Análise da relação entre o uso do solo e liberação dos nutrientes fósforo e nitrogênio nas águas superficiais do Rio das Velhas utilizando árvores de decisão e regressão múltipla

Abstract

Phosphorus and nitrogen nutrients in surface water are related to the watershed land use and land cover (LULC) due to substances generation and deposition in the soil that may reach the river by the action of rain or direct runoff. Pollution sources scattered in the basin may modify the surface water quality and are referred to as diffuse or non-point sources. Identification and quantification of pollution load from diffuse sources are hard to obtain especially when they occur jointly with point sources, as industry effluents and domestic sewage. The Velhas river watershed is an example of such a situation, due to the presence of diffuse and point sources, complicating the separation of pollutant loads from the two types of sources and, related to diffuse sources, the determination of its relationship with LULC. This thesis has as primary goal to analyze the relation between LULC and the spreading of phosphorus and nitrogen nutrients into Velhas rivers surface waters, considering potential and limitations of the analysis. The evaluated methods use mainly multiple regression and decision tree to create models. We proposed an index that relates LULC categories, such as pollution source or filter for surface water, and a hydrograph-based method for estimating diffuse and point loads. We obtained the LULC categories considering three different áreas of analysis: watershed, exclusive contribution area and buffers. The quality data selection for rainy and dry periods also use different approaches: rainy quarter, rainy period and hydrograph ascension. For establishing the relation between nutrient concentration and LULC, the decision tree results are promising, reaching classification efficiency greater than 80% on nitrate and phosphorous classes. Multiple regression showed good results in relate ammonia, nitrite and dissolved oxygen with LULC, whose determination coefficient is about 80%. The nitrate and total phosphorous parameters did not produce valid models. Analyzing Pearson coefficient, we identified the presence of LULC category regarded as pollution sources (agropastoral and urban) and as pollution filters (vegetation). Thus, we propose an index (LCPI) that relates these sources and filters substituting the individual LULC categories. Results indicate that LCPI is better than LULC to explain the relationship. In 12 out of 15 models, the coefficient of determination (R2) increased by 11 to 155% using LCPI instead of LULC. Finally, we proposed a method for estimating the diffuse and point loads in Velhas River, based on the analysis of hydrograph in storms events. Initial results showed the influence of rainfall on phosphorus and nitrate accumulation in the basin, in years with lower precipitation/flow. Thus, it was possible to obtain model for nitrate and impossible for phosphorus. The results indicated the need of jointly monitoring quality and flow, with detailing of the first hydrographs about storms events, aiming to split diffuse and point sources.