Early event detection and response can significantly reduce the societal impact of floods. Currently, early warning systems rely on gauges, radar data, models and informal local sources. However, the scope and reliability of these systems are limited. Recently, the use of social media for detecting disasters has shown promising results, especially for earthquakes. Here, we present a new database for detecting floods in real-time on a global scale using Twitter. The method was developed using 88 million tweets, from which we derived over 10,000 flood events (i.e., flooding occurring in a country or first order administrative subdivision) across 176 countries in 11 languages in just over four years. Using strict parameters, validation shows that approximately 90% of the events were correctly detected. In countries where the first official language is included, our algorithm detected 63% of events in NatCatSERVICE disaster database at admin 1 level. Moreover, a large number of flood events not included in NatCatSERVICE were detected. All results are publicly available on www.globalfloodmonitor.org.

Timely and accurate information about ongoing events are crucial for relief organizations seeking to effectively respond to disasters. Recently, social media platforms, especially Twitter, have gained traction as a novel source of information on disaster events. Unfortunately, geographical information is rarely attached to tweets, which hinders the use of Twitter for geographical applications. As a solution, geoparsing algorithms extract and can locate geographical locations referenced in a tweet's text. This paper describes TAGGS, a new algorithm that enhances location disambiguation by employing both metadata and the contextual spatial information of groups of tweets referencing the same location regarding a specific disaster type. Validation demonstrated that TAGGS approximately attains a recall of 0.82 and precision of 0.91. Without lowering precision, this roughly doubles the number of correctly found administrative subdivisions and cities, towns, and villages as compared to individual geoparsing. We applied TAGGS to 55.1 million flood-related tweets in 12 languages, collected over 3 years. We found 19.2 million tweets mentioning one or more flood locations, which can be towns (11.2 million), administrative subdivisions (5.1 million), or countries (4.6 million). In the future, TAGGS could form the basis for a global event detection system.

Abstract. Neural networks have been shown to be extremely effective rainfall-runoff models, where the river discharge is predicted from meteorological inputs. However, the question remains: what have these models learned? Is it possible to extract information about the learned relationships that map inputs to outputs, and do these mappings represent known hydrological concepts? Small-scale experiments have demonstrated that the internal states of long short-term memory networks (LSTMs), a particular neural network architecture predisposed to hydrological modelling, can be interpreted. By extracting the tensors which represent the learned translation from inputs (precipitation, temperature, and potential evapotranspiration) to outputs (discharge), this research seeks to understand what information the LSTM captures about the hydrological system. We assess the hypothesis that the LSTM replicates real-world processes and that we can extract information about these processes from the internal states of the LSTM. We examine the cell-state vector, which represents the memory of the LSTM, and explore the ways in which the LSTM learns to reproduce stores of water, such as soil moisture and snow cover. We use a simple regression approach to map the LSTM state vector to our target stores (soil moisture and snow). Good correlations (R2>0.8) between the probe outputs and the target variables of interest provide evidence that the LSTM contains information that reflects known hydrological processes comparable with the concept of variable-capacity soil moisture stores. The implications of this study are threefold: (1) LSTMs reproduce known hydrological processes. (2) While conceptual models have theoretical assumptions embedded in the model a priori, the LSTM derives these from the data. These learned representations are interpretable by scientists. (3) LSTMs can be used to gain an estimate of intermediate stores of water such as soil moisture. While machine learning interpretability is still a nascent field and our approach reflects a simple technique for exploring what the model has learned, the results are robust to different initial conditions and to a variety of benchmarking experiments. We therefore argue that deep learning approaches can be used to advance our scientific goals as well as our predictive goals.

In many semiarid regions with irrigation, the depletion rate of groundwater resources has increased substantially during the last decades. A possible reason for this is that the price that users pay for their water does not reflect its scarcity and value. An alternative way to assess the perceived value of water is calculating its shadow price, which is defined here as the marginal value produced, and relates to the efficiency gain from current reallocation. Here we determine the shadow price of water used for irrigation for the most important groundwater‐depleting countries and for four staple crops and one cash crop. To quantify the shadow price, the relation between the output and the water input is represented using production functions. We use globally available panel data on country‐specific crop yields and prices together with crop‐specific water consumption, calculated with the global hydrological model PCR‐GLOBWB, to parameterize the production function by country and crop with econometric analyses. Our results show that the variation of shadow prices for staple crops within several countries is high, indicating economically inefficient use of water resources, including nonrenewable groundwater. We also analyze the effects of reallocating irrigation water between crops, showing that changes in water allocation could lead to either an increase in the economic efficiency of water use or large reductions in irrigation water consumption. Our study thus provides a hydroeconomic basis to stimulate sustainable use of finite groundwater resources globally.