Condensation trails, or contrails, are line-shaped clouds that are produced by an aircraft engine exhaust. These contrails often impact climate significantly due to their potential warming effect. Identification of contrail formation through satellite images has been an ongoing research challenge. Traditional computer vision techniques struggle with varying imagery conditions, and supervised machine learning approaches often require a large amount of hand-labeled images. This study researches few-shot transfer learning and provides an innovative approach for contrail segmentation with a few labeled images. The methodology leverages backbone segmentation models, which are pre-trained on existing image datasets and finetuned using an augmented contrail-specific dataset. We also introduce a new loss function, SR Loss, which enhances contrail line detection by incorporating Hough transformation in model training. This transformation improves performance over generic image segmentation loss functions. The openly shared few-shot learning library, contrail-net, has demonstrated that few-shot learning can be effectively applied to contrail segmentation with the new loss function.

@en

Gathering accurate and reliable precipitation data is essential in developing countries, since it applied to a wide range of applications, from improving short term weather models to global climate change research. The most common way of acquiring rainfall measurements is the rain gauge. However, this traditional measurement equipment requires frequent visits from researchers, due to the clogging risks as well as the risk of being easily tampered with by unauthorized people. This makes it impractical and expensive to create a extensive network of these rain gauges, whilst the demand for the precipitating data remains high. A measurement equipment type which is more suitable for this remote and independent requirement for precipitation measurement in developing countries is GPS, since an aspect of GPS measurements is the possibility to use the equipment in relatively remote conditions, with little human interference necessary. In addition to this, due to the nature of GPS measurements, rain is expected to be an important component in GPS data, as a disruptive to the signal, a variable found in the Signal-to-Noise ratio (SNR). Considering the above, GPS seems like a good alternative to the traditional rain gauge for precipitationmeasurements. During this additional thesis, the question whether GPS can be used reliably for precipitation data, will be answered. During a measuring campaign in Uganda from September to November in 2018, GPS data was gathered near precipitation measurement locations from TWIGA’s School-2-School initiative. For the processing, graphs of the SNR and precipitation of the same days and locations were created and were visually inspected to see if a relationship or correlation was present. From these graphs this relationship between SNR and precipitation was not immediately clear. The nature of the SNR can be caused by a multitude of reasons and variables and unless the correlation is very strong between variables, this correlation will not be very clear from just a visual inspection of these SNR and precipitation graphs. To untangle these variables, correlationmatrices were used, where the variables can be looked at in pairs and instead of as quadruplets (or more). From the correlation matrix of the Ugandan data is is clear that correlation between precipitation and SNR is not significant, as it is similar to the correlation with a randomly generated variable. A slight correlation is present, however, with the GPS elevation angle. During processing, data from Cabauw (the Netherlands) was used as an extra data set to compensate for the small amount of rain in the Ugandan data. This processing followed the same procedure as was applied with the Ugandan data; visual inspection of the SNR and precipitation graphs and generating a correlationmatrix. In the Dutch data the relationship between the SNR and the precipitation was still small, however, larger than in the Ugandan data. Nonetheless, the correlation between the SNR and the elevation was very strong. In conclusion, from both the Ugandan and Dutch data, the correlation between precipitation and SNR is not strong enough, or in other words, using GPS data to approximate rainfall, is not achievable using the methods applied and resources used during this additional thesis.