Evaluating biases in simulated land surface albedo from CMIP5 global climate models

Abstract

Land surface albedo is a key parameter affecting energy balance and near‐surface climate. In this study, we used satellite data to evaluate simulated surface albedo in 37 models participating in the Coupled Model Intercomparison Project Phase 5 (CMIP5). There was a systematic overestimation in the simulated seasonal cycle of albedo with the highest bias occurring during the Northern Hemisphere's winter months. The bias in surface albedo during the snow‐covered season was classified into that in snow cover fraction (SCF) and albedo contrast (β1). There was a general overestimation of β1 due to the simulated snow‐covered albedo being brighter than the observed value; negative biases in SCF were not always related to negative albedo biases, highlighting the need for realistic representation of snow‐covered albedo in models. In addition, models with a lower leaf area index (LAI) tend to produce a higher surface albedo over the boreal forests during the winter, which emphasizes the necessity of improving LAI simulations in CMIP5 models. Insolation weighting showed that spring albedo biases were of greater importance for climate. The removal of albedo biases is expected to improve temperature simulations particularly over high‐elevation regions.

Climate change decreases the cooling effect from postfire albedo in boreal North America

“…Our approach and results have the potential to improve land and Earth system models, for example by providing validation data for the influence of forest structure and plant functional type composition during postfire recovery on modeled albedo. These models currently lack robust representations of fire effects in boreal environments and contain significant biases in albedo parameterizations (Li et al, ; Loranty, Berner, Goetz, Jin, & Randerson, ). Our approach also has the potential to inform land management.…”

Inferring Surface Albedo Prediction Error Linked to Forest Structure at High Latitudes

“…By comparing albedo predictions based on a forest classification that takes into account major structural differences at various successional stages of development (i.e., the enhanced ESA CCI LC product of Majasalmi et al, ) to one that does not (i.e., the original ESA CCI LC product), we were able to isolate and infer the magnitude of the albedo prediction error attributable exclusively to any difference in the representation of forest structure. Locally (i.e., at pixel scale), this structural‐related error ( ) was found to be as large as 0.19 in months with snow, which is on the same order of magnitude as that which may stem from poor climate model parameterizations of factors controlling canopy snow interception and unloading in boreal forests (Bartlett & Verseghy, ), or from deficient model parameterizations of snow metamorphosis and aging (Essery et al, , ; Y. Li et al, ). When averaged across the landscape—or a total forest area of ~611,000 km 2 —the mean absolute structural error from December to March was found to be as large as 0.02, equating to a mean absolute error in predicted RF of around 0.4 W/m 2 that is approximately equivalent to a pulse emission on the order of 273 Mt‐CO 2 ‐eq.…”

“…Our confidence to predict global climate change connected to anthropogenic greenhouse gas emissions is therefore directly tied to the skillfulness by which climate models predict (calculate) surface albedo in high‐latitude environments. Recent climate model intercomparison studies illustrate that models continue to struggle with albedo predictions in high latitude forests, particularly in the presence of snow (Boisier et al, , ; Y. Li et al, ; Loranty et al, ; Qu & Hall, ; Wang et al, ). Sources of the albedo prediction error may stem from: (i) differences in the predicted snow cover extent or snowpack physical properties; (ii) differences in the albedo parameterizations themselves (i.e., the parameters that control snow canopy interception and unloading, and snow aging and melt); (iii) differences in the vegetation mapping (i.e., biogeography); and (iv) differences in structural properties of the vegetation.…”

“…Sellers, ). While recent investigations have typically focused on the prediction error arising from differences in model parameterizations of albedo (Bartlett & Verseghy, ; Boisier et al, ; Bright et al, ; Thackeray et al, , ), model representations of snow physical attributes and/or spatial coverage (Boisier et al, ; Y. Li et al, ; Thackeray et al, ), or from differences in the vegetation mapping (Boisier et al, ), few studies have evaluated the prediction error attributable to differences in model representations of vegetation structure. Recent attribution analyses elsewhere have suggested that errors in simulated vegetation structure could be an important source of the persistent albedo prediction error seen in the current generation of climate models (Y. Li et al, ; Wang et al, ).…”

“…Wang et al () recently found strong correlations between error in predicted forest structure (leaf area index [LAI] and tree cover fraction) and error in predicted surface albedo amongst several Coupled Model Intercomparison Project Phase 5 models during winter but did not distinguish whether the error was attributed to a poor land cover/plant functional type (PFT) specification (i.e., the vegetation mapping/biogeography) or to insufficient structural parameterizations in forests (i.e., LAI and tree cover fraction). For many of the same climate models, Li et al () found a strong correlation between LAI and wintertime albedo predictions across models but did not quantify the albedo prediction error attributable to a deficiency in the LAI prediction.…”