Home >Documents >Spatial modeling of elevated groundwater nitrate ...mye/2017KarstSymposium/Canion.pdfvulnerability...

Spatial modeling of elevated groundwater nitrate ...mye/2017KarstSymposium/Canion.pdfvulnerability...

Date post:08-Aug-2020
Category:
View:1 times
Download:0 times
Share this document with a friend
Transcript:
  • Spatial modeling of elevated groundwater nitrate concentrations using random forests and regression

    kriging

    Andy CanionDean Dobberfuhl

    Lori McCloud

  • How do we find vulnerable areas to best focus projects?

    Screening Approach:Where do nitrogen sources and geologic vulnerability overlap?

    Can we use the wealth of nitrate data from well monitoring to focus efforts?

  • Header here

    • Text here

    Dataset # Wells PORFDEP Public Water System (PWS)

    499 2009 – 2013

    FDEP Water Supply Restoration (WSRP)

    968 2000 – 2015

    SJRWMD 44 2000 – 2013SWFWMD 7 2000 – 2013 USGS (Phelps 2004) 36 2001 – 2002

  • 0

    5

    10

    15

    20

    0 50 100 150 200ICUthick

    NO

    x (m

    g/L)

    L)

    0

    5

    10

    15

    20

    0 10 20 30RECH

    NO

    x (m

    g/L)

    Is there information in the data?

  • Regression Tree Approach

  • Ensemble Result

    Random Forest Algorithm

    Guo et al. 2011

  • St. Johns River Water Management District

    Probability of nitrate exceeding a chosen threshold is predicted by:

    Random Forest Classification Model

    • Well Depth• Confinement • Aquifer Depth • Recharge • Water Table Depth • Soil Hydraulic

    Conductivity

    • Soil Drainage Class • Ecoregion • Sinkhole Frequency • Land Use • Nitrogen Load • Septic Density

  • Model Area

    Nitrate Threshold

    (mg/L)

    Predictor Variable

    Subset (m)Number of Trees

    Out-of-Bag

    Error

    Area Under ROC

    Curve

    Silver 0.35 2 1,000 17.31% 0.89

    Silver 1.2 7 1,000 21.24% 0.84Wekiwa/ Blue 0.35 2 1,000 21.25% 0.86Wekiwa/ Blue 0.8 2 1,000 20.75% 0.86

    Model Diagnostics

  • Soil Hydraulic ConductivityEcoregion

    Soil Drainage ClassDepth to Water Table

    Land Use Category

    Nitrogen LoadICU/Overburden Ratio

    Sinkhole FrequencySeptic Density

    Depth to FloridanICU Thickness

    Well Depth

    Recharge

    20 30 40 50 60 70 80

    Silver Springs 1.2 mg/L Thresh

    MeanDecreaseAccuracy

    Geology / Hydrogeology

    Soils, Loading

    Model Diagnostics - Silver

  • Geology / Hydrogeology

    Soils, Loading

    Model Diagnostics – Wekiwa/Blue

    Ecoregion

    Nitrogen Load

    Land Use Category

    Soil Hydraulic Conductivity

    Septic Density

    ICU/Overburden Ratio

    Depth to Floridan

    Depth to Water Table

    ICU Thickness

    Sinkhole Frequency

    Well Depth

    Recharge

    30 40 50 60 70 80 90

    Wekiwa/Blue 0.35 mg/L Thresh

    MeanDecreaseAccuracy

  • Kriging of Random Forest Residuals

    Probability of Exceeding Concentration Threshold

    Random Forest Prediction m(s)

  • Includes kriged residuals

  • No spatial correlation in residuals – not kriged

  • Legacy Nitrate

  • Thank You

    Spatial modeling of elevated groundwater nitrate concentrations using random forests and regression krigingSlide Number 2Slide Number 3Slide Number 4Slide Number 5Slide Number 6Slide Number 7Random Forest AlgorithmSlide Number 9Slide Number 10Slide Number 11Slide Number 12Slide Number 13Slide Number 14Slide Number 15Slide Number 16Slide Number 17Thank YouExtra SlidesSlide Number 20Slide Number 21Slide Number 22Slide Number 23Slide Number 24Slide Number 25Slide Number 26Slide Number 27Slide Number 28Slide Number 29

of 18/18
Spatial modeling of elevated groundwater nitrate concentrations using random forests and regression kriging Andy Canion Dean Dobberfuhl Lori McCloud
Embed Size (px)
Recommended