1. Predicting Hydrology and Phosphorus Transport from
a Subsurface-Drained Field Using RZWQM2-P
Sami Shokrana1, Ehsan Ghane1, Zhiming Qi2
1Department of Biosystems and Agricultural Engineering, Michigan State University, USA
2Department of Bioresource Engineering, McGill University, Canada
08/31/2022
International Drainage Symposium
2. Choosing a P Loss Simulation Model
2
Models
Surface runoff P Drainage discharge P
Macropore P
Plant
uptake P
P
transformation
Farm
management
practices
DRP
loss
PP loss DRP loss PP loss
ADAPT β β β β β β β limited
APEX/EPIC β β simplified β β β β β
HYDRUS β β β β β β β limited
MACRO β β β β β β β β
ICECREAM β β simplified β β β β
PLEASE β β β β β β β limited
SWAP/ANIMO β β β β β β β β
DRAINMOD-P β β β β β β β β
RZWQM2-P β β β β β β β β
3. Literature review
3
β’ RZWQM2-P is not well-
tested
β’ Tested twice under same
soil, climate, and crop
conditions, only
fertilization was different
4. Objectives
β’ To test and validate the performance of RZWQM2-P in predicting DRP
and TP loss through drainage discharge in a clay loam soil
β’ To identify the performance of the model under high-resolution daily
flow and load data
4
5. β’ Soil type: Ziegenfuss clay
loam
β’ Drain depth: 2.68 ft (0.82 m)
β’ Drain spacing: 33 ft (10.06 m)
β’ Field slope: 0.1%
β’ Corn-Soybean rotation
β’ Commercial fertilizer
β’ Vertical till before corn
5
Methods: Site description
Blissfield Site (7.6 hectares)
6. 6
Methods: Input Data
β’ Soil water characteristic input data
οΌ gSSURGO database
οΌ Delineated into 5 layers (0 β 205 cm)
β’ Weather input data
οΌ Precipitation
οΌ Solar radiation
οΌ Wind speed
οΌ Air temperature
οΌ Relative humidity
Image source: Minh Uong (The New York Times)
Source: Shokrana and Ghane (2020)
7. 7
Methods: Calibration and Validation
October 1, 2018 June 30, 2022
3 years and 9 months
Calibration period (2 years)
September 30,
2020
Validation period (1 year and 9 months)
October 1,
2020
October 1,
2018
June 30,
2022
Calibrated hydrology parameters
ο Soil water parameters
ο Runoff parameters
ο Drainage parameters
ο Evapotranspiration parameters
ο Water table fluctuation parameters
Calibrated DRP and TP parameters
ο Macropore parameters
ο Initial P level in GW reservoir
ο Soil filtration coefficient
ο Soil replenishment rate coefficient
ο Soil detachability coefficient
8. β’ Nash-Sutcliffe Efficiency (NSE)
β’ Percent Bias (PBIAS)
8
Methods: Performance Evaluation Statistics
Statistics Hydrology DRP and TP
NSE Very good (NSE > 0.75)
Good (0.60 < NSE β€ 0.75)
Satisfactory (0.4 < NSE β€ 0.60)
Very good (NSE > 0.65)
Good (0.50 < NSE β€ 0.65)
Satisfactory (0.35 < NSE β€ 0.50)
PBIAS Very good (PBIAS < Β±10%)
Good (Β±10% < PBIAS < Β±15%)
Satisfactory (Β±15% < PBIAS < Β±25%)
Very good (PBIAS < Β±15%)
Good (Β±15% < PBIAS < Β±20%)
Satisfactory (Β±20% < PBIAS < Β±30%)
Source: Skaggs et al. (2012), Moriasi et al. (2007, 2015)
Daily time-step
12. β’ Fertilizer application rate in April 2019: 22.5 Kg/ha
β’ Fertilizer application rate in May 2020: 9.5 Kg/ha
β’ Fertilizer application rate in May 2020 was increased up to 150
Kg/ha, but still all the P were lost by surface runoff.
Discussions: Modelβs Insensitivity to Fertilizer
Application
13
Model is unable to distribute fertilizer input to
drainage discharge under surface ponding conditions
13. Discussions: Daily data vs Event-Based Data
14
β’ Sadhukhan et al. (2019):
οΌ Aggregated flow data into several flow periods/events
οΌ Each event consisted of several weeks to several months
οΌ Dampens the visibility of rapid changes in P concentration
β’ This study:
οΌ High-resolution daily flow and load data
οΌ Captures the rapid fluctuations of P concentration in drainage
discharge
Higher resolution sampling strategy is beneficial to capture the
variation and fluctuation of P concentration in drainage discharge
14. Take-Home Messages
β’ The P modelβs performance needs to be modified under surface
ponding conditions
β’ More tests with daily data are needed. The rapid changes in P
concentration are more visible with daily data
β’ RZWQM2-P is reliable in predicting drainage discharge from
agricultural fields, but more tests are necessary to validate the
performance of P module
15
16. Thank You
Sami Shokrana
Graduate Research Assistant/PhD Candidate
Department of Biosystems and Agricultural Engineering,
Michigan State University
Email: shokrana@msu.edu
Phone: +1(606)306-9453
17
Editor's Notes
Good morning, everyone. Thank you for joining todayβs presentation. All of you are aware of the eutrophication problem in the WLEB which is caused by N and P pollution. The major non-point source behind this problem is considered to be agriculture. Field-scale hydrologic models can be a useful tool to predict these nutrients losses from these agricultural fields. This study focuses on using such a field-scale model to address towards the HABs issue in the WLEB.
Todayβs presentation title is ββ¦β¦..β
My name is Sami Shokrana and I am a PhD candidate in the in the MSU. My coauthors are Dr. Ehsan Ghane, who is also my PhD advisor. The other coauthor is Dr. Zhiming Qi, who is one of the developer of the P module of the RZWQM2.
My todayβs presentation topic is ββ¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β
First step was to chose a P loss model.
before choosing the model, We decided that the model should be able to fulfil some criteria.
The Model should be able to simulate DRP and PP loss through both surface runoff and drainage discharge
The model should have a macropore component that can simulate hydrology and P loss through macropores.
The model should have a component where plant P uptake is represented
The P transformations such as mineralization and immobilization processes should be represented
Also, if the user wants to implement different beneficial management practices to see their effect on P loss, the model should also have provision to that.
Only 2 models fulfil all these criterion. We decided to go with he RZWQM2.
RZ model was first tested by Dr. Sadhukhan, Dr. Zhiming and their team. They did an excellent job in paving the path for developing future P modules for field-scale models. Since its development, the P module of the RZ has been tested twice. For those 2 studies the soil type, climate, cropping practice was the same only management practice was different. So, there is further need for testing of the P module under different soil, climate, and cropping conditions.
The objective of my research is to βEvaluate β¦β¦β¦.β.
This study will bring out the limitation of the model and help developers identify the processes and subroutines that need to be modified for an accurate prediction of P loss from subsurface-drained field.
We also want to see how the model performs under daily flow and load data.
Mention restrictive layer
Mention how weather data were collected: You can use either daily or hourly weather data. Since, we have measured hourly weather data, we used those as input for weather files. We have an on-site ATMOS-41 weather station. If there were missing weather data, we collected them eith from NOAA or a neighboring weather station called enviroweather.
NSE is a measure of how well the model predictions are whereas
PBIAS measures the avg. tendency of the simulated data to be larger or smaller than observed data
These performance statistics were collected from Dr. Skaggs paper and Dr. Moriasiβs papers
Primary vertical axis: drainage discharge
Secondary vertical axis: precipitation
Horizontal axis: timeline of the study
This is how the precipitation looked like over the study period. And this is the observed drainage discharge in response to the precipitation. And this is how the simulated drainage discharge looked like.
This red marker differentiates between the calibration and validation period
You all probably remember how wet the spring 2019 was. Farmers could not plant anything. Which crops were planted the next years?
Overall, it seems like the model did a good job in predicted all the observed peak flow events. We know that P loss is highest during the non-growing season which is from about October-March. So, it is very important that the model predicts the hydrology well during this period and the model did predict the well during this period.
The NSE value shows good model performance, while the PBIAS shows that the model has slightly overprediction bias
Primary axis has TP load, and horizontal axis has timeline of the study period
This is our measured TP load and this is the predicted TP load by the RZWQM2
Inorganic fertilizer was applied on April 2019 and on May 2020.
This red line differentiates calibration period from the validation period.
I should also mention that RZ does not directly predict TP. It predicts DRP and PP. Since TP is a combination of DRP, DUP, and PP, for this study we assumed that DUP is zero and TP = DRP + PP.
The model was able to predict well almost all the peaks of the TP loss events. The model did well when the P loss events were smaller in magnitude. But for events with high TP loss, the simulated TP could not predict those losses very well. In other words, the difference between observed and simulated TP loss were high for large TP loss events.
Our model performance statistics also agrees to that.
For the calibration period, NSE suggests a satisfactory model performance, while PBIAS is also satisfactory with an overestimation bias
For the validation period, NSE suggest satisfactory model performance but PBIAS suggest unsatisfactory simulation with underestimation bias
This time Primary axis has DRP load in Kg/ha and horizontal axis has timeline of the study.
This is our measured DRP load and this is the predicted DRP load by the RZWQM2
Inorganic fertilizer was applied on April 2019 and on May 2020.
Same as TP, the model also could not predict well the high DRP loss events both during calibration and validation period. DRP prediction was not good for most of the higher DRP loss events.
The model performance statistics for calibration period suggest unsatisfactory model performance with overestimation beyond the range of satisfaction.The result was unsatisfactory for the validation period as well
So started to think what could be the reason behind this unsatisfactory performance of the model for DRP.
We tried to understand what happened to DRP load after fertilizer application.
Fertilizer was applied on April 2019 at a rate of 22.5 Kg/ha
Again, it was applied on May 2020 at a rate of 9.5 kg/ha
So, If I apply more fertilizer, more P should be lost through drainage discharge.
To prove this hypothesis, we increase the fertilizer input 15-fold, but still the DRP input to drainage discharge did not change, All the DRP was lost through surface runoff.
Then we tried to dig deeper. We have time-lapse cameras installed in the field. We saw there was surface ponding on the field due to heavy precipitation after fertilization event. So, our assumption is that RZ-P model cannot perform well under surface ponding conditions.
The second reason for the underestimation of the model is the presence of high water table on or near the surface. We have time-lapse cameras installed in the field and cameras show that due to heavy rainfall, there was water on the surface on the surface for those events. The images are from the May 2020 and Ocr 2021 events. And you can see ponding on the field.
So, our guess is that the RZWQM2-P has limitations in simulating P loss through drainage discharge when water table is on or near the surface
Given the dynamic behavior of P, it would be better to use a higher resolution sampling strategy to capture the variation and fluctuation of P concentration in drainage discharge
First, the RZWQM2-P model did not perform well while predicting DRP loss. This unsatisfactory performance could be attributed by the underwhelming performance of the P model under ponded condition. So, the performance of the P model needs to be improved especially after fertilizer input.Second, the P model needs to be further tested with high-resolution daily data. As mentioned earlier, daily data are better than event-based data as it captures the variability and fluctuations of P concentration.
Third, Overall, the performance of RZWQM2 model in hydrology simulation was good, the performance of P model for predicting TP was satisfactory, but for DRP it was unsatisfactory. Therefore to check for the reliability of the P model, more tests are necessary.