Abstract
Societies in much of the Horn of Africa are affected by variability in two distinct rainy seasons: the March-May (MAM) “long” rains and the October-December (OND) “short” rains. A recent 5-season, La Niña-forced drought has renewed concerns about possible anthropogenic drying trends in the long rains, which had partially recovered after a multidecadal drying trend in the 1980s through 2000s. Despite observed drying, previous generations of global climate models (GCMs) have consistently projected long-term wetting due to increased greenhouse gas concentrations, an East African “Paradox” which complicates the interpretation of East African rainfall projections. We investigate the Paradox in new CMIP6 and seasonal forecast models, leveraging an improved observational record and Large Ensembles to better differentiate internal and forced trends. We find observed drying trends are at the limits of the GCM spread during the peak Paradox period, though the recent recovery is comfortably within the model spread. We find that the apparent Paradox is largely removed by prescribing sea surface temperatures (SSTs), and is likely caused by the GCM difficulties in simulating observed tropical Pacific SST trends in recent decades. In line with arguments that these SST trends are at least partially forced anthropogenically, we recommend users of future rainfall projections in East Africa consider the possibility of long-term MAM drying despite GCM wetting, and call for future model simulations that better sample the expected spread of SSTs.
Abstract
Societies in much of the Horn of Africa are affected by variability in two distinct rainy seasons: the March-May (MAM) “long” rains and the October-December (OND) “short” rains. A recent 5-season, La Niña-forced drought has renewed concerns about possible anthropogenic drying trends in the long rains, which had partially recovered after a multidecadal drying trend in the 1980s through 2000s. Despite observed drying, previous generations of global climate models (GCMs) have consistently projected long-term wetting due to increased greenhouse gas concentrations, an East African “Paradox” which complicates the interpretation of East African rainfall projections. We investigate the Paradox in new CMIP6 and seasonal forecast models, leveraging an improved observational record and Large Ensembles to better differentiate internal and forced trends. We find observed drying trends are at the limits of the GCM spread during the peak Paradox period, though the recent recovery is comfortably within the model spread. We find that the apparent Paradox is largely removed by prescribing sea surface temperatures (SSTs), and is likely caused by the GCM difficulties in simulating observed tropical Pacific SST trends in recent decades. In line with arguments that these SST trends are at least partially forced anthropogenically, we recommend users of future rainfall projections in East Africa consider the possibility of long-term MAM drying despite GCM wetting, and call for future model simulations that better sample the expected spread of SSTs.
Abstract
Atmospheric state analysis that leverages state-of-the-art data assimilation achieves high accuracy and can provide initial conditions for numerical weather prediction (NWP) and climatological reanalysis. However, the interactions between the atmosphere and the ocean have been inadequately addressed, with sea surface temperature (SST) as a boundary condition for the atmosphere. This limitation impacts the accuracy of atmospheric state analyses and the utilization of SST-sensitive observations. To address this, we developed a partially coupled data assimilation (PCDA) system for the atmosphere and SST by extending the operational global NWP system of the Japan Meteorological Agency. The PCDA system enhances the analysis variables and background error covariance matrices to include SST components and the use of microwave radiance observations sensitive to SST, particularly at low frequencies (6 to 11 GHz), which have previously been unused or absent in most NWP systems. Our numerical experiments demonstrate several key findings: (1) The PCDA system identified colder SSTs in regions with significant SST gradients, including SST fronts in the mid-latitudes, and we obtained zonally positive and negative increments in tropical instability wave regions; (2) The SST analysis produced by PCDA was consistent with independent SST analyses; (3) The system yielded a moist and warm low-level troposphere, leading to an increase in the first 24-h rain forecast near the intertropical convergence zone; and (4) PCDA globally improved the forecast accuracy of near-surface temperatures, with notable improvements in the tropics for most variables, except for mid-tropospheric temperature. In the extra-tropics, forecast accuracy improvements were observed for height and humidity, although some degradation occurred mainly in the southern hemisphere.
Abstract
Atmospheric state analysis that leverages state-of-the-art data assimilation achieves high accuracy and can provide initial conditions for numerical weather prediction (NWP) and climatological reanalysis. However, the interactions between the atmosphere and the ocean have been inadequately addressed, with sea surface temperature (SST) as a boundary condition for the atmosphere. This limitation impacts the accuracy of atmospheric state analyses and the utilization of SST-sensitive observations. To address this, we developed a partially coupled data assimilation (PCDA) system for the atmosphere and SST by extending the operational global NWP system of the Japan Meteorological Agency. The PCDA system enhances the analysis variables and background error covariance matrices to include SST components and the use of microwave radiance observations sensitive to SST, particularly at low frequencies (6 to 11 GHz), which have previously been unused or absent in most NWP systems. Our numerical experiments demonstrate several key findings: (1) The PCDA system identified colder SSTs in regions with significant SST gradients, including SST fronts in the mid-latitudes, and we obtained zonally positive and negative increments in tropical instability wave regions; (2) The SST analysis produced by PCDA was consistent with independent SST analyses; (3) The system yielded a moist and warm low-level troposphere, leading to an increase in the first 24-h rain forecast near the intertropical convergence zone; and (4) PCDA globally improved the forecast accuracy of near-surface temperatures, with notable improvements in the tropics for most variables, except for mid-tropospheric temperature. In the extra-tropics, forecast accuracy improvements were observed for height and humidity, although some degradation occurred mainly in the southern hemisphere.
Abstract
Ensemble sensitivity analysis (ESA) offers a computationally inexpensive way to diagnose sources of high-impact forecast feature uncertainty by relating a localized forecast phenomenon of interest (response function) back to early or initial forecast conditions (sensitivity variables). These information-rich diagnostic fields allow us to quantify the predictability characteristics of a specific forecast event. This work harnesses insights from a month-long dataset of ESA applied to convection-allowing model precipitation forecasts in the Central Plains of the US. Temporally-averaged and spatially-averaged sensitivity statistics are correlated with a variety of other metrics, such as skill, spread, and mean forecast precipitation accumulation. A high, but imperfect, correlation (0.81) between forecast precipitation and sensitivity is discovered. This quantity confirms the qualitatively known notion that while there is a connection between predictability and event magnitude, a high-end event does not necessarily entail a low predictability (high sensitivity) forecast. Flow regimes within this dataset are analyzed to see which patterns lend themselves to high and low predictability forecast scenarios. Finally, a novel metric known as the Error Growth Realization (EGR) Ratio is introduced. Derived by dividing the two mathematical formulations of ESA, this metric shows preliminary promise as a predictor of forecast skill prior to onset of a high-impact convective event. In essence, this research exemplifies the potential of ESA beyond its traditional use in case studies. By applying ESA to a broader dataset, we can glean valuable insight into the predictability of high-impact weather events, and in turn, work towards a collective baseline on what constitutes a high- or low-predictability event in the first place.
Abstract
Ensemble sensitivity analysis (ESA) offers a computationally inexpensive way to diagnose sources of high-impact forecast feature uncertainty by relating a localized forecast phenomenon of interest (response function) back to early or initial forecast conditions (sensitivity variables). These information-rich diagnostic fields allow us to quantify the predictability characteristics of a specific forecast event. This work harnesses insights from a month-long dataset of ESA applied to convection-allowing model precipitation forecasts in the Central Plains of the US. Temporally-averaged and spatially-averaged sensitivity statistics are correlated with a variety of other metrics, such as skill, spread, and mean forecast precipitation accumulation. A high, but imperfect, correlation (0.81) between forecast precipitation and sensitivity is discovered. This quantity confirms the qualitatively known notion that while there is a connection between predictability and event magnitude, a high-end event does not necessarily entail a low predictability (high sensitivity) forecast. Flow regimes within this dataset are analyzed to see which patterns lend themselves to high and low predictability forecast scenarios. Finally, a novel metric known as the Error Growth Realization (EGR) Ratio is introduced. Derived by dividing the two mathematical formulations of ESA, this metric shows preliminary promise as a predictor of forecast skill prior to onset of a high-impact convective event. In essence, this research exemplifies the potential of ESA beyond its traditional use in case studies. By applying ESA to a broader dataset, we can glean valuable insight into the predictability of high-impact weather events, and in turn, work towards a collective baseline on what constitutes a high- or low-predictability event in the first place.
Abstract
Earth and planetary radiometry requires spectrally dependent observations spanning an expansive range in signal flux due to variability in celestial illumination, spectral albedo, and attenuation. Insufficient dynamic range inhibits contemporaneous measurements of dissimilar signal levels and restricts potential environments, time periods, target types, or spectral ranges that instruments observe. Next-generation (NG) advances in temporal, spectral, and spatial resolution also require further increases in detector sensitivity and dynamic range corresponding to increased sampling rate and decreased field-of-view (FOV), both of which capture greater intrapixel variability (i.e., variability within the spatial and temporal integration of a pixel observation). Optical detectors typically must support expansive linear radiometric responsivity, while simultaneously enduring the inherent stressors of field, airborne, or satellite deployment. Rationales for significantly improving radiometric observations of nominally dark targets are described herein, along with demonstrations of state-of-the-art (SOTA) capabilities and NG strategies for advancing SOTA. An evaluation of linear dynamic range and efficacy of optical data products is presented based on representative sampling scenarios. Low-illumination (twilight or total lunar eclipse) observations are demonstrated using a SOTA prototype. Finally, a ruggedized and miniaturized commercial-off-the-shelf (COTS) NG capability to obtain absolute radiometric observations spanning an expanded range in target brightness and illumination is presented. The presented NG technology combines a Multi-Pixel Photon Counter (MPPC) with a silicon photodetector (SiPD) to form a dyad optical sensing component supporting expansive dynamic range sensing, i.e., exceeding a nominal 10 decades in usable dynamic range documented for SOTA instruments.
Abstract
Earth and planetary radiometry requires spectrally dependent observations spanning an expansive range in signal flux due to variability in celestial illumination, spectral albedo, and attenuation. Insufficient dynamic range inhibits contemporaneous measurements of dissimilar signal levels and restricts potential environments, time periods, target types, or spectral ranges that instruments observe. Next-generation (NG) advances in temporal, spectral, and spatial resolution also require further increases in detector sensitivity and dynamic range corresponding to increased sampling rate and decreased field-of-view (FOV), both of which capture greater intrapixel variability (i.e., variability within the spatial and temporal integration of a pixel observation). Optical detectors typically must support expansive linear radiometric responsivity, while simultaneously enduring the inherent stressors of field, airborne, or satellite deployment. Rationales for significantly improving radiometric observations of nominally dark targets are described herein, along with demonstrations of state-of-the-art (SOTA) capabilities and NG strategies for advancing SOTA. An evaluation of linear dynamic range and efficacy of optical data products is presented based on representative sampling scenarios. Low-illumination (twilight or total lunar eclipse) observations are demonstrated using a SOTA prototype. Finally, a ruggedized and miniaturized commercial-off-the-shelf (COTS) NG capability to obtain absolute radiometric observations spanning an expanded range in target brightness and illumination is presented. The presented NG technology combines a Multi-Pixel Photon Counter (MPPC) with a silicon photodetector (SiPD) to form a dyad optical sensing component supporting expansive dynamic range sensing, i.e., exceeding a nominal 10 decades in usable dynamic range documented for SOTA instruments.
Abstract
Before the satellite era, global sea-level reconstructions depend on tide-gauge records and in-situ hydrographic observations. However, the available global mean sea-level (GMSL) reconstructions, using different methods, indicate a spread in sea-level trend over 1900-2008 (1.3∼2.0 mm yr−1). With better understanding of the causes of sea-level change, here we implement an improved sea-level reconstruction, building upon Church and White (2011), and including three additional factors: the sea-level fingerprints, the sterodynamic sea level (SDSL) climate change patterns and more complete local vertical land motion (VLM) estimates. The trend of new GMSL reconstruction is 1.6 ± 0.2 mm yr−1 (90% confidence level) over 1900-2019, consistent with the sum of observation-based sea-level contributions of 1.5 ± 0.2 mm yr−1. The lower trend from the new reconstruction compared with the earlier Church and White (2011) result is mainly due to the updated VLM correction. The inclusion of sea-level fingerprints and SDSL climate change patterns are the dominant contributors for the improved skill of regional reconstruction. Despite GMSL budget closure in terms of long-term trend since 1900, our study shows discrepancies between the trends from available GMSL reconstructions and the sum of independent observation-based contributions over different periods in the 20th century, e.g., the discrepancy at the beginning of the 20th century, which could be related to possible bias in the land ice component estimate. The reconstruction methodology developed here, as tested with synthetic sea-level fields, could provide a promising way to identify potential biases in the individual sea-level components constrained by available global tide-gauge observations.
Abstract
Before the satellite era, global sea-level reconstructions depend on tide-gauge records and in-situ hydrographic observations. However, the available global mean sea-level (GMSL) reconstructions, using different methods, indicate a spread in sea-level trend over 1900-2008 (1.3∼2.0 mm yr−1). With better understanding of the causes of sea-level change, here we implement an improved sea-level reconstruction, building upon Church and White (2011), and including three additional factors: the sea-level fingerprints, the sterodynamic sea level (SDSL) climate change patterns and more complete local vertical land motion (VLM) estimates. The trend of new GMSL reconstruction is 1.6 ± 0.2 mm yr−1 (90% confidence level) over 1900-2019, consistent with the sum of observation-based sea-level contributions of 1.5 ± 0.2 mm yr−1. The lower trend from the new reconstruction compared with the earlier Church and White (2011) result is mainly due to the updated VLM correction. The inclusion of sea-level fingerprints and SDSL climate change patterns are the dominant contributors for the improved skill of regional reconstruction. Despite GMSL budget closure in terms of long-term trend since 1900, our study shows discrepancies between the trends from available GMSL reconstructions and the sum of independent observation-based contributions over different periods in the 20th century, e.g., the discrepancy at the beginning of the 20th century, which could be related to possible bias in the land ice component estimate. The reconstruction methodology developed here, as tested with synthetic sea-level fields, could provide a promising way to identify potential biases in the individual sea-level components constrained by available global tide-gauge observations.
Abstract
The implications of definitions of excessive rainfall observations on machine learning-model forecast skill is assessed using the Colorado State University Machine Learning Probabilities (CSU-MLP) forecast system. The CSU-MLP uses historical observations along with reforecasts from a global ensemble to train random forests to probabilistically predict excessive rainfall events. Here, random forest models are trained using two distinct rainfall datasets, one that is composed of fixed-frequency (FF) average recurrence intervals exceedances and flash flood reports, and the other a compilation of flooding and rainfall proxies (Unified Flood Verification System; UFVS). Both models generate 1–3 day forecasts and are evaluated against a climatological baseline to characterize their overall skill as a function of lead time, season, and region. Model comparisons suggest that regional frequencies in excessive rainfall observations contribute to when and where the ML models issue forecasts, and subsequently their skill and reliability. Additionally, the spatio-temporal distribution of observations have implications for ML model training requirements, notably, how long of an observational record is needed to obtain skillful forecasts. Experiments reveal that shorter-trained UFVS-based models can be as skillful as longer-trained FF-based models. In essence, the UFVS dataset exhibits a more robust characterization of excessive rainfall and impacts, and machine learning models trained on more representative datasets of meteorological hazards may not require as extensive training to generate skillful forecasts.
Abstract
The implications of definitions of excessive rainfall observations on machine learning-model forecast skill is assessed using the Colorado State University Machine Learning Probabilities (CSU-MLP) forecast system. The CSU-MLP uses historical observations along with reforecasts from a global ensemble to train random forests to probabilistically predict excessive rainfall events. Here, random forest models are trained using two distinct rainfall datasets, one that is composed of fixed-frequency (FF) average recurrence intervals exceedances and flash flood reports, and the other a compilation of flooding and rainfall proxies (Unified Flood Verification System; UFVS). Both models generate 1–3 day forecasts and are evaluated against a climatological baseline to characterize their overall skill as a function of lead time, season, and region. Model comparisons suggest that regional frequencies in excessive rainfall observations contribute to when and where the ML models issue forecasts, and subsequently their skill and reliability. Additionally, the spatio-temporal distribution of observations have implications for ML model training requirements, notably, how long of an observational record is needed to obtain skillful forecasts. Experiments reveal that shorter-trained UFVS-based models can be as skillful as longer-trained FF-based models. In essence, the UFVS dataset exhibits a more robust characterization of excessive rainfall and impacts, and machine learning models trained on more representative datasets of meteorological hazards may not require as extensive training to generate skillful forecasts.
Abstract
PixPol is an in-water multi-spectral polarized upwelling radiance distribution fisheye camera system. Its imaging sensors utilize a pixel-level polarizer structure allowing for polarimetric retrieval from one image capture, offering an advantage compared to other in-water polarimetric fisheye camera systems that require information from multiple images. When submerged, PixPol images a scene from which the first three Stokes parameters are derived at an angular resolution of 1° within a field of view that encompasses all azimuthal angles up to an elevation of 43° from nadir. For all viewing angles, Stokes parameter I and the linear polarization parameters, Q/I and U/I, are retrieved with an inter-pixel uncertainty of ±5%, ±0.02, and ±0.02, respectively. From these parameters, an uncertainty of ±0.01 is attained for the degree of linear polarization and ±0.9° for the angle of linear polarization. A description of the camera system, its radiometric and polarization calibration, and the associated uncertainties are described. Example images of the distribution of downwelling polarized light in the sky just above the ocean’s surface and upwelling polarized light just below the surface are provided.
Abstract
PixPol is an in-water multi-spectral polarized upwelling radiance distribution fisheye camera system. Its imaging sensors utilize a pixel-level polarizer structure allowing for polarimetric retrieval from one image capture, offering an advantage compared to other in-water polarimetric fisheye camera systems that require information from multiple images. When submerged, PixPol images a scene from which the first three Stokes parameters are derived at an angular resolution of 1° within a field of view that encompasses all azimuthal angles up to an elevation of 43° from nadir. For all viewing angles, Stokes parameter I and the linear polarization parameters, Q/I and U/I, are retrieved with an inter-pixel uncertainty of ±5%, ±0.02, and ±0.02, respectively. From these parameters, an uncertainty of ±0.01 is attained for the degree of linear polarization and ±0.9° for the angle of linear polarization. A description of the camera system, its radiometric and polarization calibration, and the associated uncertainties are described. Example images of the distribution of downwelling polarized light in the sky just above the ocean’s surface and upwelling polarized light just below the surface are provided.
Abstract
Probabilistic forecasts derived from ensemble prediction systems (EPS) have become the standard basis for many products and services produced by modern operational forecasting centres. However statistical post-processing is generally required to ensure forecasts have the desired properties expected for probability-based outputs. Precipitation, a core component of any operational forecast, is particularly challenging to calibrate due to its discontinuous nature and the extreme skew in rainfall amounts. A skillful forecasting system must maintain accuracy for low-to-moderate precipitation amounts, but preserve resolvability for high-to-extreme rainfall amounts, which, though rare, are important to forecast accurately in the interest of public safety. Existing statistical and machine-learning approaches to rainfall calibration address this problem, but each has drawbacks in design, training approaches, and/or performance. We describe RainForests, a machine-learning approach for calibrating ensemble rainfall forecasts using gradient-boosted decision trees. The model is based on the ecPoint system recently developed at ECMWF by Hewson and Pillosu (2021), but uses machine-learning models in place of the semi-subjective decision trees of ecPoint, along with some other improvements to the model structure. We evaluate RainForests on the Australian domain against some simple benchmarks, and show that it outperforms standard calibration approaches both in overall skill and in accurately forecasting high rainfall conditions, while being computationally efficient enough to be used in an operational forecasting system.
Abstract
Probabilistic forecasts derived from ensemble prediction systems (EPS) have become the standard basis for many products and services produced by modern operational forecasting centres. However statistical post-processing is generally required to ensure forecasts have the desired properties expected for probability-based outputs. Precipitation, a core component of any operational forecast, is particularly challenging to calibrate due to its discontinuous nature and the extreme skew in rainfall amounts. A skillful forecasting system must maintain accuracy for low-to-moderate precipitation amounts, but preserve resolvability for high-to-extreme rainfall amounts, which, though rare, are important to forecast accurately in the interest of public safety. Existing statistical and machine-learning approaches to rainfall calibration address this problem, but each has drawbacks in design, training approaches, and/or performance. We describe RainForests, a machine-learning approach for calibrating ensemble rainfall forecasts using gradient-boosted decision trees. The model is based on the ecPoint system recently developed at ECMWF by Hewson and Pillosu (2021), but uses machine-learning models in place of the semi-subjective decision trees of ecPoint, along with some other improvements to the model structure. We evaluate RainForests on the Australian domain against some simple benchmarks, and show that it outperforms standard calibration approaches both in overall skill and in accurately forecasting high rainfall conditions, while being computationally efficient enough to be used in an operational forecasting system.
