Expectations about weather forecasting services have changed drastically in the last decade. Location-based weather information has become one of the basic functions of a smartphone these days. But what datas feeds the meteorological service providers using the results of the measurements? It is well known that data are extracted by surface measurements and atmospheric measurements using meteorological balloons. However, perhaps less is known that in recent decades all airplanes have been required to automatically transmit meteorologically relevant measurement datas in a standardized data format to ground centers, thus increasing the availability of measurement models, the number of monitoring information.
The complex physical interactions of the atmosphere cannot be described in a deterministic way, so the model used for weather forecasting uses approximate methods when making predictions from various measurements.
Error correction has always had to be taken into account in the analysis of models, but in recent years ECMWF (European Center for Medium-Range Weather Forecasts) researchers have also identified an unpleasant recurring data error that strongly influences computational models, suggesting that aircraft back, however, due to intensive international air traffic, they were not able to identify the type of aircraft sending the erroneous data until 2020.
Due to the reduced airplane takeoff caused by the COVID-19 pandemic, last year the ECMWF identified that the fault was caused by Boeing Dreamliners, the onboard software of B787 airplane, which transmits one of the atmospheric measurements, the wind direction data, with an erroneous signal for automatic meteorological purposes in data packets.
Because these data arrive anonymously for meteorological processing, it was extremely difficult to map the true characteristics and origin of the error and to delineate the incoming data. According to the latest ECMWF seminar, the manufacturer's software repair did not solve the problem, thus ensuring the avoidance of erroneous calculations by post-correction and ignoring data from uncertain sources.
The above story illustrates that in our accelerating world, the distance between the cause of an error in the operation of an increasingly complex and interdependent IT system and the point at which the error occurs can be system-wide. The greater the distance between interconnected IT systems, the greater the resources required to detect, analyse and ultimately correct the fault.
The increasing levels and complexity of integration poses new challenges for testing specialists. More and more software relies on the services of other systems through various interfaces, just think of the already mentioned meteorological app, which can be found on most people’s smartphones.
The realization of interoperability, ie the interoperability of different IT systems, has become a key aspect in software development. This means that the functionality of a given software must be verified at multiple levels of integrity, i.e., integration tests must be included in the testing process according to the level of all cooperating systems. In the case of cooperating systems, experience shows that functions located at the boundaries of the systems tend to be less the focus of attention, so the average frequency of errors is typically higher here.
An important lesson from the above case is that during integration testing, the test conditions must be checked from the module level to the full integrity state in all separable phases, including all dependent data elements that may affect the expected operation of the test unit but come from outside the unit.
However, it is also conceivable that for meteorological data sets, Boeing's testing procedure was comprehensive and thorough at all levels and “only” one of the test environments or stub elements required to perform the tests was not properly designed and this led to false verification. .
In any case, it is surprising that due to a single sign (!) error, all meteorological service providers in the world get a reduced amount of information, and that such a seemingly trivial software bug could remain in latency for nearly a decade (the first Dreamliner was launched in 2011).
