Oleg Antonić

Spatio-temporal interpolation of climatic variables over large region of complex terrain using neural networks

Empirical models for seven climatic variables (monthly mean air temperature, monthly mean daily minimum and maximum air temperature, monthly mean relative humidity, monthly precipitation, monthly mean global solar irradiation and monthly potential evapotranspiration) were built using neural networks. Climatic data from 127 weather stations were used, comprising more than 30 000 cases for each variable. Independent estimators were elevation, latitude, longitude, month and time series of respective climatic variable observed at two weather stations (coastal and inland), which have long time-series of climatic variables (from mid last century). Goodness of fit by model was very high for all climatic variables (R>0.98), except for monthly mean relative humidity and monthly precipitation, for which it was somewhat lower (R=0.84 and R=0.80, respectively). Differences in residuals around model were insignificant between months, but significant between weather stations, both for all climatic variables. This was the reason for calculation of mean residuals for all stations, which were spatially interpolated by kriging and used as a model correction. Similarly interpolated standard deviation and standard error of residuals are estimators of the model precision and model error, respectively. Goodness of fit after the averaging of monthly values between years was very high for all climatic variables, which enables construction of spatial distributions of average climate (climatic atlas) for a given period. Presented interpolation models provide reliable, both spatial and temporal estimations of climatic variables, especially useful for dendroecological analysis.

DEM-based depth in sink as an environmental estimator

The sink as a geomorphological feature can modify a spatial distribution of soil moisture due to the water stagnation in the flood-plain region or spatial distribution of air temperature and humidity (air confluence effect) as well as soil depth in the karstic region. A new variable based on digital terrain model, depth in sink, is introduced as an environmental estimator. The ability of the new variable to explain spatial distributions of the relative presence of dominant tree species is tested in the case study for a karstic region (Europaean Silver Fir, Common Beech, Norway Spruce and Mountain Pine) and for a flood-plain region (Pedunculate Oak and Narrow-leaved Ash), both in Croatia. The general linear modelling method was used. Depth in sink was the significant estimator of the spatial distribution for all tested tree species.

Spatial distribution of main forest soil groups in Croatia as a function of basic pedogenetic factors

The model of spatial distribution of main forest soil groups in Croatia was developed as a function of basic pedogenetic factors: lithological substratum, macroclimate and relief. Used data about soil group, lithological substratum, terrain slope and aspect were collected on 1881 soil profiles. Macroclimatic data were estimated for each soil profile by spatial interpolation between meteorological stations. Feedforward neural networks were used as modelling tool. The final model has total classification correctness of 63.5% for training data set and 62.3% for independent test data set. The best result (86.4%) was achieved for fluvisols which are strongly spatially correlated with alluvial sediment in a flood plains. The worst result was achieved for luvisol (14.2%) which mainly comprised very old soils, probably developed under pedogenetic factors different from actual. The model was applied on entire Croatian territory aiming at construction of potential spatial distribution of main forest soils (without human impact), which was compared by the potential spatial distribution of major forest types modelled independently.

Modelling groundwater regime acceptable for the forest survival after the building of the hydro-electric power plant

Abstract The reservoir of the future hydro-electric power plant ‘Novo Virje’ plant on Drava river, Croatia, could increase the absolute mean groundwater level in the nearby flood-plain forest. These changes could result in a decline of some forest species, especially the pedunculate oak ( Quercus robur L.), which is very sensitive to the groundwater level disturbance. Dendroecological study was undertaken with the objective of finding acceptable groundwater level regime from the perspective of survival of nearby flood-plain forest. The study was spatially intensive (GIS based) and executed in the following steps: (1) building of the model which describes basal area increment of the pedunculate oak as a function of environmental estimators (hydrologic, soil, climatic and tree competition variables); (2) simulation of increment by the model under different groundwater level regimes and comparison of the simulation results with predefined increment minimum; and (3) designing of the envelopes of the acceptable regimes (simulated increment is larger than predefined minimum). These results are the basis for the design of hydro-technical measures aiming at protecting of flood-plain forests after the hydro-electric power plant is built.

Estimating plant species occurrence in MTB/64 quadrants as a function of DEM-based variables: a case study for Medvednica Nature Park, Croatia

Abstract Croatia is among those European countries without an Atlas of Flora produced till today as a result of constant lack of greater number of active botanists and inconsistency in gathering data in the field. Recently, a standard for collection of data, based on the Central European MTB (abbreviation of German term “Mestischblatter” that stands for a sheet of topographic map) grid was proposed and tested in the field on the “Medvednica Nature Park” on Medvednica mountain near the city of Zagreb. Using the data collected in 97 MTB/64 quadrants (presence/absence of plant species), we tested the potential of estimating species occurrence at the proposed grid by models in a function of the Digital Elevation Model (DEM)-based variables, namely altitude, terrain slope, terrain aspect, and flow accumulation potential. Because of significant spatial variability of environmental factors within MTB/64 quadrants, each one was represented by descriptive statistics (median, 5-, 25-, 75- and 95-percentiles) of DEM-based variables. Thirty-seven plant species were selected arbitrarily, on the basis of their frequency in the studied area (40–60% of all quadrants). Three methods for development of predictive model were used and compared: discriminant analyses, logistic regression, and classification trees. Yielded results suggest that spatial modelling could be probably applied in flora mapping, which would optimise fieldwork. However, decreasing of mapping unit area is recommended, especially for rare species. For larger areas, inclusion of other environmental predictors (macroclimatic, lithological, landuse) in models is probably needed.

Relationship between MODIS based Aerosol Optical Depth and PM10 over Croatia

This study analyzes the relationship between Aerosol Optical Depth (AOD) obtained from Terra and Aqua Moderate Resolution Imaging Spectroradiometer (MODIS) and ground-based PM10 mass concentration distribution over a period of 5 years (2008–2012), and investigates the applicability of satellite AOD data for ground PM10 mapping for the Croatian territory. Many studies have shown that satellite AOD data are correlated to ground-based PM mass concentration. However, the relationship between AOD and PM is not explicit and there are unknowns that cause uncertainties in this relationship.The relationship between MODIS AOD and ground-based PM10 has been studied on the basis of a large data set where daily averaged PM10 data from the 12 air quality stations across Croatia over the 5 year period are correlated with AODs retrieved from MODIS Terra and Aqua. A database was developed to associate coincident MODIS AOD (independent) and PM10 data (dependent variable). Additional tested independent variables (predictors, estimators) included season, cloud fraction, and meteorological parameters — including temperature, air pressure, relative humidity, wind speed, wind direction, as well as planetary boundary layer height — using meteorological data from WRF (Weather Research and Forecast) model.It has been found that 1) a univariate linear regression model fails at explaining the data variability well which suggests nonlinearity of the AOD-PM10 relationship, and 2) explanation of data variability can be improved with multivariate linear modeling and a neural network approach, using additional independent variables.

A global model for monthly mean hourly direct solar radiation

This paper introduces a worldwide applicable model for estimating hourly portions of monthly mean daily direct solar radiation intercepted on the horizontal surface, as a function of actual solar elevation, daily maximum solar elevation and latitude. The model fits well (R>0.99) pyranometric data measured at stations from practically entire Northern Hemisphere (0<φ<70°). Application of the model using independent data sets, including one station from the Southern Hemisphere, leads to the acceptance of the model worldwide. Model is not sensitive to the local climate differences or interannual variability and it represents an ultimate solution for daily topographic solar radiation modelling without site-specific or latitude-specific hourly radiation data.

Long-term and country scale projection of even-aged forest management: a case study for Fagus sylvatica in Croatia

Abstract Out of the 901,848 ha forests of the European beech (Fagus sylvatica L.) in Croatia, 383,473 ha are even-aged beech forests with irregular age-class structure. Potential forest management of the even-aged beech forests (including the conversion of coppice even-aged beech forests) was studied using a dynamic model system, in the context of long-term planning at the national level in Croatia. Three management policy scenarios were defined regarding the area control method, that is, planned area for annual regeneration felling was given as a function of (1) total forest area (conservative approach scenario–CAS), (2) area of stands older than 50 years (moderate approach scenario–MAS), and (3) area of stands older than 80 years (intensive approach scenario–IAS). Projections of future trends in the annual area of regeneration, age-class distribution, standing volume and annual felling, and thinning cut were obtained for each scenario. Management according to the Scenario 3 would lead to achievement of a best age-class balancing and more sustainable management in the next 50 years. The structure of the growing stock and the quality of harvested timber would also improve as a result of this approach. Intensive regeneration felling (4460 ha per year) and conversion of coppice forests (420 ha per year) is required to reach this goal.

Designing laboratory wind simulations using artificial neural networks

While experiments in boundary layer wind tunnels remain to be a major research tool in wind engineering and environmental aerodynamics, designing the modeling hardware required for a proper atmospheric boundary layer (ABL) simulation can be costly and time consuming. Hence, possibilities are sought to speed-up this process and make it more time-efficient. In this study, two artificial neural networks (ANNs) are developed to determine an optimal design of the Counihan hardware, i.e., castellated barrier wall, vortex generators, and surface roughness, in order to simulate the ABL flow developing above urban, suburban, and rural terrains, as previous ANN models were created for one terrain type only. A standard procedure is used in developing those two ANNs in order to further enhance best-practice possibilities rather than to improve existing ANN designing methodology. In total, experimental results obtained using 23 different hardware setups are used when creating ANNs. In those tests, basic barrier height, barrier castellation height, spacing density, and height of surface roughness elements are the parameters that were varied to create satisfactory ABL simulations. The first ANN was used for the estimation of mean wind velocity, turbulent Reynolds stress, turbulence intensity, and length scales, while the second one was used for the estimation of the power spectral density of velocity fluctuations. This extensive set of studied flow and turbulence parameters is unmatched in comparison to the previous relevant studies, as it includes here turbulence intensity and power spectral density of velocity fluctuations in all three directions, as well as the Reynolds stress profiles and turbulence length scales. Modeling results agree well with experiments for all terrain types, particularly in the lower ABL within the height range of the most engineering structures, while exhibiting sensitivity to abrupt changes and data scattering in profiles of wind-tunnel results. The proposed approach allows for quicker and more effective achieving targeted flow and turbulence features of the ABL wind-tunnel simulations as compared to the common trial and error procedures. This methodology is expected to enable wind-tunnel modelers a quick and time-efficient designing of ABL simulations in studies dealing with air pollutant dispersion, wind loading of structures, wind energy, and urban micrometeorology, where atmospheric flow and turbulence play a key role.