† Corresponding author. E-mail:
Project supported by the National Natural Science Foundation of China (Grant Nos. 41775027 and 41405025).
Constructing sophisticated refractivity models is one of the key problems for the RFC (refractivity from clutter) technology. If prior knowledge of the local refractivity environment is available, more accurate parameterized model can be constructed from the statistical information, which in turn can be used to improve the quality of the local refractivity retrievals. The validity of this proposal was demonstrated by range-dependent refractivity profile inversions using the adjoint parabolic equation method to the Wallops’98 experimental data.
Remote sensing of atmospheric refractivity from radar data is an effective complement for the traditional refractivity detection methods, such as radiosonde and rocketsonde.[1] The RFC technique contains three main aspects: (i) an accurate mapping between radar clutter data and refractivity, (ii) a reasonable cost function measuring the difference between the observed data and the replica field, and (iii) an efficient optimization method finding the optimal solution. Most of the previous publications focused on the third part and heuristic optimization algorithms, such as genetic algorithm,[2] support vector machine,[3] simulated annealing,[4] deep learning,[5] were used for the optimal searching.
In these inversions, a feasible measure for accuracy improvement is to reduce the nonlinearity of the problem as well as the number of the estimated parameters.[6–8] Although most commonly used refractivity parameterized models, such as log–linear evaporation duct model, segmented tri-linear model, and mixed duct model, can synoptically describe the real refraction environments,[9] these models are too simple to characterize some profiles in detail. Besides, the mathematical formulas of these models are nonlinear except for the log–linear evaporation duct model. Constructing more sophisticated refractivity models is one of the key points to improve the inversion accuracy.
This paper is an extension of the work completed by Zhao et al.,[8] where adjoint parabolic equation (PE) method has been proved to be efficient for range-dependent evaporation duct structures inversion in real-time. Here, the method is extended to general duct cases by using local a priori refractive information (supposed to be available) to formulate a more robust linear refractivity model, and the performance is validated by the Wallops’98 experimental data.
The remainder of this paper is organized as follows: Wallops’98 data, including radar data and refractivity data, are briefly introduced in Section
Detailed description about the Virginia Wallops’98 experiments and the data quality control can be found in Ref. [10] and Ref. [11]. A simple introduction about the data used in this paper is presented in this section.
The radar signals were obtained using a Space Range Radar (SPANDAR) with an operational frequency of 2.84 GHz, a beam width of 0.39°, an elevation angle of 0°, an antenna height of 30.78 m, a range bin of 600 m, and vertical polarization. The antenna pattern was approximated as a sin(x)/x pattern. Four polar plots of clutter maps (corresponding to the time of 17:40 UT, 18:10 UT, 18:32 UT, and 19:00 UT April 02, 1998) from surface-based ducting events are shown in Fig.
To mitigate the effects of point targets (including sea spikes), the radar data used in the inversions are median filtered across range (1.2 km, 3 samples) and azimuth (5°, 13 samples, i.e., the increment of 0.4°). The filtered clutter data along the 150° radial (pink lines in Fig.
The refractive environments were recorded using an instrumented helicopter and a Chessie boat provided by the Applied Physics Laboratory of Johns Hopkins University. The helicopter flew a saw-tooth up-and-down pattern in and out along the 150° radial (pink lines in Fig.
For a particular region of interest, if reliable a priori refractivity information is available (e.g., from historical measurements or outputs of numerical weather prediction models), more robust parameterized refractivity model can be constructed for this region by analyzing the principal characteristics of the provided prior information. The methods can be used for this purpose including empirical orthogonal function (EOF),[12] principal components analysis (PCA),[13] and Karhunen–Loeve (KL) transform,[14] etc.
Let there be K refractivity profiles m1(z), . . ., mK(z). It is understood that the height z is actually discrete with length N. Note the eigenvalues and eigenvectors of the covariance matrix of the data set as λ1, . . .,λN and ϕ1(z), . . . ϕN(z), where it is assumed that the eigenvalues and corresponding eigenvectors are arranged in the order λ1 > λ2 . . . > λN. Then, any profile in the data set can be expressed as
For the inversion purpose, the helicopter-based refractivity profiles were used as the basic prior information to construct the local linear empirical refractivity model. Make perturbations to each profile at every sampling height in Fig.
The refractivity fields in Fig.
Compared with the traditional parameterized refractivity models (segmented tri-linear model, mixed duct model, etc.), the primary advantage of the local empirical model is that it is completely linear, which reduces the nonlinearity of the adjoint RFC problem and simplifies the corresponding computations, although it just can be used for local inversion purposes.
The adjoint PE method for range-dependent refractivity profiles inversion was described in Ref. [8]. More detailed adjoint PE derivation can refer to Ref. [6]. How to make use of the a priori refractive information for the inversions is introduced in this Section.
The relationship between radar clutter power and refractivity parameters can be formulated as[8]
The mismatch between the predictions and the observations can be quantified by a least-squares form cost function. With the background (or a priori) information, the cost function can be modified as
For the problem in this paper, the number of the total estimated parameters is 25, i.e., NT = 4 times 6 profiles at the ranges of 0 km, 12 km, 24 km, 36 km, 48 km, and 60 km plus the nuisance parameter σ. The matrix
From the left column of Fig.
In the right column of Fig.
This paper addresses the problem of how to use the a priori refractive information to formulate a more robust linear refractivity model for real-time RFC inversions. The validity of this method was demonstrated by range-dependent refractivity profile inversions using Wallops’98 experimental data. Although the results show promise, the proposed method is just aimed at local refractivity inversion purpose. Only 320 available helicopter-based refractivity profiles were used as the basis to construct the prior information. Future work is required to evaluate the performance of the method with more real data.
[1] | |
[2] | |
[3] | |
[4] | |
[5] | |
[6] | |
[7] | |
[8] | |
[9] | |
[10] | |
[11] | |
[12] | |
[13] | |
[14] | |
[15] | |
[16] |