25d, and the second NOTAM-code-component 25e are identified from the E-field 30 in step S130. This identifying is done using the trained machine-learning algorithm 20. The machine-learning algorithm 20 comprises, for example, a single multi-output long short-term memory (LSTM) neural network NN and is trained using the method 100 as is described later (see also step S530 and step S540 in the description of FIG. 3 below). The permissible values are assigned a relative probability. The relative probability is a numerical value between zero (0) and one (1) and indicates an expected relevance of the identified permissible values for the traffic 25a, the purpose 25b, the scope 25c, the first NOTAM-code-component 25d, and the second NOTAM-code-component 25e. The relative probability can be normalized such that a sum of the relative probabilities for each of the traffic 25a, the purpose 25b, the scope 25c, the first NOTAM-code-component 25d, and the second NOTAM-code-component 25e equals one (1).

In step S140 a preferred value for one or more of the traffic 25a, the purpose 25b, the scope 25c, the first NOTAM-code-component 25d, and the second NOTAM-code-component 25e is selected. The selecting of the of the preferred value is done using a set of probabilities.

If, for example, the relative probability for the traffic 25a is identified, there are four (4) possible values for the traffic 25a (see above). In this example, the values are “I,” “IV,” “K,” or “V” and the relative probability for these possible values are, for example, [0.01, 0.2, 0.0, 0.79]. In this example, the value “V” has the highest relative probability (0.79) and is therefore selected as the preferred value for the traffic 25a.

The corrected values for the traffic 25a, the purpose 25b, the scope 25c, the first NOTAM-code-component 25d, and the second NOTAM-code-component 25e are set in step S150 to produce the corrected NOTAM-message 15c. The corrected NOTAM-messages 15c are transmitted to the users of the NOTAM-messages 15 directly. In some cases, the corrected NOTAM-messages 15c may further be reviewed by a second instance, such as a human agent or an alternative correction system, before transmission to the users. The corrected NOTAM-messages 15c may also be used for a training of the human agent by providing the corrected NOTAM-messages 15c alongside with the initial NOTAM-messages 15a and/or the validated NOTAM-messages 15b to the human agent.

The trained machine-learning algorithm 20 is, in one aspect, updated in step S160 using at least one of the corrected values for the traffic 25a, the purpose 25b, the scope 25c, the first NOTAM-code-component 25d, and the second NOTAM-code-component 25e and at least one of the input plurality of NOTAM-messages 15, the parsed input plurality of initial NOTAM-messages 15a, or the pre-processed input plurality of initial NOTAM-messages 15a. This updating of the trained machine-learning algorithm 20 further comprises validating the at least one of the corrected values for the traffic 25a, the purpose 25b, the scope 25c, the first NOTAM-code-component 25d, and the second NOTAM-code-component 25e. The updating is done, for example, periodically.

Examples of erroneous initial NOTAM-messages 15a and corrected NOTAM-messages 15c are shown below. The corrected NOTAM-messages 15c are produced using the method 10 as will be explained later.

Example 1

Initial
(A0002/21 NOTAMN

NOTAM-
Q) KZDV/QNDAS////000/999/3830N10754W005

message 15a
A) KMTJ

B) 2101021037

C) 2101042000 EST

E) /MTJ/DME U/S)

Corrected
(A0002/21 NOTAMN

NOTAM-
Q) KZDV/QNDAS/IV/BO/AE/000/999/3830N10754W025

message 15c
A) KMTJ B) 2101021037 C) 2101042000EST

E) /MTJ/DME U/S)

Explanation of
The initial NOTAM-message 15a is missing values for traffic 25a,

corrections
purpose 25b, and scope 25c. A radius is also adjusted in the corrected

version.

Example 2

Initial
(F0048/21 NOTAMR F0041/21

NOTAM-
Q) CZVR/QNDAS/IV/BO/AE/000/999/4905N12309W025

message 15a
A) CYVR B) 2101062217 C) 2101082359EST

E) VANCOUVER DME YVR 115.9 MHZ/CH106X

UNMONITORED)

Corrected
(F0048/21 NOTAMR F0041/21

NOTAM-
Q) CZVR/QNDXX/IV/BO/AE/000/999/4905N12309W025

message 15c
A) CYVR B) 2101062217 C) 2101082359EST

E) VANCOUVER DME YVR 115.9 MHZ/CH106X

UNMONITORED)

Explanation of
The second NOTAM-code-component 25 is not appropriate, because

corrections
NC45 code “AS” corresponds to “Unserviceable”, and the device here

is declared as “UNMONITORED”, so, in the corrected version, it gets

NC45 “XX” (for miscellaneous/plain language description).

Example 3

Initial
(A0020/21 NOTAMR A0014/21

NOTAM-
Q) AYPM/QWWXX/IV/NBO/W/000/999/

message 15a
A) AYPM B) 2101050100 C) 2101070600 EST

E) CTN. MANAM VOLCANO COORD S04 04 39 E145 02 21

EMITTING VA AND FU

UP TO 14000 FT MOVING WNW.PILOT REP.)

Corrected
(A0020/21 NOTAMR A0014/21

NOTAM-
Q) AYPM/QWWLW/IV/NBO/W/000/999/0444S15030E999

message 15c
A) AYPM B) 2101050100 C) 2101070600EST

E) CTN. MANAM VOLCANO COORD S04 04 39 E145 02 21

EMITTING VA AND FU

UP TO 14000 FT MOVING WNW.PILOT REP.

F) SFC G) UNL)

Explanation of
The second NOTAM-code-component 25e could be more specific and

corrections
is replaced with “LW”, which stands for “will take place” (referring to

NC23 “WW” = “Significant volcanic activity”). The coordinates &

radius field is missing altogether.

Example 4

Initial
(A0002/21 NOTAMN

NOTAM-
Q) KZTL/QXXXX////000/999/

message 15a
A) KZTL

B) 2101042118

C) 2301042118 EST

E) ROUTE ZTL.

V37 JOTTA, NC TO DOILY, VA MOCA 5900.)

Corrected
(A0002/21 NOTAMN

NOTAM-
Q) KZTL/QARXX/IV/NBO/E/000/999/3627N08048W027

message 15c
A) KZTL B) 2101042118 C) 2301042118EST

E) ROUTE ZTL.

V37 JOTTA, NC TO DOILY, VA MOCA 5900.)

Explanation of
The first NOTAM-code-component 25d could be more specific and is

corrections
replaced with “LW”, which stands for “ATS route”. The initial

NOTAM is missing values for traffic, purpose, and scope. The

coordinates & radius field is missing altogether.

Example 5

Initial
(Y0002/20 NOTAMR Y1364/19

NOTAM-
Q) ZYSH/QXXXX/IV/NBO/A/000/999/4714N12355E005

message 15a
A) ZYQQ B) 2001021223 C) 2004021223 EST

E) DUE TO OBST, VISUAL CIRCLING PROCEDURE NOT AVBL.)

Corrected
(Y0002/20 NOTAMR Y1364/19

NOTAM-
Q) ZYSH/QPKAU/V/NBO/A/000/999/4714N12355E005

message 15c
A) ZYQQ B) 2001021223 C) 2004021223EST

E) DUE TO OBST, VISUAL CIRCLING PROCEDURE NOT AVBL.)

Explanation of
The first NOTAM-code-component 25d could be more specific and is

corrections
replaced with “PK”, which stands for “VFR approach procedure. The

second NOTAM-code-component 25e could be more specific and is

replaced with “AU”, which stands for “Not available. Traffic value is

changed from “IV” (“IFR” and “VFR”) to “V” (just “VFR”).

Example 6

Initial
(G1065/20 NOTAMR G0654/20

NOTAM-
Q) OBBB/QRRXX/IV/BO/W/000/145/2607N05128E015

message 15a
A) OBBB B) 2010110806 C) 2101102359EST

E) RESTRICTED AREA OTR53, UPPER LIMIT RAISED TO

FL145.

REFER TO QATAR AIP ENR 5.1

F) SFC G) FL145)

Corrected
(G1065/20 NOTAMR G0654/20

NOTAM-
Q) OBBB/QRRCH/IV/BO/W/000/145/2607N05128E015

message 15c
A) OBBB B) 2010110806 C) 2101102359EST

E) RESTRICTED AREA OTR53, UPPER LIMIT RAISED TO

FL145.

REFER TO QATAR AIP ENR 5.1.

F) SFC G) FL145)

Explanation of
Information about a certain restricted area has changed (its vertical

corrections
limits), so it should have second NOTAM-code-component 25e “CH” =

“Changed”.

Example 7

Initial
(A1648/20 NOTAMN

NOTAM-
Q) OMAE/QWELW/IV/BO/W/000/080/2514N05401E030

message 15a
A) OMAE B) 2008140900 C) 2011141500 EST

D) DLY BTN 0900-1500

E) MIL ACT WILL TAKE PLACE WI LATERAL LIMITS OF

RESTRICTED AREA OMR 77

EXCLUDING DUBAI CTA.

F) SFC G) 8000FT AMSL)

Corrected
(A1648/20 NOTAMN

NOTAM-
Q) OMAE/QRRCA/IV/BO/W/000/080/2511N05405E042

message 15c
A) OMAE B) 2008140900 C) 2011141500EST

D) DAILY 0900-1500

E) MIL ACT WILL TAKE PLACE WI LATERAL LIMITS OF

RESTRICTED AREA

OMR77 EXCLUDING DUBAI CTA.

F) SFC G) 8000FT AMSL)

Explanation of
The first NOTAM-code-component 25d and the second NOTAM-code-

corrections
component 25e point to the wrong subject and operation (“Exercises” +

“Will take place”) and it should be “RR” = “Restricted area” + “CA” =

“Activated”.

FIG. 3 shows a flow-chart illustrating a computer-implemented method 50 for correction of the coordinates in the initial NOTAM-messages 15a. The method 50 is, for example, used in the correction of the initial NOTAM-messages 15a as described with reference to FIG. 1 (see step S140 above). In step S300 of the method 50 for correction of the coordinates, the plurality of the initial NOTAM-messages 15a are input.

The plurality of the initial NOTAM-messages 15a are parsed in step S310 (see also description of step S110 in FIG. 1). The E-field (30) of the parsed plurality of the initial NOTAM-messages 15a is pre-processed in step S320. The pre-processing of the E-field 30 in step S320 comprises:

- removing special characters, certain punctuation marks, stop-words, and numerical values (such as, apostrophes (“ ”), exclamation marks (!), question marks (?)) from the E-field 30;
- replacing abbreviations with standard nomenclature (such as, replacing the word “DANGER”/“DANGEROUS”/“DNG” and “AREA”/“ZONE” with ICAO standardized abbreviations such as “DAREA” with exception of, for example, not replacing “AS”, which is the ICAO abbreviation for “ALTOSTRATUS”, because the acronym also is a common English word, and thus ambiguous, but, for example, replacing abbreviations with standard nomenclature: replacing ICAO abbreviation “FLTCK” with “Flight check”); and
- selectively converting the text into lower case letters.

The pre-processed E-field 30 of the initial NOTAM-messages 15a is parsed in step S330 using the parser. The parsing is done, for example, using at least one of a pre-trained English chunking parser model from OpenNLP (http://opennlp.apache.org/) or a natural language processing toolkit such as a processing toolkit licensed under the Apache software license (for example, Version 2.0). This parser takes as input a sentence/string and returns a parse tree, according to a pre-defined phrase structure grammar of the English language. This parse tree is then used to extract phrases that are likely to be relevant for aviation-based items of data of the input plurality of initial NOTAM-messages 15a. This parse tree is therefore used to identify phrases that are likely to be relevant for identifying the coordinates of aeronautical entities mentioned in the E-field 30 of the initial NOTAM-messages 15a. These identified phrases are extracted in step S340.

The extracted phrases are filtered, and the aeronautical entities are identified in step S350. This identifying of the aeronautical entities is done by extracting natural identifiers out of the extracted phrases from step S340, and then performing a mixed spatial-natural identifier query in a trusted aeronautical data source, such as from European AIS Database from Eurocontrol (EAD). This identifying is done, for example, by performing a query by a natural key (designator/name and feature type) in an EAD data source.

In step S360 the coordinates are derived from the identified aeronautical entities using associated stored locations and are used for correction of the initial NOTAM-messages 15a. In step S370, the initial parsed NOTAM-message 15a is reconstructed using the derived coordinates and thereby the corrected NOTAM-message 15c is generated. This reconstructing comprises replacing initial values for the coordinates in the initial NOTAM-message 15a with the corresponding derived corrected coordinates. The reconstructing may further comprise additional modifications. For example, if the lower limit and the upper limit in the Q-field 25 are missing, these missing values for the lower limit and the upper limit are filled with values from the F-field and/or the G-field or set to default values such as, for example, 000/999. If, for example, the radius is missing in the Q-field 25 of the initial NOTAM-message 25, the radius is filled with default values based the first NOTAM-code-component 25d, the second NOTAM-code-component 25e, and the scope 25c derived from the Q-field 25.

FIG. 4 shows a flow-chart illustrating a computer-implemented method 100 for the training of the machine-learning algorithm 20. The plurality of NOTAM-messages 15 are input in step S500. The input plurality of NOTAM-messages 15 comprise initial NOTAM-messages 15a and corresponding validated NOTAM-messages 15b. The initial NOTAM-messages 15a and the corresponding validated NOTAM-messages 15b comprise at least the Q-Field 25 and an E-field 30. The Q-field 25 of the initial NOTAM-messages 15a and the corresponding validated NOTAM-messages 15b comprise items of data relating to at least one of the traffic 25a, the purpose 25b, the scope 25c, the first NOTAM-code-component 25d, and the second NOTAM-code-component 25e. The E-field 30 of the initial NOTAM-messages 15a and the corresponding validated NOTAM-messages 15b comprise a string describing the content of the NOTAM-messages 15.

The input plurality of the initial NOTAM-messages 15a and the corresponding validated NOTAM-messages 15b are parsed in step S510. The parsing comprises identifying at least the Q-field 25 and the E-field 30 (see also parsing in step S110 above).

The E-field 30 of the parsed plurality of the initial NOTAM-messages 15a are pre-processed in step S520. The pre-processing of the initial NOTAM-messages 15a comprises at least one of: