No abstract

Using 19 yr of Chinese soil moisture data from 1981 to 1999, the authors evaluate soil moisture in three reanalysis outputs: the 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) ReAnalysis (ERA-40); the National Centers for Environmental Prediction-National Center for Atmospheric Research (NCEP-NCAR) reanalysis 1 (R-1); and the NCEP-Department of Energy (DOE) reanalysis 2 (R-2) over China. R-2 shows improved interannual variability and better seasonal patterns of soil moisture than R-1 as the result of the incorporation of observed precipitation, but not for all stations. ERA-40 produces a better mean value of soil moisture for most Chinese stations and good interannual variability. Limited observations in the spring indicate a spring soil moisture peak for most of the stations. ERA-40 generally reproduced this event, while R-1 or R-2 generally did not capture this feature, either because the soil was already saturated or the deep soil layer was too thick and damped such a response. ERA-40 and R-1 have a temporal time scale comparable to observations, but R-2 has a memory of nearly 5 months for the growing season, about twice the temporal scale of the observations. The cold season tends to prolong soil moisture memory by about 3 months for R-2 and 1 month for ERA-40. The unrealistic long temporal scale of R-2 can be attributed to the deep layer of the land surface model, which is too thick and dominates the soil moisture variability. R-1 has the same land surface scheme as R-2, but shows a temporal scale close to observations, which is actually because of soil moisture nudging to a fixed climatology. This new long time series of observed soil moisture will prove valuable for other studies of climate change, remote sensing, and model evaluation.

Recognizing objects from subcategories with very subtle differences remains a challenging task due to the large intra-class and small inter-class variation. Recent work tackles this problem in a weakly-supervised manner: object parts are first detected and the corresponding part-specific features are extracted for fine-grained classification. However, these methods typically treat the part-specific features of each image in isolation while neglecting their relationships between different images. In this paper, we propose Cross-X learning, a simple yet effective approach that exploits the relationships between different images and between different network layers for robust multi-scale feature learning. Our approach involves two novel components: (i) a cross-category cross-semantic regularizer that guides the extracted features to represent semantic parts and, (ii) a cross-layer regularizer that improves the robustness of multi-scale features by matching the prediction distribution across multiple layers. Our approach can be easily trained end-to-end and is scalable to large datasets like NABirds. We empirically analyze the contributions of different components of our approach and demonstrate its robustness, effectiveness and state-of-the-art performance on five benchmark datasets. Code is available at https: //github.com/cswluo/CrossX.

The purpose of this study is to explain the unplanned longitudinal dose modulations that appear in helical tomotherapy (HT) dose distributions in the presence of irregular patient breathing. This explanation is developed by the use of longitudinal (1D) simulations of mock and surrogate data and tested with a fully 4D HT delivered plan. The 1D simulations use a typical mock breathing function which allows for more flexibility to adjust various parameters. These simplified simulations are then made more realistic by using 100 surrogate waveforms all similarly scaled to produce longitudinal breathing displacements. The results include the observation that, with many waveforms used simultaneously, a voxel-by-voxel probability of a dose error from breathing is found to be proportional to the realistically random breathing amplitude relative to the beam width if the PTV is larger than the beam width and the breathing displacement amplitude. The 4D experimental test confirms that regular breathing will not result in these modulations because of the insensitivity to leaf motion for low frequency dynamics such as breathing. These modulations mostly result from a varying average of the breathing displacements along the beam edge gradients. Regular breathing has no displacement variation over many breathing cycles. Some low frequency interference is also possible in real situations. In the absence of more sophisticated motion management, methods that reduce the breathing amplitude or make the breathing very regular are indicated. However, for typical breathing patterns and magnitudes, motion management techniques may not be required with HT because typical breathing occurs mostly between fundamental HT treatment temporal and spatial scales. A movement beyond only discussing margins is encouraged for intensity modulated radiotherapy such that patient and machine motion interference will be minimized and beneficial averaging maximized. These results are found for homogeneous and longitudinal on-axis delivery for unplanned longitudinal dose modulations.