Trend analysis of nutrient losses in agricultural catchments using Generalised Additive Model and Pettitt test

<p>The EU Water framework demands the Member States to set up water management plans with the aim that water bodies should reach good ecological status. However, the overall progress is slow and more efforts are needed to reach the goals. This study presents how long-term monitoring of water quality and agricultural management practices in four agriculturally dominated catchments in Sweden contribute to the knowledge to nutrient leaching and their behaviour in relation to catchment characteristics.</p><p>20 years of water quality data were analysed to understand nitrogen (N) and phosphorus (P) trends and variations in surface waters leaving the catchments. The catchments represent a wide range of soil types and climate and hence different farming practices. Followed to time series analysis and in order to investigate how water quality data relate to management practices, two modelling approaches were used: non-parametric Pettitt test to determine significant changes in mean values of time series data; and generalized additive model (GAM) to identify the impact of climate/ anthropogenic variables on nutrient loads as a flexible model which avoids overfitting for long time series data.</p><p>Despite the general progress in preventing deterioration of water quality, the time series analysis indicated drastic changes over years in loads leaving some of the catchments. At the same time, there were large variations in N and P loads among catchments while runoff was the only significant indicator of losses in all. Clay dominated catchment showed more fluctuations in daily TN (0-40 mg/l) and TP (0-3.2 mg/l) concentrations, and also very high values of P (>0.07 mg PO₄P/l) compared to other catchments. On the other hand, sandy loam catchment was more consistent in losses despite the high values of N (>7 mg NO₃N/l). Although nutrients were washed into water bodies after the first heavy rain following a prolonged drought period, Pettitt model, which is insensitive against spikes, proved that permanent change points in P or N loads was not always following immediately after a change point in runoff or rainfall. Finally, GAM modelling did not generate a direct relationship between a single management practice and trend of nutrient concentration, and demonstrated the complexity in analysing the commutative impact of temporal/spatial factors that influence nutrients loads.</p><p>The study results proved that having long term water quality record is of great importance to show the pattern of nutrient trends, signal any undesirable changes, and to observe the impact of environment including cumulative impact of management practices on nutrient mobilization/retention. Therefore, continuing monitoring is critical to support the EU WFD 3^rd cycle actions. However, better information on peak load events is needed as climate and especially precipitation&#8211;runoff evenest are changing. In addition, application of technologies such as sensors or remote sensing will increase accuracy of the measurement by providing high temporal data regardless of any logistic complications.</p>