The quantification of the sediment carrying capacity of a river is a difficult task that has received much attention. For sand-bed rivers especially, several sediment transport functions have appeared in the literature based on various concepts and approaches; however, since they present a significant discrepancy in their results, none of them has become universally accepted. This paper employs three machine learning techniques, namely artificial neural networks, symbolic regression based on genetic programming and an adaptive-network-based fuzzy inference system, for the derivation of sediment transport formulae for sand-bed rivers from field and laboratory flume data. For the determination of the input parameters, some of the most prominent fundamental approaches that govern the phenomenon, such as shear stress, stream power and unit stream power, are utilized and a comparison of their efficacy is provided. The results obtained from the machine learning techniques are superior to those of the commonly-used sediment transport formulae and it is shown that each of the input combinations tested has its own merit, as they produce similarly good results with respect to the data-driven technique employed.

Rainfall erosivity concerns the ability of rainfall to cause erosion on the surface of the earth. The difficulty in modeling the distribution, the size, and the terminal velocity of raindrops in relation to the detachment of soil particles led to the use of more tractable rainfall indices. Thus, in the universal soil loss equation (USLE), the coefficient of rainfall erosivity, R, was introduced. This coefficient is based on the product of the rainfall kinetic energy of a storm and its maximum 30-minute intensity. An important problem in the application of USLE and its revisions in various parts of the world concerns the computation of R, which requires pluviograph records with a length of at least 20 years. For this reason, empirical equations have been developed that are based on coarser rainfall data, such as daily, monthly, or yearly, which are available on larger spatial and temporal extents. However, the lack of denser data is dealt more effectively by means of machine learning methods. Computational systems for this purpose were recently developed based on feed-forward neural networks, yielding significantly better results.

A future variation of precipitation characteristics, due to climate change, will affect the ability of rainfall to precipitate soil loss. In this paper, the monthly and annual values of rainfall erosivity (R) in Greece are calculated, for the historical period 1971–2000, using precipitation records that suffer from a significant volume of missing values. In order to overcome the data limitations, an intermediate step is applied using the calculation of monthly erosivity density, which is more robust to the presence of missing values. Spatial Quantile Regression Forests, a data driven algorithm that imitates kriging without the need of strict statistical assumptions, was utilized and validated, in order to create maps of R and its uncertainty using error propagation. The monthly average precipitation for the historical period 1971–2000 estimated by five (5) Global Circulation Models-Regional Climatic Models were validated against observed values and the one with the best performance was used to estimate projected changes of R in Greece for the future time period 2011–2100 and two different greenhouse gases concentration scenarios. The main findings of this study are: (a) the mean annual R in Greece is 1039 MJ·mm/ha/h/y, with a range between 405.1 and 3160.2 MJ·mm/ha/h/y. The highest values are calculated at the mountain range of Pindos and the lowest at central Greece; (b) the monthly R maps adhere to the spatiotemporal characteristics of precipitation depth and intensities over the country; (c) the projected R values, as an average over Greece, follow the projected changes of precipitation of climatic models, but not in a spatially homogenous way.