Accurate short term rain predictions are important for flood early warning systems, emergency services, energy management and other services that that make weather dependent decisions. Recently introduced machine learning models suffer from blurry and unrealistic predictions at longer lead times, causing poor performance on the rarer heavy rainfall events. The objective of this research was to explore how the loss function in a recurrent, convolutional neural network (TrajGRU, X. Shi et al. (2017)) can be modified, to get sharper and more realistic predictions, without worsening its performance. Six perceptive loss functions, which should better represent how an image is perceived, are thoroughly analysed to understand the reasons behind their functioning in the context of precipitation nowcasting. It was shown that these perceptive losses can lead to an improvement in sharpness for the first lead time, but that the blurriness for longer lead times arises due to the imbalance in the dataset and the uncertainty of the model with respect to the location of the rain. This thesis gets concluded with recommendations on how to deal with these two problems.

Oceans play a vital role in the regulation of Earth's intricate climate system. The majority of gas and energy exchanges occurring at the ocean-atmosphere interface are driven by small-scale, O(1 km), coupled processes. Despite its importance, too few observations have been able to capture the ocean-atmosphere coupling at scales smaller than O(10 km). This has led to poor parameterisation in climate models. In an effort to enhance our understanding of the ocean-atmosphere coupling, this study aims to test and improve upon two separate Marine Atmospheric Boundary Layer (MABL) characterisation methodologies (called algorithm 2A and algorithm 2B) put forward by Young et al., (2000). Algorithms 2A and 2B relate processed Synthetic Aperture Radar (SAR) image properties to a specific atmospheric state through use of surface-layer similarity theory and a combination of surface-layer and mixed-layer similarity theory respectively. Results of both methods indicate significant inherent limitations. Algorithm 2A's utility is curbed by uncertainty introduced during the estimation of Convective Boundary Layer depth Zi and spectral power-law extrapolation. Algorithm 2B suffers from uncertainty introduced by the dimensionless energy dissipation rate psi. As a result of these (and other) uncertainties, the estimated atmospheric instability can be off by a factor 2 or more. Further analyses suggest shortcomings in the applicability of the Geophysical Model Function (GMF). It is hypothesised that both employed GMFs underestimate the horizontal wind-field variance at scales relevant to turbulent convection, which subsequently manifests itself as (part of) the observed average 50% overestimation of absolute Obukhov length and subsequent 33% underestimation of the atmospheric instability. Additional research is required to support and quantify the GMF-induced underestimation hypothesis. Inspired by algorithm 2B, a third method (algorithm 2C) is developed which circumvents major uncertainties inherent to both algorithm 2A and 2B. However, due to limitations of its own, algorithm 2C is incapable of replacing algorithms 2A or 2B for a large range of atmospheric instabilities. If improved upon and successfully employed, spaceborne characterisation of the MABL could benefit climate studies by providing a wealth of continuous and global atmospheric-state measurements on scales previously unavailable.

The ongoing large scale adoption of wind power increases the associated risks related to the variability. An essential way to mitigate these risks for a utility company is to forecast production accurately. This study aims to create insight into the potential of deep learning models for both forecast quality and value on the ultra-short-term wind power forecasting (UST-WPF) horizon.

The status quo at Eneco, a Dutch utility company, for UST-WPF is a numerical weather prediction (NWP) based model with a rudimentary ultra-short-term (UST) correction with real-time power data. The methodology followed during this thesis was the development of four UST-WPF models for Princess Amalia Wind Farm (PAWP) with a 16 programme time unit (PTU) forecast horizon and a forecast frequency of 1 PTU. Both model 1 and model 2 only use real-time data and are based on a multilayer perceptron (MLP) and a long short-term memory (LSTM) architecture, respectively. The other two proposed models are a multivariate combination of these two respective models with the Eneco model.

The accuracy of the four proposed models was compared to two benchmark models: a Persistence and the Eneco model. Additionally, a novel framework was designed to evaluate the forecast value relative to the Eneco model on a variable forecast horizon. This study indicates that the proposed deep learning models can contribute both in quality and value up to 9 PTUs ahead.

Several factors play a role in the ongoing process of the acceptance and eventual universal use of LiDAR remote sensing technology in wind site assessment. Especially turbulence parameter 푇퐼 is still giving food for thought. Three important aspects of LiDAR measured 푇퐼 that are considered key are picked out and feature in this work, which attempts to contribute to the handling of LiDAR measured turbulence. Firstly, the accuracy in measuring several wind characteristics compared to met masts,which have been the standard in wind science and industry for the past centuries, is an issue. This thesis evaluates a LiDAR met mast comparison for a Dutch onshore wind farm site to support the handling of the biases between both sources, with a focus on 푇퐼. It finds LiDAR suitable for basic wind measurements, but it still struggles with measuring turbulence accurately. Patterns in biases are exposed and explained and propositions of an alteration of the internal LiDAR correction factor 퐶, as well as the use of a new turbulence parameter called transience 휏m, are made. The second factor addressed in this thesis is an indirect one and originates from the fact that turbulence, and so 푇퐼, plays an important role in the fatigue lifetime damage of a wind turbine. It is tried to answer the question what influence small biases in 푇퐼 have on the fatigue lifetime damage of the tower and blades of a common wind turbine by means of a simulation based sensitivity study. Increments with an order of magnitude found in the first part of this work form an input for this. It is found that the fatigue damage due to stresses from certain bending moments is higher than others and that the blades and tower are affected differently by turbulence. The third factor is related to the inherent flexibility of LiDAR devices. This forms a big advantage over met masts in general and also possibly paves the way for short-term measurement campaigns by LiDAR. It is proven that 푇퐼 shows intra-annual variability, which would therefore require extrapolation techniques. The Measure Correlate Predict (MCP) methodology is chosen to conduct a case study on the application of several MCP algorithms on 푇퐼 data, containing several experiments. It was found that the linear regression method (LR) is unsuitable for making long-term turbulence predictions, but that the variance ratio method (VR) shows promising results on the used performance metrics. It seems that it does not matter whether the short-term data length is 3 or 6 months and that the picked season does not harm the final result significantly.

The increasing penetration of weather-dependent energy sources brings additional challenges to the operation of the power system. Wind power forecasting is a valuable resource for these power operators: a tool that aids the decision-making process and facilitates risk management. On the other hand, the progress of machine learning and their success in different fields attracted research into wind power probabilistic forecasting. The objective of the research was to enhance wind power forecasting through data-driven or machine learning models, proposing a new framework as an alternative to the existing ones, validated in different climate regions using grid-like topology data (weather images). The project focused on the European Energy Markets 2020 Conference (EEM20) Forecasting Competition. The setting emulates the day-ahead electricity market and the location consisted of all four electricity price regions in Sweden, where three main sets of data were provided. Three main challenges are present in this competition, namely the large volume of data (the curse of dimensionality), the lack of information regarding wind turbine availability, and the growing evolution of wind installed capacity between data sets. The methodology to create a framework consisted of evaluating different machine learning approaches, feature engineering, and tuning the final model for every region. The final framework is called the Smooth Pinball Hybrid Neural Network (SPinHy-NN). The results showcased a suitable CNN architecture, providing more accurate forecasts compared to MLP and k-means clustering. The quantile cross-over problem has been evaluated through the crossing loss and the number of crossings metrics. The analysis shows that a margin parameter can correct this undesired behavior. The conclusions validate the SPinHy-NN as a generalized framework for different climates through the final score achieved post-competition, managing to capture the spatial patterns in most cases. Moreover, the framework can be adapted to reach a desirable trade-off between accuracy, sharpness, and consistency. Recommendations aim to extend this research by employing satellite imaging, further feature engineering, novel neural networks, and exploring spatiotemporal models to capture atmospheric dynamics.

In the coming decade, the Dutch government plans to broadly expand its wind farm real estate over the North Sea region. This is a significant undertaking because, from literature, it is understood that wind turbines can affect the local environment. These impacts include, but are not limited to, changes in variables such as wind speed and turbulent kinetic energy. To accurately capture these effects and improve power forecast accuracy, weather prediction models numerically represent the physical effects of a wind farm (i.e., a wind farm's behaviour as a momentum sink and a source of turbulent kinetic energy) by incorporating a wind farm parameterization in the model. This study looked at two simulations (with and without a wind farm parameterization) from HARMONIE-AROME, an operational weather model, and assessed the performance of the wind farm parameterization by using the simulation results for the Belgian wind farm zone in the North Sea, and comparing it with observational data from Sentinel-1 SAR, floating LiDARs, measured power as well as results from a large-eddy simulation. The data was also composited based on wind direction and atmospheric stability and analysed. Most importantly, it was found, after comprehensive analysis, that the wind farm parameterization caused a marked improvement in the results of the model especially in the region downwind of a wind farm where the effects are most severely experienced. Specifically, the simulation with the wind farm parameterization reduced the overall wind bias to -0.028 m/s from 0.602 m/s when compared with observational data for the selected region. The overall power bias was also found to be approximately 1.92\%. Several hypotheses based on existing literature were also tested, and along with the composite analysis, indicated that the simulation with the wind farm parameterization performed well in most stability regimes with the exception of stable conditions. It was recommended that this be further explored to determine the cause. Recommendations for expanding the study to look at wind farm effects on surface fluxes in offshore regions have also been made.