The Pannonian Basin in southeastern Europe is heavily used for rain-fed agriculture. The region experienced several droughts in the last years, causing major yield losses. Ongoing climate change, characterised by increasing temperatures and potential evapotranspiration, and by changes in precipitation distribution will likely increase the frequency and intensity of drought episodes in the future. Hence, ongoing monitoring of droughts and estimation of their impact on agriculture is necessary to adapt agricultural practices to changing weather and climate extremes. Several regional initiatives, projects and online tools have been established to facilitate drought monitoring and management in the Pannonian Basin. However, reliable systems to forecast potential drought impacts on plant productivity and agricultural yields at monthly to seasonal scales are only in their infancy, as plant response to climatic extremes is still poorly understood. With the increasing availability of high-resolution and long-term Earth Observation (EO) data and recent progress in machine learning and artificial intelligence, further improvements in drought monitoring and impact prediction capacities are expected. Here we review the current state of drought monitoring in the Pannonian Basin, identify EO-based variables to potentially improve regional drought impact monitoring and outline future perspectives for seasonal forecasts of drought impacts on agriculture.

Radio signals transmitted by Global Navigation Satellite System (GNSS) satellites experience tropospheric delays. While the hydrostatic part, referred to as zenith hydrostatic delay (ZHD) when mapped to the zenith direction, can be analytically modelled with sufficient accuracy, the wet part, referred to as zenith wet delay (ZWD), is much more difficult to determine and needs to be estimated. Thus, there exist several ZWD models which are used for various applications such as positioning and climate research. In this study, we present a data-driven, global model of the spatial ZWD field, based on the Extreme Gradient Boosting (XGBoost). The model takes the geographical location, the time, and a number of meteorological variables (in particular, specific humidity at several pressure levels) as input, and can predict ZWD anywhere on Earth as long as the input features are available. It was trained on ZWDs at 10718 GNSS stations and tested on ZWDs at 2684 GNSS stations for the year 2019. Across all test stations and all observations, the trained model achieved a mean absolute error of 6.1 mm, respectively, a root mean squared error of 8.1 mm. Comparisons of the XGBoost-based ZWD predictions with independently computed ZWDs and baseline models underline the good performance of the proposed model. Moreover, we analysed regional and monthly models, as well as the seasonal behaviour of the ZWD predictions in different climate zones, and found that the global model exhibits a high predictive skill in all regions and across all months of the year.

Climate models' outputs are affected by biases that need to be detected and adjusted to model climate impacts. Many climate hazards and climate-related impacts are associated with the interaction between multiple drivers, i.e. by compound events. So far climate model biases are typically assessed based on the hazard of interest, and it is unclear how much a potential bias in the dependence of the hazard drivers contributes to the overall bias and how the biases in the drivers interact. Here, based on copula theory, we develop a multivariate bias-assessment framework, which allows for disentangling the biases in hazard indicators in terms of the underlying univariate drivers and their statistical dependence. Based on this framework, we dissect biases in fire and heat stress hazards in a suite of global climate models by considering two simplified hazard indicators: the wet-bulb globe temperature (WBGT) and the Chandler burning index (CBI). Both indices solely rely on temperature and relative humidity. The spatial pattern of the hazard indicators is well represented by climate models. However, substantial biases exist in the representation of extreme conditions, especially in the CBI (spatial average of absolute bias: 21 ^∘C) due to the biases driven by relative humidity (20 ^∘C). Biases in WBGT (1.1 ^∘C) are small compared to the biases driven by temperature (1.9 ^∘C) and relative humidity (1.4 ^∘C), as the two biases compensate for each other. In many regions, also biases related to the statistical dependence (0.85 ^∘C) are important for WBGT, which indicates that well-designed physically based multivariate bias adjustment procedures should be considered for hazards and impacts that depend on multiple drivers. The proposed compound-event-oriented evaluation of climate model biases is easily applicable to other hazard types. Furthermore, it can contribute to improved present and future risk assessments through increasing our understanding of the biases' sources in the simulation of climate impacts.

Global navigation satellite systems (GNSS) provide globally distributed station coordinate time series that can be used for a variety of applications such as the definition of a terrestrial reference frame. A reliable estimation of the coordinate time series trends gives valuable information about station movements during the measured time period. Detecting discontinuities of various origins in such time series is crucial for accurate and robust velocity estimation. At present, there is no fully automated standard method for detecting discontinuities. Instead, discontinuity-catalogues are frequently used, which provide information about when a device was changed or an earthquake occurred. However, it is known that these catalogues suffer from incompleteness. This study investigates the suitability of machine learning classification algorithms that are fully data-driven to detect discontinuities caused by earthquakes in station coordinate time series without the need for external information. For this study, Japan was selected as a testing area. Ten different machine learning algorithms have been tested. It is found that Random Forest achieves the best performance with an F1 score of 0.77, a recall of 0.78, and a precision of 0.76. Overall, 525 of 565 recorded earthquakes in the test data were correctly classified. It is further highlighted that splitting the time series into chunks of 21 days leads to the best performance. Furthermore, it is beneficial to combine the three (normalized) components of the GNSS solution into one sample, and that adding the value range as an additional feature improves the result. Thus, this work demonstrates how it is possible to use machine learning algorithms to detect discontinuities in GNSS time series.

Earthquakes and tsunamis can trigger acoustic and gravity waves that could reach the ionosphere, generating electron density disturbances, known as traveling ionospheric disturbances. These perturbations can be investigated as variations in ionospheric total electron content (TEC) estimated through global navigation satellite systems (GNSS) receivers. The VARION (Variometric Approach for Real-Time Ionosphere Observation) algorithm is a well-known real-time tool for estimating TEC variations. In this context, the high amount of data allows the exploration of a VARION-based machine learning classification approach for TEC perturbation detection. For this purpose, we analyzed the 2015 Illapel earthquake and tsunami for its strength and high impact. We use the VARION-generated observations (i.e., dsTEC/dt) provided by 115 GNSS stations as input features for the machine learning algorithms, namely, Random Forest and XGBoost. We manually label time frames of TEC perturbations as the target variable. We consider two elevation cut-off time series, namely, 15° and 25°, to which we apply the classifier. XGBoost with a 15° elevation cut-off dsTEC/dt time series reaches the best performance, achieving an F1 score of 0.77, recall of 0.74, and precision of 0.80 on the test data. Furthermore, XGBoost presents an average difference between the labeled and predicted middle epochs of TEC perturbation of 75 s. Finally, the model could be seamlessly integrated into a real-time early warning system, due to its low computational time. This work demonstrates high-probability TEC signature detection by machine learning for earthquakes and tsunamis, that can be used to enhance tsunami early warning systems.