An operational land cover and land cover change toolbox: processing open‐source data with open‐source software

Abstract

Accurate and up‐to‐date land cover maps are vital for underpinning evidence‐based landscape management decision‐making. However, the technical skills required to extract tailored information about land cover dynamics from these open‐access geospatial data often limit their use by those making landscape management decisions. Using Dartmoor National Park as an example, we demonstrate an open‐source toolkit which uses open‐source software (QGIS and RStudio) to process freely available Sentinel‐2 and public LiDAR data sets to produce fine scale (10 m2 grain size) land cover maps. The toolbox has been designed for use by staff within the national park, for example, enabling land cover maps to be updated as required in the future. An area of 945 km2 was mapped using a trained random forest classifier following a classification scheme tailored to the needs of the national park. A 2019 land cover map had an overall user's accuracy of 79%, with 13 out of 17 land cover classes achieving greater than 70% accuracy. Spatially, accuracy was related via logistical regression to blue band surface reflectance in the spring and topographic slope derived from LiDAR (1 m resolution), with greater accuracy in steeper terrain and areas exhibiting higher blue reflectance. Between an earlier (2017–2019) and later (2019–2021) time frame, 8% of pixels changed, most of the change by area occurred in the most common classes. However, the largest proportional increase occurred in Upland Meadows, Lowland Meadows and Blanket Bog, all habitats subject to restoration efforts. Identifying areas of change enables future field work to be better targeted. We discuss the application of this mapping to land management within the Dartmoor national park and of the potential of tailored land cover and land cover change mapping, via this toolbox, to evidence‐based environmental decision‐making more widely.

Mapping the Land Suitability Rating of Arabica Coffee Crops: A Geographical Indication Factor-Based Approach

“…It is highly expected that this research will contribute to the development of GIS, and the management of data and information generated can help map Arabica coffee land development planning logically and sustainably to increase the income of coffee farmers and surrounding communities while increasing regional income. This also answers the importance of the chosen research location because the existing phenomenon that the higher the area, the lower the income level of farmers can be refuted [12][13][14][15].…”

“…2, used in field surveys to identify and inventory biophysical and non-biophysical environmental conditions so that data or parameters in geographical indication units can be separated appropriately and correctly. The technical skills required to extract information relevant to land cover dynamics from geospatial data [13].…”

Rapid, repeatable landscape‐scale mapping of tree, hedgerow, and woodland habitats (THaW), using airborne LiDAR and spaceborne SAR data

“…Finally, features known to result in false positives (via iterative examination of trial output data) were removed as a raster mask using other spatial land cover datasets, including mapping of buildings and scree/outcrop rock formations (Gatis et al, 2022). These features were buffered and co‐processed to autonomously remove these false positives from the finalized dataset (Figure 2‐G).…”

“…These areas form a contiguous landscape block covering more than 3800 km 2 . This extent has an elevation range between sea level and 621 meters and demonstrates a diversity of landscape character types and habitats including urban, intensively managed farmland, commercial forestry, primary travel corridors, and open upland moorland with blanket bog habitats (Gatis et al, 2022). As such, this area provides a sufficient breadth of land cover types to test the robustness of the mapped THaW habitats and the detection of canopy change.…”

Multi-Stage Semantic Segmentation Quantifies Fragmentation of Small Habitats at a Landscape Scale

“…In this study, we have demonstrated that Machine Learning LC annotation was able to efficiently scale to large areas; after training and testing on just over four square kilometres of LC image patches, we could predict the LC of 1439 km 2 of land within days on a desktop computer (Figure 8). Compared to the visual interpretation approach [9,10] or low-resolution automated LC maps [21], our method facilitates a range of new applications that require detailed knowledge of the landscape's land cover [53]. For example, these include fire risk modelling [8], climate change vulnerability assessments [4], tree planting planning [54], biodiversity monitoring [55,56] and habitat mapping [57][58][59].…”