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]

• Overall, REDD+ interventions were found to be aligned with deforestation drivers. • Local interventions predominantly target small-scale drivers. • No single data acquisition method suffices to assess multifaceted deforestation drivers. • Multidisciplinary data sources complement each other in information on drivers. Efforts to reduce emissions from deforestation and forest degradation and enhancing forest carbon stocks (REDD+) have evolved over the past decade. Early REDD+ programs and local/subnational projects used various interventions (i.e. enabling measures, disincentives and incentives), implemented by government, the commercial and non-commercial private sector, but are currently understudied vis-à-vis their effectiveness to address site-specific drivers of deforestation and forest degradation (DD). We assess how well REDD+ interventions addressed DD at five project sites in Peru (1), Brazil (1), Vietnam (1) and Indonesia (2). Our study design includes an integrated assessment of remotely sensed, spatially modelled, and locally reported drivers. First, we observe follow-up land use from high resolution imagery as proxy for direct deforestation drivers. Second, spatial Random Forest modelling of DD drivers allows for influence quantification of topographic, climatic and proximity variables at each site. Third, we report direct and indirect DD drivers from pre-intervention surveys and semi-structured interviews with five REDD+ implementers, 40 villages and 1200 households. Data gathered included perceived changes in forest cover and quality, and their causes. We found general agreement between observed, modelled and reported local DD drivers, yet some were inadequately addressed by interventions. Intra-site differences in drivers underscores the importance of analysing micro-level DD drivers. Our interdisciplinary approach reveals the complexities of local direct and indirect DD drivers, and the complementarity of remotely sensed, spatially modelled and locally reported methods for driver identification. A better understanding of the alignment between DD drivers and REDD+ interventions is vital for practitioners and policy makers to enhance the effectiveness, efficiency, equity and co-benefits of REDD+ at the local level. [ABSTRACT FROM AUTHOR]

Over the past decade, several global maps of above-ground biomass (AGB) have been produced, but they exhibit significant differences that reduce their value for climate and carbon cycle modelling, and also for national estimates of forest carbon stocks and their changes. The number of such maps is anticipated to increase because of new satellite missions dedicated to measuring AGB. Objective and consistent methods to estimate the accuracy and uncertainty of AGB maps are therefore urgently needed. This paper develops and demonstrates a framework aimed at achieving this. The framework provides a means to compare AGB maps with AGB estimates from a global collection of National Forest Inventories and research plots that accounts for the uncertainty of plot AGB errors. This uncertainty depends strongly on plot size, and is dominated by the combined errors from tree measurements and allometric models (inter-quartile range of their standard deviation (SD) = 30–151 Mg ha−1). Estimates of sampling errors are also important, especially in the most common case where plots are smaller than map pixels (SD = 16–44 Mg ha−1). Plot uncertainty estimates are used to calculate the minimum-variance linear unbiased estimates of the mean forest AGB when averaged to 0.1∘. These are used to assess four AGB maps: Baccini (2000), GEOCARBON (2008), GlobBiomass (2010) and CCI Biomass (2017). Map bias, estimated using the differences between the plot and 0.1∘ map averages, is modelled using random forest regression driven by variables shown to affect the map estimates. The bias model is particularly sensitive to the map estimate of AGB and tree cover, and exhibits strong regional biases. Variograms indicate that AGB map errors have map-specific spatial correlation up to a range of 50–104 km, which increases the variance of spatially aggregated AGB map estimates compared to when pixel errors are independent. After bias adjustment, total pantropical AGB and its associated SD are derived for the four map epochs. This total becomes closer to the value estimated by the Forest Resources Assessment after every epoch and shows a similar decrease. The framework is applicable to both local and global-scale analysis, and is available at https://github.com/arnanaraza/PlotToMap. Our study therefore constitutes a major step towards improved AGB map validation and improvement. • A framework to assess global above-ground biomass maps at 0.1° spatial scale is developed and demonstrated. • Uncertainty assessment of reference data is a prerequisite prior to map comparisons. • Spatial uncertainty modelling reveals regional map biases and map-specific spatial correlation of errors. • Map-based AGB totals become closer to the value estimated by the Forest Resource Assessment over time. [ABSTRACT FROM AUTHOR]

The use of spectral data is seen as a fast and non-destructive method capable of monitoring pasture biomass. Although there is great potential in this technique, both end users and sensor manufacturers are uncertain about the necessary sensor specifications and achievable accuracies in an operational scenario. This study presents a straightforward parametric method able to accurately retrieve the hyperspectral signature of perennial ryegrass (Lolium perenne) canopies from multispectral data collected within a two-year period in Australia and the Netherlands. The retrieved hyperspectral data were employed to generate optimal indices and continuum-removed spectral features available in the scientific literature. For performance comparison, both these simulated features and a set of currently employed vegetation indices, derived from the original band values, were used as inputs in a random forest algorithm and accuracies of both methods were compared. Our results have shown that both sets of features present similar accuracies (root mean square error (RMSE) ≈490 and 620 kg DM/ha) when assessed in cross-validation and spatial cross-validation, respectively. These results suggest that for pasture biomass retrieval solely from top-of-canopy reflectance (ranging from 550 to 790 nm), better performing methods do not rely on the use of hyperspectral or, yet, in a larger number of bands than those already available in current sensors. [ABSTRACT FROM AUTHOR]

For decades scientists have produced maps of biological, ecological and environmental variables. These studies commonly evaluate the map accuracy through cross-validation with the data used for calibrating the underlying mapping model. Recent studies, however, have argued that cross-validation statistics of most mapping studies are optimistically biased. They attribute these overoptimistic results to a supposed serious methodological flaw in standard cross-validation methods, namely that these methods ignore spatial autocorrelation in the data. They argue that spatial cross-validation should be used instead, and contend that standard cross-validation methods are inherently invalid in a geospatial context because of the autocorrelation present in most spatial data. Here we argue that these studies propagate a widespread misconception of statistical validation of maps. We explain that unbiased estimates of map accuracy indices can be obtained by probability sampling and design-based inference and illustrate this with a numerical experiment on large-scale above-ground biomass mapping. In our experiment, standard cross-validation (i.e., ignoring autocorrelation) led to smaller bias than spatial cross-validation. Standard cross-validation was deficient in case of a strongly clustered dataset that had large differences in sampling density, but less so than spatial cross-validation. We conclude that spatial cross-validation methods have no theoretical underpinning and should not be used for assessing map accuracy, while standard cross-validation is deficient in case of clustered data. Model-free, design-unbiased and valid accuracy assessment is achieved with probability sampling and design-based inference. It is valid without the need to explicitly incorporate or adjust for spatial autocorrelation and perfectly suited for the validation of large scale biological, ecological and environmental maps. • Both standard and spatial cross-validation methods may provide biased estimates of map accuracy. • Unbiased estimates of map accuracy indices can be obtained by probability sampling and design-based inference. • Spatial cross-validation techniques should not be used for map accuracy assessment. [ABSTRACT FROM AUTHOR]