Reply on RC2

The mountainous snow cover is highly variable at all temporal and spatial scales. Snowpack models only imperfectly represent this variability, because of uncertain meteorological inputs, physical parameterisations, and unresolved terrain features. In-situ observations of the height of snow (HS), despite their limited representativeness, could help constrain intermediate and large scale modelling errors by means of data assimilation. In this work, we assimilate HS observations from an in-situ network of 295 stations covering the French Alps, Pyrenees and Andorra, over the period 2009–2019. In view of assimilating such observations into a spatialised snow cover modelling framework, we investigate whether such observations can be used to correct neighbouring snowpack simulations. We use CrocO, an ensemble data assimilation framework of snow cover modelling, based on a Particle Filter suited to the propagation of information from observed to unobserved areas. This ensemble system already benefits from meteorological observations, assimilated within SAFRAN analysis scheme. CrocO also proposes various localisation strategies to assimilate snow observations. These approaches are evaluated in a Leave-One-Out setup against the operational deterministic model and its ensemble open-loop counterpart, both running without HS assimilation. Results show that intermediate localisation radius of 35–50 km yield a slightly lower root mean square error (RMSE), and a better Spread-Skill than the strategy assimilating all the observations from a whole mountain range. Significant continuous ranked probability score (CRPS) improvements of about 13 % are obtained in the areas where the open-loop modelling errors are the largest, e.g. the Haute-Ariège, Andorra and the Extreme Southern Alps. Over these areas, weather station observations are generally sparser, resulting in more uncertain meteorological analyses, and therefore snow simulations. In-situ HS observations thus shows an interesting complementarity with meteorological observations to better constrain snow cover simulations over large areas.

Abstract. The mountainous snow cover is highly variable at all temporal and spatial scales. Snowpack models only imperfectly represent this variability, because of uncertain meteorological inputs, physical parameterizations, and unresolved terrain features. In situ observations of the height of snow (HS), despite their limited representativeness, could help constrain intermediate and large-scale modeling errors by means of data assimilation. In this work, we assimilate HS observations from an in situ network of 295 stations covering the French Alps, Pyrenees, and Andorra, over the period 2009–2019. In view of assimilating such observations into a spatialized snow cover modeling framework, we investigate whether such observations can be used to correct neighboring snowpack simulations. We use CrocO, an ensemble data assimilation framework of snow cover modeling, based on a particle filter suited to the propagation of information from observed to unobserved areas. This ensemble system already benefits from meteorological observations, assimilated within SAFRAN analysis scheme. CrocO also proposes various localization strategies to assimilate snow observations. These approaches are evaluated in a leave-one-out setup against the operational deterministic model and its ensemble open-loop counterpart, both running without HS assimilation. Results show that an intermediate localization radius of 35–50 km yields a slightly lower root mean square error (RMSE), and a better spread–skill than the strategy of assimilating all the observations from a whole mountain range. Significant continuous ranked probability score (CRPS) improvements of about 13 % are obtained in the areas where the open-loop modeling errors are the largest, e.g., the Haute-Ariège, Andorra, and the extreme southern Alps. Over these areas, weather station observations are generally sparser, resulting in more uncertain meteorological analyses and, therefore, snow simulations. In situ HS observations thus show an interesting complementarity with meteorological observations to better constrain snow cover simulations over large areas.

Multimodal ultra-short-term probabilistic solar power forecasting with generative AI and Transformer

To improve control over the drying process and the quality of the dried products, it is essential to comprehend the drying kinetics model. This study aim to find the best thin-layer drying kinetics model for box-shaped candied mangoes dried in a rotary oven. The source of the data is Rozana’s (2016) research into the moisture content and drying times of candied mangoes at 45 and 50°C in a rotating oven. The Lewis, Page, Henderson and Pabis, Logarithmic, Two Term, Two Term Exponential, Diffusion Approximation, Midilli et al., and Modified Henderson and Perry models are among the kinetic models used. The moisture ratios of mangoes collected during drying at various time intervals were used to fit drying models in order to predict the moisture ratio. The model with the highest coefficient of determination (R2), the lowest root mean square error (χ2), and the lowest root mean square error (RMSE) is the best drying kinetics model. The Midilli et al. model was determined to be the most appropriate drying model for box-shaped candied mangoes dried in a rotating oven with R2 of 0,9989, χ2 of 0,0001, and an RMSE value of 0,0074 for a temperature of 45°C; and R2 is 0,9973, χ2 is 0,0002, and RMSE is 0,0137 at 50°C. The Midilli et al. model equations obtained is MR=1,026 exp(-0,0007t1,0706)+0,0001t at 45°C and MR=1,0129exp(-0,0003t1,3568)+0,0002t at 50°C.

ABSTRACTThis paper deals with seasonal climate forecasting using as many as 13 coupled ocean—atmosphere models. An analysis of the individual model, multimodel ensemble, and FSU synthetic superensemble (FSUSSE) climate forecasts was performed for monthly and seasonal forecasts of precipitation, sea surface temperature (SST) and surface air temperature over the Euro-Mediterranean region including the land areas of Europe, North Africa and the Near East, during the period 1989–2001. In the FSUSSE methodology, forecasts are obtained by a weighted combination of the individual coupled ocean—atmosphere model forecasts based on a training period. The set of 13 individual climate forecast models utilized in this research is comprised of the seven models in the European suite of DEMETER models, a suite of four Florida State University models, the Australian POAMA model and the NCAR CCM3 model.The FSUSSE forecasts of seasonal precipitation anomalies were found to have the lowest root mean square (RMS) errors in comparison to the models in the multimodel ensemble, and their ensemble mean. However, the anomaly correlation (AC) coefficient results for seasonal precipitation anomaly forecasts by the FSUSSE were less impressive. The equitable threat scores for the FSUSSE forecasts of seasonal precipitation were found to be better than the various models in the multimodel ensemble, but those scores for the forecasts of positive seasonal anomalies were found to be worse than most of the models in the multimodel ensemble.The FSUSSE seasonal forecasts for SST and surface air temperature for the season considered (winter) were found to be excellent for both AC coefficients and RMS errors in the forecast anomalies.

Electric power grids are increasingly penetrated by generation from photovoltaic (PV) systems. The volatility of solar irradiance can therefore cause problematic fluctuations of grid voltage and frequency. For local low voltage power grid sections highly accurate nowcasts of the short-term local PV-power generation may enable higher PV penetration by employing short-term flexibility options while saving investments in grid-infrastructure. Systems of up to four All-Sky-Imagers (ASI) have been used successfully to nowcast direct and global irradiance as well as the production of solar power plants. In Oldenburg and the surrounding area these systems are enhanced by interconnecting 34 All-Sky-Imagers in a measurement network called Eye2Sky. Intermediate results can be combined and exchanged between systems with overlapping observation areas but also between systems up to 100 km apart. To demonstrate the potential of the approach three state-of-the-art 2-ASI-systems from the network are selected. Nowcasts of the individual systems are averaged weighting each prediction by its expected accuracy in the single points of the forecast domain. In a validation period of 86 days the network outperforms a single 2-ASI-system. For all lead times greater than zero, significantly lower root mean square errors are found compared to the 2-ASI system. For highly variable time intervals the root mean square error is reduced by up to 50 %. The network´s combined nowcast is found to reduce forecast errors related to the often uncertain cloud edge positions and further systematic errors found in single 2-ASI-systems.

Abstract. Italy is a territory characterized by complex topography with the Apennines mountain range crossing the entire peninsula and its highest peaks in central Italy. Using the latter as our area of interest and the snow seasons 2018/19, 2019/20 and 2020/21, the goal of this study is to investigate the ability of a simple single-layer and a more sophisticated multi-layer snow cover numerical model to reproduce the observed snow height, snow water equivalent and snow extent in the central Apennines, using for both models the same forecast weather data as meteorological forcing. We here consider two well-known ground surface and soil models: (i) Noah LSM, an Eulerian model which simulates the snowpack as a bulk single layer, and (ii) Alpine3D, a multi-layer Lagrangian model which simulates the snowpack stratification. We adopt the Weather Research and Forecasting (WRF) model to produce the meteorological data to drive both Noah LSM and Alpine3D at a regional scale with a spatial resolution of 3 km. While Noah LSM is already online-coupled with the WRF model, we develop here a dedicated offline coupling between WRF and Alpine3D. We validate the WRF simulations of surface meteorological variables in central Italy using a dense network of automatic weather stations, obtaining correlation coefficients higher than 0.68, except for wind speed, which suffered from the model underestimation of the real elevation. The performances of both WRF–Noah and WRF–Alpine3D are evaluated by comparing simulated and measured snow height, snow height variation and snow water equivalent, provided by a quality-controlled network of automatic and manual snow stations located in the central Apennines. We find that WRF–Alpine3D can predict better than WRF–Noah the snow height and the snow water equivalent, showing a correlation coefficient with the observations of 0.9 for the former and 0.7 for the latter. Both models show similar performances in reproducing the observed daily snow height variation; nevertheless WRF–Noah is slightly better at predicting large positive variations, while WRF–Alpine3D can slightly better simulate large negative variations. Finally we investigate the abilities of the models in simulating the snow cover area fraction, and we show that WRF–Noah and WRF–Alpine3D have almost equal skills, with both models overestimating it. The equal skills are also confirmed by Jaccard and the average symmetric surface distance indices.

Arctic sea ice and snow affect the energy balance of the global climate system through the radiation budget. Accurate determination of the snow cover over Arctic sea ice is significant for the retrieval of the sea ice thickness (SIT). In this study, we developed a new snow depth retrieval method over Arctic sea ice with a long short-term memory (LSTM) deep learning algorithm based on Operation IceBridge (OIB) snow depth data and brightness temperature data of AMSR-2 passive microwave radiometers. We compared climatology products (modified W99 and AWI), altimeter products (Kwok) and microwave radiometer products (Bremen, Neural Network and LSTM). The climatology products and altimeter products are completely independent of the OIB data used for training, while microwave radiometer products are not completely independent of the OIB data. We also compared the SITs retrieved from the above different snow depth products based on Cryosat-2 radar altimeter data. First, the snow depth spatial patterns for all products are in broad agreement, but the temporal evolution patterns are distinct. Snow products of microwave radiometers, such as Bremen, Neural Network and LSTM snow depth products, show thicker snow in early winter with respect to the climatology snow depth products and the altimeter snow depth product, especially in the multiyear ice (MYI) region. In addition, the differences in all snow depth products are relatively large in the early winter and relatively small in spring. Compared with the OIB and IceBird observation data (April 2019), the snow depth retrieved by the LSTM algorithm is better than that retrieved by the other algorithms in terms of accuracy, with a correlation of 0.55 (0.90), a root mean square error (RMSE) of 0.06 m (0.05 m) and a mean absolute error (MAE) of 0.05 m (0.04 m). The spatial pattern and seasonal variation of the SITs retrieved from different snow depths are basically consistent. The total sea ice decreases first and then thickens as the seasons change. Compared with the OIB SIT in April 2019, the SIT retrieved by the LSTM snow depth is superior to that retrieved by the other SIT products in terms of accuracy, with the highest correlation of 0.46, the lowest RMSE of 0.59 m and the lowest MAE of 0.44 m. In general, it is promising to retrieve Arctic snow depth using the LSTM algorithm, but the retrieval of snow depth over MYI still needs to be verified with more measured data, especially in early winter.

For numerous hydrological applications, information on snow water equivalent (SWE) and snow liquid water content (LWC) are fundamental. In situ data are much needed for the validation of model and remote sensing products; however, they are often scarce, invasive, expensive, or labor‐intense. We developed a novel nondestructive approach based on Global Positioning System (GPS) signals to derive SWE, snow height (HS), and LWC simultaneously using one sensor setup only. We installed two low‐cost GPS sensors at the high‐alpine site Weissfluhjoch (Switzerland) and processed data for three entire winter seasons between October 2015 and July 2018. One antenna was mounted on a pole, being permanently snow‐free; the other one was placed on the ground and hence seasonally covered by snow. While SWE can be derived by exploiting GPS carrier phases for dry‐snow conditions, the GPS signals are increasingly delayed and attenuated under wet snow. Therefore, we combined carrier phase and signal strength information, dielectric models, and simple snow densification approaches to jointly derive SWE, HS, and LWC. The agreement with the validation measurements was very good, even for large values of SWE (>1,000 mm) and HS (> 3 m). Regarding SWE, the agreement (root‐mean‐square error (RMSE); coefficient of determination (R2)) for dry snow (41 mm; 0.99) was very high and slightly better than for wet snow (73 mm; 0.93). Regarding HS, the agreement was even better and almost equally good for dry (0.13 m; 0.98) and wet snow (0.14 m; 0.95). The approach presented is suited to establish sensor networks that may improve the spatial and temporal resolution of snow data in remote areas.

Refining the global branched glycerol dialkyl glycerol tetraether (brGDGT) soil temperature calibration

Predictive maintenance has gained importance across various industries, including the automotive sector. It is very challenging to detect vehicle failures in advance due to the intricate composition of various components and sensors. The vehicle’s reliability is of utmost importance for ensuring the absence of fatalities or malfunctions to foster economic development. This study introduces an innovative method for developing a predictive framework for vehicle engines with faster and higher decision accuracy. The framework is specifically designed to recognize patterns and abnormalities that may suggest prospective engine problems in real-time and allow proactive maintenance. We assessed the performance of the developed vehicular engine health monitoring systems using a deep learning model based on essential measures like root mean square error, root mean square deviation, mean absolute error, accuracy, confusion matrix, and area under the curve. In this case, the deep learning models are developed by following ensemble techniques using the most prominently used machine learning techniques. Significantly, Stacked Model 1 outperformed other stacked models (Models 2 and 3) and achieved an impressive AUC value of 0.9702 with a low root mean square error (RMSE) of 0.3355, a high accuracy rate of 0.9470, and a precision of 0.9486. It happens due to the effective incorporation of different approaches into Stacked Model 1, which signifies a significant advancement in predicting vehicular engine failures. The model can be used in real-time monitoring systems to continuously monitor the health of vehicular engines and provide early warnings of potential failures, thereby reducing maintenance costs and improving safety.

Rainfalls are one of the most important climate variables that today impact significantly different sectors like agriculture, energy, industry, and so on. Agriculture is one of the most sensitive sectors to climate change because rainfalls in this case, directly affect the positive progress of corps activity. In this case, forecasting rainfalls would help farmers to effectively survive the increasing occurrence of extreme weather events, plan their farming activities, and reduce costs. On the other hand, circular economy (CE) promises a strategy to support sustainable and regenerative agriculture by supporting the sustainable management of water based on water resources. This paper aims to determine the best method for forecasting a natural phenomenon such as the rainfall, that today in Albania, as a result of the unpredictable flows that it often has, is a major problem in the field of agriculture. In this study, the rainfall model based on statistical methods, Auto-Regressive Integrated Moving Average (ARIMA), Error, Trend & Seasonal (ETS) and deep learning models, Long Short-Term Memory Network (LSTM), and Deep Forward Neural Network (DFNN) was developed. The study area that will be used for rainfall forecasting is Albania with a time interval between January 1901 and December 2022. The period that will be used for prediction will be January 2023- December 2024. The performance of each of the models used has been evaluated by using Root Mean Square Error (RMSE) where we also used the comparison of training and validation loss curves to analyze and avoid the model overfitting in the training phase. The results showed that from the comparison between ARIMA and ETS, ETS has the minimum prediction error value while between LSTM and DFNN, DFNN has the best performance in the evaluation metrics (RMSE) and with the best training and validation loss curves. From the final comparison, ETS was better than the DFNN model with the lowest root mean square error (RMSE). ETS was the best model and provided higher accuracy in precipitation forecast.

Abstract. Stream water temperatures influence water quality, with effects on aquatic biodiversity, drinking-water provision, electricity production, agriculture, and recreation. Therefore, stakeholders would benefit from an operational forecasting service that would support timely action. Deep-learning models are well-suited to providing probabilistic forecasts at individual stations of a monitoring network. Here, we train and evaluate several state-of-the-art models using 10 years of data from 54 stations across Switzerland. Static catchment features, time of the year, meteorological observations from the past 64 d, and their ensemble forecasts for the following 32 d are included as predictors in the models to estimate daily maximum water temperature over the next 32 d. Results show that the temporal fusion transformer (TFT) model performs best, with a continuous rank probability score (CRPS) of 0.70 °C averaged over all lead times, stations, and 90 forecasts distributed over 1 year. The TFT is followed by the recurrent neural network encoder–decoder, with a CRPS of 0.74 °C, and the neural hierarchical interpolation for time series, with a CRPS of 0.75 °C. These deep-learning models outperform other simpler models trained at each station: random forest (CRPS = 0.80 °C), multi-layer perceptron neural network (CRPS = 0.81 °C), and autoregressive linear model (CRPS = 0.96 °C). The average CRPS of the TFT degrades from 0.38 °C at lead a time of 1 d to 0.90 °C at a lead time of 32 d, largely driven by the uncertainty of the meteorological ensemble forecasts. In addition, TFT water temperature predictions at new and ungauged stations outperform those from the other models. When analyzing the importance of model inputs, we find a dominant role of observed water temperature and future air temperature, while including precipitation and time of the year further improves predictive skill. Operational probabilistic forecasts of daily maximum water temperature are generated twice per week with our TFT model and are publicly available at https://www.drought.ch/de/impakt-vorhersagen-malefix/wassertemperatur-prognosen/ (last access: 20 March 2025). Overall, this study provides insights into the extended-range predictability of stream water temperature and into the applicability of deep-learning models in hydrology.

Uncertainty propagation within a water level ensemble prediction system

Assessment of gridded datasets of various near surface temperature variables over Heihe River Basin: Uncertainties, spatial heterogeneity and clear-sky bias