**4. Discussion**

The overall trend in Lithuania during the last five decades has been increases in the areas of forest and built-up lands and decreasing areas of producing land, meadow/pasture, wetlands, and other land uses. Nevertheless, the development trends for the proportions of producing land and meadow/pasture changed their trajectories several times. The breakpoints in the development of key agricultural land uses were linked with important dates in Lithuanian history. This suggests that the land-use development trends could be impacted by political processes in and around the land managemen<sup>t</sup> and use relationships. Three periods were singled out with potentially differing land-use conditions. The first period (1971–1990) we associate with the development of large agricultural enterprises under the condition of a planned economy, as Lithuania was one of the former Soviet Union republics. Restructuring of agriculture started in 1991–1992. The reform of the national agrarian sector took place since the restoration of independence, which resulted in introducing private land ownership, together with changed overall principles of agriculture and land management. This was followed by a period of European Union and state budget support allocated to agriculture and rural development, since Lithuania joined the EU in 2004 [50–52]. The stable overall increase in forest could be explained by command-andcontrol forest governance restricting radical changes, strict deforestation control, and the aspiration to preserve domestic forest resources [22,53,54]. Thus, the trends observed in our study are associated, first of all, with political and social factors rather than natural conditions. This is supported by Ribokas and Milius [55], who argued that nearly all legal, economic, and social land managemen<sup>t</sup> reforms in Lithuania were neither consistent nor unambiguous.

Even though Lithuania is a relatively small country with rather smoothly changing geographic conditions, we could still observe statistically significant patterns in the landuse distribution and changes. The increase in forest was largest in southwest Lithuania, potentially due to the fast increase during 1971–1990. Since 2005, however, forest increased the most in northeastern Lithuania and the hilly municipalities in the western part of the country. We explain this by the intensive afforestation of abandoned land or land not used for agriculture. The trajectories of producing land development were different during the periods analysed. If taking into consideration the last five decades, the overall decrease in producing land in the hilly areas of western and eastern Lithuania could be explained by the fast decrease in producing land in 1990–2005. These areas are less favourable for agriculture, and the presence of abandoned agricultural land is more common here. However, the development of producing land proportion was radically different in these areas during other periods, i.e., 1971–1990 and 2005–2015. Development trajectories of meadow/pasture were, at least in principle, the opposite to those of producing land. The most rapid reduction in meadow/pasture during the whole period analysed was in the flat central and northern municipalities with the most fertile soil for agriculture. The fastest decrease in meadow/pasture was seen here since 2005. Usually, producing land is converted into meadow/pasture, and vice versa. Similar changes were also noted by Aleknaviˇcius [56], who reported that the area of producing land in Lithuania decreased by, on average, 18,900 ha annually in 1948–1989 and by even more—51,800 ha—during 1990–2005, with large areas of producing land converted into meadow/pasture. The total area of agricultural land was reported to have shrunk by 2.35% during 2007–2017 [57]. The decreasing area of agricultural land was explained by increasing forest and new housing areas, especially in hilly western regions [58,59]. The forest area of Lithuania is reported to have increased during the period since 1950 [41,60]. Usually, the largest increase in forest proportion is found in regions least favourable for agriculture. The largest areas of new forests emerged in southeastern Lithuania, while the slowest increase om forest was in the least forested municipalities. Some forest loss was also reported [41] since the 1950s, associated with forest transformation into agricultural land, or less frequently into scrubland or water bodies. The latter transformation was related with the construction of large artificial reservoirs. It should be noted that all of the national studies mentioned above, except for Jukneliene and Mozgeris [ ˙ 41], did not use spatial statistics to support their findings on land-use distribution patterns. Similar forest and agricultural land changes were reported in neighbouring countries, e.g., in Poland [61].

The available land use and land-use change patterns are usually associated with interactions between socioeconomic and cultural land managemen<sup>t</sup> conditions, biophysical constraints, and land-use history [62]. To specify the interactions, we have chosen the multiple regression. Our focus in the current study was on the characterization—or, at least, identification—of the most important biophysical and socioeconomic drivers of land use in Lithuania. Usually, the candidate drivers are suggested based on a literature review and expert knowledge. We introduced one extra criterion: the driver needs to be described using easily available data. In addition to census data, we gathered study information available from the Spatial Information Portal of Lithuania. The majority of such spatial information was captured during the last few decades; thus, this could have impacts on the performance of the regression models developed for the earlier periods covered in our study. The best regression model, in terms of *R*2, was developed to explain the changes in forest proportion during the whole period (i.e., 1971–2015). However, the development of forest was very smooth during the whole period. Shorter periods resulted

in better performance of the regression models if modelling the proportion changes of meadow/pasture and, partly, the proportion of producing land. In all cases, the Akaike Information Criterion values for models with a shorter time period were higher than those for the land-use change from 1971 to 2015. In addition to the availability and quality of historical explanatory driver variables, multiple regression in land-use change analyses can be used for relatively short time periods, i.e., one or two decades [63]. We should also emphasise that we did not aim to elaborate the overall best regression models, i.e., the focus was on testing all potential driver variables in all potential combinations, taking into consideration, of course, the statistical significance and multicollinearity of factors and properties of model residuals.

If taking a closer look at the performance of each tested candidate driver variable, the importance of the forest proportion at the beginning of each period stands out. We could consider the abundance of forest in the municipality as a key indicator of landscape stability [64]. In 2019 forest covered 33.7% of Lithuania [60], and a political objective was set to increase this figure to at least 35% by the year 2030 [65]. Assuming that the annual forest area increase rate during the period from 1971 until 2015 was 0.085% (0.108% during the last decade), this objective could be achieved by increasing the country's forest area by at least 0.118% per year. This challenging task would impact the development of other land uses, both considering the models suggested in the current study and the practice of afforesting abandoned or unsuitable agricultural land [65]. We identified the soil productivity grade as an important factor shaping land-use changes, even though there was some scepticism regarding using the crop production potential of the land for exploring land-use change patterns [66,67]. Soil productivity grade was most strongly correlated with the change trends of producing land and meadow/pasture proportion (Table 3). It was a statistically significant contributor in models explaining, e.g., forest changes (the factor was significant in 69% and 61% of all cases tested for the periods 1990–2005 and 2005–2015, respectively) and grassland changes (98% and 97%). Population is usually reported as an important factor influencing land-use distribution [68–74]. We did not directly use the statistics on, e.g., the ratio between the urban and rural population; however, we integrated the factors that were used to specify the rural population in the recent FP7 RURALJOBS project [47]. However, neither population density nor the share of population within a specified driving distance of cities was found to be among the most important factors. The reason could also be the reference date of the population data—e.g., the population density in 2011 was a significant factor in nearly 70% of cases tested to describe forest area changes after 2005. Land reclamation is considered an important factor that has been shaping Lithuanian landscapes in the second half of the 20th century [75–77]. It should be emphasized that the facilities available for land reclamation in Lithuania influence the land use—e.g., afforestation of agricultural lands, is dependent on the presence or absence of land with a functioning land reclamation system [78]. In our study, the intensity of land reclamation in the municipality is an important factor for explaining changes in producing land and meadow/pasture. The topography of the landscape is usually closely related to the land use and land-use change patterns [62,79,80]. However, this attribute is scale-dependent; thus, relatively coarse-scale elevation data sources were used to reveal the general trends. Even though Lithuania can be characterised as a lowland country (cf. Figure 1), there are differences in the land use and land-use change patterns observed between the hilly and relatively flat municipalities. Topography-related factors are, therefore, more effective at explaining changes in agricultural land. In Lithuanian municipalities, the soil productivity is inversely correlated with the average altitude ((Pearson's correlation coefficient −0.579 (*N* = 51))), slope steepness (Pearson's correlation coefficient −0.552 ( *N* = 51)), and diversity of elevation conditions, expressed as a standard deviation of altitude (Pearson's correlation coefficient −0.333 ( *N* = 51)) or slope steepness (Pearson's correlation coefficient −0.510 (*N* = 51)). The land-use change transitions usually involve conversion from producing land into meadow and pasture or vice versa, usually on land less suitable for growing crops.


