Mapping soil water holding capacity over large areas to predict the potential production of forest stands

Ecological studies need environmental descriptors to establish the response of species or communities to ecological conditions. Soil water resource is an important factor but is poorly used by plant ecologists because of the lack of accessible data. We explore whether a large number of plots with basic soil information collected within the framework of forest inventories allows the soil water holding capacity (SWHC) to be mapped with enough accuracy to predict tree species growth over large areas. We first compared the performance of available pedotransfer functions (PTFs) and showed significant differences in the prediction quality of SWHC between the PTFs selected. We also showed that the most efficient class PTFs and continuous PTFs compared had similar performance, but there was a significant reduction in efficiency when they were applied to soils different from those used to calibrate them. With a root mean squared error (RMSE) of 0.046 cm3 cm-3 (n = 227 horizons), we selected the Al Majou class PTFs to predict the SWHC in the soil horizons described in every plot, thus allowing 84% of SWHC variance to be explained in soils free of stone (n = 63 plots). Then, we estimated the soil water holding capacity by integrating the stone content collected at the soil pit scale (SWHC') and both the stone content at the soil pit scale and rock outcrop at the plot scale (SWHC") for the 100.307 forest plots recorded in France within the framework of forest inventories. The SWHC" values were interpolated by kriging to produce a map with 1 km² cell size, with a wider resolution leading to a decrease in map accuracy. The SWHC" given by the map ranged from 0 to 148 mm for a soil down to 1 m depth. The RMSE between map values and plot estimates was 33.9 mm, the best predictions being recorded for soils developed on marl, clay, and hollow silicate rocks, and in flat areas. Finally, the ability of SWHC' and SWHC" to predict Ecological studies need environmental descriptors to establish the response of species or communities to ecological conditions. Soil water resource is an important factor but is poorly used by plant ecologists because of the lack of accessible data. We explore whether a large number of plots with basic soil information collected within the framework of forest inventories allows the soil water holding capacity (SWHC) to be mapped with enough accuracy to predict tree species growth over large areas. We first compared the performance of available pedotransfer functions (PTFs) and showed significant differences in the prediction quality of SWHC between the PTFs selected. We also showed that the most efficient class PTFs and continuous PTFs compared had similar performance, but there was a significant reduction in efficiency when they were applied to soils different from those used to calibrate them. With a root mean squared error (RMSE) of 0.046 cm3 cm-3 (n = 227 horizons), we selected the Al Majou class PTFs to predict the SWHC in the soil horizons described in every plot, thus allowing 84% of SWHC variance to be explained in soils free of stone (n = 63 plots). Then, we estimated the soil water holding capacity by integrating the stone content collected at the soil pit scale (SWHC') and both the stone content at the soil pit scale and rock outcrop at the plot scale (SWHC") for the 100.307 forest plots recorded in France within the framework of forest inventories. The SWHC" values were interpolated by kriging to produce a map with 1 km² cell size, with a wider resolution leading to a decrease in map accuracy. The SWHC" given by the map ranged from 0 to 148 mm for a soil down to 1 m depth. The RMSE between map values and plot estimates was 33.9 mm, the best predictions being recorded for soils developed on marl, clay, and hollow silicate rocks, and in flat areas. Finally, the ability of SWHC' and SWHC" to predict height growth for Fagus sylvatica, Picea abies and Quercus petraea is discussed. We show a much better predictive ability for SWHC" compared to SWHC'. The values of SWHC" extracted from the map were significantly related to tree height growth. They explained 10.7% of the height growth index variance for Fagus sylvatica (n = 866), 14.1% for Quercus petraea (n = 877) and 10.3% for Picea abies (n = 2067). The proportions of variance accounted by SWHC" were close to those recorded with SWHC" values estimated from the plots (11.5, 11.7, and 18.6% for Fagus sylvatica, Quercus petraea and Picea abies, respectively). We conclude that SWHC" can be mapped using basic soil parameters collected from plots, the predictive ability of the map and of data derived from the plot being close. Thus, the map could be used just as well for small areas as for large areas, directly or indirectly through water balance indices, to predict forest growth and thus production, today or in the future, in the context of an increasing drought period linked to a global change of climatic conditions.