Keywords
model prediction robustness, climate change, PEST, RZWQM, PDSI
Start Date
25-6-2018 2:00 PM
End Date
25-6-2018 3:20 PM
Abstract
Making projections with models is always difficult and climate change poses a particularly challenging problem for biophysical models which tend to be over-parameterized and have poor predictive power when extrapolations are beyond the range of the calibration. It is common practice to assume that if a calibrated model replicates observations reasonably well, predictions under other conditions will also be reasonably good. Unfortunately, this assumption is not always correct, as we show for prediction of nitrate loss from a tile-drained, corn-soybean experiment in Northern Iowa.
The RZWQM is a biophysical process model that simulates plant growth and movement of water and nutrients in agricultural systems. Using experimental data over 12 years, we investigated the robustness of RZWQM predictions of crop yield, subsurface drainage flow, and nitrate-N loss of multiple model calibrations using the PEST parameter estimation software.
Model prediction robustness was found to be related to the range of soil moisture conditions in the calibration data. Calibration data representing a particular range of Palmer Drought Severity Index (PDSI) allow a calibration able to predict performance in years exhibiting a similar range of PDSI. We found that the addition of a single year’s data identified by PDSI to a five-year calibration improved Nash-Sutcliffe model efficiency coefficient (NSE) from -0.22 to 0.7 and achieved nearly all of the improvement possible using all available observations. The range of PDSI was found to be a suitable measure of the information content of hydrologic calibration data for RZWQM and useful in assessing the range of robust prediction.
Modelers must use extreme caution when making projections for conditions beyond the range of available calibration data. Indicators of model projection robustness are necessary to build confidence in extrapolations under a changing climate. We propose and demonstrate the PDSI as an robustness indicator for projections of the movement of water in biophysical models.
A metric for evaluating the ability of the RZWQM model to project the impact of climate change
Making projections with models is always difficult and climate change poses a particularly challenging problem for biophysical models which tend to be over-parameterized and have poor predictive power when extrapolations are beyond the range of the calibration. It is common practice to assume that if a calibrated model replicates observations reasonably well, predictions under other conditions will also be reasonably good. Unfortunately, this assumption is not always correct, as we show for prediction of nitrate loss from a tile-drained, corn-soybean experiment in Northern Iowa.
The RZWQM is a biophysical process model that simulates plant growth and movement of water and nutrients in agricultural systems. Using experimental data over 12 years, we investigated the robustness of RZWQM predictions of crop yield, subsurface drainage flow, and nitrate-N loss of multiple model calibrations using the PEST parameter estimation software.
Model prediction robustness was found to be related to the range of soil moisture conditions in the calibration data. Calibration data representing a particular range of Palmer Drought Severity Index (PDSI) allow a calibration able to predict performance in years exhibiting a similar range of PDSI. We found that the addition of a single year’s data identified by PDSI to a five-year calibration improved Nash-Sutcliffe model efficiency coefficient (NSE) from -0.22 to 0.7 and achieved nearly all of the improvement possible using all available observations. The range of PDSI was found to be a suitable measure of the information content of hydrologic calibration data for RZWQM and useful in assessing the range of robust prediction.
Modelers must use extreme caution when making projections for conditions beyond the range of available calibration data. Indicators of model projection robustness are necessary to build confidence in extrapolations under a changing climate. We propose and demonstrate the PDSI as an robustness indicator for projections of the movement of water in biophysical models.
Stream and Session
D2: Extreme Events: Improve Predictability, Increase Resilience, and Address Knowledge Gaps