A comparison of two numerical weather prediction methods for diagnosing fast‐physics errors in climate models

Artificial intelligence and numerical weather prediction models: A technical survey

Forecasting the weather is an increasingly data-intensive exercise. Numerical weather prediction (NWP) models are becoming more complex, with higher resolutions, and there are increasing numbers of different models in operation. While the forecasting skill of NWP models continues to improve, the number and complexity of these models poses a new challenge for the operational meteorologist: how should the information from all available models, each with their own unique biases and limitations, be combined in order to provide stakeholders with well-calibrated probabilistic forecasts to use in decision making? In this paper, we use a road surface temperature example to demonstrate a three-stage framework that uses machine learning to bridge the gap between sets of separate forecasts from NWP models and the ‘ideal’ forecast for decision support: probabilities of future weather outcomes. First, we use quantile regression forests to learn the error profile of each numerical model, and use these to apply empirically derived probability distributions to forecasts. Second, we combine these probabilistic forecasts using quantile averaging. Third, we interpolate between the aggregate quantiles in order to generate a full predictive distribution, which we demonstrate has properties suitable for decision support. Our results suggest that this approach provides an effective and operationally viable framework for the cohesive post-processing of weather forecasts across multiple models and lead times to produce a well-calibrated probabilistic output.This article is part of the theme issue ‘Machine learning for weather and climate modelling’.

The presence of cities affects local ecosystems by changing the surface energy balance and introducing anthropogenic heat, greenhouse gases, and air pollutants into the atmosphere. Currently, over 50% of the global population resides in urban areas, which increases their exposure to risks related to extreme weather and climate change. To better forecast weather and address climate adaptation and mitigation challenges, numerical weather prediction (NWP) models are beginning to integrate urban physics. However, the absence of high-quality and comprehensive urban morphology data such as building heights and footprints is still a challenge. This study introduces a high-resolution (500 m) global database developed using open data sources to support regional and global weather and climate modeling and enhance the representation of urbanization in NWP models. WUMPOD contains following urban morphological parameters; MH (mean building height), STDH (standard deviation of building height), HGT (area weighted mean building height), LP (building plan area fraction), LB (building surface to plan area fraction), LF (frontal aspect ratio for cardinal directions of 0, 45, 90, and 135 degrees), LC (complete aspect ratio), H2W (height to width ratio), HI [0-75 m] (percentage distribution of building heights at every 5 m up to 75 m), Z0M, ZDM (Roughness length and displacement height according to Macdonald et al. (1998) for 0, 45, 90, and 135 directions), Z0S, ZDS (Roughness length and displacement height according to Grimmond and Oke (1999)), Z0R, and ZDR (Roughness length and displacement height according to Raupach (1994) for 0, 45, 90, and 135 directions) and urban fraction (from WorldCover 2021). The World Urban Morphological Parameter Open Dataset (WUMPOD) is available at https://doi.org/10.5281/zenodo.10039127. The dataset will be incorporated in the development of a global ancillary and a contribution to the UK Met Office Momentum partnership.

<div><span>Snow as a major part of the cryosphere is an important component of Earth’s hydrological cycle and energy balance. Understanding the microstructural, macrophysical, thermal and optical properties of the snowpack is essential for integration into numerical models and there is a great need for accurate snow data at different spatial and temporal resolutions to address the challenges of changing snow conditions.</span></div><div><span>Physical snow properties are currently determined by traditional ground-based measurements as well as remote sensing, over a range of temporal and spatial scales, following considerable developments in instrument technology over recent years. </span></div><div><span>Data assimilation (DA)</span><span> methods are widely used</span> <span>to combine data from different observations</span><span> with numerical model using uncertainties </span><span>of observed and modeled variables  to produce an optimal estimate. DA provides a reliable improvement of the initial states of the numerical model and a benefit for hydrological and snow model forecasts. </span></div><div> </div><div><span>European efforts to harmonize approaches for validation, and methodologies of snow measurement practices, instrumentation, algorithms and data assimilation techniques were coordinated by the European Cooperation in Science and Technology (COST) Action ES1404 “HarmoSnow”, entitled, “A European network for a harmonized monitoring of snow for the benefit of climate change scenarios, hydrology and numerical weather prediction” (2014-2018) .</span></div><div><span>One of the key objectives of the action was “Advance the application of snow DA in numerical weather prediction (NWP) and hydrological models, and show its benefit for weather and hydrological forecasting as well as other applications.” </span></div><div><span>One key result from COST HarmoSnow is a better knowledge about the diversity of usage of snow observations in DA, forcing, monitoring, validation, or verification within NWP, hydrology, snow and climate models. The main parts of this knowledge are retrieved from a COST HarmoSnow survey exploring the common practices on the use of snow observations in different modeling environments. We will show results from the survey and their implications towards standardized and improved usage of snow observations in various data assimilation applications.</span></div>

Warm conveyor belts (WCBs) associated with extratropical cyclones transport air from the lower troposphere into the tropopause region and contribute to upper-level ridge building and the formation of blocking anticyclones. Recent studies indicate that this constitutes an important source and magnifier of forecast uncertainty and errors in numerical weather prediction (NWP) models. However, a systematic evaluation of the representation of WCBs in NWP models has yet to be determined. Here, we employ the logistic regression models developed in Part I to identify the inflow, ascent, and outflow stages of WCBs in the European Centre for Medium-Range Weather Forecasts (ECMWF) subseasonal reforecasts for Northern Hemisphere winter in the period January 1997 to December 2017. We verify the representation of these WCB stages in terms of systematic occurrence frequency biases, forecast reliability, and forecast skill. Systematic WCB frequency biases emerge already at early lead times of around 3 days with an underestimation for the WCB outflow over the North Atlantic and eastern North Pacific of around 40% relative to climatology. Biases in the predictor variables of the logistic regression models can partially explain these biases in WCB inflow, ascent, or outflow. Despite an overconfidence in predicting high WCB probabilities, skillful WCB forecasts are on average possible up to a lead time of 8–10 days with more skill over the North Pacific compared to the North Atlantic region. Our results corroborate that the current limited forecast skill for the large-scale extratropical circulation on subseasonal time scales beyond 10 days might be tied to the representation of WCBs and associated upscale error growth.

In this study we examine the determination accuracy of both the mean radiant temperature (Tmrt) and the Universal Thermal Climate Index (UTCI) within the scope of numerical weather prediction (NWP), and global (GCM) and regional (RCM) climate model simulations. First, Tmrt is determined and the so-called UTCI-Fiala model is then used for the calculation of UTCI. Taking into account the uncertainties of NWP model (among others the HIgh Resolution Limited Area Model HIRLAM) output (temperature, downwelling short-wave and long-wave radiation) stated in the literature, we simulate and discuss the uncertainties of Tmrt and UTCI at three stations in different climatic regions of Europe. The results show that highest negative (positive) differences to reference cases (under assumed clear-sky conditions) of up to -21°C (9°C) for Tmrt and up to -6°C (3.5°C) for UTCI occur in summer (winter) due to cloudiness. In a second step, the uncertainties of RCM simulations are analyzed: three RCMs, namely ALADIN (Aire Limitée Adaptation dynamique Développement InterNational), RegCM (REGional Climate Model) and REMO (REgional MOdel) are nested into GCMs and used for the prediction of temperature and radiation fluxes in order to estimate Tmrt and UTCI. The inter-comparison of RCM output for the three selected locations shows that biases between 0.0 and ±17.7°C (between 0.0 and ±13.3°C) for Tmrt (UTCI), and RMSE between ±0.5 and ±17.8°C (between ±0.8 and ±13.4°C) for Tmrt (UTCI) may be expected. In general the study shows that uncertainties of UTCI, due to uncertainties arising from calculations of radiation fluxes (based on NWP models) required for the prediction of Tmrt, are well below ±2°C for clear-sky cases. However, significant higher uncertainties in UTCI of up to ±6°C are found, especially when prediction of cloudiness is wrong.

<p>Weather forecasting in the Arctic and Antarctic is a challenge above all due to rarity of observations to be assimilated in numerical weather prediction (NWP) models. As observations are expensive and logistically challenging, it is important to evaluate the benefit that additional observations could bring to NWP.</p><p>Considering the Arctic, in this study the effects of the spatial coverage of the network on numerical weather prediction were evaluated by comparing radiosonde observations from land station taken from Integrated Global Radiosonde Archive (IGRA) and radiosonde observations from expeditions in the Arctic Ocean with operational analyses and background fields (12‐hr forecasts) of the European Centre for Medium Range Weather Forecasts (ECMWF). The focus was on 850 hPa level temperature for the period January 2016 – September 2018. Comparison of the analyses and background fields showed that radiosoundings had a remarkable impact on improving operational analyses but the impact had a large geographical variation. In particular, radiosonde observations from islands (Jan Mayen and Bear Island) in the northern North Atlantic and from Arctic expeditions substantially improved analyses suggesting that those observations were critical for the quality of analyses and forecasts. Comparison of two cases with and without assimilation of radiosonde sounding data from expeditions of Icebreaker Oden in 2016 and 2018 in the central Artic Ocean showed that satellite observations were not able to compensate for the large spatial gap in the radiosounding network. In the areas where the network is reasonably dense, the density of the sounding network was not the most critical factor for the quality of background fields. Instead, the quality of background field was more related to how radiosonde observations were utilized in the assimilation and to the quality of those observations.</p><p>Considering the Antarctic, we applied radiosonde sounding and Unmanned Aerial Vehicles (UAV) observations from an RV Polarstern cruise in the ice-covered Weddell Sea in austral winter 2013 to evaluate the impact of their assimilation in the Polar version of the Weather Research and Forecasting (Polar WRF) model. Our experiments revealed small or moderate impacts of radiosonde and UAV data assimilation. In any case, the assimilation of sounding data from both radiosondes and UAVs improved the analyses of air temperature, wind speed, and humidity at the observation site for most of the time. Further, the impact on the results of 5-day long Polar WRF experiments was often felt over distances of at least 300 km from the observation site. All experiments succeeded in capturing the main features of the evolution of near-surface variables, but the effects of data assimilation varied between different cases. Due to the limited vertical extent of the UAV observations, the impact of their assimilation was limited to the lowermost 1-2 km layer, and assimilation of radiosonde data was more beneficial for modelled sea level pressure and near-surface wind speed. Considering the perspectives for technological advance, atmospheric soundings applying UAV have a large potential to supplement conventional radiosonde sounding observations.</p><p>The differences in the results obtained for the Arctic and Antarctic are discussed.</p>

The Arctic is the most rapidly warming region on Earth, and it is facing considerable environmental change. These changes are beginning to alter the way humanity uses and exploits the Arctic region, with commercial activities such as tourism, fishing, mineral and oil extraction, and shipping on the increase. Increased human activity in the Arctic has implications for safety and environmental conservation, and can present difficulties for local communities who rely on subsistence hunting and fishing, and community resupply. Environmental prediction for the Arctic is therefore becoming increasingly important, but our ability to accurately predict the Arctic atmosphere–sea-ice–ocean system is relatively immature, with short-term forecasts significantly less accurate than those produced for the mid-latitudes. There are numerous challenges with respect to forecasting systems in the Arctic: there is a relatively sparse in-situ observing network; some satellite-based observations are hard to use because of the snow and ice-covered surfaces; it is difficult and expensive to make detailed research-quality observations; and the dominant physical processes are different to those at mid-latitudes and over the tropics. To address these issues, the Year of Polar Prediction (YOPP1) was developed as one of the key elements of the WMO's Polar Prediction Project2. On Wednesday 21 November 2018, the Royal Meteorological Society hosted a meeting at the Society of Chemical Industry (SCI) in London to discuss recent advances and challenges associated with the understanding and prediction of key Arctic processes using numerical models and observations – including new insights provided by several field campaigns and modelling experiments carried out under the YOPP umbrella. The first talk was entitled: What are the challenges and priorities for improved prediction and climate monitoring of the Arctic? (Irina Sandu, ECMWF). Irina explained that in order to improve Arctic predictions and reanalyses, which constitute a great tool for Arctic climate monitoring, work is needed in three areas: (i) enhanced coupled modelling, (ii) data assimilation methods and (iii) the effective use of observations in the numerical weather prediction systems. Arctic regions pose specific challenges for each of these three areas because model errors are large, model uncertainty is not necessarily well represented, and in-situ observations are sparse. Moreover, while the Arctic is very rich in terms of remote sensing observations from polar orbiting satellites, some of these observations are difficult to use in data assimilation because of ambiguous signal properties, particularly over snow and sea-ice and in cloudy situations. Irina introduced the challenges and priorities in each of these three key areas and described recent and ongoing efforts to evaluate and improve predictions in the Arctic, and beyond, made in the framework of the YOPP modelling activities and the EU-APPLICATE3 project. She highlighted the fact that, compared to previous similar initiatives, YOPP puts an additional emphasis on numerical experimentation, in a concerted effort to exploit observations for model improvement and drive developments in data assimilation and the design of observing systems. The next talk presented was: What are the limitations of Arctic sea ice remote sensing products, and what opportunities can they provide for improving predictive skill of Arctic forecasts? (Ed Blockley, Met Office). Ed explained that, in response to declining sea-ice cover, human activity in the Arctic is increasing, and that accurate forecasts of Arctic sea ice – on a variety of different timescales – are therefore becoming increasingly important for the safety of human activities in the region. He then provided a brief overview of how sea ice has been traditionally initialised in short-range forecasting and seasonal prediction systems, and explained that most operational systems initialise sea ice purely using passive microwave sea ice concentration products. He gave an overview of the passive microwave remote sensing products typically used for initialisation of sea ice concentration, as well as the (relatively new) satellite-derived sea ice thickness products currently available. Information was provided about how these observations are derived, what the uncertainties and limitations are, and what new avenues of research are currently underway to improve measurements of Arctic sea ice from space. Ed also illustrated the potential impact of assimilating satellite sea ice thickness products into our operational prediction systems. Using the Met Office GloSea seasonal prediction system, he showed how sea ice thickness values inferred from the CryoSat-2 satellite altimetry could be used within the initialisation process to considerably improve forecasts of Arctic summer sea ice on seasonal timescales. Next up was: Chasing the source of the AMOC (Atlantic Meridional Overturning Circulation) atmosphere–ocean coupling in the Iceland and Greenland Seas (Ian Renfrew, University of East Anglia) Ian described a coordinated meteorological and oceanographic field campaign over the Iceland and southern Greenland Seas that took place in February and March 2018. The aim of the campaign was to characterise the atmospheric forcing and the ocean response of coupled atmosphere–ocean processes – in particular, cold-air outbreaks in the vicinity of the marginal-ice-zone, and their triggering of oceanic heat loss and the generation of dense waterbody body masses. He described how the team observed the spatial structure and variability of surface flux fields in the region, and the weather systems that dictate these fluxes, through the first meteorological field campaign in the Iceland Sea. This was done as part of a coupled atmosphere–ocean field campaign in winter 2018 involving a rare wintertime research cruise, airborne observations and a host of ocean and atmosphere observing systems. In-situ observations of air–sea interaction processes were made from several platforms. Ian presented some highlights from the field campaign and introduced some early findings from the post-campaign analysis – most notably a comprehensive set of observations of a marine cold air outbreak which were obtained via radiosonde releases and the research aircraft. Mixing it up: What connects Arctic clouds and sea ice? was a talk presented by Ian Brooks (University of Leeds). Ian explained that the single largest source of uncertainty in climate models arises from the representation of clouds; this is particularly true in the Arctic where, for much of the year, low-level boundary-layer clouds dominate the surface energy budget. The time-evolution and properties of these clouds are intimately linked their interactions with the surface and boundary layer structure. The properties of both clouds and boundary layer (and indeed the surface) are essentially sub-grid-scale in climate and weather forecast models, and thus many important features are poorly represented by current parameterisations. Ian emphasised that observations show that the summertime boundary-layer over sea ice is often decoupled below cloud base, inhibiting turbulent mixing – and hence the transport of aerosols and water vapour – between cloud and the surface. This decoupling is not reproduced by models, and the controls on this turbulent structure are poorly understood. He provided a review of the state of knowledge, alongside some of the recent research on Arctic boundary layer processes and their effects on clouds and the surface energy budget. A unique feature of polar boundary layers is the occurrence of a humidity inversion, where absolute humidity increases across the top of the boundary layer. This allows boundary-layer clouds to extend into the temperature inversion and results in a weakening of turbulence generated by cloud-top radiative cooling. Finally, he showed how recent field measurements have improved the parameterisation of surface turbulent fluxes over marginal sea ice, which will ultimately affect the model representation of the evolution of sea ice. The view from above Arctic snow at 89-325GHz: What can surface emissivity on these channels tell us about snowpack stratigraphy? was the next talk by Chawn Harlow ( Met Office). Chawn introduced Measurements of Arctic Cloud, Snow and Sea Ice in the Marginal Ice ZonE (MACSSIMIZE) – an airborne campaign carried out by the Met Office in March 2018 that focussed on evaluation of snow microwave emission models in the 89–325GHz frequency range. He described how the campaign included a ground-based component to collect snow pit measurements co-located with the airborne radiometric measurements, and that these were now being used to improve snow physical models for use in future Met Office numerical weather prediction systems. An explanation of the main motivation for MACSSIMIZE followed, both from the point of view of snow remote sensing and for improving atmospheric data assimilation, along with a description of the emissivity retrievals obtained during the campaign at 89, 118, 157, 183.3, 243 and 325GHz. Chawn then described how the MACSSIMIZE observations will be used to evaluate the snow thermodynamic and emissivity models used at the Met Office in the short term, and how they will eventually be used to improve assimilation of microwave atmospheric sounding data over snow-covered surfaces in the polar regions. Finally, Richard Essery (University of Edinburgh) presented: Why tundra snow is upside down in models, and why it matters?. Richard described how Earth System and Numerical Weather Prediction models are beginning to use more sophisticated representations of terrestrial snow, drawing on snow physics models that were first developed for forecasting avalanche risk. He explained that, having been developed for deep mid-latitude mountain snow, these models neglect important physical processes occurring in shallow tundra snow subjected to high winds and low temperatures. Understanding these processes is important for representing the thermal properties of Arctic snow and exploiting information from microwave remote sensing over snow-covered surfaces. Richard provided an overview of the ground-based component of the YOPP-endorsed MACSSIMIZE campaign, carried out at Trail Valley Creek in northwestern Canada. He presented results obtained from the snow pit measurements within MACSSIMIZE and drew attention to the existence of a granular depth hoar layer located below the wind-slab, which is a common feature of Arctic tundra snow. The resulting vertical density profile – with low-density depth hoar below a high-density wind-slab layer – is not reproduced by current snow models. Field instruments that can replace visual and subjective measurements of snow structure are now becoming available and can provide data for improving models. With speakers focussing on operational forecasting, Earth observation, and on observation and modelling of key Arctic processes such as clouds, sea ice and snow, this meeting illustrated the broad range of the UK's expertise and interest in Arctic prediction, and highlighted the significant contribution that scientists in the UK are making to YOPP. This RMetS meeting was convened by Ed Blockley and Chawn Harlow (Met Office) and was chaired by Ed Blockley (Met Office), with Jonathan Day (ECMWF) as rapporteur. The event was sponsored by the Centre for Polar Observation and Modelling (CPOM).

Lakes influence the structure of the atmospheric boundary layer and, consequently, the local weather and local climate. Their influence should be taken into account in the numerical weather prediction (NWP) and climate models through parameterisation. For parameterisation, data on lake characteristics external to the model are also needed. The most important parameter is the lake depth. Global database of lake depth GLDB (Global Lake Database) is developed to parameterise lakes in NWP and climate modelling. The main purpose of the study is to upgrade GLDB by use of indirect estimates of the mean depth for lakes in boreal zone, depending on their geological origin. For this, Tectonic Plates Map, geological, geomorphologic maps and the map of Quaternary deposits were used. Data from maps were processed by an innovative algorithm, resulting in 141 geological regions where lakes were considered to be of kindred origin. To obtain a typical mean lake depth for each of the selected regions, statistics from GLDB were gained and analysed. The main result of the study is a new version of GLDB with estimations of the typical mean lake depth included. Potential users of the product are NWP and climate models.

Abstract. Broadband short-wave (SW) surface direct and diffuse irradiances are not typically within the set of output variables produced by numerical weather prediction (NWP) models. However, they are frequently requested for solar energy applications. In order to compute them, a detailed representation of the aerosol optical properties is important. Nonetheless, NWP models typically oversimplify aerosol representation or even neglect their effect. In this work, a flexible method to account for the SW aerosol optical properties in the computation of broadband SW surface direct and diffuse irradiances is presented. It only requires aerosol optical depth at 0.55 μm and knowledge of the type of predominant aerosol. Other parameters needed to consider spectral aerosol extinction, namely, Angström exponent, aerosol single-scattering albedo and aerosol asymmetry factor, are parameterized. The parameterization has been tested using the Rapid Radiative Transfer Model for climate and weather models (RRTMG) SW scheme of the Weather Research and Forecasting (WRF) NWP model for data over the continental US. In principle, it can be adapted to any other SW radiative transfer band model. It has been verified against a control experiment and using data from five radiometric stations in the contiguous US. The control experiment consisted of a clear-sky evaluation of the RRTMG solar radiation estimates obtained in WRF when RRTMG is driven with ground-observed aerosol optical properties. Overall, the verification has shown satisfactory results for both broadband SW surface direct and diffuse irradiances. The parameterization has proven effective in significantly reducing the prediction error and constraining the seasonal bias in clear-sky conditions to within the typical observational error expected in well maintained radiometers.

Improving the Turbulence Coupling between High Resolution Numerical Weather Prediction Models and Lagrangian Particle Dispersion Models

For effective forecasting of tropical cyclone (TC) it is very important to have accurate initial structure of cyclone in numerical models. Assimilation of local data such as Doppler Weather Radar (DWR) data in numerical weather prediction (NWP) models has potential to improve the initial and boundary condition for the prediction of land falling cyclones. Xiao et al. (2005, 2007) carried out study on the assimilation of DWR radial wind and reflectivity into NWP model using the 3-dimensional variational data assimilation (3DVAR) system for the heavy rainfall events. A number of case studies on the positive impact of DWR radial wind and reflectivity observations in the assimilation cycle of Advanced Regional Prediction System (ARPS) were documented by Xue et al. (2003). Kun Zhao and Ming Xue (2009) studied the impact of radar data on the analysis and prediction of the structure, intensity and track of land falling Hurricane Ike-2008, at a cloud-resolving resolution. The hurricane landfall, intensification and weakening during the simulation period are well captured by assimilating both airborne Doppler radar reflectivity and wind data (Zhaoxia et al., 2009). The Bratseth successive correction technique and cloud analysis are part of ARPS Model developed by Center for Analysis and Prediction of Storms (CAPS), Oklahoma University, USA (Bratseth, 1986; Brewster, 1996). The ARPS data assimilation system (ADAS) and cloud analysis technique have capability to be applied for assimilation of radar data in Weather Research and Forecasting (WRF) Model.

Since 1970, tropical cyclone (TC) track forecasts have improved steadily in the Atlantic basin. This improvement has been linked primarily to advances in numerical weather prediction (NWP) models. Concurrently, with few exceptions, the development and operational use of statistical track prediction schemes have experienced a relative decline. Statistical schemes provided the most accurate TC track forecasts until approximately the late 1980s. In this note, it is shown that increased reliance on the global NWP models does not always guarantee the best forecast. Here, Hurricane Ivan is used from the 2004 Atlantic TC season as a classical example, and reminder, of how strong climatological signals still can add substantial value to TC track forecasts, in the form of improved accuracy and increased timeliness at minimal computational cost. In an 8-day period in early September 2004, Hurricane Ivan was repeatedly, and incorrectly, forecast by 12 operational NWP models to move with a significant northward (poleward) component. It was found that the mean 24-h trajectory forecasts of a consensus of five commonly used NWP track prediction aids had a statistically significant right-of-track bias. Furthermore, the official track forecasts, which relied heavily on erroneous numerical guidance over this period, were also found to have significant poleward trajectory errors. At the same time, a climatology-based prediction technique, drawn entirely from the historical record of motion characteristics of TCs in geographical locations similar to Ivan, correctly and consistently indicated a more westward motion component, had a small directional spread, and was supported by a large number of archived cases. This climatological signal was in conflict with the deterministic NWP model output, and it is suggested that the large errors in the official track forecast for TC Ivan could have been reduced considerably by taking into greater account such a strong climatological signal. The potential impact of such an error reduction is a saving of lives and billions of dollars in both actual damage and unnecessary evacuations costs, for just this one hurricane. We also suggest that this simple strategy of examining the strength of the climatological signal be considered for all TCs to identify cases where the NWP and official forecasts differ significantly from strong, persistent climatological signals.

Machine learning techniques are increasingly used in weather forecasting, to either improve results from numerical weather prediction (NWP) models or emulate some aspects of the forward integration part of NWP models using a purely data-driven approach. So far most of the techniques presented have focused on producing deterministic forecasts, as many standard machine learning techniques lack the ability to express uncertainty.   Whilst methods that attempt to fully emulate NWP models are both very complex and computationally expensive, in this talk I will show how much simpler and computationally cheaper techniques can be used to create a probabilistic forecast from a single global high-resolution forecast by post-processing. Specifically, I will focus on the relatively novel technique of a Bayesian Neural Network and show how it can predict the distribution of the forecast error relative to its own analysis. By adding these error distributions to the original forecast, we can create a probabilistic forecast. Time permitting, I will discuss how more advanced techniques can be applied to take these distributions and generate spatially consistent ensemble members.  This methodology is particularly useful for NWP models at very high resolutions where running an ensemble is too computationally expensive and for machine learning approaches where no uncertainty information is available. In this talk, I will show how this methodology can be successfully applied to both ECMWF’s high-resolution forecast and a purely data-driven weather forecast model being run at ECMWF, for both the surface variable of 2m temperature and the atmospheric variable of Z500. These probabilistic forecasts have been verified using standard metrics and, in the case of the high-resolution numerical forecast, have as good as or better CRPS scores than the ECMWF ensemble forecast for the lead times tested. Moreover, these probabilistic forecasts are reliable with spread-skill ratios close to one. Hence, this novel machine learning post-processing technique has the potential to produce probabilistic forecasts that are valuable and useful to forecasters. 

Recent studies have shown the unique value of satellite-observed land surface thermal infrared (TIR) information (e.g., skin temperature) and the feasibility of assimilating land surface temperature (LST) into land surface models (LSMs) to improve the simulation of land-atmosphere water and energy exchanges. In this study, two different types of LST assimilation techniques are implemented and the benefits from the techniques are compared. One of the techniques is to directly assimilate LST using ensemble Kalman filter (EnKF) data assimilation (DA) utilities. The other is to use the Atmosphere-Land Exchange Inversion model (ALEXI) as an “observation operator” that converts LST retrievals into the soil moisture (SM) proxy based on the ratio of actual to potential evapotranspiration (fPET), which is then assimilated into an LSM. While most current studies have shown some success in both directly the assimilating LST and assimilating ALEXI SM proxy into offline LSMs, the potential impact of the assimilation of TIR information through coupled numerical weather prediction (NWP) models is unclear. In this study, a semi-coupled Land Information System (LIS) and Weather Research and Forecast (WRF) system is employed to assess the impact of the two different techniques for assimilating the TIR observations from NOAA GOES satellites on WRF model forecasts. The NASA LIS, equipped with a variety of LSMs and advanced data assimilation tools (e.g., the ensemble Kalman Filter (EnKF)), takes atmospheric forcing data from the WRF model run, generates updated initial land surface conditions with the assimilation of either LST- or TIR-based SM and returns them to WRF for initializing the forecasts. The WRF forecasts using the daily updated initializations with the TIR data assimilation are evaluated against ground weather observations and re-analysis products. It is found that WRF forecasts with the LST-based SM assimilation have better agreement with the ground weather observations than those with the direct LST assimilation or without the land TIR data assimilation.