An annual land cover dataset for the Baltic Sea Region with crop types and peat bogs at 30 m from 2000 to 2022

Abstract

We present detailed annual land cover maps for the Baltic Sea region, spanning more than two decades (2000-2022). The maps provide information on eighteen land cover (LC) classes, including eight general LC types, eight major crop types and grassland, and two peat bog-related classes. Our maps represent the first homogenized annual dataset for the region and address gaps in current land use and land cover products, such as a lack of detail on crop sequences and peat bog exploitation. To create the maps, we used annual multi-temporal remote sensing data combined with a data encoding structure and deep learning classification. We obtained the training data from publicly available open datasets. The maps were validated using independent field survey data from the Land Use/Cover Area Frame Survey (LUCAS) and expert annotations from high-resolution imagery. The quantitative and qualitative results of the maps provide a reliable data source for monitoring agricultural transformations, peat bog exploitation, and restoration activities in the Baltic Sea region and its surrounding countries.

Fig. 1

(a) The Baltic Sea region; (b) Thematic details of CORINE Land Cover and Continental Europe Land Cover (44 classes) compared to Baltic Sea Region Land Cover Plus (18 classes) maps for 2018; upper example: an area in Germany (center coordinate 52.89 N, 10.85E) dominated by agricultural land, which is oversimplified by existing LC products, whereas our maps reveal the diverse land use in agriculture; bottom example: our map distinguishes unexploited and exploited peat bogs in Estonia (center coordinate 58.54 N, 24.37E). High resolution images are taken from Google Earth.

Fig. 2

Overall workflow.

Fig. 3

(a) Annual satellite availability in the study area. (b) Annual clear sky observations (CSO) from 2000 to 2022.

Fig. 4

(Top): Temporal encoding process converts annual time-series information to 52 feature spaces in one band (NDVI is used as an example, scaling by 10,000). (Bottom): Examples of spectral NDVI feature spaces for three LC types: built-up, broadleaf forest and unexploited peat bog in three different years (2006, 2012, 2018). Red boxes in the high resolution image chips represent a 30 m × 30 m pixel over a corresponding land cover type.

Fig. 5

Simplified 1-Dimensional Convolutional Neural Network (1D-CNN) architecture for land cover classification with temporal encoding input. Details of the network architecture is provided in Supplement File 1.

Fig. 6

Post processing. (a) Spatial filtering (applying to both Level 1 and Level 2 maps); (b) Temporal filtering (for Level 1 maps); (c) Spatial-temporal filtering (for Level 2 maps).

Fig. 7

Examples of Level 2 land cover maps (2015) before and after post-processing.

Fig. 8

Annual crop statistics of Denmark from 2009 to 2022 compared to estimated areas from BSRLC+.

Fig. 9

Peat bog exploitation in Estonia over two decades (2000–2022). (Left): Visual assessments showed similar patterns of exploited peatbog between high resolution images from Google Earth and the classification from BSRLC+ in three different years: 2000, 2010 and 2020. (Right): Estimated peatbog exploitation by year derived from the maps.