Search

To link to full-text access for this article, visit this link: http://dx.doi.org/10.1016/j.jeconom.2005.03.003 Byline: Atsushi Inoue (a), Lutz Kilian (b)(c) Abstract: It is standard in applied work to select forecasting models by ranking candidate models by their prediction mean squared error (PMSE) in simulated out-of-sample (SOOS) forecasts. Alternatively, forecast models may be selected using information criteria (IC). We compare the asymptotic and finite-sample properties of these methods in terms of their ability to mimimize the true out-of-sample PMSE, allowing for possible misspecification of the forecast models under consideration. We show that under suitable conditions the IC method will be consistent for the best approximating model among the candidate models. In contrast, under standard assumptions the SOOS method, whether based on recursive or rolling regressions, will select overparameterized models with positive probability, resulting in excessive finite-sample PMSEs. Author Affiliation: (a) Department of Agricultural and Resource Economics, North Carolina State University, Raleigh, NC 27695, USA (b) CEPR, London, UK (c) Department of Economics, University of Michigan, Ann Arbor, MI 48109, USA Article History: Accepted 21 March 2005

Determination of the scale of the interaction between precipitation and topography is important for the accurate interpolation of rainfall in mountainous areas and also provides insight into the physical processes involved. In this paper, trivariate thin-plate smoothing splines are used to investigate the scale of interaction between monthly precipitation and topography by interpolating monthly rainfall over three subregions of the Australian continent, incorporating different climatic conditions and rainfall types. The interpolations are based upon elevations derived from digital elevation models (DEMs) of various resolutions. All of the DEMs are local averages of version 2.0 of the 9-s-resolution DEM of Australia. The results suggest that the optimal scale of the interaction between precipitation and topography, as it pertains to the elevation-dependent interpolation of monthly precipitation in Australia, is between 5 and 10 km. This is in agreement with results of similar studies that addressed daily precipitation over Switzerland.

The transformation of a real, continuous variable into an event probability is reviewed from the Bayesian point of view, after which a Gaussian model is employed to derive an explicit expression for the probability. In turn, several scalar (one-dimensional) measures of performance quality and reliability diagrams are computed. It is shown that if the optimization of scalar measures is of concern, then prior probabilities must be treated carefully, whereas no special care is required for reliability diagrams. Specifically, since a scalar measure gauges only one component of performance quality - a multidimensional entity - it is possible to find the critical value of prior probability that optimizes that scalar measure; this value of 'prior probability' is often not equal to the 'true' value as estimated from group sample sizes. Optimum reliability, however, is obtained when prior probability is equal to the estimate based on group sample sizes. Exact results are presented for the critical value of 'prior probability' that optimize the fraction correct, the true skill statistic, and the reliability diagram, but the critical success index and the Heidke skill statistic are treated only graphically. Finally, an example based on surface air pressure data is employed to illustrate the results in regard to precipitation forecasting.

Precipitation estimates from radar systems are a crucial component of many hydrometeorological applications, from flash flood forecasting to regional water budget studies. For analyses on large spatial scales and long timescales, it is frequently necessary to use composite reflectivities from a network of radar systems. Such composite products are useful for regional or national studies, but introduce a set of difficulties not encountered when using single radars. For instance, each contributing radar has its own calibration and scanning characteristics, but radar identification may not be retained in the compositing procedure. As a result, range effects on signal return cannot be taken into account. This paper assesses the accuracy with which composite radar imagery can be used to estimate precipitation in the convective environment of Florida during the summer of 1991. Results using Z = 300[R.sup.1.4] (WSR-88D default Z-R relationship) are compared with those obtained using the probability matching method (PMM). Rainfall derived from the power law Z-R was found to be highly biased (+90%-110%) compared to rain gauge measurements for various temporal and spatial integrations. Application of a 36.5-dBZ reflectivity threshold (determined via the PMM) was found to improve the performance of the power law Z-R, reducing the biases substantially to 20%-33%. Correlations between precipitation estimates obtained with either Z-R relationship and mean gauge values are much higher for areal averages than for point locations. Precipitation estimates from the PMM are an improvement over those obtained using the power law in that biases and root-mean-square errors are much lower. The minimum timescale for application of the PMM with the composite radar dataset was found to be several days for area-average precipitation. The minimum spatial scale is harder to quantify, although it is concluded that it is less than 350 [km.sup.2]. Implications relevant to the WSR-88D system are discussed.

The National Weather Service intends to begin routinely issuing long-lead forecasts of 3-month mean U.S. temperature and precipitation by the beginning of 1995. The ability to produce useful forecasts for certain seasons and regions at projection times of up to 1 yr is attributed to advances in data observing and processing, computer capability, and physical understanding - particularly, for tropical ocean - atmosphere phenomena. Because much of the skill of the forecasts comes from anomalies of tropical SST related to ENSO, we highlight here long-lead forecasts of the tropical Pacific SST itself, which have higher skill than the U.S. forecasts that are made largely on their basis. The performance of five ENSO prediction systems is examined: Two are dynamical [the Cane-Zebiak simple coupled model of Lamont-Doherty Earth Observatory and the nonsimple coupled model of the National Centers for Environmental Prediction (NCEP)]; one is a hybrid coupled model (the Scripps Institution for Oceanography - Max Planck Institute for Meteorology system with a full ocean general circulation model and a statistical atmosphere); and two are statistical (canonical correlation analysis and constructed analogs, used at the Climate Prediction Center of NCEP). With increasing physical understanding, dynamically based forecasts have the potential to become more skillful than purely statistical ones. Currently, however, the two approaches deliver roughly equally skillful forecasts, and the simplest model performs about as well as the more comprehensive models. At a lead time of 6 months (defined here as the time between the end of the latest observed period and the beginning of the predictand period), the SST forecasts have an overall correlation skill in the 0.60s for 1982-93, which easily outperforms persistence and is regarded as useful. Skill for extra-tropical surface climate is this high only in limited regions for certain seasons. Both types of forecasts are not much better than local higher-order autoregressive controls. However, continual progress is being made in understanding relations among global oceanic and atmospheric climate-scale anomaly fields. It is important that more real-time forecasts be made before we rush to judgement. Performance in the real-time setting is the ultimate test of the utility of a long-lead forecast. The National Weather Service's plan to implement new operational long-lead seasonal forecast products demonstrates its effectiveness in identifying and transferring 'cutting edge' technologies from theory to applications. This could not have been accomplished without close ties with, and the active cooperation of, the academic and research communities.

An automatic tracking method that can monitor single cloud movements and complex dynamic systems has been developed to study African convective systems and precipitation parameters. Thetracking method, which involves Meteostat infrared images, follows cloud labeling and intersection principles in observing occurences of cloud separation and merging. Moreover, the method has proven its applicability in the fields of climatology, meteorology and hydrology. In addition, the manual tracking system exhibits simpler parameters than automatic tracking systems.

This paper describes the Tropical Rainfall Measuring Mission (TRMM) standard algorithm that estimates the vertical profiles of attenuation-corrected radar reflectivity factor and rainfall rate. In particular, this paper focuses on the critical steps in the algorithm. These steps are attenuation correction, selection of the default drop size distribution model including vertical variations, and correction for the nonuniform beam-filling effect. The attenuation correction is based on a hybrid of the Hitschfeld-Bordan method and a surface reference method. A new algorithm to obtain an optimum weighting function is described. The nonuniform beam-filling problem is analyzed as a two-dimensional problem. The default drop size distribution model is selected according to the criterion that the attenuation estimates derived from the model and the independent estimates from the surface reference with the nonuniform beam-filling correction are consistent for rain over ocean. It is found that the drop size distribution models that are consistent for convective rain over ocean are not consistent over land, indicating a change in the size distributions associated with convective rain over land and ocean, respectively.

Promising agreement over land and sea has been obtained between NEXRAD 3-GHz radar observations of precipitation rate and retrievals based on simultaneous passive observations at 50-191 GHz from the Advanced Microwave Sounding Unit (AMSU) on the NOAA-15 meteorological satellite. A neural network with three hidden nodes and one linear output node operated on 15 km resolution data at 183 [+ or -] 1 and 183 [+ or -] 7 GHz, plus the cosine of scan angle, to produce estimates that match well the morphology of NEXRAD hurricane and frontal precipitation data smoothed to 15-km resolution. A second neural network operated on the same three parameters used in the first network, but smoothed to 50-km resolution, plus spatially-filtered cold perturbations detected in three AMSU tropospheric temperature-sounding channels (channels 4-6), which also have 50-km resolution. Comparison with the same NEXRAD data smoothed to 50-km resolution yielded root mean square (rms) discrepancies for two frontal systems and two passes over Hurricane Georges of ~1.1 mm/h, and [+ or -]1.4 dB for those precipitation events over 4 mm/h. Only 8.9% of the total AMSU-derived rainfall was in areas where AMSU saw more than 1- mm/h and NEXRAD saw less than 1-mm/h, and only 6.2% of the total NEXRAD-derived rainfall was in areas where NEXRAD saw more than 1-mm/h and AMSU saw less than 1-mm/h. Index Terms--Measurement, microwave propagation, microwave radiometry, microwave spectroscopy, precipitation, remote sensing.

Areal rainfall estimation from ground sensors is essential as a direct input to various hydrometeorological models or as a validation of remote sensing estimates. More critical than the estimation itself is the assessment of the uncertainty associated with it. In tropical regions knowledge on this topic is especially scarce due to a lack of appropriate data. It is proposed here to assess standard estimation errors of the areal rainfall in the Sahel, a tropical region of notoriously unreliable rainfall, and to validate those errors using the data of the EPSAT-Niger experiment. A geostatistical framework is considered to compute theoretical variances of estimation errors for the event-cumulative rainfall, and rain gauge networks of decreasing density are used for the validation. As a result of this procedure, charts giving the standard estimation error as a function of the network density, the area, and the rainfall depth are proposed for the Safielian region. An extension is proposed for larger timescales (decade, month, and season). The seasonal error is estimated as a product of the error at the event scale by a reduction coefficient, which is a function of the number K of recorded events and the probability distribution function of the point storm rain depth. For a typical network of 10 stations regularly dispatched over a 1 [degree] x 1 [degree] square, the relative estimation error decreases from 14% for an average storm rain depth of 16 mm to 5% for an average August rainfall of 160 mm. For a density comparable to that of the operational rain gauge network of southern Niger and similar Sahelian regions, the standard errors are, respectively, 26% at the event scale and 10%-15% at the monthly scale, depending on the number of events recorded during the month. The areas considered here are 1 [degree] x 1 [degree] and smaller, which makes a comparison with results obtained in previous studies for other regions of the world difficult since the reference area most often used in these studies is either 2.5 [degrees] x 2.5 [degrees] or 5 [degrees] x 5 [degrees]. Further work is thus needed to extend the results presented here to larger spatial scales.