Reduce Model Bias Research Articles

Nepal Ambient Monitoring and Source Testing Experiment (NAMaSTE): emissions of particulate matter and sulfur dioxide from vehicles and brick kilns and their impacts on air quality in the Kathmandu Valley, Nepal

Abstract. Air pollution is one of the most pressing environmental issues in the Kathmandu Valley, where the capital city of Nepal is located. We estimated emissions from two of the major source types in the valley (vehicles and brick kilns) and analyzed the corresponding impacts on regional air quality. First, we estimated the on-road vehicle emissions in the valley using the International Vehicle Emissions (IVE) model with local emissions factors and the latest available data for vehicle registration. We also identified the locations of the brick kilns in the Kathmandu Valley and developed an emissions inventory for these kilns using emissions factors measured during the Nepal Ambient Monitoring and Source Testing Experiment (NAMaSTE) field campaign in April 2015. Our results indicate that the commonly used global emissions inventory, the Hemispheric Transport of Air Pollution (HTAP_v2.2), underestimates particulate matter emissions from vehicles in the Kathmandu Valley by a factor greater than 100. HTAP_v2.2 does not include the brick sector and we found that our sulfur dioxide (SO2) emissions estimates from brick kilns are comparable to 70 % of the total SO2 emissions considered in HTAP_v2.2. Next, we simulated air quality using the Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) for April 2015 based on three different emissions scenarios: HTAP only, HTAP with updated vehicle emissions, and HTAP with both updated vehicle and brick kilns emissions. Comparisons between simulated results and observations indicate that the model underestimates observed surface elemental carbon (EC) and SO2 concentrations under all emissions scenarios. However, our updated estimates of vehicle emissions significantly reduced model bias for EC, while updated emissions from brick kilns improved model performance in simulating SO2. These results highlight the importance of improving local emissions estimates for air quality modeling. We further find that model overestimation of surface wind leads to underestimated air pollutant concentrations in the Kathmandu Valley. Future work should focus on improving local emissions estimates for other major and underrepresented sources (e.g., crop residue burning and garbage burning) with a high spatial resolution, as well as the model's boundary-layer representation, to capture strong spatial gradients of air pollutant concentrations.

The Impact of Rimed Ice Hydrometeors on Global and Regional Climate.

Rimed hydrometeors (graupel or hail) are added to a stratiform cloud scheme for global models and tested in a variety of configurations. Off‐line tests compare well to other cloud microphysics schemes with rimed ice used in mesoscale models. Tests in single column and climate mode show expected production of small amounts of rimed ice in the middle troposphere and at high latitudes. The overall climate impacts of rimed ice (hail or graupel) at 100‐km horizontal grid spacing are small. There are some changes to partitioning between cloud ice and snow that affect upper troposphere water budgets and clouds. High‐resolution simulations are conducted with a global but regionally refined grid at 14 km over the Contiguous United States. High‐resolution simulations show local production of graupel with realistic size and number concentrations. The maximum graupel frequency at high resolution is over Western U.S. mountain ranges. Differences in total precipitation with the addition of rimed ice in 8‐year simulations are statistically significant only for orographic precipitation over the Cascade and Rocky mountains, reducing model biases when rimed ice is included. Rimed ice slightly improves summer precipitation intensity relative to observations. Thus, while the global climate impact of rimed ice in stratiform clouds may be negligible, there are potentially important and systematic regional effects, particularly for orographic precipitation. Rimed ice in cumulus clouds is not yet treated but is an important next step.

More realistic land‐use and vegetation parameters in a regional climate model reduce model biases over China

AbstractAccurate vegetation cover data are important for realistic simulation of regional climate. The default vegetation parameters from Global Land Cover 2000, currently incorporated into global climate models and used in regional climate model RegCM, are not realistic for China, which may have contributed to serious bias in surface climate simulation. In this study, a new set of vegetation parameters considering the Plant Functional Type (PFT) fractions and the corresponding monthly leaf area index (PFT_LAI), were developed based on the land cover and MODIS LAI data sets. The regional climate model RegCM4.5 coupled with the land surface model CLM4.5 were utilized to test the performance of the new vegetation parameters by comparing simulations with observations using different surface parameters. The surface energy balance was analysed to examine the effects of changed vegetation parameters on regional climate. The results showed that the new parameters were more accurate than the GLC2000 parameters when describing the distribution of crops, grassland, and forests over China. The improved vegetation parameters reduced model biases for winter air temperature and precipitation over southern China by 0.9°C and 8%, respectively, and reduced the winter temperature and summer precipitation biases over northeastern China by approximately 0.7°C and 8%, respectively. More accurate surface albedo are the main reasons for reductions in model bias. However, certain biases, such as the cold and dry bias over the Tibetan Plateau, still remained in the simulation results using our new vegetation data.

Modeling phenological controls on carbon dynamics in dryland sagebrush ecosystems

Dryland ecosystems play an important role in determining how precipitation anomalies affect terrestrial carbon fluxes at regional to global scales. Thus, to understand how climate change may affect the global carbon cycle, we must also be able to understand and model its effects on dryland vegetation. Dynamic Global Vegetation Models (DGVMs) are an important tool for modeling ecosystem dynamics, but they often struggle to reproduce seasonal patterns of plant productivity. Because the phenological niche of many plant species is linked to both total productivity and competitive interactions with other plants, errors in how process-based models represent phenology hinder our ability to predict climate change impacts. This may be particularly problematic in dryland ecosystems where many species have developed a complex phenology in response to seasonal variability in both moisture and temperature. Here, we examine how uncertainty in key parameters as well as the structure of existing phenology routines affect the ability of a DGVM to match seasonal patterns of leaf area index (LAI) and gross primary productivity (GPP) across a temperature and precipitation gradient. First, we optimized model parameters using a combination of site-level eddy covariance data and remotely-sensed LAI data. Second, we modified the model to include a semi-deciduous phenology type and added flexibility to the representation of grass phenology. While optimizing parameters reduced model bias, the largest gains in model performance were associated with the development of our new representation of phenology. This modified model was able to better capture seasonal patterns of both leaf area index (R2 = 0.75) and gross primary productivity (R2 = 0.84), though its ability to estimate total annual GPP depended on using eddy covariance data for optimization. The new model also resulted in a more realistic outcome of modeled competition between grass and shrubs. These findings demonstrate the importance of improving how DGVMs represent phenology in order to accurately forecast climate change impacts in dryland ecosystems.

State-parameter estimation approach for data-driven wildland fire spread modeling: Application to the 2012 RxCADRE S5 field-scale experiment

Data assimilation is an emerging and powerful tool towards real-time flame front monitoring for wildland fire applications. The key idea is to regularly update the state and/or parameters of a fire spread model using observed firelines in order to improve a forecast on future fire locations. The merits of combining state estimation and parameter estimation through a hybrid state-parameter estimation algorithm are demonstrated through the 2012 RxCADRE S5 field-scale controlled burn experiment. For state estimation, we adopt a cost-effective Luenberger observer formulation to reconstruct a complete view of the burning state at a given time. For parameter estimation, we use an ensemble transform Kalman filter to solve the inverse modeling problem consisting of inferring more realistic wind conditions given observations of the actual burning state. The data-driven model relies on a front shape similarity measure derived from image segmentation theory to quantify position errors. We show that the hybrid approach provides an efficient framework to address all sources of model uncertainties and to select burning scenarios that are most likely to occur. Parameter estimation is a key component of the data-driven model by reducing model bias. Using the fire spread model in forecast mode is then an asset to accurately track the flame front dynamics at future lead times.

Real-time simulation of surface water and groundwater with data assimilation

Data assimilation (DA) has proven to be a useful technique in real-time hydrological modeling and forecasting. Jointly assimilating both surface water and groundwater data has promising application value for hydrological simulations in areas where surface water and groundwater are closely linked; however, such studies have not been intensively reported. In addition, the role of the quality of precipitation forecast has not been fully addressed in real-time forecasting using a coupled surface water - groundwater model, where the model evaluation includes both deterministic and probabilistic forecasts. In the present study, we use the MIKE SHE hydrological model code in conjunction with the Ensemble Transform Kalman Filter DA technique. The study area is a small urbanized catchment in Denmark. The model is run in simulated real-time using historical numerical weather prediction forecasts. The results show that DA can significantly reduce model bias and thereby improve model performance for both surface water and groundwater simulations. Comparing the impact of DA and rainfall forecast quality, it is found that, for streamflow forecasts, the most important factor is the quality of the rainfall data; whereas for groundwater head forecasts, the initial state at time of forecast is more important. We also find that inclusion of rainfall forecast uncertainty may be important for simulating a single event, however, it is not vital if long-term average model performance is of interest.

Multi-Model Approaches for Improving Seasonal Ensemble Streamflow Prediction Scheme with Various Statistical Post-Processing Techniques in the Canadian Prairie Region

Hydrologic models are an approximation of reality, and thus, are not able to perfectly simulate observed streamflow because of various sources of uncertainty. On the other hand, skillful operational hydrologic forecasts are vital in water resources engineering and management for preparedness against flooding and extreme events. Multi-model techniques can be used to help represent and quantify various uncertainties in forecasting. In this paper, we assess the performance of a Multi-model Seasonal Ensemble Streamflow Prediction (MSESP) scheme coupled with statistical post-processing techniques to issue operational uncertainty for the Manitoba Hydrologic Forecasting Centre (HFC). The Ensemble Streamflow Predictions (ESPs) from WATFLOOD and SWAT hydrologic models were used along with four statistical post-processing techniques: Linear Regression (LR), Quantile Mapping (QM), Quantile Model Averaging (QMA), and Bayesian Model Averaging (BMA)]. The quality of MSESP was investigated from April to July with a lead time of three months for the Upper Assiniboine River Basin (UARB) at Kamsack, Canada. While multi-model ESPs coupled with post-processing techniques improve predictability (in general), results suggest that additional avenues for improving the skill and value of seasonal streamflow prediction. Next steps towards an operational ESP system include adding more operationally used models, improving models calibration methods to reduce model bias, increasing ESP sample size, and testing ESP schemes at multiple lead times, which, once developed, will not only help HFCs in Canada but would also help Centers South of the Border.

Reducing model bias in a deep learning classifier using domain adversarial neural networks in the MINERvA experiment

We present a simulation-based study using deep convolutional neural networks (DCNNs) to identify neutrino interaction vertices in the MINERvA passive targets region, and illustrate the application of domain adversarial neural networks (DANNs) in this context. DANNs are designed to be trained in one domain (simulated data) but tested in a second domain (physics data) and utilize unlabeled data from the second domain so that during training only features which are unable to discriminate between the domains are promoted. MINERvA is a neutrino-nucleus scattering experiment using the NuMI beamline at Fermilab. A-dependent cross sections are an important part of the physics program, and these measurements require vertex finding in complicated events. To illustrate the impact of the DANN we used a modified set of simulation in place of physics data during the training of the DANN and then used the label of the modified simulation during the evaluation of the DANN. We find that deep learning based methods offer significant advantages over our prior track-based reconstruction for the task of vertex finding, and that DANNs are able to improve the performance of deep networks by leveraging available unlabeled data and by mitigating network performance degradation rooted in biases in the physics models used for training.

Sensitivity of the Bay of Bengal upper ocean to different winds and river input conditions

Abstract The sensitivity of the Bay of Bengal (BoB) upper ocean circulation and thermohaline structure to varying wind strengths and river salinity conditions is investigated using a set of long-term mesoscale simulations. The Regional Ocean Modeling System (ROMS) simulations differ in their forcing fields for winds (strong vs. weak) and in their representations of river input salinity conditions (seasonally varying estuarine salinity vs. zero salinity). The sensitivities are analyzed in terms of the responses of the surface circulation, thermohaline structure, freshwater plume dispersion, and the coastal upwelling along the western boundary. All the simulations reproduce the main broad-scale features of the Bay, while their magnitudes and variabilities depend on the forcing conditions. The impact of stronger wind is felt at greater depths for temperature than for salinity throughout the domain; however, the impact is realized with vertical distributions that are different in the northern than in the southern Bay. As expected, the stronger wind-induced enhanced mixing lowers (enhances) the upper ocean temperature (salinity) by 0.2 °C (0.3 psu), and weakens the near-surface stratification. Moreover, stronger wind enhances eddy activity, strengthens the springtime Western Boundary Current (WBC) and enhances coastal upwelling during spring and summer along the east coast of India. The fresher river input reduces the surface salinity and hence enhances the spreading and intensity of the freshwater plume, stratification, and barrier layer thickness. The lower salinity simulation leads to an eddy-dominant springtime WBC, and enhances the freshness, strength, and southward extent of the autumn East India Coastal Current (EICC). The stronger wind simulations appear to prevent the spreading of the freshwater plume during the summer monsoon due to enhanced mixing. Fresher river input reduces the overall surface salinity by ~0.4 psu; however, it significantly underestimates the salinity near the river mouths, whereas the estuarine salinity river input simulations are closer to reality. These results highlight the importance of river input salinity and realistic strong winds in reducing model biases of high-resolution simulations for the Bay of Bengal.

Moving beyond run‐off calibration—Multivariable optimization of a surface–subsurface–atmosphere model

AbstractSpatially distributed hydrological models are traditionally calibrated and evaluated against few spatially aggregated observations such as river discharge. This model evaluation approach does not enable an assessment of the model predictive capabilities of other hydrological states and fluxes nor does it give any insight into the model ability to mimic the spatial patterns within a catchment. The current study explores a multivariable optimization of a complex coupled surface–subsurface–atmosphere model at the catchment scale in an attempt to move beyond simple run‐off calibration. The model is evaluated against five independent observational data sets of discharge (Q), hydraulic head (h), actual evapotranspiration (ET), soil moisture (SM), and remotely sensed land surface temperature (LST). It is shown that a balanced optimization can be achieved where errors on objective functions for all five observation data sets can be reduced simultaneously. Additionally, the multivariable calibration proved more robust, compared with calibration against Q and h only, during the validation period, even for Q and h. The current parameterization and calibration framework was mainly suitable for reducing model biases and allowed only limited improvements in the spatio‐temporal patterns of the model simulations. This points towards development of better parametrization schemes that will allow simulated spatial patterns to adjust during calibration. Additionally, analysis showed that systematic spatial patterns in the errors of the LST maps could be a very valuable diagnostic tool for assessing deficiencies in the model structure, spatial parameterization, or process description.

Deep-Structured Machine Learning Model for the Recognition of Mixed-Defect Patterns in Semiconductor Fabrication Processes

Semiconductor manufacturers aim to fabricate defect-free wafers in order to improve product quality, increase yields, and reduce costs. Typically, wafer defects form spatial patterns that provide useful information, helping to identify problems and faults during the fabrication process. Machine learning (ML) methods have been used to classify these defects in order to locate the root causes of failure. This paper proposes a novel deep-structured ML approach as an extension of our previous randomized general regression network (RGRN) model, to identify and classify both single-defect and mixed-defect patterns. The principal motivation for this paper is that a shallow-structured RGRN performs well on single-pattern defects, achieving an accuracy of 99.8%, but performs poorly when a wafer has mixed-defect patterns. The proposed approach improves RGRN performance, particularly on mixed-pattern defects, by incorporating a novel information gain (IG)-based splitter as well as deep-structured ML. A spatial filter is applied to remove random noise and reduce model bias during training. During the first detection stage, the splitter generates unique rules that are built using the IG theory and splits the defects data into single-defect and mixed-defect patterns. Single-defect patterns are then classified by RGRN, whereas mixed-defect patterns are fed into the deep-structured ML model for further classification. This combination improves the ability of the proposed approach to classify diverse defect patterns and achieve a better overall performance. Our experimental results demonstrate that the proposed approach achieves an overall detection accuracy of 86.17% on a dataset that contains real data representing both single-defect and mixed-defect patterns, as commonly found in real manufacturing scenarios, outperforming existing ML-based models.

Variance analysis of forecasted streamflow maxima in a wet temperate climate

Coupling global climate models, hydrologic models and extreme value analysis provides a method to forecast streamflow maxima, however the elusive variance structure of the results hinders confidence in application. Directly correcting the bias of forecasts using the relative change between forecast and control simulations has been shown to marginalize hydrologic uncertainty, reduce model bias, and remove systematic variance when predicting mean monthly and mean annual streamflow, prompting our investigation for maxima streamflow. We assess the variance structure of streamflow maxima using realizations of emission scenario, global climate model type and project phase, downscaling methods, bias correction, extreme value methods, and hydrologic model inputs and parameterization. Results show that the relative change of streamflow maxima was not dependent on systematic variance from the annual maxima versus peak over threshold method applied, albeit we stress that researchers strictly adhere to rules from extreme value theory when applying the peak over threshold method. Regardless of which method is applied, extreme value model fitting does add variance to the projection, and the variance is an increasing function of the return period. Unlike the relative change of mean streamflow, results show that the variance of the maxima’s relative change was dependent on all climate model factors tested as well as hydrologic model inputs and calibration. Ensemble projections forecast an increase of streamflow maxima for 2050 with pronounced forecast standard error, including an increase of +30(±21), +38(±34) and +51(±85)% for 2, 20 and 100 year streamflow events for the wet temperate region studied. The variance of maxima projections was dominated by climate model factors and extreme value analyses.

Sensitivity of Ensemble Forecast Verification to Model Bias

This study demonstrates how model bias can adversely affect the quality assessment of an ensemble prediction system (EPS) by verification metrics. A regional EPS [Global and Regional Assimilation and Prediction Enhanced System-Regional Ensemble Prediction System (GRAPES-REPS)] was verified over a period of one month over China. Three variables (500-hPa and 2-m temperatures, and 250-hPa wind) are selected to represent “strong” and “weak” bias situations. Ensemble spread and probabilistic forecasts are compared before and after a bias correction. The results show that the conclusions drawn from ensemble verification about the EPS are dramatically different with or without model bias. This is true for both ensemble spread and probabilistic forecasts. The GRAPES-REPS is severely underdispersive before the bias correction but becomes calibrated afterward, although the improvement in the spread’s spatial structure is much less; the spread–skill relation is also improved. The probabilities become much sharper and almost perfectly reliable after the bias is removed. Therefore, it is necessary to remove forecast biases before an EPS can be accurately evaluated since an EPS deals only with random error but not systematic error. Only when an EPS has no or little forecast bias, can ensemble verification metrics reliably reveal the true quality of an EPS without removing forecast bias first. An implication is that EPS developers should not be expected to introduce methods to dramatically increase ensemble spread (either by perturbation method or statistical calibration) to achieve reliability. Instead, the preferred solution is to reduce model bias through prediction system developments and to focus on the quality of spread (not the quantity of spread). Forecast products should also be produced from the debiased but not the raw ensemble.

Niño4 as a Key Region for the Interannual Variability of the Western Pacific Warm Pool

AbstractThe Western Pacific Warm Pool (WPWP) plays an important role in the global climate through modulating deep convections, ENSO, monsoon onsets, etc. Due to the vast spatial range and huge heat storage of the WPWP, near‐real‐time monitoring of its three‐dimensional variations remains challenging. Based on Argo observations and three reanalysis data sets, we find that the Niño4 sea surface temperature (SST) index captures the interannual variability of the WPWP well. The Niño4 SST can explain approximately half of the variance of the WPWP heat content and almost all the variance of the east‐west migration of the WPWP. An assessment of 31 CMIP5 models also reveals that models with larger interannual spectral powers and amplitudes of the Niño4 SST tend to simulate larger variations in the heat content and east‐west migration of the WPWP. A surface heat budget analysis further shows that the Niño4 SST and WPWP are physically connected through basin‐scale horizontal advections of mean temperatures by anomalous horizontal currents, which dominate the interannual variations of both the Niño4 SST and WPWP. Our results indicate that the Niño4 SST can efficiently estimate the interannual WPWP changes and a reliable predictor of the onset time of the South China Sea summer monsoon and Bay of Bengal summer monsoon, without the need to calculate the eastern boundary location and heat content of the WPWP. Moreover, a better simulation of the SST and horizontal currents in the Niño4 region can help to reduce model bias when reproducing the WPWP's interannual variabilities.

Nonhydrostatic Simulations of Tide‐Induced Mixing in the Halmahera Sea: A Possible Role in the Transformation of the Indonesian Throughflow Waters

AbstractThe Indonesian Throughflow (ITF) waters are significantly transformed within the Indonesian Archipelago and consequently influence the large‐scale ocean circulation such as Agulhas and Leeuwin Currents. Existing ocean general circulation models (OGCMs) are, however, incapable of reproducing the transformation of the ITF waters, since tidal forcing is neglected in such models. In the present study, we first conduct high‐resolution nonhydrostatic three‐dimensional numerical experiments focusing on the transformation of the ITF waters in the Halmahera Sea which is thought to be the most important bottleneck in simulating the ITF water mass properties. It is shown that intensive vertical mixing induced by breaking of internal tides in the shallow regions in the Halmahera Sea dilutes the ITF waters, significantly reducing model biases found in the existing OGCMs. We next evaluate quantitatively the effect of tide‐induced vertical mixing on the transformation of the ITF waters. It is shown that tide‐induced vertical mixing dominates the transformation of the ITF waters, although some supplementary processes such as horizontal mixing associated with the submesoscale eddies resulting from tidal interaction with land configurations cannot be ignored.

Evaluation of a multi-scale WRF-CAM5 simulation during the 2010 East Asian Summer Monsoon

The Weather Research and Forecasting model with Chemistry (WRF-Chem) with the physics package of the Community Atmosphere Model Version 5 (CAM5) has been applied at multiple scales over Eastern China (EC) and the Yangtze River Delta (YRD) to evaluate how increased horizontal resolution with physics designed for a coarser resolution climate model impacts aerosols and clouds, and the resulting precipitation characteristics and performance during the 2010 East Asian Summer Monsoon (EASM). Despite large underpredictions in surface aerosol concentrations and aerosol optical depth, there is good spatial agreement with surface observations of chemical predictions, and increasing spatial resolution tends to improve performance. Model bias and normalized root mean square values for precipitation predictions are relatively small, but there are significant differences when comparing modeled and observed probability density functions for precipitation in EC and YRD. Increasing model horizontal resolution tends to reduce model bias and error for precipitation predictions. The surface and column aerosol loading is maximized between about 32°N and 42°N in early to mid-May during the 2010 EASM, and then shifts north while decreasing in magnitude during July and August. Changing model resolution moderately changes the spatiotemporal relationships between aerosols, cloud properties, and precipitation during the EASM, thus demonstrating the importance of model grid resolution in simulating EASM circulation and rainfall patterns over EC and the YRD. Results from this work demonstrate the capability and limitations in the aerosol, cloud, and precipitation representation of WRF-CAM5 for regional-scale applications down to relatively fine horizontal resolutions. Further WRF-CAM5 model development and application in this area is needed.

Incorporating remote sensing-based ET estimates into the Community Land Model version 4.5

Abstract. Land surface models bear substantial biases in simulating surface water and energy budgets despite the continuous development and improvement of model parameterizations. To reduce model biases, Parr et al. (2015) proposed a method incorporating satellite-based evapotranspiration (ET) products into land surface models. Here we apply this bias correction method to the Community Land Model version 4.5 (CLM4.5) and test its performance over the conterminous US (CONUS). We first calibrate a relationship between the observational ET from the Global Land Evaporation Amsterdam Model (GLEAM) product and the model ET from CLM4.5, and assume that this relationship holds beyond the calibration period. During the validation or application period, a simulation using the default CLM4.5 (CLM) is conducted first, and its output is combined with the calibrated observational-vs.-model ET relationship to derive a corrected ET; an experiment (CLMET) is then conducted in which the model-generated ET is overwritten with the corrected ET. Using the observations of ET, runoff, and soil moisture content as benchmarks, we demonstrate that CLMET greatly improves the hydrological simulations over most of the CONUS, and the improvement is stronger in the eastern CONUS than the western CONUS and is strongest over the Southeast CONUS. For any specific region, the degree of the improvement depends on whether the relationship between observational and model ET remains time-invariant (a fundamental hypothesis of the Parr et al. (2015) method) and whether water is the limiting factor in places where ET is underestimated. While the bias correction method improves hydrological estimates without improving the physical parameterization of land surface models, results from this study do provide guidance for physically based model development effort.

Refining multi-model projections of temperature extremes by evaluation against land–atmosphere coupling diagnostics

Abstract. The Earth's land surface and the atmosphere are strongly interlinked through the exchange of energy and matter. This coupled behaviour causes various land–atmosphere feedbacks, and an insufficient understanding of these feedbacks contributes to uncertain global climate model projections. For example, a crucial role of the land surface in exacerbating summer heat waves in midlatitude regions has been identified empirically for high-impact heat waves, but individual climate models differ widely in their respective representation of land–atmosphere coupling. Here, we compile an ensemble of 54 combinations of observations-based temperature (T) and evapotranspiration (ET) benchmarking datasets and investigate coincidences of T anomalies with ET anomalies as a proxy for land–atmosphere interactions during periods of anomalously warm temperatures. First, we demonstrate that a large fraction of state-of-the-art climate models from the Coupled Model Intercomparison Project (CMIP5) archive produces systematically too frequent coincidences of high T anomalies with negative ET anomalies in midlatitude regions during the warm season and in several tropical regions year-round. These coincidences (high T, low ET) are closely related to the representation of temperature variability and extremes across the multi-model ensemble. Second, we derive a land-coupling constraint based on the spread of the T–ET datasets and consequently retain only a subset of CMIP5 models that produce a land-coupling behaviour that is compatible with these benchmark estimates. The constrained multi-model simulations exhibit more realistic temperature extremes of reduced magnitude in present climate in regions where models show substantial spread in T–ET coupling, i.e. biases in the model ensemble are consistently reduced. Also the multi-model simulations for the coming decades display decreased absolute temperature extremes in the constrained ensemble. On the other hand, the differences between projected and present-day climate extremes are affected to a lesser extent by the applied constraint, i.e. projected changes are reduced locally by around 0.5 to 1 °C – but this remains a local effect in regions that are highly sensitive to land–atmosphere coupling. In summary, our approach offers a physically consistent, diagnostic-based avenue to evaluate multi-model ensembles and subsequently reduce model biases in simulated and projected extreme temperatures.

Modeling the diurnal variability of agricultural ammonia in Bakersfield, California, during the CalNex campaign

Abstract. NH3 retrievals from the NASA Tropospheric Emission Spectrometer (TES), as well as surface and aircraft observations of NH3(g) and submicron NH4(p), are used to evaluate modeled concentrations of NH3(g) and NH4(p) from the Community Multiscale Air Quality (CMAQ) model in the San Joaquin Valley (SJV) during the California Research at the Nexus of Air Quality and Climate Change (CalNex) campaign. We find that simulations of NH3 driven with the California Air Resources Board (CARB) emission inventory are qualitatively and spatially consistent with TES satellite observations, with a correlation coefficient (r2) of 0.64. However, the surface observations at Bakersfield indicate a diurnal cycle in the model bias, with CMAQ overestimating surface NH3 at night and underestimating it during the day. The surface, satellite, and aircraft observations all suggest that daytime NH3 emissions in the CARB inventory are underestimated by at least a factor of 2, while the nighttime overestimate of NH3(g) is likely due to a combination of overestimated NH3 emissions and underestimated deposition.Running CMAQ v5.0.2 with the bi-directional NH3 scheme reduces NH3 concentrations at night and increases them during the day. This reduces the model bias when compared to the surface and satellite observations, but the increased concentrations aloft significantly increase the bias relative to the aircraft observations. We attempt to further reduce model bias by using the surface observations at Bakersfield to derive an empirical diurnal cycle of NH3 emissions in the SJV, in which nighttime and midday emissions differ by about a factor of 4.5. Running CMAQv5.0.2 with a bi-directional NH3 scheme together with this emissions diurnal profile further reduces model bias relative to the surface observations. Comparison of these simulations with the vertical profile retrieved by TES shows little bias except for the lowest retrieved level, but the model bias relative to flight data aloft increases slightly. Our results indicate that both diurnally varying emissions and a bi-directional NH3 scheme should be applied when modeling NH3(g) and NH4(p) in this region. The remaining model errors suggest that the bi-directional NH3 scheme in CMAQ v5.0.2 needs further improvements to shift the peak NH3 land–atmosphere flux to earlier in the day. We recommend that future work include updates to the current CARB NH3 inventory to account for NH3 from fertilizer application, livestock, and other farming practices separately; adding revised information on crop management practices specific to the SJV region to the bi-directional NH3 scheme; and top-down studies focused on determining the diurnally varying biases in the canopy compensation point that determines the net land–atmosphere NH3 fluxes.

Improving the twilight model for polar cap absorption nowcasts

During Solar Proton Events (SPE), energetic protons ionize the polar mesosphere causing HF radiowave attenuation, more strongly on the dayside where the effective recombination coefficient, αeff, is low. Polar cap absorption (PCA) models predict the 30 MHz cosmic noise absorption, A, measured by riometers, based on real-time measurements of the integrated proton flux-energy spectrum, J. However, empirical models in common use cannot account for regional and day-to-day variations in the day- and nighttime profiles of αeff(z) or the related sensitivity parameter, m=A/√J. Large prediction errors occur during twilight when m changes rapidly, and due to errors locating the rigidity cutoff latitude. Modeling the twilight change in m as a linear or Gauss error-function transition over a range of solar-zenith angles (χl < χ < χu) provides a better fit to measurements than selecting day or night αeff profiles based on the Earth-shadow height. Optimal model parameters were determined for several polar cap riometers for large SPEs in 1998-2005. The optimal χl parameter was found to be most variable, with smaller values (as low as 60°) post-sunrise compared with pre-sunset, and with positive correlation between riometers over a wide area. Day and night values of m exhibited higher correlation for closely spaced riometers. A nowcast simulation is presented in which rigidity boundary latitude and twilight model parameters are optimized by assimilating age-weighted measurements from 25 riometers. The technique reduces model bias, and root-mean-squared errors are reduced by up to 30% compared with a model employing no riometer data assimilation.