Land use and cover hotspot analysis and change detection in Belarus

Note: This project was completed with the support of the Alliance of International Science Organizations (ANSO) Joint Research Cooperation Program under the Belt and Road Initiative.

Project Background

Data preparation and fusion were performed using Sentinel Hub and Google Earth Engine. Training samples were labeled using publicly available datasets, publications, and visual interpretation. The supervised learning method OBIA-RF (Object-Oriented Random Forest) was then used to classify and extract land cover features from the agricultural and forestry datasets. Finally, mapping and statistical analysis were conducted. Soil ecological potential was assessed by combining meteorological and soil data.

This article mainly introduces the use of the method mentioned in the previous article. Belarusian land use classification based on object-oriented extractionFurther hotspot analysis and change detection of LULC (land use and land cover) data results are conducted to identify land use change hotspots and change types.

△Figure 1. Flowchart of Land Use Classification Hotspots and Change Detection

Basic Project Process

After obtaining LULC data from 2014 to 2023 using object-oriented land use classification, the area of ​​the land use polygon data was calculated and the center point was extracted. The area was used as the weight value, and spatial kernel density analysis was applied to the center point of each polygon to comprehensively evaluate the number and area of ​​various land use distribution polygons within a certain search distance, and the distribution hotspots of forest and farmland were obtained respectively.

Next, the quantity and types of changes in land cover types were determined, and a mapping table was established, as shown in the diagram above. These changes were categorized into six types: forest loss, farmland loss, forest growth, farmland growth, forest to farmland conversion, and farmland to forest conversion. A reclassification table was then created based on these categories, and statistics on changes over every two years were compiled to produce a change distribution map.

Project Results

△ Figure 2. Distribution of arable land hotspots in Belarus, 2014–2023

△ Figure 3. Distribution map of forest hotspots in Belarus, 2014–2023

Comparing Figures 2 and 3, a clear spatial clustering of arable land is observed. It is mainly concentrated in southwestern Minsk, Grodno, western Brest, Mogilev, and eastern Gomel. Arable land is scarce in Gomel and northern and southern Vitebsk. Forest distribution is relatively even compared to arable land, with only a few hotspots, indicating that forest resources are widely distributed throughout the region, with some large, contiguous forest areas.

Agriculture in Belarus is primarily concentrated in fertile plains, particularly in the south and central regions, which are better suited for large-scale mechanized farming. Consequently, agricultural plots in these areas generally exhibit high continuity and consolidation, with typical Belarusian farms (especially state-owned farms) being quite large, often ranging from hundreds to thousands of hectares. This aligns with the distribution of arable land hotspots as depicted in the arable land hotspot map.

From 2014 to 2019, the distribution of arable land hotspots remained relatively stable with little inter-year variation. However, between 2020 and 2023, the hotspots showed significant spatial shifts. This may be due to several factors: a substantial increase in precipitation in Belarus in recent years, leading to fewer cloudless days throughout the year, which affected satellite imagery quality. Furthermore, variations in precipitation and temperature resulting in changes in phenological periods reduced the consistency of crop types and growth stages, causing classification errors and consequently affecting kernel density statistics. In the long term, the distribution of agricultural resources in Belarus remains relatively stable. Changes in forest resources are even less pronounced; forest area has shown only minor fluctuations over the past decade, with both total area and spatial hotspot distribution remaining relatively stable.

△ Figure 4. Distribution map of land cover type changes in Belarus from 2014 to 2023

The above figure shows the changes in land use and cover types every two years from 2014 to 2023. The changes in forests and cultivated land were relatively small from 2014 to 2016, mainly involving a decrease and increase in forest area in some regions, and a small amount of farmland-forest conversion in urban areas. A larger-scale farmland-forest conversion occurred from 2016 to 2018. This may be related to slightly less accurate classification of low shrubs, complex crop rotation areas, and some economic forests, as well as the impact of data normalization across different years, which confused the classification results and led to changes in land cover. However, overall, it still reflects the key areas of land cover type change and the trends over a long period. Combined with annual area statistics, it can reflect a relatively objective distribution of forest and farmland resources, providing an objective and effective reference for the formulation and implementation of macro and regional agricultural policies.