The knowledge of soil properties is of paramount importance for agriculture and forest management. The content in organic material (SOM) and texture fractions (TF) (clay, silt and sand percentages) are two of the most important properties, being determinant for biogeochemical cycles and soil fertility. These soil properties depend on different factors such as climate, relief, lithology or vegetation development, among others. The existent relations among these features produce different pedogenetic conditions, with their related physicochemical processes, which explain spatial variability of soil properties, allowing its modelling through statistical procedures. The aim of this work was twofold: i) the mapping of SOM and TF of the peninsular Spain; and, ii) the evaluation of new features derived from remote sensing which are related to seasonal cycles of vegetation applied to digital soil mapping.
The SOM and TF measurements were obtained from the Spanish National Inventory for Soil Erosion, which comprised more than 10000 samples from most superficial soil horizon, which were sampled between 2003 and 2016. Different features were used as predictors in the mapping such as: DEM-derived terrain features (elevation, slope, aspect), climatic features (cumulative precipitation, temperature seasonal averages), lithology, and different variables describing the land surface phenology (LSP) extracted from remotely sensed vegetation indices time series (date of the start of spring, date of the end of senescence, growing season length, end of the growing season, length of the growing season, maximum peak, and large seasonal integral as a proxy of productivity). The LSP variables were obtained from time series of NDVI that were computed from the MODIS weekly surface reflectance product (MOD09Q1 v6) at a spatial resolution of 250 for the entire study period. The performance of two different algorithms was evaluated for the predicting modelling of SOM and TF within the framework of a high dimensional feature space: the hitherto applied regression kriging and Random Forest. The evaluation was based on the accuracy of models, but also in the method’s capacity for handling numerous features and perform feature selection.

Area A - Board 18

Soil erosion is a naturally dynamic phenomenon, but human activities, human-induced forces and climate change have further accelerated this process. As a consequence, the mean annual soil loss rate in Europe exceeds the average soil formation rate. That means soil can be actually considered as a non-renewable resource. Therefore, understanding spatial patterns and temporal trends of soil loss could support government land use policies and strategies for reducing this overlooked natural hazard.
Among the factors driving soil erosion dynamics, meteorological forcings and land cover/land use vary in time and space, while other factors like soil properties (i.e. their pedology) and topography can be considered constant over time. Soil erosion can be reduced acting on land cover/land use, but, this factor is the most complex and expensive to be continuatively monitored by national or regional agencies through field campaign or ad hoc surveys. However, satellites for Earth Observation can help a lot in monitoring spatial and temporal land cover changes and the cover management factor can be effectively estimated through spectral indices (e.g. NDVI).
This paper describes an ad hoc implementation of the Revised Universal Soil Loss Equation (RUSLE) model to provide dynamic maps of soil erosion. In this work, we evaluate the benefits of integrating Landsat and Sentinel-2 time series to study soil erosion in an alpine river basin of Italy. Thus, reducing the revisit time up to 5 days allows to consider the effects on soil erosion of land use/land cover modification caused by extreme meteorological events, that otherwise would be missed if estimating soil erosion through institutional product of land use/land cover.
Besides, the high revisit time provided by the combination of the two sensors could provide snow cover maps of the study area, useful for quantifying the contribution to erosion of soil covered by snow.

Results demonstrate that the ESA’ Sentinel-2 twin satellites could effectively enhance the estimate of the cover management factor, both spatially and temporally.

Area A - Board 19

Soil surface roughness affects different hydrological processes such as surface depression storage, water infiltration, overland flow velocity, organization of overland flow (connectivity) and consequently soil erosion by water as well. Surface roughness also is an important component governing the scattering of microwaves and thus for understanding RADAR backscatter in remote sensing. In agricultural soils, surface roughness is mainly an anthropogenic factor determined by the type of tillage and management with an oriented component due to tillage implements and a random component representing soil clods or aggregates. Further, surface roughness is a property with a high spatial variability, since the same type of tillage can result in surfaces with different roughness depending on the physical characteristics of the soil. Additionally, surface roughness is susceptible to change throughout time due to the action of meteorological agents (e.g., precipitation, wind and temperature changes). In recent years, advances in surveying have greatly improved the spatial resolution, extent, and availability of surface elevation datasets. In this sense, laser-based technologies, also known as laser scanning or LiDAR (light detection and ranging), have been used for high-resolution surface roughness measurements in agricultural soils. Therefore, multi-temporal laser-based surface roughness characterization and quantification could be very useful for understanding and modelling hydrology and soil erosion processes relevant for different applications.

With this aim, terrestrial laser scanner (TLS) measurements were obtained on four experimental plots during a full agricultural year in Pamplona (Navarre, Spain). The TLS used in this study was the FARO Focus 3D. The scans resolution was set to 0.018º (20480 3D pixel in 360º), so for the maximum range distance of 6 m, the sampling interval was ~1.8 mm. The four experimental plots (4 x 4 m) were created using different tillage treatments, so as to represent different surface roughness conditions typical of agricultural soils. For each of the four plots, four scans were measured (i.e., one from each side of the plot). The roughness measurements were carried out on nine different dates during the agricultural year (starting in September). In total, 144 scans with approximately 20 million points were obtained.

The data processing was very laborious due to the high volume of TLS data acquired. Raw scans were first filtered to avoid the so-called mixed-pixels measurements using a self-implemented algorithm that showed very good results for our data. Then, the four scans of every experimental plot were co-registered and merged into a single point cloud trough two steps. In the first step, an approximate co-registration was applied using the point pairs picking. In the second step, a fine co-registration was applied using then the iterative closest point (ICP) algorithm. The results obtained after the ICP-based co-registration were very satisfactory, with standard deviation of the point-to-plane residuals below 2 mm in all the experimental plots and in the different dates of measurements. The co-registration of the different multi-temporal point clouds was processed taking into account only the elements with no spatial variability during the study of period (e.g., auxiliary spheres, reference sticks). Finally, to obtain DEMs, all point clouds were interpolated into rasters with a raster cell of 5 mm x 5 mm, using the mean elevation value. The maximum and minimum values were excluded in the interpolation to avoid outliers.

The multi-temporal surface roughness analysis was focused on two different aspects. On the one hand, the behavior of different roughness parameters (e.g., standard deviation of the heights, correlation length) during the agricultural year was analyzed. On the other hand, multi-temporal DEM changes on the different experimental plots were analyzed to detect different hydrological and erosive processes. Although this is still an ongoing study, preliminary results already showed that high-resolution, multi-temporal roughness information derived from TLS will be very useful in understanding and modeling the temporal behavior of agricultural soils.

Area A - Board 22