An Available Water Capacity Pedotransfer Function using Random Forest - 2020 Cornell Soil Health Model
Harvard Dataverse (Africa Rice Center, Bioversity International, CCAFS, CIAT, IFPRI, IRRI and WorldFish)
View Archive InfoField | Value | |
Title |
An Available Water Capacity Pedotransfer Function using Random Forest - 2020 Cornell Soil Health Model
|
|
Identifier |
https://doi.org/10.7910/DVN/U5DAEP
|
|
Creator |
Amsili, Joseph
Harold van Es Robert Schindelbeck |
|
Publisher |
Harvard Dataverse
|
|
Description |
In late 2018, the Cornell Soil Health lab determined that AWC, a valuable, but time-intensive measurement, could be accurately predicted. A CASH database containing 7,885 soil samples was used to develop a Random Forest model to predict AWC from a suite of measured parameters, including % sand, % silt, % clay, Organic Matter, Active Carbon also known as Permanganate Oxidizable Carbon (POxC), Respiration, Wet Aggregate Stability, Potassium, Magnesium, Iron, and Manganese. The Random Forest model was able to explain more variation in AWC than alternative multiple linear regression models. Specifically, the RF model was able to explain 71.8 % of the variation in AWC with a low average root mean square error (RMSE) value. The RMSE value was only 3% of actual AWC values, which is equivalent to the sensitivity of the laboratory method. Therefore, our predicted values had no more error than the original raw AWC data. |
|
Subject |
Agricultural Sciences
AWC Available Water Capacity Random Forest Soil Health Indicator |
|
Contributor |
Amsili, Joseph
|
|