11 September 2020 in Nature Communications
Mapping aboveground forest biomass is central for assessing the global carbon balance. However, current large-scale maps show strong disparities, despite good validation statistics of their underlying models. Here, we attribute this contradiction to a flaw in the validation methods, which ignore spatial autocorrelation (SAC) in data, leading to overoptimistic assessment of model predictive power. To illustrate this issue, we reproduce the approach of large-scale mapping studies using a massive forest inventory dataset of 11.8 million trees in central Africa to train and validate a random forest model based on multispectral and environmental variables. A standard nonspatial validation method suggests that the model predicts more than half of the forest biomass variation, while spatial validation methods accounting for SAC reveal quasi-null predictive power. This study underscores how a common practice in big data mapping studies shows an apparent high predictive power, even when predictors have poor relationships with the ecological variable of interest, thus possibly leading to erroneous maps and interpretations.
1 November 2022 in Environmental Research Letters
Clean water is key for sustainable development. However, large gaps in monitoring data limit our understanding of global hotspots of poor water quality and their evolution over time. We demonstrate the value added of a data-driven approach (here, random forest) to provide accurate high-frequency estimates of surface water quality worldwide over the period 1992–2010. We assess water quality for six indicators (temperature, dissolved oxygen, pH, salinity, nitrate-nitrite, phosphorus) relevant for the sustainable development goals. The performance of our modeling approach compares well to, or exceeds, the performance of recently published process-based models. The model’s outputs indicate that poor water quality is a global problem that impacts low-, middle- and high-income countries but with different pollutants. When countries become richer, water pollution does not disappear but evolves. Water quality exhibited a signif icant change between 1992 and 2010 with a higher percentage of grid cells where water quality shows a statistically significant deterioration (30%) compared to where water quality improved (22%).
8 July 2020 in Scientific Data
Forest biomass is key in Earth carbon cycle and climate system, and thus under intense scrutiny in the context of international climate change mitigation initiatives (e.g. REDD+). In tropical forests, the spatial distribution of aboveground biomass (AGB) remains, however, highly uncertain. There is increasing recognition that progress is strongly limited by the lack of field observations over large and remote areas. Here, we introduce the Congo basin Forests AGB (CoFor-AGB) dataset that contains AGB estimations and associated uncertainty for 59,857 1-km pixels aggregated from nearly 100,000 ha of in situ forest management inventories for the 2000 – early 2010s period in five central African countries. A comprehensive error propagation scheme suggests that the uncertainty on AGB estimations derived from c. 0.5-ha inventory plots (8.6–15.0%) is only moderately higher than the error obtained from scientific sampling plots (8.3%). CoFor-AGB provides the first large scale view of forest AGB spatial variation from field data in central Africa, the second largest continuous tropical forest domain of the world.
12 April 2013 in Journal of Multivariate Analysis
1 January 2010 in Ecological Modelling