Challenges and progress in the detailed estimation of sediment export in agricultural watersheds in Navarra (Spain) after two decades of experience

<p>In 1993, the Government of Navarre (northern Spain) began the installation and operation of a network of experimental watersheds in order to assess, among other aspects, soil erosion in representative agricultural areas of the territory. Initially, sediment sampling at the outlet of each of the five basins was performed on a daily basis, despite which it was possible to get a highly fruitful and novel knowledge on sediment export (Merch&#225;n et al., 2019). However, in the last 16 years, with the aim of studying sediment export in more detail, the sampling frequency was increased so that the behavior of the sedimentogram at the event level could be known. In these cases, when the amount of sediment was large enough, the sediment texture was also determined. In addition, from the beginning of the observations, a turbidimeter was used to record turbidity data every 10 minutes. The aim of this work is to deepen the knowledge of sediment export dynamics in representative agricultural watersheds of Navarre by analyzing the database described above and focusing in specific events. To do this, first, the entire database was represented in graphs that include variables such as sediment texture, samples taken per event, daily mean precipitation, turbidity, flow rate, etc. Next, events with a minimum of six samples were selected and the linear relationship between turbidity and sediment concentration was analyzed using simple linear regressions, as this is the method used in similar works. Subsequently, these same event data were added to set up monthly samples where again linear regressions were performed. Apart from the simple linear analysis, where the linear relationship with turbidity was analyzed as the only predictor variable, different artificial intelligence methods have been explored, such as the generalized linear model (GLM), support vector machine (SVM), multivariate adaptive regression splines (MARS) and random forest (RF), adding additional variables such as accumulated precipitation, and season or water level in the analysis. The results from all these statistical studies have been disappointing, since no pattern or generalization has been found to predict sediment concentration from the variables considered. These results suggest that the sediment export behavior of small agricultural watersheds is particularly complex and controlled by spatially and temporally varying variables. It is evident that at least some of these variables have not been taken into account in the study. The high variability found in sediment textures supports the hypothesis that the erosive behavior of watersheds is of great complexity. We believe that the consideration of variables such as vegetation on slopes and channels and its evolution can be helpful in the analysis.</p><p>&#160;</p><p>Merch&#225;n, D., Luquin, E., Hern&#225;ndez-Garc&#237;a, I., Campo-Besc&#243;s, M. A., Gim&#233;nez, R., Casal&#237;, J., & Del Valle de Lersundi, J. (2019). Dissolved solids and suspended sediment dynamics from five small agricultural watersheds in Navarre, Spain: A 10-year study. <em>CATENA</em>, <em>173</em>, 114&#8211;130. https://doi.org/10.1016/J.CATENA.2018.10.013</p>