Search

Estimates of the area of land cover classes or land change are frequently calculated from land cover classification maps by counting the pixels labeled as each class in the map. This procedure is known to produce biased estimates of area for many widely used classification algorithms, including random forests. Poststratification estimation using the mapped classes as strata has been proposed to obtain unbiased estimates of the class areas. Still, the method requires additional sampling units, which may not be available or be the most efficient method depending on the application. Alternatively, consistent estimates of class areas can be obtained using class membership probabilities estimates from a random forest classification. This article demonstrates that, for a large sample and proper set of explanatory variables, the error of the predicted class membership probabilities obtained from a random forest classification converges to zero. Therefore, the expected class areas calculated from these probabilities converge to the population class areas. On average, the relative error of the expected class proportions computed by class membership probabilities from a random forests model was 40% points lower than the proportions estimated by pixel counting. Our proposed approach is also comparable to the area-adjusted method, which is currently considered the best practice by the remote sensing community. We recommend that class probability estimates area always retained and used for calculating expected class areas or area proportions based on our results. Our method reduces bias compared to statistics calculated by pixel counting and circumvents the need for poststratification area estimates under certain conditions. [ABSTRACT FROM AUTHOR]

Reference data for large-scale land cover map are commonly acquired by visual interpretation of remotely sensed data. To assure consistency, multiple images are used, interpreters are trained, sites are interpreted by several individuals, or the procedure includes a review. But little is known about important factors influencing the quality of visually interpreted data. We assessed the effect of multiple variables on land cover class agreement between interpreters and reviewers. Our analyses concerned data collected for validation of a global land cover map within the Copernicus Global Land Service project. Four cycles of visual interpretation were conducted, each was followed by review and feedback. Each interpreted site element was labelled according to dominant land cover type. We assessed relationships between the number of interpretation updates following feedback and the variables grouped in personal, training, and environmental categories. Variable importance was assessed using random forest regression. Personal variable interpreter identifier and training variable timestamp were found the strongest predictors of update counts, while the environmental variables complexity and image availability had least impact. Feedback loops reduced updating and hence improved consistency of the interpretations. Implementing feedback loops into the visually interpreted data collection increases the consistency of acquired land cover reference data. [ABSTRACT FROM AUTHOR]

Highlights • Experts largely disagree on visually interpreted shadows. • A risk-based shadow detection method is presented. • Difference in shadow detected automatically and by the experts is within expectations. Abstract Land cover identification and area quantification are key aspects in determining support payments to farmers under the European Common Agricultural Policy. Agricultural land is monitored using the Land Parcel Identification System and visual image interpretation. However, shadows covering reference parcel boundaries can hinder effective delineation. Visual interpretation of shadows is labor intensive and subjective, while automated methods give reproducible results. In this paper we compare shadow detection on satellite imagery obtained by expert photointerpretation to a proposed automated, data-driven method. The latter automated method is a thresholding approach employing both panchromatic and multispectral imagery, where the former has a finer spatial resolution than the latter. Thresholds are determined from automatically generated training data using a risk-based approach. Comparison of the total shadow area per scene showed that more pixels were labelled as shadow by the automatic procedure than by visual interpretation. However, the union of shadow area independently identified by twelve experts on a subscene was larger than the automatically determined shadow area. The limited intersection of the shadow areas identified by the experts demonstrated that experts strongly disagreed in their interpretations. The shadow area labelled by the automated method was in between the intersection and the union of the areas interpreted by experts. Furthermore, the automated shadow detection method is reproducible and reduces the interpretation effort and skill required. [ABSTRACT FROM AUTHOR]

Global scale land cover (LC) mapping has interested many researchers over the last two decades as it is an input data source for various applications. Current global land cover (GLC) maps often do not meet the accuracy and thematic requirements of specific users. This study aimed to create an improved GLC map by integrating available GLC maps and reference datasets. We also address the thematic requirements of multiple users by demonstrating a concept of producing GLC maps with user-specific legends. We used a regression kriging method to integrate Globcover-2009, LC-CCI-2010, MODIS-2010 and Globeland30 maps and several publicly available GLC reference datasets. Overall correspondence of the integrated GLC map with reference LC was 80% based on 10-fold cross-validation using 24,681 sample sites. This is globally 10% and regionally 6–13% higher than the input map correspondences. Based on LC class presence probability maps, expected LC proportion maps at coarser resolution were created and used for characterizing mosaic classes for land system modelling and biodiversity assessments. Since more reference datasets are becoming freely accessible, GLC mapping can be further improved by using the pool of all available reference datasets. LC proportion information allow tuning LC products to specific user needs. [ABSTRACT FROM PUBLISHER]

Along with the creation of new maps, current efforts for improving global land cover (GLC) maps focus on integrating maps by accounting for their relative merits, e.g., agreement amongst maps or map accuracy. Such integration efforts may benefit from the use of multiple GLC reference datasets. Using available reference datasets, this study assesses spatial accuracy of recent GLC maps and compares methods for creating an improved land cover (LC) map. Spatial correspondence with reference dataset was modeled for Globcover-2009, Land Cover-CCI-2010, MODIS-2010 and Globeland30 maps for Africa. Using different scenarios concerning the used input data, five integration methods for an improved LC map were tested and cross-validated. Comparison of the spatial correspondences showed that the preferences for GLC maps varied spatially. Integration methods using both the GLC maps and reference data at their locations resulted in 4.5%-13% higher correspondence with the reference LC than any of the input GLC maps. An integrated LC map and LC class probability maps were computed using regression kriging, which produced the highest correspondence (76%). Our results demonstrate the added value of using reference datasets and geostatistics for improving GLC maps. This approach is useful asmore GLC reference datasets are becoming publicly available and their reuse is being encouraged. [ABSTRACT FROM AUTHOR]

Land cover identification and area quantification are key aspects of implementing the European Common Agriculture Policy. Legitimacy of support provided to farmers is monitored using the Land Parcel Identification System (LPIS), with land cover identification performed by visual image interpretation. While the geometric orthoimage quality required for correct interpretation is well understood, little is known about the photometric quality needed for LPIS applications. This paper analyzes the orthoimage quality characteristics chosen by authors as being most suitable for visual identification of agricultural fields. We designed a survey to assess users' preferred brightness and contrast ranges for orthoimages used for LPIS purposes. Survey questions also tested the influence of a background color on the preferred orthoimage brightness and contrast, the preferred orthoimage format and color composite, assessments of orthoimages with shadowed areas, appreciation of image enhancements and, finally, consistency of individuals' preferred brightness and contrast settings across multiple sample images. We find that image appreciation is stable at the individual level, but preferences vary across respondents. We therefore recommend that LPIS operators be enabled to personalize photometric settings, such as brightness and contrast values, and to choose the displayed band combination from at least four spectral bands. [ABSTRACT FROM AUTHOR]

Accurate and high-resolution land cover (LC) information is vital for addressing contemporary environmental challenges. With the advancement of satellite data acquisition, cloud-based processing, and deep learning technology, high-resolution Global Land Cover (GLC) map production has become increasingly feasible. With a growing number of available GLC maps, a comprehensive evaluation and comparison is necessary to assess their accuracy and suitability for diverse uses. This particularly applies to maps lacking statistically robust accuracy assessment or sufficient reported detail on the validation procedures. This study conducts a comparative independent validation of recent 10 m GLC maps, namely ESRI Land Use/Land Cover (LULC), ESA WorldCover, and Google and World Resources Institute (WRI)'s Dynamic World, examining their spatial detail representation and thematic accuracy at global, continental, and national (for 47 larger countries) levels. Since high-resolution map validation is impacted by reference data uncertainty owing to geolocation and labelling errors, five validation approaches dealing with reference data uncertainty were evaluated. Of the considered approaches, validation using the sample label supplemented by majority label within the neighborhood is found to produce more reasonable accuracy estimates compared to the overly optimistic approach of using any label within the neighborhood and the overly pessimistic approach of direct comparison between the map and reference labels. Overall global accuracies of the maps range between 73.4% ± 0.7% (95% confidence interval) to 83.8% ± 0.4% with WorldCover having the highest accuracy followed by Dynamic World and ESRI LULC. The quality of the maps varies across different LC classes, continents, and countries. The maps' spatial detail representation was assessed at various homogeneity levels within a 3 × 3 kernel. Although considered as high-resolution maps, this study reveals that ESRI LULC and Dynamic World have less spatial detail than WorldCover. All maps have lower accuracies in heterogenous landscapes and in some countries such as Mozambique, Tanzania, Nigeria, and Spain. To select the most suitable product, users should consider both the map's accuracy over the area of interest and the spatial detail appropriate for their application. For future high-resolution GLC mapping, producers are encouraged to adopt standardized LC class definitions to ensure comparability across maps. Additionally, the spatial detail and accuracy of GLC maps in heterogeneous landscapes and over some countries are the key features that should be improved in future versions of the maps. Independent validation efforts at regional and national levels, as well as for LC changes, should be strengthened to enhance the utility of GLC maps at these scales and for long-term monitoring. • Three recent 10 m global land cover maps were validated and compared. • Overall accuracy ranges from 73.4% to 83.8%; WorldCover highest. • Important map accuracy differences at multiple scales and land cover types. • Substantial contrast between the maps' ability to represent spatial detail. • Study emphasizes potentials and limitations of recent maps for users and producers. [ABSTRACT FROM AUTHOR]