Where σ^sup 2^^sub f^ and σ^sup bop da 2^^sub a^ are the ten diversities of the prospect and examines, respectively, at each grid point, sigma grade, and calendar day
Empirical Correction of a Coupled Land-Atmosphere ModelABSTRACT
This paper inspects empirical recommendations for fixing the unfairness of a coupled land-atmosphere model and exams the theory which a prejudice correction could enhance the maneuver of such versions. The correction methodologies searched into encompass 1) meditation ways and means, 2) nudging based on long-term biases, and three) nudging based on disposition mistakes. The prior strategy involves foreseeing the disposition mistakes of prognostic variables based on short forecasts-say direct times of 24 h or less-and so therefore subtracting the climatological mean value of the disposition mistakes at any time step. By any and all evaluate, the very best correction method is found to be nudging based on disposition mistakes. This approach significantly eliminates biases within the long-term predicts of warmness and soil dampness, and maintains the discrepancy of the prospect meadow, unlike meditation ways and means. Disposition mistakes appraised from ten 1-day predicts yielded quite as valid corrections as disposition mistakes appraised from all hours in a couple of weeks, implying which the strategy is really small to undertake by new age benchmarks. Disappointingly, none of the ways and means searched into incessantly developed the occasional miscalculation discrepancy of the model, even though this finding can be model based primarily. Nonetheless, the empirical correction strategy is disputed to be deserving eventhough it improves merely the unfairness, since the strategy has merely marginal effects on the mathematical speed and depicts prospect miscalculation in the way of a disposition miscalculation which may be likened right to other clauses within the disposition equations, that in turn offers indications as about the source of the prospect miscalculation.
(ProQuest: ... connotes formulae omitted.)
1. Unveiling
As well as that about the beyond method, we give consideration to two others: 1) meditation and 2) nudging based on long-term biases. Within the meditation strategy, the state is rested toward a useful resource state for a price that's selected empirically. In nudging based on long-term biases, the state is nudged for a price which opposes the biases which improve beyond the time scale of interest. The family member merits over these methodologies are straightforward to appreciate. Meditation ways and means adaptively amend the state toward the climatology, but also humid deviations to the climatology. Nudging based on long-term biases opposes the biases which inevitably have the biggest amplitude, but the dynamical reaction to such nudging can be highly nonlinear and nonlocal, and so cannot really behave as prepared. Nudging based on disposition mistakes grabs the swiftest expanding mistakes within the first stages of development, before nonlinear and nonlocal effects become vital, but the disposition mistakes which contribute to long-term biases might not be detectable within the day after day variations of disposition mistakes. Nudging based on disposition mistakes has nil tunable parameters, despite the fact that the other two ways and means engage tunable parameters linked with meditation proportions.
Our inspection of empirical correction methodologies can be considered more decisive than former studies in numerous features. First, we exploit corrections about the coupled system - that's, to both the atmospheric and land items of the model. 2nd, our sample size is big; the correction coefficients are appraised from Ten years of each day predicts, and after that examined on 10 independent years. 3rd, one or more correction method is searched into. 4th, the affect of the empirical correction strategy on distinct time scales is tested, from hours to seasons.
Within the after part, we review the dynamical model and datasets use within this learn. The empirical correction methodologies are spoken about in depth in part 3, and our evaluates of prospect performance are spoken about in part 4. Our main results are spoken about in part 5. We sum up with a synopsis and dialog. In a imminent confederate paper (Zhao et al. 2008, manuscript submitted to J. Hydrometeor., hereafter ZDD), we report upon our results from applying the empirical correction technique to the soil and atmospheric components separately, that sheds light on dynamical interactions amidst the soil and ecosystem.
2. Dynamical model and informations
The dataset used to authenticate the atmospheric prospect 's the Countrywide Centres for Ecological PredictionNational Centre for Atmospheric Research (NCEPNCAR) reanalysis product (Kalnay et al. 1996). To evade mistakes within the atmospheric grounds arising from inside the interpolation of informations about the model grid, we extracted the reanalysis informations in spectral form about the same 28 sigma grades as were use within the info assimilation system. Moreover,, such as T62L28, with very similar topography.
The dataset used to authenticate the soil prospect 's the Universal Offline Terrain Dataset (Silver; Dirmeyer and Tan 2001), edition 2., and to offer informations at 6-hourly intervals. We give consideration to predicts merely in the course of the summer season of June- Aug (JJA). The empirical correction clauses were appraised in the course of the period 1982-91, whilst the empirically repaired model was ascertained exploiting informations from inside the independent period 1992-2001.
3. Guesstimate of the correction coefficients
To clarify the empirical correction methodologies, take into account the dynamical model
... (1)
., a proportion of alter) calculated from inside the dynamical model, namely an overall flow model. We try to find a compelling e in ways that the model
... (2)
has less miscalculation. The despondent indication will allow e to be described as the disposition miscalculation. The disposition miscalculation is appraised as the incline of the least squares row fit amidst prospect mistakes and direct time. Definitely the right makes use of the 6-, A dozen-, 18-, and 24-h prospect mistakes, and is carried out at each grid point and sigma grade individually. We discover which model (2) is capricious if ever the each day disposition miscalculation is used. Stable integrations may just be manufactured by smoothing the disposition mistakes subsequently trying the recipe
... (3)
where f^sub N^ 's the appraised disposition miscalculation on the Mh day, and N* is known as a adjusted incessant. This equation is an autoregressive model for e with a decorrelation lifetime of N* hours (where the "decorrelation time" 's the quantity of the autocorrelations total direct times). Results for N* = 1 and 2 were indistinguishable from each other, whilst those for N* = 5 were methodically worse (in a mean square miscalculation sensation); thereby, results are presented exclusively for N* = 2.
Preliminary conditions manufactured by the beyond procedure may just be interpreted as model alleges nudged toward the immediate diagnostic. So it is suitable to call a preliminary sistuation formulated this way an assimilation. This assimilation is reminiscent of the incremental diagnostic upgrade strategy for Schubert et al. (1993). In ZDD, we talk about researches within which merely subsets of variables were assimilated, as tabulated in Table 1.
Because the reanalysis ain't "truth," the appraised disposition miscalculation isn't the true disposition miscalculation, but fairly the disposition discrepancy amidst the reanalysis and the model. Thus, any prejudice within the reanalysis are going to contaminate the appraised e term. But still, we'll imply that the empirical correction strategy drastically diminishes biases over direct times from that the disposition mistakes were appraised, proposing which the actual disposition mistakes are caught by the guesstimate procedure.
We give consideration to quite a few easy parameterizations of the disposition mistakes. The initial, that identifies nudging based on disposition mistakes, 's the state-independent brand of the form
ε = «εN», (4)
where the brackets stand for a period average. Two models of time averaging are thought out: http://oh-nikkireed.com/ the each day average of each one calendar day as time passes 1982-91, showed by the term "each day" in Table 1, and the four week period average of each one thirty day period as time passes 1982-91, showed by the term "four week period" in Table 1. These runs are described as "nudging, disposition" in Table 1.
The instant parameterization, that identifies meditation, is of the form
ε = (x^sub c^ - x)/τ^sub R^, bop nam (5)
where τ^sub R^ is known as a meditation time scale and x^sub c^ is known as a "climatological" meadow taken to be the four week period average meadow for the 1982-91 period. A worth of τ^sub R^ = 5 hours was selected since this is comparable about the time scale of the mistake maturity of atmospheric variables. The resulting runs are described as "loosen up to climatology" in Table 1.
The 3rd parameterization, that identifies nudging based on long-term biases, is given by
ε = -b^sub M^/τ^sub R^, (6)
where b^sub M^ 's the four week period mean uncorrected prospect minus the four week period mean reanalysis. The accompanied runs are described as "nudging, longer term" in Table 1.
Lastly, we think about a state-dependent correction brand of the form
ε = ax + b. (7)
where a and b are selected to attenuate [left angle bracket](^egr;^sub N^ - ax - b)^sup 2^[right angle bracket]) at each grid point individually. The resulting model is signaled "linear, circulation based primarily" in Table 1.
We tried to contain a correction which counted on the day time, by fitting prospect mistakes at 6, A dozen, 18, and 24 h about the sine and cosine functions with 24-h stages. Sadly, the resulting empirically repaired model functioned worse than did versions with corrections which were independent of the day time. This poor performance is probable as a result of the figure which 6 h is too short to solve the diurnal cycle (it's only one harmonic broke up from inside the Nyquist frequency). These runs won't be spoken about within the remainder of the paper.
4. Maneuver evaluates
To evade overfitting, correction clauses were appraised from predicts all through 1982-91, and maneuver was appraised from predicts all through 1992-2001. Maneuver would be analyzed by the mean square miscalculation. Time mean square miscalculation may just be decomposed into prejudice and occasional portions as
[left angle bracket](x^sub f^-x^sub a^)^sup 2^[right angle bracket] = b^sup 2^ + [left angle bracket]r^sup 2^[right angle bracket], (8)
where x^sub f^ and x^sub a^ are the prospect and diagnostic grounds, respectively; the brackets [left angle bracket][right angle bracket] denote a period average whilst holding the direct time adjusted; and
b = [left angle bracket]x^sub f^- x^sub a^[right angle bracket], r = x^sub f^ - x^sub a^ - [left angle bracket]x^sub f^ - x^sub a^[right angle bracket], (9)
are the unfairness and occasional components, respectively. The handiness of this decomposition is that in case the model is linear, so therefore adding state-independent correction clauses influences merely the unfairness, that in turn may just be excluded afterwards the real thing. Because all predicts within the 1992-2001 testing period were initialized on 1 June, the time mean is efficiently a ten calendar day mean.
The root-mean-square miscalculation (RMSE) is labeled as
RMSE = [[left angle bracket](x^sub f^ -x^sub a^)^sup 2^[right angle bracket]]1/2, (10)
where brackets [] denote a region average. If ever the prospect has nil maneuver, so therefore the prospect and diagnostic are statistically independent and the mistake discrepancy equates to the quantity of the person diversities. Thus, a maneuver score may just be regained by normalizing the mistake discrepancy by the quantity of the diversities as
... (11)
. We also characterize a normalized occasional squared miscalculation as
... (A dozen)
The maneuver of local weather prophecies is analyzed by applying the beyond metrics to immediate alleges x^sub f^ and x^sub a^. The maneuver over local weather forcasting may just be analyzed by translating the alleges x^sub f^ and x^sub a^ as four week period or in season mean alleges, with the brackets [left angle bracket][right angle bracket] defining 10-yr averages over these alleges. We'll exploit these metrics separately about the over all meadow and anomaly meadow, labeled as the exact amount meadow minus the ten average meadow.
, adequate to approximately 130 hPa, and on vertical cross parts of zonal mean amounts. Results for land variables are indicated on the instant soil grade, that has depths ranging from 4 to fourteen cm, relying on the plant life. Results for other soil layers simulate those for the instant soil grade, with the exception of an overall lessen by mistake amplitude with depth.
5. Results
a. Comparability of alternative empirical correction necessary arrangements
We have now show the performance of numerous empirical correction necessary arrangements.., RMSEs) of chosen necessary arrangements are represented in Fig. 2. Most often, results for U0 and V0 are similar and thus merely V0 is represented. First, realize that the time scale of the mistake maturity is approximately 5 hours for V0 and more time for other variables, proposing which the time scale τ^sub R^ use within meditation and nudging necessary arrangements must be nil smaller than 5 hours.
2nd, realize that the mistakes for diverse runs are similar, with the exception of warmness mistakes within the rush adequate to nudging based on long-term biases (the "v rush" showed by grayish triangles). Within the latter rush, the correction scheme increases the miscalculation within the atmospheric warmness within the 2nd week of the integration, in comparison to other necessary arrangements. Moreover, the rush turns into numerically capricious afterwards 15 July. (The "kink" at 1 July is as a result of the sudden augment within the four week period prejudice throughout the month border.) The performance of this scheme may very well be developed by tuning the time scale τ^sub R^, but such tuning seemingly wouldn't enhance the mistakes at the 2 time scale and at more time time scales all at once, on account of their contradictory relationships about the miscalculation of the uncorrected model. Similar results are located at other model grades. These results imply that simply compelling a model with the rescaled and sign-reversed four week period biases doesn't enhance the model predicts in the least time scales. Accordingly, we give nil further consideration to this scheme.
3rd, realize that mistakes for the linear,.,., the p rush). In line with this,, implying a somewhat petite linear connection amidst prospect miscalculation and regional state. Similar results were discovered at other model grades. Because linear, flow-dependent correction necessary arrangements engage more parameters without clean gains in prospect maneuver, we sum up that they're not deserving, at the minimum for this model, and we give consideration to them nil further.
4th,.,., the ? rush). In comparison, meditation has a tendency to degrade the mistakes in land variables, particularly within the first week. This early degradation is unquestionably as a result of the figure which the soil variables evolve on slower time scales, implying which the 5-day meditation time is too short for land variables. We suspect which a lengthier meditation time, declare 10-15 hours, can enhance the mistakes of the soil variables, but can degrade mistakes within the wind variables. Purportedly, applying distinct meditation proportions to distinct variables can enhance the all in all prospect, but this entails further trials and tuning.
The decomposition of the mistake into over all, prejudice, and occasional components is represented in Fig. 3. Nudging based on disposition mistakes incessantly diminishes the unfairness more than do meditation ways and means (the miscroscopic exclusions to this govern within the wind variables aren't known to be statistically elemental). In comparison, meditation incessantly diminishes the occasional miscalculation more than nudging based on disposition mistakes. The cause of this discrepancy would be spoken about pretty soon.
b. Nudging based on disposition mistakes
The worldwide mean square mistakes of versions with four week period modernized nudging clauses were virtually indistinguishable from those based on each day modernized nudging clauses (p vs . m runs). This consequence, plus the marginal change for the better as a result of linear, flow-dependent corrections, signifies that the disposition miscalculation is dominated by a virtually incessant term, at the minimum within the variables ST, SW, and TO. Funnily, empirically fixing all variables except the wind variables UO and VO handed virtually very similar mean square mistakes to runs within which all variables were nudged, proposing Find Out More which empirical correction of winds has comparatively minor benefit.
The spatial structure of the correction clauses for atmospheric warmness (top panels in Fig. 7) differs from which of the long-term biases (top left panel in Fig. 6). Thus, the compelling for taking away biases in atmospheric warmness isn't a mirror photo of the long-term biases, originating from the variation amidst short- and longterm biases.
c. Maneuver of in season mean predicts
At present take into account the prospect maneuver of anomalies to the 1992-2001 mean, represented in Fig. 9 and analyzed by NRSE in (A dozen). First realize that the NRSE valuations are much smaller than those without subtraction of the cli- matology. This consequence signifies that prejudice correction as a result of empirical correction ain't as valid as after-the-fact correction. We also applied a cross-validated after-the- figure correction, within which the climatology being sandwich- tracted is appraised from all years except the prospect yr, but this diagnostic simply raised the mistake ampli- tudes without converting the order of the mistakes. 2nd,., ? runs) improves NRSE 24hr in soil tem- perature but seldom in atmospheric warmness or soil humidity, despite the fact that meditation improves NRSE 24hr in soil humidity and seldom in atmospheric warmness. Blended results are also regained if pattern relationship is used to evaluate maneuver.
Atlases of the relationship amidst the in season mean prospect and confirmation at each grid cellular (not represented) are really similar for repaired and uncorrected predicts, implying which the empirical correction has just a little effect on the regional relationship maneuver of in season mean predicts.
The beyond results imply that the searched into correction ways and means don't could result in homogeneous improvements in in season anomaly predicts family member to easy postprocessing ways and means, at the minimum for this model and with our correction method. Moreover, empirical correction doesn't evade the requirement for postprocessing ways and means, because further improvements in maneuver may just be made by subtracting the climatological mean prejudice from inside the empirically repaired prospect.
d. Sensitivity to sample size and preliminary sistuation
6. Synopsis and dialog
This paper searched into empirical correction methodologies based on adding compelling clauses to an atmospheric GCM, and examined the theory which a prejudice correction could develop prospect maneuver. We discovered that the very best empirical correction scheme was nudging based on disposition mistakes. This approach involves foreseeing the disposition mistakes of prognostic variables based on short predicts - declare direct times less than 24 h - and after that subtracting these disposition mistakes at any time step. We targeted mostly on state-independent corrections within which the correction term equates to (minus) the climatological mean disposition miscalculation. This approach drastically reduced biases in long-term predicts of warmness and soil dampness, and shielded the discrepancy of the prospect meadow, unlike meditation ways and means. Nudging based on long-term biases degraded the maneuver on 2-week time scales and yielded mathematical instabilities. Linear, regional, flow-dependent correction clauses produced nil detectable change for the better in a mean square miscalculation sensation compared against nudging based on disposition mistakes. Nudging based on disposition mistakes was quite as valid if clauses were modernized four week period fairly than each day, as well as if corrections for the momentum equations were omitted. These results stand for that the majority of prejudice rises from mistakes within the model thermodynamics.
The prospect maneuver of empirically repaired versions was searched into in depth. Within the first 5 hours, minor or nil change for the better in occasional miscalculation discrepancy was tracked down. Over Two weeks, the occasional miscalculation saturates and the advance within the mean square miscalculation rises mostly from prejudice correction. If ever the climatological a technique of the predicts aren't deducted, so therefore nudging based on disposition mistakes explicitly boosts the mean square mistakes of the four week period and in season implies. If ever the climatological a technique of the predicts are deducted, so which merely anomalies are likened, so therefore empirical correction ways and means don't incessantly enhance the prospect maneuver. Furthermore, change for the better as a result of subtracting the mean prejudice is superior to which as a result of empirical correction solitary. Results for universal averages and in season implies were represented, but exam of regional geographic averages and four week period implies produced the equivalent final thoughts. Also, the repaired model didn't always enhance the maneuver on a point-by-point basis. These results direct us to summarize which, for our especial model and correction methodologies, the principal benefit of empirical correction ways and means is to minimize biases; they don't incessantly develop prospect maneuver of long-term averages.
Famously, the mean disposition miscalculation calculated from ten 1-day predicts produced almost the equivalent elimination in mean square miscalculation as disposition mistakes appraised from 10 year of each day informations in every month. But still, the disposition mistakes appraised from preliminary conditions drawn straight up from inside the reanalysis were about one factor of five less than those from inside the nudged preliminary conditions, even though they had similar structures. This amplitude variance is in line with the belief that the nudged preliminary conditions drift off of the reanalysis, hence raising the extent of the plain disposition mistakes. It is certainly ironic which less accurate preliminary conditions could result in better correction coefficients. Further researches with correction coefficients which were either zoomed or damped yielded worse predicts, proposing which our guesstimate method is optimal.
Our main conclusion - which nudging based on disposition mistakes fails to further improve occasional miscalculation discrepancy concurs with some studies but contradicts others. Sadly, distinct studies use distinct versions, informations, and strategies, so comparability ain't straightforward.
Nudging based on disposition mistakes doesn't straight up attribute step-by-step mistakes to precise physiological proceedings. Nonetheless, Klinker and Sardeshmukh (1992) used parallels amidst disposition mistakes and the inclinations from physiological parameterizations to infer which a chief source of miscalculation during their model arose from inside the specification of orographically induced gravity wave drag. Equally, our strategy may very well be useful in style development thus it depicts the mistake in the way of a disposition which may be likened right to other clauses within the disposition equation.
[Useful resource]
REFERENCES
Achatz, U., and G. Branstator, 1999: A two-layer model with empirical linear corrections and reduced order for studies of internal weather variability. J. Atmos. Sci., 56, 3140-3160.
Danforth, C M., E. Kalnay, and T. Miyoshi, 2007: Foreseeing and fixing universal local weather model miscalculation. Friday. Wea. Rev., 135, 281-299.
bop nam DelSole, T., and A. Y. Hou, 1999: Empirical correction of a dynamical model. Section I: Imperative issues. Friday. Wea. Rev., 127, 2533-2545.
Dirmeyer, P. A., and L. Tan, 2001: A multi-decadal universal landsurface informations set of state variables and fluxes. Diet coke Tech. Rep. 102, 43 pp. [Completely ready from inside the Centre for OceanLand-Atmosphere Studies, 4041 Powder Mill Rd., Suite 302, Calverton, MD 20705.]
Faller, A. J., and D. K. Shelter, 1975: Mathematical corrections to mathematical forcasting equations. Friday. Wea. Rev., 103, 845-855.
_____, and C. E. Schemm, 1977: Mathematical corrections to mathematical forcasting equations. II. Friday. Wea. Rev., One hundred and five, 37-56.
Johansson, A., and S. Saha, 1989: Simulation of step-by-step miscalculation effects and their elimination in a very simple brand of the ambience. Friday. Wea. Rev., 117, 1658-1675.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc, 77, 437-471.
Kirtman, B. P., J. Shukla, B. Huang, Z. Zhu, and E. K. Schneider, 1997: Multiseasonal prophecies with a coupled tropical ocean-global ecosystem system. Friday. Wea. Rev., 125, 789-808.
Klinker, E., and P. D. Sardeshmukh, 1992: The prognosis of mechanized dissipation within the ecosystem from large-scale balance requisites. J. Atmos. Sci., 49, 608-626.
Leith, C. E., 1978: Goal techniques for local weather forcasting. Annu. Rev. Fluid Mech., 10, 107-128.
Manabe, S., R. J. Stouffer, M. J. Spelman, and K. Bryan, 1991: Transient answers of a coupled ocean-atmosphere model to uninterrupted alters of atmospheric Carbon dioxide. Section I: Yearly mean reaction. J. Weather, 4, 785-818.
Misra, V., and Coauthors, 2007: Validating and empathetic the ENSO simulation in two coupled weather versions. Tellus, 59A, 292-308,.
Phillips, T. J., and P. J. Glecker, 2006: Appraisal of continental precipitation in Twentieth century weather simulations: The utility of multimodel statistics. Essential fluids Resour. Res., 42, W03202,.
Saha, S., 1992: Reaction of the NMC MRF model to systematicerror correction within integration. Friday. Wea. Rev., 120, 345-360.
Schubert, S. D., R. B. Rood, and J. Pfaendtner, 1993: An assimilated dataset for planet science applications. Bull. Amer. Meteor. Soc, 74, 2331-2342.
Yang, X., T. DelSole,. Pan, 2008: Empirical correction of the NCEP Universal Prospect System. Friday. Wea. Rev., in squeeze.
Yang,., and J. L. Anderson, 2000: Correction of step-by-step mistakes in coupled GCM predicts. J. Weather, 13, 2072-2085.
[Author Network]
vi da nam TIMOTHY DELSOLE
George Mason College, Fairfax, Virginia, and Centre for Ocean-Land-Atmosphere Studies, Calverton, Maryland
MEI ZHAO AND PAUL A. DIRMEYER
Centre for Ocean-Land-Atmosphere Studies, Calverton, Maryland
BEN P. KIRTMAN
Rosenstiel School of Underwater and Atmospheric Sciences, College of Miami, Miami, Florida, and Centre for
Ocean-Land-Atmosphere Studies, Calverton, Maryland
(Manuscript earned 15 Aug 2007, in final form 11 April 2008)
[Author Network]
Corresponding author address: Timothy DelSole, Centre for Ocean-Land-Atmosphere Studies, 4041 Powder Mill Rd., Suite 302, Calverton, MD 20705-3106.
