The article analyses the seasonal nitrogen variation in drainage water, when the plants in the field crop rotation are fertilized with slurry during different seasons. The investigations were carried out in 2001-2003 in Juodkiškis experimental site of the Lithuanian Water Management Institute. The investigations established that the largest amounts of nitrogen are leached out in spring and in winter. In the autumn fertilized variant 38.8% more of this element was leached out in winter and spring compared with the variant fertilized in spring. During autumn nitrogen leaching was also 21% higher from the variant fertilized in autumn. It was established that the fertilization rate and dissolved inorganic nitrogen (DIN) supply in soil have influence on the concentrations of this element. During the cold season nitrogen concentrations in drainage water, when plants had been fertilized with slurry in spring, were more affected by the supply of dissolved inorganic nitrogen in soil more compared with the rate of fertilization; and if fertilization had been performed in autumn - the concentrations were more affected by the fertilization rate. In warm season both the fertilization rate and the supply of dissolved inorganic nitrogen in soil had similar influence on the concentrations of nitrogen in the drainage water in both treatments. Meteorological conditions affect nitrogen leaching a lot. During the cold season a greater amount of nitrogen is leached out when the air temperature is higher and during the warm season - when more precipitation falls.

The main objective of this study is to demonstrate the use of artificial neural network (AN) modelling tool to predict the risk of phosphorus (P) loss from the fields to nearest water body. The attention is drawn to AN as an alternative approach to the P index calculation for prediction of the P losses. The specific tasks of this study were to determine risk classes of P loss by linking together source and transport factors that accelerate P losses and to evaluate AN model performance for predicting risk classes via nutrient transport. AN was trained with a Levenberg-Marquardt algorithm, and Scaled Conjugate Gradient algorithm was used to estimate the possible risk of P losses from agricultural land. Two small agricultural watersheds in Auce and Bauska were chosen to determine field parameters, and expert's evaluation was used for description of the risk classes' of P loss. Finally these values were used as inputs for the neural network model. The model was trained and validated by assessing its predictive performance on a testing set of data excluded from the training set. The research results highlight the capabilities of AN to predict risk for a particular field and suggest that future research on application of other algorithms is required.

The paper deals with monitoring results of nitrate nitrogen (NO3 --N) run-off in three small agricultural catchments in Latvia (Berze, Mellupite, and Vienziemite) during the period of 1995 - 2007. Continuous flow measurements and water sampling were carried out in two scales - catchment and drainage field. Water quality data was analyzed statistically to identify outliers at various intensity agricultural production systems. The results indicated that with increase of agriculture intensity outlying values are higher and scattered from the rest of the data set thereby the risk of NO3 --N leaching is higher. It can be explained by application of different rates of organic and inorganic fertilization. To analyze water discharge data, cumulative distribution was used. The results show that main part of the water discharge is observed from late autumn to spring, whereas in summer period it is low and stable. The dependence of NO3 --N concentrations on the discharge is expressed by Spearman's correlation coefficient - at catchment scale it is 0.37 in Vienziemite site, 0.39 - in Berze, and 0.44 in Mellupite. Calculated correlation coefficients are statistically reliable.

The investigations were carried out in the Lithuanian Agricultural University Water Research Institute land plots in the Endocalcari Endohypogleyic Cambisols (CMg-n-w-can). The basis of the investigation is 3 variants field experiment. Each variant consists of 3 in 0.54 ha drainage. The traditional arable farming is applied in variant I. In the variant II the land is not being cultivated, but in spring the perennial ryegrass (‘Lolium perenne L’) is being seeded into the spring barley and kept till spring. In the variant III the land is not being cultivated after the harvest and left for the rest till spring. The variant II is distinguished by the minimal mineral nitrogen content. Applied to cultivated and uncultivated land, the min N reserves are increased 51 - 83 and 33 - 40 and 11 - 101 and (38 - 134%) (to 9.5 - 14.3 mg kgE-1 and 152 - 68 and 154 - 61 kg haE-1). The average investigation of N concentration in the drainage water shows, that the minimum concentration of this element was in the second variant. Applied to the traditional farming and uncultivated land, the N concentration is increased by (30 - 42% to 34 and 37 mg lE-1). By average data the min N, leaching by drainage water in the variant II was minimal and about 27 kg haE-1. Applied to the arable farming and uncultivated land, the min N leching is increased (30 - 55%) (to 35 - 42 kg haE-1).

Yasso07 soil carbon model was used to estimate soil carbon balance in dry forest site types (6 site types in total) in Latvia and the results were compared with data from Biosoil2012 soil surveys. Litter input, chemical quality and climatic data are required to run the model. Three different scenarios were used for climate data input – steady climate, climate change + 0.025 °C annually and climate change + 0.05 °C annually. Forest mineral soil is a carbon sink for the whole modelled period - the years of 1990 – 2030. Under steady climate, the average carbon removal is 0.6 t CO2 haE-1 yrE-1, under climate change (+ 0.025 °C) scenario 0.4 t CO2 haE-1 yrE-1, but under climate change (+ 0.05 °C) scenario 0.3 t CO2 haE-1 yrE-1. CO2 removal at the beginning of the period (1990) was 0.35 – 0.38 t CO2 haE-1 yrE-1. Carbon stock modelled with Yasso07 is lower than estimated in Biosoil2012 soil surveys. Differences between modelled and Biosoil2012 results vary from 2 t C haE-1 in the poorest and 41 t CO2 haE-1 in the third poorest site type. Carbon stock modelled with Yasso07 increases from the poorest to the most fertile site type while Biosoil2012 shows an increase from the poorest to the third poorest, and a decrease from the third poorest to the most fertile site type. Underestimation and different trends between Yasso07 and measured carbon stock may be explained by inappropriate equations and models used to estimate non-woody biomass. It is necessary to improve accuracy of input data for non-woody biomass by elaborating national equations and models in order to include Yasso07 in the national GHG inventory.