To date, long short-term memory (LSTM) networks have been successfully applied to a key problem in hydrology: the prediction of runoff. Unlike traditional conceptual models, LSTM models are built on concepts that avoid the need for our knowledge of hydrology to be formally encoded into the model. The question, then, is how we can still make use of our domain knowledge and traditional practices, not to build the LSTM models themselves, as we do for conceptual models, but to use them more effectively. In the present paper, we adopt this approach, investigating how we can use information concerning the hydrologic characteristics of catchments for LSTM runoff models. In this first application of LSTM in a French context, we use 361 gauged catchments with very diverse hydrologic conditions from across France. The catchments have long time series of at least 30 years. Our main directions for investigation include (a) the relationship between LSTM performance and the length of the LSTM input sequence within different hydrologic regimes, (b) the importance of the hydrologic homogeneity of catchments when training LSTMs on a group of catchments, and (c) the interconnected influence of the local tuning of the two important LSTM hyperparameters, namely the length of the input sequence and the hidden unit size, on the performance of group-trained LSTMs. We present a classification built on three indices taken from the runoff, precipitation, and temperature regimes. We use this classification as our measure of homogeneity: catchments within the same regime are assumed to be hydrologically homogeneous. We train LSTMs on individual catchments (local-level training), on catchments within the same regime (regime-level training), and on the entire sample (national-level training). We benchmark local LSTMs using the GR4J conceptual model, which is able to represent the water gains/losses in a catchment. We show that LSTM performance has the highest sensitivity to the length of the input sequence in the Uniform and Nival regimes, where the dominant hydrologic process of the regime has clear long-term dynamics; thus, long input sequences should be chosen in these cases. In other regimes, this level of sensitivity is not found. Moreover, in some regimes, almost no sensitivity is observed. Therefore, the size of the input sequence in these regimes does not need to be large. Overall, our homogeneous regime-level training slightly outperforms our heterogeneous national-level training. This shows that the same level of data adequacy with respect to the complexity of representation(s) to be learned is achieved in both levels of training. We do not, however, exclude a potential role of the regime-informed property of our national LSTMs, which use previous classification variables as static attributes. Last but not least, we demonstrate that the local selection of the two important LSTM hyperparameters (the length of the input sequence and the hidden unit size) combined with national-level training can lead to the best runoff prediction performance.

Over the last decade, large-sample approaches, i.e., based on large catchment sets, have become increasingly popular in hydrological studies. Efforts were made to assemble and disseminate national catchment datasets. This article aims to make a contribution to the construction of a large international database of catchments by proposing the CAMELS-FR dataset, a contribution to the CAMELS (Catchment Attributes and MEteorology for Large-sample Studies) initiative. The first version presented here gathers hydroclimatic data and physical attributes for a set of 654 catchments in France. These catchments cover a wide spectrum of hydroclimatic conditions (from oceanic to continental, mountainous, or Mediterranean conditions) and are considered to have limited human influence. Data include time series of daily streamflow (with at least 30 years over the 1970–2021 period, also aggregated to monthly and yearly time steps) and of 11 catchment-scale daily climate variables (including precipitation, potential evaporation, and air temperature), as well as a total of 255 catchment attributes organized into 10 classes (e.g., geology, soil, land cover). River flow time series were quality-checked. Along with the database itself, two graphical tools are proposed, namely dynamic graphs to visualize time series and graphical fact sheets to summarize the main catchment characteristics. Care was taken to provide as many metadata as possible to help users interpret their results based on this dataset. We intend to update the database regularly to include new available data and account for end users' feedback. CAMELS-FR (Delaigue et al., 2024c) is available at https://doi.org/10.57745/WH7FJR.

Over the last decades, different methods have been used by hydrologists to extend observed hydro-climatic time series, based on other data sources, such as tree rings or sedimentological datasets. For example, tree ring multi-proxies have been studied for the Caniapiscau Reservoir in northern Québec (Canada), leading to the reconstruction of flow time series for the last 150 years. In this paper, we applied a new hydro-climatic reconstruction method on the Caniapiscau Reservoir and compare the obtained streamflow time series against time series derived from dendrohydrology by other authors on the same catchment and study the natural streamflow variability over the 1881–2011 period in that region. This new reconstruction is based not on natural proxies but on a historical reanalysis of global geopotential height fields, and aims firstly to produce daily climatic time series, which are then used as inputs to a rainfall–runoff model in order to obtain daily streamflow time series. The performances of the hydro-climatic reconstruction were quantified over the observed period, and showed good performances, in terms of both monthly regimes and interannual variability. The streamflow reconstructions were then compared to two different reconstructions performed on the same catchment by using tree ring data series, one being focused on mean annual flows and the other on spring floods. In terms of mean annual flows, the interannual variability in the reconstructed flows was similar (except for the 1930–1940 decade), with noteworthy changes seen in wetter and drier years. For spring floods, the reconstructed interannual variabilities were quite similar for the 1955–2011 period, but strongly different between 1880 and 1940. The results emphasize the need to apply different reconstruction methods on the same catchments. Indeed, comparisons such as those above highlight potential differences between available reconstructions and, finally, allow a retrospective analysis of the proposed reconstructions of past hydro-climatological variabilities.

Haiti, a Caribbean country, is highly vulnerable to hydroclimatic hazards due to heavy rainfall, which is partly linked to tropical cyclones. Additionally, its steep slopes generate flash floods, particularly in small catchments. Moreover, the hydrology of this region remains poorly understood and understudied. Unfortunately, there is no accessible database for the scientific community to use in this country. To fill this gap, hydroclimatic data were collected to create the first historical database in Haiti. This database, called Simbi (guardian of rivers, freshwater, and rain in Haitian mythology), includes 156 monthly rainfall series over the period 1905–2005, 59 daily rainfall series over the period 1920–1940, 70 daily streamflow series, and 23 monthly temperature series, not necessarily continuous, over the period 1920–1940. It also provides simulated streamflow series over the period 1920–1940 using the GR2M and GR4J rainfall–runoff models for 24 catchments and 49 attributes covering a wide range of topographic, climatic, geological, land use, hydrogeological, and hydrological signature indices. Simbi is the first open-access hydro-meteorological dataset for Haiti and will contribute to a better knowledge of hydrological risk in Haiti. Several sources of uncertainty associated with Simbi are acknowledged, including data quality (historical data), digitisation of paper archives, identification of relevant rain gauges, and rainfall–runoff models. It is important to consider these uncertainties when using Simbi. The database will be regularly updated to include additional historical data that will be digitised in the future. It will thus contribute toward better knowledge of the hydrology of Haitian catchments and will enable the implementation of various hydrological calculations useful for designing structures or flow forecasting. Simbi is an open-access database and is available for download at https://doi.org/10.23708/02POK6 (Bathelemy et al., 2023).