Optimal resolution of soil properties maps varies according to their geographical extent and location

• Published in: Geoderma 2022-04, Vol.412 (115723), p.115723-37, Article 115723
• Main Authors: Piedallu, Christian, Pedersoli, Eloise, Chaste, Emeline, Morneau, François, Seynave, Ingrid, Gégout, Jean-Claude
• Format: Article
• Language: English
• Subjects: Agricultural sciences; Digital soil mapping; Environmental Engineering; Environmental predictors; Environmental Sciences; Life Sciences; Map accuracy; Resolution; Scale; Soil properties; Soil study
• ISSN: 0016-7061; 1872-6259
• DOI: 10.1016/j.geoderma.2022.115723

•Large-scale model or map performance significantly decreases above 1 km resolution.•Fine resolution is more accurate for heterogeneous landscapes and for reduced extents.•Aggregation of the cell values of predictive maps can improve their precision.•Different spatial resolution levels should be compared before soil mapping. The important development of digital soil mapping (DSM) these last decades has led to a large number of maps of soil properties with increasingly finer raster size. Map resolution is mostly determined by expert knowledge or by matching with the resolution of existing data, while scale is recognized as a major issue. Using the pH and the C/N ratio describing the surface horizon of forest soils and estimated by bioindication, we evaluated the effect of resolution changes on model and map performance for different geographical extents. Using 40,663 plots from the national forest inventory and 25 environmental variables calculated at eight different spatial resolution levels (50, 100, 250, 500, 1000, 8000, 16,000, and 50,000 m), we modeled and mapped pH and C/N over a vast and diversified area of 91,000 km2 in the north-east of France. The models highlighted the importance of geology in pH and C/N spatial variations, and to a lesser extent the importance of stand type, climate and topography, with a slight influence of data resolution on predictor selection. On the contrary, the accuracy of model or map performance decreased significantly above 1000 m resolution. Significant performance differences were observed according to the location and the size of the geographical extent. Globally, the more heterogeneous environmental characteristics and the smaller the geographical extent, the better fine spatial resolution performed. In addition, the aggregation of fine-resolution pH or C/N maps at a coarser cell size improved map performance as compared to the direct use of the coarse-resolution predictors. The impact of resolution changes on map accuracy varies according to the mapping procedure, the local environment, and the geographical extent, and should be evaluated in DSM studies to optimize map accuracy.