13.  Nimac I. and M. Perčec Tadić: Complete and homogeneous monthly air temperature series for the construction of 1981-2010 climatological normals in Croatia.

Providing climatological normals is one of the most important tasks for national meteorological services. Estimating the statistical characteristics of climate variables from incomplete and inhomogeneous data can result in biased estimations; thus, it is necessary to fill in missing values and remove inhomogeneities. Though it is very important, the homogenization procedure is still not a part of data quality-check procedures. In this work, monthly temperature data from 39 meteorological stations in Croatia for the period 1981-2010 were examined for missing data and inhomogeneities. Stations were divided into three climatic regions, and homogenization was performed for each one separately. The performance of the homogenization method was tested by: (1) comparison of correlation coefficients amongst stations and (2) changes in rotated principal components for datasets before and after homogenization. Obtained homogeneity breaks were compared with metadata and published literature. Changes in the statistical characteristics of temperature climate normals between 1981 and 2010 (e.g., long-term means and decadal trends) were observed at annual and seasonal scales between original and homogenized series. The significance of the changes in mean was tested using the Student's t-test, while the significance of trends was tested with the Mann-Kendall test. The homogenization software used was the R package, climatol.

Keywords: completeness, homogeneity, monthly air temperature, principal component analysis, climatological normal

[ PDF]

14.  Krušić, J., M. Marjanović, M. Samardžić-Petrović, B. Abolmasov, K. Andrejev and A. Miladinović: Comparison of expert, deterministic and Machine Learning approach for landslide susceptibility assessment in Ljubovija Municipality, Serbia.

Landslide Susceptibility Assessment is becoming a very productive research area, wherein different modeling approaches are practiced to delineate zones of the high-low likelihood of landslide occurrence. However, there is no strong consensus on which approach is the most adequate. The reason behind the lack of the general view on the performance of different approaches could be partially explained by the particularity of each study. To evaluate the efficiency of different approaches they need to be applied under the same conditions for the same study area. Herein, we examined three different approaches, including expert, deterministic and Machine Learning, on the example of Ljubovija Municipality in western Serbia. The study area has been known as susceptible to landslides, and represents good ground for assessing the chosen methods. It is represented by complex geology, prone to landslides that are commonly hosted in thick weathering crust of Paleozoic formations, composed of schists and meta-sediments. Under extreme triggering conditions, such as the one that unfolded in May 2014, these thick weathering crusts saturate, and give way to a variety of landslide and flash-flood processes that we will be focusing on in this study. The application of the expert-approach, through Analytical Hierarchy Process provided a rough assessment map. The deterministic model, which couples simple infinite slope and hydrological model, provided us with lower quality results, when compared to the expert-based one. This could be explained by the assumptions used in the model are too simplistic to generically model a wide range of landslide typology. Finally, Machine Learning approach, using the Random Forest algorithm, provided significantly better results and showed that it can cope with versatile landslide typology over larger scales. Its AUC performance is about 0.75 which is considerably outperforming the AUC values of the other two models, which were up to 0.55, i.e. at the level of random guess.

Keywords: landslide susceptibility, Analytical Hierarchy Process, deterministic, Machine Learning, Random Forest

[ PDF]

15.  Dumitrescu A., M.-V. Birsan and I.-A. Nita: A Romanian daily high-resolution gridded dataset of snow depth (2005-2015).

This study presents the spatial interpolation procedure from snow depth measurements at weather stations implying the following stages: (1) Spatial interpolation at 1 km × 1 km resolution of the mean multiannual values (2005-2015) corresponding to each month, computed from the data extracted from the climatological database; (2) Computation of the daily deviations against the multiannual monthly mean for every day and year over 2005-2015 and their spatial interpolation; (3) Spatio-temporal datasets were obtained through merging the two surfaces obtained in stages 1 and 2. The anomalies were considered to be the ratio between the daily snow depth values and the climatology. The spatial variability of the data used in the first stage was accounted for through the use of a series of predictors derived from the digital elevation model (DEM). To plot the maps with the climatological normals (multiannual means), the Regression-Kriging (RK) spatial interpolation method was used. In order to choose the optimum method applied in spatializing deviations, four interpolation methods were tested using a cross-validation procedure: Multiquadratic, Ordinary Kriging (separated and pooled variograms) and 3d Kriging.

Keywords: snowpack, spatial interpolation, Kriging, multiquadratic, cross-validation, Romania

[ PDF]

16.  Todorović Drakul, M., M. Samardžić Petrović, S. Grekulović, O. Odalović and D. Blagojević: Modelling extreme values of the total electron content: Case study of Serbia.

This paper is dedicated to modeling extreme TEC values at the territory of Serbia. For the extreme TEC values, we consider the maximum values from the peak of the 11-year cycle of solar activity in the years 2013, 2014 and 2015 for the days of the winter and summer solstice and autumnal and vernal equinox. The average TEC (Total Electron Content) values between 10 and 12 UT (Universal Time) were treated. As the basic data for all processing, we used GNSS (Global Navigation Satellite System) observation obtained by three permanent stations located in the territory of Serbia. Those data, we accept as actual, i.e. as a "true TEC values". The main objectives of this research were to examine the possibility to use two machine learning techniques: neural networks and support vector machine. In order to emphasize the quality of applied techniques, all results are adequately compared to the TEC values obtained by using International Reference Ionosphere global model. In addition, we separately analyzed the quality of techniques throughout temporal and spatial-temporal approach.

Keywords: TEC, machine learning, ionosphere, GNSS, IRI

[ PDF]