**3. Results**

Agricultural landscapes dominate in Lithuania. Land-use types contributing the most to the carbon accumulation in the LULUCF sector (forest, producing land, and meadow/pasture) covered, in 1971, rather similar proportions of the country's area—each around 28–30%. The areas of other land uses accounted for less than 12%. Even though Lithuania is a small country, the land-use proportions in different parts of the country differed. Additionally, if taking into consideration only two years, i.e., 1971 and 2015, one could state that the areas of forest, producing land, and built-up land did increase, while the proportion of meadow/pasture, wetlands, and other land decreased. However, the trajectories of specific land-use development during shorter periods experienced notable changes.

Even though there is no statistically significant global autocorrelation in values of forest proportion in Lithuanian municipalities, the southeastern and western parts of the country are more forested (Figure 3). Lower forest proportions are found in northern and central municipalities, where producing land dominates. The global spatial autocorrelation of agricultural land proportions in Lithuanian municipalities—both producing land and pasture/meadow—was statistically significant at practically all the time points used for the analysis. Producing land dominated in the northern and central municipalities, with lower forest proportions. Larger proportions of pasture/meadow were reported in municipalities with higher forest proportions, but not along the southeastern border of the country with overall forest dominance. The proportions of other land uses in Lithuanian municipalities are notably lower and usually do not exhibit global spatial autocorrelation. The Anselin Local Moran's *I* statistic was used to explore the spatial clusters of features with high or low values, as well as the spatial outliers. Two clusters of municipalities with low proportions of forest area that were stable over time and neighboured by municipalities with low values were identified. They practically overlapped with the high–high clusters of producing land abundance. It should be noted that the high–high cluster of producing land proportion in the northern part of Lithuania was the highest one among all clusters identified in this study, made up of 4–7 municipalities. This cluster also overlapped with the low–low cluster of pasture/meadow. Municipalities in the eastern part of Lithuania made up the low–low cluster of producing land proportions, which partly overlapped with a high–high cluster of wetlands that was stable over time. A high–high cluster of meadow/pasture was identified in the western part of the country, in the lowland associated with the Nemunas Delta area. Spatial outliers were usually small, i.e., including just one municipality and associated with municipalities with forest proportions that were different from their neighbourhoods. Local clusters of proportions of built-up areas were also small and dispersed throughout the whole country. Local spatial clusters and outliers of other land exhibited rather random occurrence patterns over time; however, the low proportions of that land-use type in the municipalities should be kept in mind.

The areas of forest and built-up land increased in Lithuania since 1971, while the areas of producing land, pasture/meadow, wetlands, and other land went down—this is suggested by, respectively, the positive and negative values of the slope of the linear trend (Table 1). Stable development trajectories were followed by the proportions of forest, wetland, built-up land, and other land during the whole period under assessment; however, the areas of producing land and pasture/meadow did both increase or decrease during specific periods. Thus, the areas of producing land were increasing at the cost of a decrease in pasture/meadow from 1971 to 1990. By the end of this period, the area of producing land was at its highest level—36%. The area of producing land decreased since 1990, with the proportion of pasture/meadow increasing to be level with the areas of key agricultural land-use types in 2005, at a level of 28%. Finally, the trajectories as they were since 1971 were repeated after 2005.


**Table 1.** Trends of change in proportion of land-use types across the whole of Lithuania (significance level of slope: \*\*\*, 0.001; \*\*, 0.01; and \*, 0.05).

**Figure 3.** Area proportions of land-use types in Lithuanian municipalities during different periods since 1971. Statistically significant values of Global Moran's *I* statistic are in bold. Linear shades identify statistically significant hotspots, cold spots, and spatial outliers based on the Anselin Local Moran's *I* statistic.

Furthermore, the spatial patterns of changes in three land-use types in Lithuanian municipalities were analysed, i.e., forest, producing land, and pasture/meadow, and are presented in Figure 4. The slope of the linear trend of forest proportion changed both during the whole period (1971–2015) and in all three shorter spans in an interval between –0.5 and 0.5, suggesting rather slow development. Statistically significant global spatial autocorrelation in the slope values was observed only for 1971–1990. Even though the slope values were low, there were some spatial clusters and outliers identified, such as the low–low cluster suggesting aggregation of municipalities with decreasing forest proportion during 1971–2015 in the central part of the country and some southwestern municipalities since 1990 or the high–high cluster in 1971–1990 in municipalities along the border of the former Soviet Union and Poland. The slope of a linear trend for the development of forest proportion in 1971–2015 was statistically significant in practically all the municipalities. However, if shorter periods were taken into consideration, usually only positive slope values were statistically significant at the level of the municipality. The trends of producing land changes in the municipalities were inverse to the ones of pasture/meadow. This refers both to the value of the slope of linear trend and the types and the location of spatial clusters. The areas of producing land increased most intensively in 1971–1990 in the eastern and western parts of the country, resulting in statistically significant global spatial autocorrelation and local spatial clusters. However, since the restoration of independence in Lithuania in 1990, the proportion of producing land started to decrease, with the most intensive drop in the municipalities, where the increase was faster before 1990. Opposite trends could be reported for the development of pasture/meadow. Finally, since 2005, agricultural land uses changed their trajectories once again. Even though there is no statistically significant global autocorrelation in the value of the slope for the proportion of producing land—the area of this land-use type was increasing practically all municipalities, with some small spatial clustering effects—the decrease in pasture/meadow was faster in the central part of Lithuania (with the highest global Moran's *I* statistic among all the cases estimated). If the whole period of 1971 to 2015 is taken into consideration, the value of the linear slope for producing land and pasture/meadow was usually statistically nonsignificant for most of the municipalities, suggesting large fluctuations in land-use type proportions over the time. However, if taking into consideration shorter periods, the slope of linear trend was statistically significant in the majority of municipalities—e.g., for 1971–1990, there were just six municipalities with nonsignificant slope values for both producing land and pasture/meadow, or 10 and eight municipalities, respectively, for the period 1990–2005.

To explain the land-use change trends in Lithuanian municipalities during different periods of the last half-century, we used information available from different GIS databases and multiple linear regression. If taking into consideration the whole period (1971–2015), the best explained variable was the slope of steadily increasing forest proportion (Table 2). The best regression models explained 65% of the variance of the slope of forest proportion changes. The figures for producing land and pasture/meadow were, respectively, 40% and 37%. When considering a shorter period, the percentage of variance explained by forest change models decreased but increased in models for meadow and pasture. In the case of producing land, the coefficient of determination only increased in 1971–1990.

**Figure 4.** Slope of linear trend in changes of area proportions of main land-use types in Lithuanian municipalities during different periods since 1971. Statistically significant values of Global Moran's *I* statistic are in bold. Linear shades identify statistically significant hotspots, cold spots, and spatial outliers based on the Anselin Local Moran's *I* statistic. Dotted areas identify the statistical significance of the slope in a certain municipality.

The proportion of the time that each candidate explanatory variable was detected to be statistically significant, testing all potential combinations of variables, is illustrated in Figure 5. Usually, there were more variables with larger significance when modelling the change trends of forest. The abundance of land-use types in the beginning of each analysed period was among the most significant factors in most of the tested cases. Forest proportion in the municipality also had an impact on the development trends of other land uses. Soil productivity was another factor often present in the models. Terrain-related attributes played a more important role in forest and grassland change models. Topographic details participated in forest change models. It should be noted that the impacts of explanatory variables were similar in the forest and grassland change models but opposite when modelling the producing land development.





*Land* **2021**, *10*, 619

**Figure 5.** The proportions of times that each candidate explanatory variable was statistically significant when testing all potential regression models. Blank means that the variable was not included in the exploratory regression runs.
