IT-SNOW: a snow reanalysis for Italy blending modeling, in-situ data, and satellite observations (2010–2021)

Gabellani

Delogu

et al. 2023

Earth Syst. Sci. Data

Self Cite

Abstract. We present IT-SNOW, a serially complete and multi-year snow reanalysis for Italy (∼ 301 × 103 km2) – a transitional continental-to-Mediterranean region where snow plays an important but still poorly constrained societal and ecological role. IT-SNOW provides ∼ 500 m daily maps of snow water equivalent (SWE), snow depth, bulk snow density, and liquid water content for the initial period 1 September 2010–31 August 2021, with future updates envisaged on a regular basis. As the output of an operational chain employed in real-world civil protection applications (S3M Italy), IT-SNOW ingests input data from thousands of automatic weather stations, snow-covered-area maps from Sentinel-2, MODIS (Moderate Resolution Imaging Spectroradiometer), and H SAF products, as well as maps of snow depth from the spatialization of over 350 on-the-ground snow depth sensors. Validation using Sentinel-1-based maps of snow depth and a variety of independent, in situ snow data from three focus regions (Aosta Valley, Lombardy, and Molise) show little to no mean bias compared to the former, and root mean square errors are of the typical order of 30–60 cm and 90–300 mm for in situ, measured snow depth and snow water equivalent, respectively. Estimates of peak SWE by IT-SNOW are also well correlated with annual streamflow at the closure section of 102 basins across Italy (0.87), with ratios between peak water volume in snow and annual streamflow that are in line with expectations for this mixed rain–snow region (22 % on average and 12 % median). Examples of use allowed us to estimate 13.70 ± 4.9 Gm3 of water volume stored in snow across the Italian landscape at peak accumulation, which on average occurs on 4 March ± 10 d. Nearly 52 % of the mean seasonal SWE is accumulated across the Po river basin, followed by the Adige river (23 %), and central Apennines (5 %). IT-SNOW is freely available at https://doi.org/10.5281/zenodo.7034956 (Avanzi et al., 2022b) and can contribute to better constraining the role of snow for seasonal to annual water resources – a crucial endeavor in a warming and drier climate.

Section: Data Formatmentioning

confidence: 99%

“…IT-SNOW is available in an open-access framework at https://doi.org/10.5281/zenodo.7034956 (CC BY-NC 4.0; see Avanzi et al, 2022b).…”

Section: Code and Data Availabilitymentioning

confidence: 99%

IT-SNOW: a snow reanalysis for Italy blending modeling, in situ data, and satellite observations (2010–2021)

Gabellani

Delogu

et al. 2023

Earth Syst. Sci. Data

Self Cite

“…These 27 snow-depth sensors were chosen based on a geographical-diversity criterion to guarantee heterogeneity, especially with regard to the Aosta Valley data (Figure 2). This second dataset include data from 3 years: 2018, 2020 and 2022, which were chosen due to their significantly different accumulation patterns (deep snowpacks in 2018, somewhat average snowpacks in 2020, and extraordinarily low snowpacks in 2022, see Avanzi et al (2022a)). No prior processing was available for these data, thus we proceeded with our own manual classification to assign codes as in Table 1.…”

Section: Other Italian Datamentioning

confidence: 99%

“…The period of record goes from August 2003 to September 2021, thus covering a variety of snow seasons across 18 years of data (Avanzi et al, 2022b). The elevation range of these sensors goes from 545 m to 2842 m asl, with an average elevation of 2007 m asl that is representative of average elevations across the Italian Alps where the bulk of sensors are located (Avanzi et al, 2021b).…”

Section: Introductionmentioning

confidence: 99%

A Random Forest approach to quality-checking automatic snow-depth sensor measurements

Blandini

Gabellani

et al. 2023

Preprint

Abstract. State-of-the-art snow sensing technologies currently provide an unprecedented amount of data from both remote sensing satellites and ground sensors, but their assimilation into dynamic models is bounded to data quality, which is often low − especially in mountain, high-elevation, and unattended regions where snow is the predominant land-cover feature. To maximize the value of snow-depth measurements, we developed a Random Forest classifier to automatize the quality assurance/quality control (QA/QC) procedure of near-surface snow depth measurements collected through ultrasonic sensors, with particular reference to differentiate snow cover from grass or bare ground data and to detecting random errors (e.g., spikes). The model was trained and validated using a split-sample approach of an already manually classified dataset of 18 years of data from 43 sensors in Aosta Valley (north-western Italian Alps), and then further validated using 3 years of data from 27 stations across the rest of Italy (with no further training or tuning). The F1 score was used as scoring metric, being it the most suited to describe the performances of a model in case of a multi-class imbalanced classification problem. The model proved to be both robust and reliable in the classification of snow cover vs. grass/bare ground in Aosta Valley (F1 values above 90 %), yet less reliable in rare random-error detection, mostly due to the dataset imbalance (samples distribution: 46.46 % snow, 49.21 % grass/bare ground, 4.34 % error). No clear correlation with snow-season climatology was found in the training dataset, which further suggests robustness of our approach. The application across the rest of Italy yielded F1 scores on the order of 90 % for snow and grass/bare ground, thus confirming results from the testing region and corroborating model robustness and reliability, with again a less skillful classification of random errors (values below 5 %). This machine learning algorithm of data quality assessment will provide more reliable snow ground data, enhancing their use in snow models.

“…The climate in the region shows a transition from alpine and cold, with a bimodal precipitation annual cycle and peaks in autumn and spring, to temperate with a dry season and most of the precipitation falling in winter (Beck et al, 2018;Crespi et al, 2018) (Figure 1b). Snow contribution to streamflow generation can be relevant, especially at high elevations in the northern and western part of the catchment, where the mean annual ratio between peak snow water equivalent and the annual cumulative Q can exceed 60% (Avanzi et al, 2022). Subsequently, Q usually has two peaks, one in autumn caused by heavy rainfall events and one in spring caused by rainfall events and snowmelt, with a low-flow period during summer.…”

Section: Study Areamentioning

confidence: 99%

Parameter transferability of a distributed hydrological model to droughts

Bruno

Duethmann

et al. 2022

Preprint

Abstract. Hydrological models often have issues in simulating streamflow (Q) during droughts, because of hard-to-capture feedback mechanisms across precipitation deficit, actual evapotranspiration (ET), and terrestrial water storage anomalies (TWSA). To gain more insights into these performance drops and move toward more robust hydrological models in the anthropogenic era, we evaluated Q, ET, and TWSA simulations during droughts of different severity and their sensitivity to the climatic conditions of the calibration period. We used the distributed hydrological model Continuum over the heavily human-affected Po river basin (northern Italy, period 2010–2022) and independent ground- and remote sensing-based datasets of Q, ET, and TWSA as benchmarks. Across the 38 study sub-catchments, Continuum simulated Q comparably well during wet years (2014 and 2020) and moderate droughts (2012 and 2017) with mean KGE = 0.59±0.32 during wet years and = 0.55±0.25 during moderate droughts. The model simulated well Q for the outlet section of the basin also for the severe 2022 drought (KGE = 0.82). However, performances for 2022 declined across the other sub-catchments (mean KGE = 0.18±0.69, meaning the model still preserved some skill over a climatological mean). The model properly represented seasonality of Q, ET, and TWSA over the basin, as well as a declining trend in TWSA. We explained the performance drops in 2022 with an increased uncertainty in ET anomalies, in particular in human-affected croplands. Calibrating during a moderate drought (2017) did not improve model performances during the severe 2022 drought (mean KGE = 0.18±0.63), pointing to the fairly unique conditions of this period in terms of hydrological processes and human interference on the hydrological cycle. By highlighting increased uncertainty of hydrological models specifically during severe droughts which are expected to increase in frequency, these findings provide relevant guidelines for assessments of model robustness in a changing climate and so for informing water management, disaster risk reduction, and climate change adaptation strategies.