Title: Developing Regression Models for Sediment Yield and Nutrient Loss From Biomass Crops
1Developing Regression Models for Sediment Yield
and Nutrient Loss From Biomass Crops
- E. Z. Nyakatawa1, D.A. Mays1 , and V.R.
Tolbert2, - 1Department of Plant and Soil Science, Alabama
AM University, P.O. Box 1208, Normal, AL 35762
2Bioenergy Feedstock Development Programs,
Oakridge National Laboratory, Oakridge, TN 37831
2INTRODUCTION
Soil Erosion An environmental problem.
- Siltation of rivers, dams, and lakes.
- Pollution of surface and ground water resources.
3- Eutrophication of rivers and lakes.
- NO3- and soluble P (H2PO4-) in runoff water.
- NH4 and P attached to sediment.
4Evaluating the sustainability of growing biomass
crops.
- Information needed
- How much runoff occurs.
- How much soil is lost as eroded sediment.
- Amount of nutrients, esp. N (NO3- and NH4 ) P
are lost in runoff water and eroded sediment.
5STUDY
- Co-operation between Tennessee Valley Authority
- (TVA), DOEs Bio-energy Feedstock Development
- Program (BFDP), Oakridge TN and Alabama
- AM University, Huntsville AL.
- Objective
- To assess the environmental impacts of bio-energy
- crop production on soil and water quality.
6- Location
- Winfred Thomas Agricultural Research Station
(WTARS), Hazel Green AL, from 1995 to 1999. - Soil type
- Decatur silt loam (clayey, kaolinitic thermic,
Typic Paleudults).
7TREATMENT FACTORS
- No-till Corn (NTC)
- Switchgrass cover crop (SWG)
- Sweetgum with cover crop (SGC)
- Sweetgum without cover crop (SWC)
8FIELD PLAN
- Eight runoff plots (four 0.18 ha and 0.43 ha)
established. - Treatments randomly assigned to plots with each
treatment getting one large and one small plot
(see diagram).
9Plot layout, WTARS
10FIELD PLAN
- Eight runoff plots (four 0.18 ha and 0.43 ha)
established. - Treatments randomly assigned to plots with each
treatment getting one large and one small plot
(see diagram). - Earth berm 50 cm high around each plot to
isolate them from surrounding fields.
11DATA COLLECTED
- Rainfall.
- Runoff volume.
- Sediment yield.
- Nutrient loss.
- (nitrate, ammonium, and P).
12Runoff and sediment collecting point
13Setup for collecting runoff and sediment V flume
14Housing for data loggers
15Data collection and storage
16Data collection and storage
17Sweet gum plot
18Switchgrass Plot
19Corn plot
20RESULTS AND DISCUSION
21Rainfall and runoff, WTARS 1995-1999
22Runoff NH4 as affected by no-till corn (NTC),
switchgrass (SWG), sweetgum wo/cover (SWC), and
sweetgum w/cover (SGC) treatments, WTARS, AL
23Runoff NO3 as affected by no-till corn,
switchgrass, sweetgum wo/cover, and sweetgum
w/cover treatments, WTARS, AL
24Runoff P as affected by no-till corn,
switchgrass, sweetgum wo/cover, and sweetgum
w/cover treatments, WTARS, AL
25Sediment yield as affected by no-till corn,
switchgrass, sweetgum wo/cover, and sweetgum
w/cover treatments, WTARS, AL
26Sediment NO3 as affected by no-till corn,
switchgrass, sweetgum wo/cover, and sweetgum
w/cover treatments, WTARS, AL
27Sediment P as affected by no-till corn,
switchgrass, sweetgum wo/cover, and sweetgum
w/cover treatments, WTRS, AL
28- Fitting regression models
- To relate sediment yield, NH4 and NO3
- losses to rainfall and runoff.
- Factors affecting sediment yield and
- nutrient loss
- Climate
- Soil type
- Topography (slope)
- Land use
29- In this study
- Climate, soil type, and slope are similar for all
treatments. - Land use different as defined by each
treatment. - Therefore, interested in models for each
treatment to be able to compare land use systems.
30Strategy for Developing Regression Models
Start
Exploratory data analysis
Develop tentative regression models
Is one or more models suitable?
NO
Revise models/new ones?
YES
Identify most suitable model
Make inferences based on model
Stop
31- Statistics for developing regression
- models.
- Several regression analyses procedures.
- Proc Reg, Proc Glm, Proc rsreg, Proc Catmod etc
- Each procedure has special features for handling
similar or different data types. - Choice - largely on applicability of each proc
and degree of complexity of data. - However, individual preference also plays big
part.
32- However, no matter how artistic model
development can be, reality must be kept at
manageable proportions !! - Only a limited number of of independent or
predictor variables (parameters) should be
included in the model. - Considerations in choosing model parameters
- Contribution to reducing variation in dependent
variable. - Degree to which parameter can be obtained
efficiently (quickness, cost, accuracy). - Importance of parameter to the study.
33Scatter plots showing relationships between
sediment yield, rainfall and runoff volume
No-till corn
Switch grass
Sweetgum
Sweetgum w/cover
34Scatter plots showing relationships between NO3
loss, rainfall and runoff volume
No-till corn
Switch grass
Sweetgum
Sweetgum w/cover
35Scatter plots showing relationships between P
loss, rainfall and runoff volume
No-till corn
Switch grass
Sweetgum
Sweetgum w/cover
36Scatter plots showing relationships between
runoff volume and rainfall
No-till corn
Switch grass
Sweetgum
Sweetgum w/cover
37- In this study
- Proc REG choice over Proc GLM.
- Proc REG more diagnostic tools and
capabilities. - SAS ver. 6.02 introduced two new model-selection
methods for fitting the best model. - Adgjusted R2 (ADRSQ) and Mallows CP statistic.
38SAS Procedure
- Proc rsquare cp mse adjrsq by trt by season
trt - rsquare regression SS/corrected total sum of
squares - this finds a no.of models with the highest R
square. - cp Mallows Cp
- (Residual SSp)/ (Res MS from full model) 2p-n
- this finds a no.of models with the lowest Cp.
- adjusted R2 (ADRSQ)
- 1 Error Mean Square/Total MS 1 (1-
R2)(n-1)/(n-p) - this finds a no.of models with the highest
adjusted R square - MSE mean square error.
39SAS Model Statement
- Model sediment NH4 NO3 X1 X2 X3 Y1 Y2
Y3 Z1 - Where
- X1 rainfall
- X2 rainfallrainfall
- X3 rainfallrainfallrainfall
- Y1 runoff
- Y2 runoffrunoff
- Y3 runoffrunoffrunoff
- Z1 rainfallrunoff.
40Regression Models for Dependent Variable
SEDIMENT Number R-square Adjusted C(p)
MSE Variables in Model Model
R-square 1 0.06656206
0.05018596 1.11168 349275.40 X1
. ... . .
1 0.01977926 0.00258240
1.58914 366780.67 Y1 ------------------------
--------------------------------------------------
----------------------------- 2
0.08346260 0.05072912 0.08737 349075.66
X2 X3 . . .. .
. . ------------------------------------
--------------------------------------------------
------------------ 3 0.09370310
0.04426873 1.32144 351451.34 X1 X2 X3
. . . .. . .
--------------------------------------
--------------------------------------------------
----------------- 4 0.10033782
0.03369618 2.93841 355339.19 X1 Y1 Y2
Y3 . . . . .. .
. ----------------------------------
--------------------------------------------------
--------------------- 5 0.11168467
0.02788134 4.28335 357477.48 X2 X3 Y1
Y2 Y3 . . . . .. .
. ---------------------------------
--------------------------------------------------
--------------------- 6 0.11506849
0.01296101 6.08799 362964.14 X2 X3 Y1
Y2 Y3 Z1 -----------------------------------------
--------------------------------------------------
------------- 7 0.11659270
-.00465929 8.00000 369443.65 X1 X2 X3
Y1 Y2 Y3 Z1 --------------------------------
--------------------------------------------------
-----------------
41-----------------------------TRT
NTC------------------------------
---------------------------TRT NTC
WINTER-----------------------------
----------------------------TRT NTC SPRING
-----------------------------
--------------------------TRT NTC SUMMER
----------------------------
42-------------------------------- TRT SGC
-------------------------------
-------------------------- TRT SGC WINTER
-----------------------------
------------------------- TRT SGC SPRING
-----------------------------
------------------------ TRT SGC SUMMER
----------------------------
43-------------------------------- TRT SWC
-------------------------------
-------------------------- TRT SWC WINTER
-----------------------------
------------------------- TRT SWC SPRING
-----------------------------
------------------------ TRT SWC SUMMER
----------------------------
44-------------------------------- TRT SWG
-------------------------------
-------------------------- TRT SWG WINTER
-----------------------------
------------------------- TRT SWG SPRING
-----------------------------
------------------------ TRT SWG SUMMER
-----------------------------
45SUMMARY
- For each treatment and dependent variable, best
model varies with season (most likely reflecting
rainfall variation). - Seasonal changes also affect nutrient chemistry
hence behavior (minerilization, immobilization,
microbial activity) - Therefore, have to develop models separately for
each season..these can be called in succession
during a simulation. - Within treatments and seasons, model parameters
seem to change depending on amount rainfall and
initial soil moisture content (affects runoff,
and behavior of soil particles) .