Propagation of climate model biases to biophysical modelling can complicate assessments of climate change impact in agricultural systems

Abstract

Regional climate model (RCM) simulations are being increasingly used for climate change impact assessments, but their application is challenging due to considerable biases inherited from global climate model (GCM) simulations and generated from dynamical downscaling processes. This study assesses the biases in NARCliM (NSW and ACT regional climate modelling) simulations and quantifies the consequence of the climate biases in the downstream assessment of climate change impact on wheat crop system, using the Agricultural Production System sIMulator (APSIM). Results showed that post-processing bias-corrected temperature and rainfall data from NARCliM had small annual mean biases but large biases in the crop growing season (CGS). During the CGS, the mean bias error of rainfall was generally positive for rainfall probability and negative for intensity, which subsequently resulted in APSIM simulating negative biases for runoff and deep drainage and positive bias in soil evaporation. Bias in soil water balance and water availability resulted in less plant transpiration and less N uptake, ultimately, leading to large negative biases in crop yields. A simple bias correction of the simulated crop yield driven by RCMs could result in a largely consistent distribution with those generated with APSIM simulations forced by observed climate. Our results showed that RCM simulation biases could confound with the climate change signal and produced an unreliable estimate of the effects of the changes in climate and farm management variables on crop yields. The results suggested that RCM simulations with the current bias correction on the RCM-simulated outputs applied on an annual basis were inadequate for climate change assessments which involve biophysical models. Our study highlights the need for improved RCM simulations by eliminating the systemic biases associated with rainfall characteristics, although suitable post-processing bias correction on a seasonal or monthly basis may result in improved RCM simulations for agricultural impacts of climate change.

The implication of spatial interpolated climate data on biophysical modelling in agricultural systems

“…The contribution of the biases in gridded climate variables to the biases in biophysical modelled outputs was quantified by the regression coefficients in Table . The different MBEs of climate variables had different effects on the MBEs in modelled outputs because different functionalities of climate variables were implemented in the biophysical model (Liu et al ., ). Regression analysis showed that the MBE in climate variables could account for an average of 71–92% (R 2 : .71 ± .20 ~ .92 ± .06) of the variance in the MBEs for APSIM‐simulated phenological parameters.…”

“…This could result in misrepresentation of production in spatial analyses including climate change impact assessments. This finding is similar to climate biases from other sources such as dynamical downscaling (Liu et al ., ).…”

“…Our study suggests that post‐processing bias‐correction with the gridded data can result in inherent biases (i.e., more frequent and less intensive rainfall bias from the gridded data treated as observations). This may explain, in part, the large biases that have been found in the bias corrected dynamical downscaling data (Liu et al ., ). To avoid such additional biases, regional modelling outputs may be interpolated to observational data and bias‐correction can be undertaken on station sites using observed station data.…”

“…To assess the impact of different climate input data on agricultural crop model outputs we selected an important agricultural area, the Murray‐Riverina cropping region (MRC), for the study domain. The MRC is located in Southern New South Wales and covers an area of 125,551 km 2 (Liu et al ., ). The region is characterized by a semi‐arid climate with a long term annual rainfall of 495 mm and long term average temperature of 9.3°C (daily minimum temperature) and 22.0°C (daily maximum temperature).…”

“…Similarly, we defined the consequent modelled biases as the difference between the APSIM‐simulated outputs forced by gridded data ( X g ) and those forced by observations ( X o ). We therefore defined mean bias error (MBE) as the difference in the mean over the assessed period between observed and gridded climate data for the mean values of APSIM simulated outputs forced with different climate data (Liu et al ., ).…”