Rainfall erosivity estimation via several methods, and water erosion modeling at Peixe Angical reservoir-TO

Abstract

The erosion process undergoes changes as land use and land cover change through the conversion of the forest to pasture and / or agricultural crops. In addition to vegetation cover, other factors such as rainfall erosivity –which is the potential of rain to cause erosion– are indispensable in erosion modeling. This thesis was divided into chapters, where in the first chapter the package RainfallErosivityFactor was developed within the programming language R, and later made available on CRAN of R. This package is a tool to analyze rain data, such as total precipitation, depth and number of erosive and non-erosive rains, and to determine the rainfall erosivity, with monthly and annual outputs. In this way, the rainfall erosivity factor is calculated accurately and effectively. An example was provided for Pirassununga, SP, Brazil, using a 7-year rainfall data set with an interval of 10 minutes between measurements. The results can be processed in the R environment itself for statistical analysis, construction of graphs, or application of geostatistics. The average rainfall erosivity for Pirassununga was 9,512.9 MJ mm ha-1 h-1 year-1. In view of the above, conservationist practices must be adopted to minimize the impacts of the erosion process. In the second chapter, the method of determining rainfall erosivity proposed by Wishcmeier and Smith was compared with other methods used in different parts of the world to estimate rainfall erosivity, in order to select a consistent method to replace the Wischmeier and Smith method for conditions without considering the intensity of the rain. The tested methods included: Modified Fournier, MF; Modified Fournier by Zhang, MF-Z; Modified Fournier by Men, MF-M; Rainfall Disaggregation, RD; TRMM satellite with modified Fournier coefficient, TRMM-F; and TRMM Satellite with monthly precipitation, TRMM-M. The analyzes were performed according to the Additive Main Effects and Multiplicative Interaction (AMMI) model and Scott-Knott cluster tests. The evaluated methods behaved differently for the rainy, dry, monthly and annual periods. The MF method proved to be able to consistently replace the Wischmeier and Smith method. It is emphasized that the methods based on the TRMM satellite can be a plausible alternative for locations without precipitation information. Finally, in the third chapter, the objective was to model soil losses in the drainage basin of the Peixe Angical reservoir, Brazil, in addition to assessing the level of importance of the RUSLE factors. The sediment export values were also calculated to identify areas where soil conservation practices were needed. To estimate soil losses for the 1990, 2000, 2010 and 2017 chronological scenario, the RUSLE model was coupled with GIS. The evaluation of the level of importance of each RUSLE factor was carried out using a non-parametric machine learning algorithm, Random Forest. The level of importance of the RUSLE factors was assessed in the following order: C> K> LS> R. Water erosion in the drainage basin of the Peixe Angical reservoir has increased over the years due to changes in land use, although soil losses in most of the basin were classified as very low (<2.5 Mg ha-1 year-1). The impact of change in land use must be minimized with soil conservation practices, enabling sustainable development.