Search

This study explores coupled land–atmosphere data assimilation (DA) for improving weather and hydrological forecasts by assimilating soil moisture (SM) data. This study integrates a land DA component into a global atmospheric DA system of the Nonhydrostatic ICosahedral Atmospheric Model and the local ensemble transform Kalman filter (NICAM-LETKF) and performs both strongly and weakly coupled land–atmosphere DA experiments. We explore various types of coupled DA experiments by assimilating atmospheric observations and SM data simultaneously. The results show that analyzing atmospheric variables by assimilating SM data improves the SM analysis and forecasts and mitigates a warm bias in the lower troposphere where a dry SM bias exists. On the other hand, updating SM by assimilating atmospheric observations has detrimental impacts due to spurious error correlations between the atmospheric observations and land model variables. We also find that assimilating SM by strongly coupled DA is beneficial in the Sahel and equatorial Africa from May to October. These regions are characterized by seasonal variations in the precipitation patterns and benefit from updates in the atmospheric variables through SM DA during periods of increased precipitation. Additionally, these regions coincide with those identified in the previous studies, where a global initialization of SM would enhance the prediction skill of seasonal precipitation.

The control simulation experiment (CSE) is a recently developed approach to investigate the controllability of dynamical systems, extending the well-known observing system simulation experiment (OSSE) in meteorology. For effective control of chaotic dynamical systems, it is essential to exploit the high sensitivity to initial conditions for dragging a system away from an undesired regime by applying minimal perturbations. In this study, we design a CSE for reducing the number of extreme events in the Lorenz-96 model. The 40 variables of this model represent idealized meteorological quantities evenly distributed on a latitude circle. The reduction of occurrence of extreme events over 100-year runs of the model is discussed as a function of the parameters of the CSE: the ensemble forecast length for detecting extreme events in advance, the magnitude and localization of the perturbations, and the quality and coverage of the observations. The design of the CSE is aimed at reducing weather extremes when applied to more realistic weather prediction models.

The success of ensemble data assimilation systems substantially depends on localization, which is required to mitigate sampling errors caused by modeling background error covariances with undersized ensembles. However, finding an optimal localization is highly challenging, as covariances, sampling errors, and appropriate localization depend on various factors. Our study investigates vertical localization based on a unique convection-permitting 1000-member ensemble simulation; 1000-member ensemble correlations serve as truth for examining vertical correlations and their sampling error. We discuss requirements for vertical localization by deriving an empirical optimal localization (EOL) that minimizes the sampling error in 40-member subsample correlations with respect to the 1000-member reference. Our analysis covers temperature, specific humidity, and wind correlations on various pressure levels. Results suggest that vertical localization should depend on several aspects, such as the respective variable, vertical level, or correlation type (self- or cross-correlations). Comparing the empirical optimal localization with common distance-dependent localization approaches highlights that finding suitable localization functions bears substantial room for improvement. Furthermore, we examine approaches for achieving positive semi-definiteness for covariance localization that hardly affect the sampling error reduction. Finally, we discuss the gain of combining different localization approaches with an adaptive statistical sampling error correction.

A previous study proposed an adaptive observation error inflation (AOEI) method for an ensemble Kalman filter (EnKF)-based atmospheric data assimilation system to assimilate all-sky infrared brightness temperatures. Brightness temperature differences between clear- and cloudy-sky radiances are large, and observation-minus-forecast differences (or innovations) are therefore likely to be large around boundaries between clear- and cloudy-sky regions. The AOEI method mitigates these discrepancies by adaptively inflating observation errors. Ocean frontal regions have similar characteristics to the borders between clear- and cloudy-sky regions with large innovations. Consequently, we have implemented the AOEI with an EnKF-based regional ocean data assimilation system, in which the assimilation interval is set to 1 d to utilize frequent satellite observations. We conducted sensitivity experiments to investigate the impacts of the AOEI on salinity structure, geostrophic balance, and accuracy. A control run, in which the AOEI is not applied, shows the degradation of low-salinity North Pacific Intermediate Water around the Kuroshio Extension region, where the innovation amplitude and forecast ensemble spread are large in association with the fronts and eddies. The resulting large temperature and salinity increments weaken the density stratification, leading to large vertical diffusivity. As a result, the low-salinity water in the intermediate layer is lost through strong vertical diffusion. When the AOEI is used, the salinity structure in the ocean interior is preserved because the AOEI suppresses the salinity degradation by reducing the temperature and salinity increments. We also demonstrate that the AOEI provides significant improvement of the geostrophic balance and the analysis accuracy of temperature, salinity, and surface-flow fields.

A particle filter (PF) is an ensemble data assimilation method that does not assume Gaussian error distributions. Recent studies proposed local PFs (LPFs), which use localization, as in the ensemble Kalman filter, to apply the PF efficiently for high-dimensional dynamics. Among others, Penny and Miyoshi (2016) developed an LPF in the form of the ensemble transform matrix of the local ensemble transform Kalman filter (LETKF). The LETKF has been widely accepted for various geophysical systems, including numerical weather prediction (NWP) models. Therefore, implementing the LPF consistently with an existing LETKF code is useful. This study develops a software platform for the LPF and its Gaussian mixture extension (LPFGM) by making slight modifications to the LETKF code with a simplified global climate model known as Simplified Parameterizations, Primitive Equation Dynamics (SPEEDY). A series of idealized twin experiments were accomplished under the ideal-model assumption. With large inflation by the relaxation to prior spread, the LPF showed stable filter performance with dense observations but became unstable with sparse observations. The LPFGM showed a more accurate and stable performance than the LPF with both dense and sparse observations. In addition to the relaxation parameter, regulating the resampling frequency and the amplitude of Gaussian kernels was important for the LPFGM. With a spatially inhomogeneous observing network, the LPFGM was superior to the LETKF in sparsely observed regions, where the background ensemble spread and non-Gaussianity were larger. The SPEEDY-based LETKF, LPF, and LPFGM systems are available as open-source software on GitHub (https://github.com/skotsuki/speedy-lpf, last access: 16 November 2022) and can be adapted to various models relatively easily, as in the case of the LETKF.

This study develops an ensemble Kalman filter (EnKF)-based regional ocean data assimilation system in which the local ensemble transform Kalman filter (LETKF) is implemented with version 1.0 of the Stony Brook Parallel Ocean Model (sbPOM) to assimilate satellite and in situ observations at a daily frequency. A series of sensitivity experiments are performed with various settings of the incremental analysis update (IAU) and covariance inflation methods, for which the relaxation-to-prior perturbations and spread (RTPP and RTPS, respectively) and multiplicative inflation (MULT) are considered. We evaluate the geostrophic balance and the analysis accuracy compared with the control experiment in which the IAU and covariance inflation are not applied. The results show that the IAU improves the geostrophic balance, degrades the accuracy, and reduces the ensemble spread, and that the RTPP and RTPS have the opposite effect. The experiment using a combination of the IAU and RTPP results in a significant improvement for both balance and analysis accuracy when the RTPP parameter is 0.8–0.9. The combination of the IAU and RTPS improves the balance when the RTPS parameter is ≤0.8 and increases the analysis accuracy for parameter values between 1.0 and 1.1, but the balance and analysis accuracy are not improved significantly at the same time. The experiments with MULT inflating the forecast ensemble spread by 5 % do not demonstrate sufficient skill in maintaining the balance and reproducing the surface flow field regardless of whether the IAU is applied or not. The 11 d ensemble forecast experiments show consistent results. Therefore, the combination of the IAU and RTPP with a parameter value of 0.8–0.9 is found to be the best setting for the EnKF-based ocean data assimilation system.

In numerical weather prediction (NWP), sensitivity to initial conditions brings chaotic behaviors and an intrinsic limit to predictability, but it also implies an effective control in which a small control signal grows rapidly to make a substantial difference. The Observing Systems Simulation Experiment (OSSE) is a well-known approach to study predictability, where “nature” is synthesized by an independent NWP model run. In this study, we extend the OSSE and design the control simulation experiment (CSE), where we apply a small signal to control “nature”. Idealized experiments with the Lorenz-63 three-variable system show that we can control “nature” to stay in a chosen regime without shifting to the other, i.e., in a chosen wing of Lorenz's butterfly attractor, by adding small perturbations to “nature”. Using longer-lead-time forecasts, we achieve more effective control with a perturbation size of less than only 3 % of the observation error. We anticipate our idealized CSE to be a starting point for a realistic CSE using the real-world NWP systems, toward possible future applications to reduce weather disaster risks. The CSE may be applied to other chaotic systems beyond NWP.

Abstract Systematic biases in numerical weather prediction models cause forecast deviation from reality. While model biases also affect data assimilation and degrade the analysis accuracy, observation information incorporated through data assimilation can provide information for detecting and alleviating such biases. In this study, the application of machine learning to model bias correction is demonstrated, emphasizing the effectiveness of recurrent neural networks. Idealized experiments are performed using the two‐scale coupled Lorenz‐96 model as the true system and single Lorenz‐96 model as the imperfect forecast model, to compare the effectiveness of bias correction methods based on various architectures of neural networks and simple linear regression. The neural networks generally outperformed linear regression, and recurrent neural networks showed the best ability in finding the systematic bias component from the analysis increment data. Bias correction using the recurrent neural networks also gives the most significant improvement in reducing the error growth rate in extended range forecasts. The results suggest that including past time series of the forecast variables improve model bias correction when limited information of the observation is incorporated through data assimilation.

Non-Gaussian forecast error is a challenge for ensemble-based data assimilation (DA), particularly for more nonlinear convective dynamics. In this study, we investigate the degree of the non-Gaussianity of forecast error distributions at 1 km resolution using a 1000-member ensemble Kalman filter, and how it is affected by the DA update frequency and observation number. Regional numerical weather prediction experiments are performed with the SCALE (Scalable Computing for Advanced Library and Environment) model and the LETKF (local ensemble transform Kalman filter) assimilating phased array radar observations every 30 s. The results show that non-Gaussianity develops rapidly within convective clouds and is sensitive to the DA frequency and the number of assimilated observations. The non-Gaussianity is reduced by up to 40 % when the assimilation window is shortened from 5 min to 30 s, particularly for vertical velocity and radar reflectivity.

Strontium Ruthenate, Sr2RuO4, displays a remarkable number of intriguing physical phenomena, from superconductivity, to strain-induced ferromagnetism. Here, using low-energy muon spectroscopy, Fittipaldi et al. demonstrate the existence of unconventional magnetism at the surface of Sr2RuO4 in its normal state and without any applied strain.

Abstract We present the first ever real‐time numerical weather prediction system with 30‐s update cycles at a 500‐m grid spacing for the prediction of convective precipitation in the subsequent 30 min using a new‐generation multi‐parameter phased array weather radar. The system comprises a regional atmospheric model known as the SCALE and the local ensemble transform Kalman filter (LETKF). To accelerate the SCALE‐LETKF system, data transfer between the two aforementioned components is performed using a memory copy instead of a file I/O. A complete real‐time workflow including domain nesting and observational data transfer is constructed. A real‐time test in July and August 2020 showed that the system is fast enough for a real‐time application of 30‐s forecast‐analysis cycles and 30‐min prediction. The development includes a new thinning method considering the spatially correlated observation errors in the dense radar data. This new thinning method is effective in two past case studies in the summer of 2019.

The objective of this study is to evaluate the spectral difference between and within 11 tree species belonging to Brazilian Atlantic Forest located in the countryside of São Paulo State, Brazil. Tree species with different development stages may have different reflectance spectra because of the structural and phenological influence on it. Tree structure can affect the crown spectra due to the differences in geometry of view and density of shadow and sunlight pixels. Additionally, reflectance spectra can be similar between tree species belonging from different species, which affect the classification accuracy of these targets. In this sense, we evaluate the spectral difference between highly diverse Atlantic Forest using two different features extracted from individual tree crowns (ITCs). Mean reflectance spectra and mean normalized reflectance spectra of each ITC were used. They were computed from hyperspectral images acquired with sensor onboard unmanned aerial vehicle. Differences between tree species and within-species were calculated. Wilcoxon-Mann-Whitney test was applied in order to statistically evaluate whether the reflectance spectra were different. Results showed that the normalized values are more propense to increase the classification accuracies than using the mean values since it reduces the variability within-species and present a higher number of different values between-species.

The classification of tree species can significantly benefit from high spatial and spectral information acquired by unmanned aerial vehicles (UAVs) associated with advanced feature extraction and classification methods. Different from the traditional feature extraction methods, that highly depend on user’s knowledge, the convolutional neural network (CNN)-based method can automatically learn and extract the spatial-related features layer by layer. However, in order to capture significant features of the data, the CNN classifier requires a large number of training samples, which are hardly available when dealing with tree species in tropical forests. This study investigated the following topics concerning the classification of 14 tree species in a subtropical forest area of Southern Brazil: i) the performance of the CNN method associated with a previous step to increase and balance the sample set (data augmentation) for tree species classification as compared to the conventional machine learning methods support vector machine (SVM) and random forest (RF) using the original training data; ii) the performance of the SVM and RF classifiers when associated with a data augmentation step and spatial features extracted from a CNN. Results showed that the CNN classifier outperformed the conventional SVM and RF classifiers, reaching an overall accuracy (OA) of 84.37% and Kappa of 0.82. The SVM and RF had a poor accuracy with the original spectral bands (OA 62.67% and 59.24%) but presented an increase between 14% and 21% in OA when associated with a data augmentation and spatial features extracted from a CNN.

The classification of tree species can significantly benefit from high spatial and spectral information acquired by unmanned aerial vehicles (UAVs) associated with advanced classification methods. This study investigated the following topics concerning the classification of 16 tree species in two subtropical forest fragments of Southern Brazil: i) the potential integration of UAV-borne hyperspectral images with 3D information derived from their photogrammetric point cloud (PPC); ii) the performance of two machine learning methods (support vector machine – SVM and random forest – RF) when employing different datasets at a pixel and individual tree crown (ITC) levels; iii) the potential of two methods for dealing with the imbalanced sample set problem: a new weighted SVM (wSVM) approach, which attributes different weights to each sample and class, and a deep learning classifier (convolutional neural network – CNN), associated with a previous step to balance the sample set; and finally, iv) the potential of this last classifier for tree species classification as compared to the above mentioned machine learning methods. Results showed that the inclusion of the PPC features to the hyperspectral data provided a great accuracy increase in tree species classification results when conventional machine learning methods were applied, between 13 and 17% depending on the classifier and the study area characteristics. When using the PPC features and the canopy height model (CHM), associated with the majority vote (MV) rule, the SVM, wSVM and RF classifiers reached accuracies similar to the CNN, which outperformed these classifiers for both areas when considering the pixel-based classifications (overall accuracy of 84.4% in Area 1, and 74.95% in Area 2). The CNN was between 22% and 26% more accurate than the SVM and RF when only the hyperspectral bands were employed. The wSVM provided a slight increase in accuracy not only for some lesser represented classes, but also some major classes in Area 2. While conventional machine learning methods are faster, they demonstrated to be less stable to changes in datasets, depending on prior segmentation and hand-engineered features to reach similar accuracies to those attained by the CNN. To date, CNNs have been barely explored for the classification of tree species, and CNN-based classifications in the literature have not dealt with hyperspectral data specifically focusing on tropical environments. This paper thus presents innovative strategies for classifying tree species in subtropical forest areas at a refined legend level, integrating UAV-borne 2D hyperspectral and 3D photogrammetric data and relying on both deep and conventional machine learning approaches.

We previously performed local ensemble transform Kalman filter (LETKF) experiments with up to 10 240 ensemble members using an intermediate atmospheric general circulation model (AGCM). While the previous study focused on the impact of localization on the analysis accuracy, the present study focuses on the probability density functions (PDFs) represented by the 10 240-member ensemble. The 10 240-member ensemble can resolve the detailed structures of the PDFs and indicates that non-Gaussianity is caused in those PDFs by multimodality and outliers. The results show that the spatial patterns of the analysis errors are similar to those of non-Gaussianity. While the outliers appear randomly, large multimodality corresponds well with large analysis error, mainly in the tropical regions and storm track regions where highly nonlinear processes appear frequently. Therefore, we further investigate the life cycle of multimodal PDFs, and show that they are mainly generated by the on–off switch of convective parameterization in the tropical regions and by the instability associated with advection in the storm track regions. Sensitivity to the ensemble size suggests that approximately 1000 ensemble members are necessary in the intermediate AGCM-LETKF system to represent the detailed structures of non-Gaussian PDFs such as skewness and kurtosis; the higher-order non-Gaussian statistics are more vulnerable to the sampling errors due to a smaller ensemble size.

Spaceborne precipitation radars, such as the Tropical Rainfall Measuring Mission (TRMM) and the Global Precipitation Measurement (GPM) Core Observatory, have been important platforms to provide a direct measurement of three-dimensional precipitation structure globally. Building upon the success of TRMM and GPM Core Observatory, the Japan Aerospace Exploration Agency (JAXA) is currently surveying the feasibility of a potential satellite mission equipped with a precipitation radar on a geostationary orbit. The quasi-continuous observation realized by the geostationary satellite radar would offer a new insight into meteorology and would advance numerical weather prediction (NWP) through their effective use by data assimilation. Although the radar would be beneficial, the radar on the geostationary orbit measures precipitation obliquely at off-nadir points. In addition, the observing resolution will be several times larger than those on board TRMM and GPM Core Observatory due to the limited antenna size that we could deliver. The tilted sampling volume and the coarse resolution would result in more contamination from surface clutter. To investigate the impact of these limitations and to explore the potential usefulness of the geostationary satellite radar, this study simulates the observation data for a typhoon case using an NWP model and a radar simulator. The results demonstrate that it would be possible to obtain three-dimensional precipitation data. However, the quality of the observation depends on the beam width, the beam sampling span, and the position of precipitation systems. With a wide beam width and a coarse beam span, the radar cannot observe weak precipitation at low altitudes due to surface clutter. The limitation can be mitigated by oversampling (i.e., a wide beam width and a fine sampling span). With a narrow beam width and a fine beam sampling span, the surface clutter interference is confined to the surface level. When the precipitation system is located far from the nadir, the precipitation signal is obtained only for strong precipitation.

The objective of this work was the comparison of two different classification approaches to detect four different tree species of a highly diverse tropical Atlantic Forest area. In order to achieve the objective, images were acquired with the Rikola hyperspectral camera onboard the UX4 UAV. The study area is in the Western part of São Paulo State, a tropical Atlantic Forest area protected by governmental laws, which contains areas already deforested in the past and which are currently in regeneration. The tested approaches were one based only in the pixel values and other one based in regions. After the image acquisition, the images were radiometrically and geometrically processed. In addition, an airborne laser scanning point cloud was used to calculate the canopy height model of the area, which was used to detect the individual tree crowns with the superpixels method. Those superpixels were used to the region-based classification and to feature extraction. A total of 28 features were extracted where 25 correspond to the spectral bands acquired with the Rikola camera and three correspond to the three first principal components of the images. The features were extracted from the 91 samples recognized during a field work. From the total of samples, 19 were separated to validate the classification results. The chosen classifier was the Random Forests and the results presented a kappa coefficient of 18.20% and 36.57% for the pixel-based and region-based classifications showing that the second one had a better performance.

The objective of this study was to evaluate the impact of reducing the radiometric information of hyperspectral images. The image data was collected originally with 32 bits and rescaled to 8 and 16 bit/pixel. The images were acquired with a Rikola Hyperspectral Camera attached to an Unmanned Aerial Vehicle (UAV). After the geometric and radiometric processing of the images, a mosaic was obtained with pixels representing reflectance factor coded in 32 bits. Using the minimum and maximum values of each spectral band, a linear equation was thus applied to reduce the radiometric resolution of the original mosaic, from 32 bits to 8 bits and from 32 bits to 16 bits. Following, the Normalized Root Mean Square Error (NRMSE %) and the Mean Absolute Percentage Error (MAPE %) were used to evaluate the results, showing that for the 8 bits mosaic, the loss of information was higher. For this radiometric resolution rescaling, the MAPE % achieved values until 22.486 % and the highest NRMSE % value was 0.455 % while, for the 16 bits mosaics, the highest MAPE % and NRMSE % values were 0.069 % and 0.002 %, respectively. Finally, it can be inferred that the impact of radiometric transformation can be considered as negligible for the hyperspectral mosaic with 16 bits of radiometric resolution, which presented lower values of NRMSE % and MAE % and could not affect the mosaic analysis.

To successfully assimilate data from a new observing system, it is necessary to develop appropriate data selection strategies, assimilating only the generally useful data. This development work is usually done by trial and error using observing system experiments (OSEs), which are very time and resource consuming. This study proposes a new, efficient methodology to accelerate the development using ensemble forecast sensitivity to observations (EFSO). First, non-cycled assimilation of the new observation data is conducted to compute EFSO diagnostics for each observation within a large sample. Second, the average EFSO conditionally sampled in terms of various factors is computed. Third, potential data selection criteria are designed based on the non-cycled EFSO statistics, and tested in cycled OSEs to verify the actual assimilation impact. The usefulness of this method is demonstrated with the assimilation of satellite precipitation data. It is shown that the EFSO-based method can efficiently suggest data selection criteria that significantly improve the assimilation results.

High spatial resolution remote sensing images acquired by drones are highly relevant data source in many applications. However, strong variations of radiometric values are difficult to correct in hyperspectral images. Honkavaara et al. (2013) presented a radiometric block adjustment method in which hyperspectral images taken from remotely piloted aerial systems – RPAS were processed both geometrically and radiometrically to produce a georeferenced mosaic in which the standard Reflectance Factor for the nadir is represented. The plants crowns in permanent cultivation show complex variations since the density of shadows and the irradiance of the surface vary due to the geometry of illumination and the geometry of the arrangement of branches and leaves. An evaluation of the radiometric quality of the mosaic of an orange plantation produced using images captured by a hyperspectral imager based on a tunable Fabry-Pérot interferometer and applying the radiometric block adjustment method, was performed. A high-resolution UAV based hyperspectral survey was carried out in an orange-producing farm located in Santa Cruz do Rio Pardo, state of São Paulo, Brazil. A set of 25 narrow spectral bands with 2.5 cm of GSD images were acquired. Trend analysis was applied to the values of a sample of transects extracted from plants appearing in the mosaic. The results of these trend analysis on the pixels distributed along transects on orange tree crown showed the reflectance factor presented a slightly trend, but the coefficients of the polynomials are very small, so the quality of mosaic is good enough for many applications.

Tree species classification provides valuable information to forest monitoring and management. The high floristic variation of the tree species appears as a challenging issue in the tree species classification because the vegetation characteristics changes according to the season. To help to monitor this complex environment, the imaging spectroscopy has been largely applied since the development of miniaturized sensors attached to Unmanned Aerial Vehicles (UAV). Considering the seasonal changes in forests and the higher spectral and spatial resolution acquired with sensors attached to UAV, we present the use of time series of images to classify four tree species. The study area is an Atlantic Forest area located in the western part of São Paulo State. Images were acquired in August 2015 and August 2016, generating three data sets of images: only with the image spectra of 2015; only with the image spectra of 2016; with the layer stacking of images from 2015 and 2016. Four tree species were classified using Spectral angle mapper (SAM), Spectral information divergence (SID) and Random Forest (RF). The results showed that SAM and SID caused an overfitting of the data whereas RF showed better results and the use of the layer stacking improved the classification achieving a kappa coefficient of 18.26 %.

We developed a data assimilation system based on a particle filter approach with the spatially explicit individual-based dynamic global vegetation model (SEIB-DGVM). We first performed an idealized observing system simulation experiment to evaluate the impact of assimilating the leaf area index (LAI) data every 4 days, simulating the satellite-based LAI. Although we assimilated only LAI as a whole, the tree and grass LAIs were estimated separately with high accuracy. Uncertain model parameters and other state variables were also estimated accurately. Therefore, we extended the experiment to the real world using the real Moderate Resolution Imaging Spectroradiometer (MODIS) LAI data and obtained promising results.

A local particle filter (LPF) is introduced that outperforms traditional ensemble Kalman filters in highly nonlinear/non-Gaussian scenarios, both in accuracy and computational cost. The standard sampling importance resampling (SIR) particle filter is augmented with an observation-space localization approach, for which an independent analysis is computed locally at each grid point. The deterministic resampling approach of Kitagawa is adapted for application locally and combined with interpolation of the analysis weights to smooth the transition between neighboring points. Gaussian noise is applied with magnitude equal to the local analysis spread to prevent particle degeneracy while maintaining the estimate of the growing dynamical instabilities. The approach is validated against the local ensemble transform Kalman filter (LETKF) using the 40-variable Lorenz-96 (L96) model. The results show that (1) the accuracy of LPF surpasses LETKF as the forecast length increases (thus increasing the degree of nonlinearity), (2) the cost of LPF is significantly lower than LETKF as the ensemble size increases, and (3) LPF prevents filter divergence experienced by LETKF in cases with non-Gaussian observation error distributions.

In this paper, we propose the design and implementation of an ensemble data assimilation (DA) framework for weather prediction at a high resolution and with a large ensemble size. We consider the deployment of this framework on the data throughput of file input/output (I/O) and multi-node communication. As an instance of the application of the proposed framework, a local ensemble transform Kalman filter (LETKF) was used with a Non-hydrostatic Icosahedral Atmospheric Model (NICAM) for the DA system. Benchmark tests were performed using the K computer, a massive parallel supercomputer with distributed file systems. The results showed an improvement in total time required for the workflow as well as satisfactory scalability of up to 10 K nodes (80 K cores). With regard to high-performance computing systems, where data throughput performance increases at a slower rate than computational performance, our new framework for ensemble DA systems promises drastic reduction of total execution time.

The most widely used methods of data assimilation in large-scale oceanography, such as the Simple Ocean Data Assimilation (SODA) algorithm, specify the background error covariances and thus are unable to refine the weights in the assimilation as the circulation changes. In contrast, the more computationally expensive Ensemble Kalman Filters (EnKF) such as the Local Ensemble Transform Kalman Filter (LETKF) use an ensemble of model forecasts to predict changes in the background error covariances and thus should produce more accurate analyses. The EnKFs are based on the approximation that ensemble members reflect a Gaussian probability distribution that is transformed linearly during the forecast and analysis cycle. In the presence of nonlinearity, EnKFs can gain from replacing each analysis increment by a sequence of smaller increments obtained by recursively applying the forecast model and data assimilation procedure over a single analysis cycle. This has led to the development of the "running in place" (RIP) algorithm by Kalnay and Yang (2010) and Yang et al. (2012a,b) in which the weights computed at the end of each analysis cycle are used recursively to refine the ensemble at the beginning of the analysis cycle. To date, no studies have been carried out with RIP in a global domain with real observations. This paper provides a comparison of the aforementioned assimilation methods in a set of experiments spanning seven years (1997–2003) using identical forecast models, initial conditions, and observation data. While the emphasis is on understanding the similarities and differences between the assimilation methods, comparisons are also made to independent ocean station temperature, salinity, and velocity time series, as well as ocean transports, providing information about the absolute error of each. Comparisons to independent observations are similar for the assimilation methods but the observation-minus-background temperature differences are distinctly lower for LETKF and RIP. The results support the potential for LETKF to improve the quality of ocean analyses on the space and timescales of interest for seasonal prediction and for RIP to accelerate the spin up of the system.

We present sensitivity tests for a global aerosol assimilation system utilizing AERONET observations of AOT (aerosol optical thickness) and AAE (aerosol Ångström exponent). The assimilation system employs an ensemble Kalman filter which requires tuning of three numerical parameters: ensemble size nens, local patch size npatch and inflation factor ρ. In addition, experiments are performed to test the impact of various implementations of the system. For instance, we use a different prescription of the emission ensemble or a different combination of observations. The various experiments are compared against one-another and against independent AERONET and MODIS/Aqua observations. The assimilation leads to significant improvements in modelled AOT and AAE fields. Moreover remaining errors are mostly random while they are mostly systematic for an experiment without assimilation. In addition, these results do not depend much on our parameter or design choices. It appears that the value of the local patch size has by far the biggest impact on the assimilation, which has sufficiently converged for an ensemble size of nens=20. Assimilating AOT and AAE is clearly preferential to assimilating AOT at two different wavelengths. In contrast, initial conditions or a description of aerosol beyond two modes (coarse and fine) have only little effect. We also discuss the use of the ensemble spread as an error estimate of the analysed AOT and AAE fields. We show that a very common prescription of the emission ensemble (independent random modification in each grid cell) can have trouble generating sufficient spread in the forecast ensemble.

We present a global aerosol assimilation system based on an Ensemble Kalman filter, which we believe leads to a significant improvement in aerosol fields. The ensemble allows realistic, spatially and temporally variable model covariances (unlike other assimilation schemes). As the analyzed variables are mixing ratios (prognostic variables of the aerosol transport model), there is no need for the extra assumptions required by previous assimilation schemes analyzing aerosol optical thickness (AOT). We describe the implementation of this assimilation system and in particular the construction of the ensemble. This ensemble should represent our estimate of current model uncertainties. Consequently, we construct the ensemble around randomly modified emission scenarios. The system is tested with AERONET observations of AOT and Angström exponent (AE). Particular care is taken in prescribing the observational errors. The assimilated fields (AOT and AE) are validated through independent AERONET, SKYNET and MODIS Aqua observations. We show that, in general, assimilation of AOT observations leads to improved modelling of global AOT, while assimilation of AE only improves modelling when the AOT is high.

We have developed an advanced data assimilation system for a global aerosol model with a four-dimensional ensemble Kalman filter in which the Level 1B data from the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) were successfully assimilated for the first time, to the best of the authors' knowledge. A one-month data assimilation cycle experiment for dust, sulfate, and sea-salt aerosols was performed in May 2007. The results were validated via two independent observations: 1) the ground-based lidar network in East Asia, managed by the National Institute for Environmental Studies of Japan, and 2) weather reports of aeolian dust events in Japan. Detailed four-dimensional structures of aerosol outflows from source regions over oceans and continents for various particle types and sizes were well reproduced. The intensity of dust emission at each grid point was also corrected by this data assimilation system. These results are valuable for the comprehensive analysis of aerosol behavior as well as aerosol forecasting.