Purdue University Graduate School
Pauline Welikhe Dissertation_05_06_2020edited.pdf (3.7 MB)

Evaluating the Effects of Legacy Phosphorus on Dissolved Reactive Phosphorus Losses in Tile-Drained Systems

Download (3.7 MB)
posted on 2020-05-07, 12:25 authored by Pauline Kageha WelikhePauline Kageha Welikhe

Eutrophication due to phosphorus (P) enrichment continues to be a primary water quality concern affecting freshwater and marine estuaries around the world. Excessive anthropogenic P inputs, driven by the need to meet the rising food and energy demands of a growing and increasingly urbanized population, have resulted in the buildup of P creating legacy (historical) P pools in agricultural landscapes. There is growing evidence that remobilization of accumulated legacy P can interfere with conservation efforts aimed at curbing eutrophication and improving water quality. Less is known about the magnitude and effects of these legacy P pools on dissolved reactive P (DRP) losses in tile-drained systems. This dissertation consists of three separate inquiries into how legacy P may affect DRP losses in tile drains. In the first inquiry, we examined the possibility of developing a suitable pedo-transfer function (pedoTF) for estimating P sorption capacity (PSC). Subsequent PSC-based indices (Phosphorus Saturation Ratio (PSR) and Soil Phosphorus Storage Capacity (SPSC)) were evaluated using daily water quality data from an in-field laboratory. The pedoTF derived from soil aluminum and organic matter accurately predicted PSC (R2 = 0.60). Segmented-line models fit between PSR and soluble P (SP) concentrations in both desorption assays (R² = 0.69) and drainflows (R² = 0.66) revealed apparent PSR thresholds in close agreement at 0.21 and 0.24, respectively. Linear relationships were observed between negative SPSC values and increasing SP concentrations (R² = 0.52 and R2 =0.53 respectively), and positive SPSC values were associated with very low SP concentrations in both desorption assays and drainflows. Zero SPSC was suggested as a possible environmental threshold. Thus, PSC-based indices determined using a pedoTF could estimate the potential for SP loss in tile drains. Also, both index thresholds coincided with the critical soil test P level for agronomic P sufficiency (22 mg kg-1 Mehlich 3 P) suggesting that the agronomic threshold could serve as an environmental P threshold. In the second inquiry, PSC- based indices in addition to other site characteristics present in a P index (PI), were used as inputs in the development of a multi-layer feed-forward artificial neural network (MLF-ANN). The MLF-ANN was trained, tested, and validated to evaluate its performance in predicting SP loss in tile drains. Garson’s algorithm was used to determine the weight of each site characteristic. To assess the performance of ANN-generated weights, empirical data from an in-field laboratory was used to evaluate the performance of an unweighted PI (PINO), a PI weighted using Lemunyon and Gilbert weights (PILG), and an ANN-weighted PI (PIANN) in estimating SP losses in tile effluent. The MLF-ANN provided reliable predictions of SP concentrations in tile effluent (R2 = 0.99; RMSE = 0.0024). Soil test P, inorganic fertilizer application rate (FPR), SPSC, PSR, and organic P fertilizer application rate (OPR), with weights of 0.279, 0.233, 0.231, 0.097, and 0.084, respectively, were identified as the top five site characteristics with the highest weights explaining SP loss in tile discharge. These results highlighted the great contribution of both contemporary and legacy P sources to SP concentrations in tile discharge. Also, PIANN was the only PI with a significant exponential relationship with measured annual SP concentrations (R2 = 0.60; p < 0.001). These findings demonstrated that MLF-ANNs coupled with Garson’s algorithm, can accurately quantify weights for individual site characteristics and develop PIs with a strong correlation with measured SP in tile discharge. Finally, in the third inquiry, we compared DRP loads and flow-weighted mean DRP (FDRP) concentrations in P source and P sink soils and evaluated the predominant DRP concentration – discharge (C-Q) behavior in these soils on a daily and event scale. At the daily scale, C-Q patterns were linked to the soil P status whereby a chemostatic and dilution behavior was observed for P source and P sink soils, respectively. At the event scale, C-Q patterns were linked to soil P status, flow path connectivity, and mixing of event water, matrix water, and rising shallow groundwater. The predominant anti-clockwise rotational pattern observed on P source soils suggested that, as the discharge event progressed, contributions from P poor waters including matrix and shallow groundwater resulted in lower DRP concentrations on the rising limb compared to the falling limb. However, the variable flushing and dilution behavior observed on the rising limb suggested that, in addition to discharge and soil P status, rapid exchanges between P pools, the magnitude of discharge events (Q), and the relative number of days to discharge peak (Drel) also regulated DRP delivery. On the other hand, the predominant non-hysteretic C-Q behavior in P sink soils suggest that DRP loss from these soils can be discounted. Our collective results highlight the need for nutrient and conservation practices focused on P drawdown, P sequestration, and P supply close to the crop needs, which will likely be required to convert P sources to sinks and to avoid the conversion of P sinks to sources.


Degree Type

  • Doctor of Philosophy


  • Agronomy

Campus location

  • West Lafayette

Advisor/Supervisor/Committee Chair

Dr. Sylvie Brouder

Additional Committee Member 2

Dr. Jeffrey Volenec

Additional Committee Member 3

Dr. Ronald Turco

Additional Committee Member 4

Dr. Margaret Gitau