pueden generar "artifact", "ojos de buey", etc .. en su salidas.Bye
...En realidad, los modelos pueden cometer errores grandes; X, decodificado por el modelo como otra Y, puede llevar a inexactitudes bastante importantes. Un ejemplo son los llamados “boguscanes”, huracanes falsos originados por modelos demasiado ansiosos en generar un ciclón tropical a partir de una elemento inofensivo....
Computer models are a great help in predicting the weather, but just like human beings, they aren't perfect. Now that hurricane season is at hand, forecasters are watching the models closely to see if any tropical cyclones spin up. The models often provide several days' warning that a new cyclone will form. But forecasters have to sort out the real cyclones from the fake ones. It turns out that some computer models love to generate phony hurricanes. They're so common that forecasters now call them "boguscanes." I'm Dave Thurlow for the Weather Notebook.What's behind a boguscane? For one thing, the models aren't yet detailed enough to include realistic thunderstorms, and you need these to kick off an actual tropical storm. Sometimes all it takes is a little moisture for the model to generate a boguscane, even when the other ingredients for tropical action aren't there. It's also hard for a model to tell what's going on in the middle of the ocean. Showers and thunderstorms can block the view of a satellite, so a model may start out with a fuzzy impression of the real atmosphere. Right now there's no magic way to tell which hurricanes are bogus and which are real.Forecasters play it safe by looking at several models that cover the same time period. If a hurricane develops in each model, it's probably for real. But if it only shows up in one model, then forecasters recognize it as a figment of that computer model's bogus imagination.
Most of the bad-forecast cases that we have observed are related to problems associated with boguscanes. In addition to the boguscanes, a sudden spurious deepening and intensifying of the target TC seems to result in bad forecasts. In this sense the performance of current RDAPS in terms of TC prediction is not satisfactory. This does not mean that TC prediction by RDAPS is useless because when the model produces an extremely bad forecast, model failure like boguscane can be discerned by the forecaster on duty by examining the forecast synoptic field. Besides, the current bogus work certainly proves to be of help in improving the forecast when the original RDAPS produces normal forecasts.
...En cambio, spline ofrece un resultado más continuo, suave y adaptable (si se ajustan adecuadamente sus parámetros se suele llegar a una interpolación de mayor precisión), pero debe controlarse que no genere salidas de rango excesivas, causadas habitualmente por datos muy distintos en relación a su proximidad....
Abstract The use of spaceborne RADAR interferometry has considerably increased since the ERS-1 launch. In this paper we verify the atmospheric refraction hypothesis. We show that the presence of large height variations and a difference of refractive profiles between the two imagings create interferometric artifacts that have to be dealt with in the Digital Elevation Model generation process. Moreover, water content horizontal gradients, clear air turbulences and ionospheric phenomena also create local artifacts on interferograms. We conclude that one needs more than one interferogram to solve this problem. ConclusionWe have shown that the presence of large height variations and a difference of refractive profiles between the two imagings create interferometric artifacts that can reach one fringe. This phenomenon has to be dealt with when processing data to construct Digital Elevation Models. Moreover, water content horizontal gradients, clear air turbulences and ionospheric phenomena also create local artifacts on interferograms. It seems that these artifacts appear on most interferograms but that their orders of magnitude vary drastically according to the meteorological situation. In order to remove these artifacts on Digital Elevation Models, one needs more than one interferogram to process data.
Model Errors The accuracy of the a model depends on many factors, which can be roughly grouped as: * Time Step Errors: Since the equations represent the change over an "infinitesimal" time but the actual time step is finite, errors will occur with larger time steps creating larger errors. The actual time step used is, like much else in numerical modeling, a compromise between the ideal and the practical. * Initial Condition Errors: The model requires "Initial Conditions" to start from, needing "knowledge" of the full three-dimensional atmosphere i.e. all prognostic variables at every grid point! This is obtained by combining available observations - notably, but not limited to, observed soundings - with previous model predictions to produce an initial "analysis". This analysis must be done carefully while respecting the dynamic constraints of the equations, i.e. is not a matter of simple interpolation of observations, since a large error introduced between observing points would cause large and unrealistic changes immediately after startup as the equations tried to adjust to an unrealistic initial imbalance. Correct initial conditions can be important to model predictions, but are less important at later times and closer to the surface. Initial conditions are most important for predicting the progress of disturbances above the BL, such as frontal movement. * Lateral Boundary Condition Errors: A 20 km resolution model cannot cover the entire world, so such a "fine-mesh" model is "nested" inside a "coarse-mesh" model, having larger-spacing and larger domain, which provides "neighboring" cells along the lateral (outside) boundary of the fine-mesh model. The coarse-grid model typically covers the entire globe (though sometimes there will be multiple "nestings" to get to the global model). Errors in the lateral boundary conditions will, over time, reach further into the fine-mesh grid so lateral BC errors can be reduced by increasing the size of the fine-mesh domain - but of course that takes additional grid points and computational time and given finite constraints on each the choice of a lateral boundary is always a trade off. The model domain used by the RUC model can be viewed here. By comparison, the NAM grid is much larger, reaching from the North Pole to the Equator! * Surface Boundary Condition Errors: The model must know the type of surface the atmosphere is interacting with since its roughness, vegetation type, soil type and water content, etc. all affect model predictions. This is the one true boundary the atmosphere has and creates the "Boundary Layer" (BL) so of course the predicted BL is especially affected by surface boundary errors! A major difficulty is that "average" conditions for all of surface quantities must be known for each of the surface model grid cells while in reality these quantities vary over much smaller scales so trying to, first, know what the actual surface consists of and, secondly, create a meaningful average for each grid cell are both subject to much error. Additionally, certain variables such as vegetation and soil moisture vary with time. Further, trying to predict soil moisture in detail would require calculations as intensive as that for the atmosphere itself, so greatly simplified equations are used instead. * Inadequate Spatial Resolution Errors: The use of differential equations assumes/requires that a feature is "resolved" by the grid resolution. If the effect of a lake or mountain ridge or whatever upon the atmosphere is to be predicted, the model must adequately "know" about the existence of that lake or ridge or whatever. Realistic resolution requires a minimum of four grid points inside such a feature. Modelers keep trying, within the bounds of available computer power, to use finer and finer grid resolutions since with present grid spacings there is still much that is not being resolved, particularly if the forcing is controlled by topography. There are also many atmospheric features such as convergence-created upward motions which are smaller than can be resolved with present model grids and so are not well predicted. This error can be thought of as error that occurs because the PDE equations assume changes over an infinitesimally small distance whereas the model can only estimate changes over a finite distance - if in reality changes occurs over a smaller distance than the grid can effectively resolve, the finite-difference equations then cannnot accurately represent the actual differences existing in the atmosphere. * Model "Noise" Errors: One result of the limited resolution resulting from use of a finite-spacing grid is that model "noise" develops, particularly at the smallest resolveable scales. Physically this is because energy in the atmosphere tends to be generated at relatively large scales and then break down into successively smaller eddies. The "differential" PDEs can and do simulate this behavior but only when resolution of atmospheric eddies is adequate, which is not true when the eddy size becomes comparable to the grid spacing - so in the model energy breaks down until it reaches the smallest model scale (i.e. the smallest resolvable eddy size) and is then trapped there. As a result, model predictions often vary at the smallest model scale in a saw-tooth manner, e.g. a forecast variable will be too large at one cell, then too small at the neighboring cell, and then again too large at the next cell, etc. This is typically reduced by "numerical filtering", but too much filtering also throws away part of the true signal so a compromise is required (as is typical in numerical modeling!). Often one can see evidence of this model noise not being fully controlled when a "bullseye" pattern appears in a BLIPMAP, with the value at one grid point being much larger/smaller than its surrounding neighbors. Decreasing model grid spacing helps to reduce model noise, but it will always exist to some degree. * Model Topography Errors: Generally model conditions represent "average" conditions over the extent of its grid cell, but the effect of surface elevation on the model is somewhat different since the topography used by a model is typically smoothed to a coarser resolution than that of the model grid spacing and can differ significantly from the actual topography, particularly when the actual surface elevation changes abruptly. The reason for this degradation is the model noise problem discussed above: if the topography were to be fully resolved then much noise would be generated at the very smallest scale, aggravating the normal model noise build-up problem at that scale, so to avoid this the very smallest scales are filtered out of the topography. Note that this means that 8 model points are now required to resolve a ridge, so resolution of surface elevation influences requires a finer grid spacing than for many other atmospheric influences. Another terrain factor is that models often use an "envelope topography" to produce better velocity predictions - but that can result in worsened BL predictions, such as for BL top. The idea behind an "envelope topography" is that in reality velocities on either side of a mountain ridge, the Sierras being a good example, are separated by a relatively high ridge; but if one uses elevations averaged over 20 km (or larger!) cells then the ridge largely disappears, so flows at a level which are not interacting in reality will be interacting in the model. Therefore a weighting is employed which pushs model topography toward the higher elevations that actually exist over each grid cell rather than to a simple average. However, other parameters such as surface temperature and the BL driven by it do depend upon the average elevation over a grid cell, so use of an envelope topography makes those predictions less accurate! Sometimes a compromise solution is attempted - for example, the RUC model has two topographies, a "normal" (envelope) topography used for most calculations and a "minimum" topography used for surface temperature adjustments. Additional discussion of differences between the smoothed model topography and the real topography, with two illustrations, can be found on the Grid Orientation webpage. * Parameterization Errors: "Parameterization" refers to model terms which cannot be obtained from fundamental principles so instead are computed from approximated equations. For example, a model which has a 20 km resolution cannot resolve many small clouds, yet those clouds affect the atmosphere through effects such as release of heat aloft in condensation, reduction of solar radiation reaching the surface, etc. Since these effects are important but can't be predicted explicitly by model equations/resolution, they must be "parameterized". Parametrization is the "voodoo" part of numerical modeling since it tries to predict complex processes using necessarily over-simplified assumptions. Many cloud forecast terms fall into this category. [In one sense this might be described as an "inadequate resolution" error, but here the resolution increase required to obtain fundamnetally correct equations is so large that it cannot be achieved in the foreseeable future, if ever, so is a unique problem. An example familiar to soaring pilots is the formation of small puffy cumulus clouds - these start out is small wisps of visible vapor and even the cell resolution needed to resolve such wisps is well beyond anything presently possible but to predict this truly correctly one would have to resolve down to the cloud droplet scale! A non-cloud example which is theoretically more do-able but still impossible in practice is predicting the upward transfer of heat in the BL - in reality this occurs though eddies such as thermals and downdrafts which would need to be resolved, but that would take a grid spacing of less than 50 m so instead it is parameterized as a grid-average vertical transfer.] * End User Errors: I cannot resist adding this after noting that some users misuse the predictions by incorrectly applying the model-produced forecasts, apparently due to a lack of appropriate knowledge.
5 Respuestas
4226 Vistas