is expected over diel cycles of stratification and mixing. Monin-Obukhov similarity theory (MOST) predicts an enhancement in ε during heating (buoyancy flux, β+) relative to that for shear (u*w 3/κz where u*w is water friction velocity, κ is von Karman constant, z is depth). To verify and expand predictions, we quantified ε in the upper 0.25 m and below from profiles of temperature-gradient microstructure in combination with time series meteorology and temperature in a tropical reservoir for winds <4 m s−1. Maximum likelihood estimates of near-surface ε during heating were independent of wind speed and high, ∼5 × 10−6 m2 s−3, up to three orders of magnitude higher than predictions from u*w 3/κz, increased with heating, and were ∼10 times higher than during cooling. k, estimated using near-surface ε, was ∼10 cm hr−1, validated with k obtained from chamber measurements, and 2–5 times higher than computed from wind-based models. The flux Richardson number (Rf) varied from ∼0.4 to ∼0.001 with a median value of 0.04 in the upper 0.25 m, less than the critical value of 0.2. We extend MOST by incorporating the variability in Rf when scaling the influence of β+ relative to u*w 3/κz in estimates of ε, and by extension, k, obtained from time series meteorological and temperature data.more » « less
-
Abstract. Machine learning is quickly becoming a commonly used technique for wind speed and power forecasting. Many machine learning methods utilize exogenous variables as input features, but there remains the question of which atmospheric variables are most beneficial for forecasting, especially in handling non-linearities that lead to forecasting error. This question is addressed via creation of a hybrid model that utilizes an autoregressive integrated moving-average (ARIMA) model to make an initial wind speed forecast followed by a random forest model that attempts to predict the ARIMA forecasting error using knowledge of exogenous atmospheric variables.Variables conveying information about atmospheric stability and turbulence as well as inertial forcing are found to be useful in dealing with non-linear error prediction. Streamwise wind speed, time of day, turbulence intensity, turbulent heat flux, vertical velocity, and wind direction are found to be particularly useful when used in unison for hourly and 3 h timescales. The prediction accuracy of the developed ARIMA–random forest hybrid model is compared to that of the persistence and bias-corrected ARIMA models. The ARIMA–random forest model is shown to improve upon the latter commonly employed modeling methods, reducing hourly forecasting error by up to 5 % below that of the bias-corrected ARIMA model and achieving an R2 value of 0.84 with true wind speed.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.3390/rs13071328
