Increased harvested carbon of cropland in China

Crop harvested carbon (HC) is one of the most important components of the carbon cycle in cropland ecosystems, with a significant impact on the carbon budget of croplands. China is one of the most important crop producers, however, it is still unknown on the spatial and temporal variations of HC. This study collected statistical data on crop production at the province and county levels in China for all ten crop types from 1981 to 2020 and analyzed the magnitude and long-term trend of harvested crop carbon. Our results found a substantial increase of HC in cropland from 0.185 Gt C yr−1 in 1981 to 0.423 Gt C yr−1 in 2020 at a rate of 0.006 Gt C yr−1. The results also highlighted that the average annual carbon sink removal from crop harvesting in China from 1981 to 2020 was 0.32 Gt C yr−1, which was comparable to the net carbon sink of the entire terrestrial ecosystems in China. This study further generated a gridded dataset of HC from 2001 to 2019 in China by using jointly the statistical crop production and distribution maps of cropland. In addition, a model-data comparison was carried out using the dataset and results from seven state-of-the-art terrestrial ecosystem models, revealing substantial disparities in HC simulations in China compared to the dataset generated in the study. This study emphasized the increased importance of HC for estimating cropland carbon budget, and the produced dataset is expected to contribute to carbon budget estimation for cropland ecosystems and the entire China.


Introduction
As one of the most important terrestrial ecosystems, cropland plays a crucial role in regulating the terrestrial carbon cycle (Bondeau et al 2007).Globally, approximately 12% of the ice-free land surface is occupied by cropland, with large spatiotemporal variations influenced by factors such as population changes, crop product prices, and socioeconomic factors (Yuan et al 2018).The vegetation gross primary productivity of croplands contributes to 12%-16% of global vegetation gross primary productivity (Cai et al 2014), and also partly controls the seasonal fluctuations of the terrestrial carbon cycle (Zeng et al 2014).Previous studies have also emphasized the positive relationship between the growth of cropland production and the seasonal variations in atmospheric CO 2 concentration (Gray et al 2014).Furthermore, over the past few decades, there has been a notable expansion of cropland at the expense of forests, grasslands, and other ecosystems (Winkler et al 2021), leading to an increased contribution to the global carbon cycle (Erb et al 2017).Consequently, it is imperative to accurately estimate the magnitude and changes in the carbon cycle for agricultural land areas (Osborne et al 2010).
In crop ecosystems, during harvest and removal from cropland, the carbon in grains will be emitted to the atmosphere in a short period along with human and livestock consumption.It is widely acknowledged that the carbon derived from harvested grain significantly influences the carbon budget of the cropland ecosystem (Smith et al 2010).Measurements of eddy covariance in 17 European cropland ecosystems showed that the cropland transitioned from carbon sink to carbon source after accounting for harvested carbon (HC) at crop harvest (Moors et al 2010).Regional studies also highlighted that HC dominates the carbon budget not only for cropland but also at the entire country level (Ciais et al 2010).A recent study estimated that annual carbon losses from cropland harvesting were 0.44 Gt C yr −1 within the continental United States, which was 2.56 times of annual mean net carbon sink (0.17 Gt C yr −1 ) (Liu et al 2020).According to statistical yield datasets, the average crop harvest in 25 European countries was 275 and 239 g C m −2 yr −1 during the period 1990-1999, representing approximately 40% of net vegetation production (Ciais et al 2010).The global cropland harvest in 2000 was 3.2 Gt C yr −1 , which is around half of the total biomass harvest or the global terrestrial net carbon sink (Wirsenius 2003, Krausmann et al 2008, Friedlingstein et al 2022).
The carbon harvested significantly affects the carbon budget of cropland ecosystems, and it has been the most common land management-related process considered in current terrestrial ecosystem models (Gervois et al 2004, Lokupitiya et al 2009, Drewniak et al 2013, Pongratz et al 2018).To integrate agriculture into a comprehensive land biosphere model, Kucharik and Brye (2003) incorporated a crop growth model (i.e.Erosion-Productivity Impact Calculator) into the Integrated BIosphere Simulator (IBIS) (Foley et al 1996, Yuan et al 2014) to simulate the agricultural process.More recently, the community Land Model Version 5 (CLM5.0)introduced a crop module to accurately represent crop distributions and management, including dynamic major crop distributions, fertilization, and irrigation management (Lombardozzi et al 2020).
Although previous research has made valuable efforts, there is still great uncertainty in accurately quantifying harvested carbon in cropland (HCC) ecosystems (Smith et al 2010, Lun et al 2016).For example, the CLM5.0 crop model can reproduce the observed long-term trend of crop yields with relative accuracy before 1990 but showed a substantial underestimation after 1990 (Lombardozzi et al 2020).There are still numerous challenges for terrestrial ecosystem models to simulate crop yield.Specifically, models need accurately represented the multiple crop properties (e.g.planting density, crop variety, crop breeding), field management practices (e.g.irrigation, tillage, fertilization), and impacts of climate change (e.g.drought, flood, heatwave), which largely affect crop yield and crop-carbon cycle interactions (Yuan et al 2016).Most of the above factors highly depend on human decisions and activities, which strongly challenge the model's ability (Cheng et al 2014).
Few terrestrial ecosystem models, therefore, can represent crop properties and management practices due to the complicated interactions between human decisions and biogeochemical cycle processes (Zhang et al 2018).Therefore, many models used the globally fixed ratio of removed carbon with large uncertainties in simulating regional and global carbon budgets (Stocker et al 2011, Malyshev et al 2015).In contrast, a large volume of statistical data is more readily available, which provides a good opportunity to generate a dataset of HC instead of relying on terrestrial ecosystem models (National Bureau of Statistics of China 2020).
As one of the largest crop producers globally, China contributes approximately 610 million tons of grain annually, representing 20% of the total global crop production (FAO 2021).In addition, the cropland accounts for about 13% of total land area in China (Chen et al 2022b), and plays an important role in determining terrestrial carbon budget.However, it is still unknown about the magnitude and trend of HC in China due to limited information and insufficient model capability.This study first collected agricultural census data at the province and county levels from 1981 to 2020 and investigate the magnitude and long-term trend of harvested crop carbon.In addition, we generated a gridded dataset of harvested crop carbon from 2001 to 2019 based on the existing crop distribution maps in China.Finally, based on this newly produced gridded dataset, we examined the model performance of seven state-of-the-art terrestrial ecosystem models in reproducing harvested crop carbon.

Agricultural statistical data
This study collected province and county-level statistical data on the planting area, yield, and production of various crops from agricultural yearbooks of various regions (National Bureau of Statistics of China 2020).We collected province-level statistical data of 10 crop types for all investigated 31 provinces from 1981 to 2020, which was used to analyze the longterm trend of harvested crop carbon in China.To generate the gridded dataset of harvested crop carbon, i.e.Terrestrial Ecosystem Disturbance-HCC (TED-HCC), we collected county-level statistical data of crop production from 2001 to 2019 to match with the distribution map of cropland (see section S2).Province-level data was obtained from the agricultural section of the National Statistical Yearbooks, and county-level crop production data was sourced from the sections on key economic indicators of counties (cities, districts) in the statistical yearbooks of each province.For the county without statistical production through the period of 2001-2019, we conducted a gap-filling method to fill the missing county-level statistical production.(1) For a given county, the missing statistical production was less than three years continuously, then a linear equation was used to fill the missing statistical production based on the statistics of the adjacent two years.( 2) If the years of missing statistical production were more than three years continuously, the mean ratio between county-level and province-level statistical production in the other years was calculated, and calculated the statistical production of this given year for this given county based on the mean ratio and province-level production of this year.

Estimation of harvested crop carbon
Harvested crop carbon was estimated by considering crop production data from provincial or county statistical sources.Given the significant variations in carbon content among different crops, this study employed specific carbon content values of each crop (table 1) (Baes et al 1984) to calculate the HC based on their statistical production.The investigated crops include ten types: grain (wheat, corn, and rice), oilseeds, cotton, hemp, sugar beet, sugarcane, tea, tobacco, fruit, and vegetables.Furthermore, we allocated county-level HC of 10 crops to the pixel level in each county, with a focus on wheat, maize, and rice, which are key for determining harvested crop carbon and account for 74% of China's total crop production.Using the ChinaCropArea1km dataset, we assigned county-level HC to corresponding 1 × 1 km pixels for these grains.This method, assuming higher HC in pixels with greater leaf area index, applies an equation (see section S2) to calculate the allocated HC at each pixel.For crops like sugarcane and oilseeds without specific distribution maps, we used the China Land Cover Dataset for allocation, after removing areas of wheat, maize, and rice.Due to the absence of comprehensive crop distribution maps and to reduce uncertainties, the gridded datasets were aggregated into a 10 × 10 km resolution for comparison with ecosystem model simulations of HC.

Generation of gridded HC dataset and model comparison
To examine the performance of state-of-the-art terrestrial ecosystem models in reproducing the spatial and temporal variations of harvested crop carbon, this study generated a gridded dataset of HCC ecosystems from 2001 to 2019, referred to as TED-HCC, as a component of the TED dataset.The gridded dataset was generated through the combined utilization of statistical data, crop distribution datasets, and satellite-based vegetation index.The detailed method of generating gridded HC dataset was introduced in the supplementary (see section S1).
This study collected simulations of harvested crop carbon from the Global Carbon Budget (Friedlingstein et al 2022) To meticulously analyze interannual variations, we employed linear detrending on the HC outputs from each terrestrial ecosystem model.This method meticulously subtracted the calculated linear trend from the original data series, enabling us to isolate and focus on the cyclical fluctuations inherent in the data.

Magnitude of HC and its spatial-temporal changes
Based on the province-level statistical data, we first analyzed the magnitude of HC and its spatiotemporal patterns.During the past 40 years, the mean HC of the cropland ecosystem in China was determined to be 0.32 Gt C yr −1 in China (figure 1).The HC in China showed substantial temporal change during the past 40 years.Specifically, total HC increased by about 2.3 times, from 0.185 Gt C yr −1 in 1981 to 0.423 Gt C yr −1 in 2020 (figure 1).All crop types showed an increasing trend, with grain types (i.e.rice, wheat, and maize) contributing the most to China's total harvested growth (68.3%)(figure 2(b)) because of their larger share (figure 2(a)).
Notably, there were significant variations in the distribution of HC across different crop types.The grain, including maize, rice, and winter wheat, accounted for the largest proportion of total HC at 75.9% through 1981-2020 (figure 2(a)).In addition, sugarcane, vegetables, and oilseeds contributed shares of 11.1%, 5.2%, and 4.3% respectively.Collectively, these four crop types accounted for  more than 96% of the total HC (figure 2(a)).Other six types of crops (i.e.cotton, hemp, sugar beet, tobacco, tea, and fruits) only shared 3.7% of the total HC (figure 2(a)).Of these, hemp, tobacco and tea accounted for the smallest share, all at 0.2%.
Large differences of harvested crop carbon also were found among the various provinces.The largest five provinces, i.e.Guangxi, Henan, Shandong, Sichuan, and Heilongjiang, contributed more than 36% of total HC in China (figures 2(c) and S1).While, these  five provinces also contributed the largest increased shares, and contributed more than 48.5% increases of HC in China (figure 2(d)).
Both planting area and yields have had a significant impact on harvested crop carbon over the past 40 years (figure 3).Most provinces increased in planting area and yields, which jointly promoted an increase in crop HC.Only a few provinces, such as Zhejiang, Guangdong, Fujian, Beijing, Shanghai, Sichuan, showed a decline in HC in the last 40 years (figure 3).As figure 3 shown, only Fujian province, the decreased HC was attributed to the decreases of both planted area and yield (figure 3), and other 5 mainly due to a decrease in planted area (figure 3).There were several provinces with decreased planting areas, but the HC still showed the increased trend, e.g.Shanxi, Jiangsu (figure 3).

Gridded dataset of HC
Based on the statistical dataset, this study generated a gridded dataset of HC in China with a spatial resolution of 10 × 10 km (i.e.TED-HCC dataset).Notably, the dataset excludes Hongkong, Macau, and Taiwan due to the lack of statistical data.Figure 4 reveals that the HC was mainly distributed in Northeast, North, and Southwest China.The intensity of HC varies from 1 g C m −2 yr −1 to 467 g C m −2 yr −1 (as depicted in figure 4(a)).Through the past two decades (2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018)(2019), the intensity of HC showed an increasing trend over more than 42.1% of regions (34.4% with a significant upward trend), and only 18.4% showed a decreased trend (5.1% with a significant downward trend) (figure 4(b)).It should be noticed that the decreasing trend mainly occurred in the Southeast region.It should be noticed that the availability of statistical crop production highly determined the accuracy of TED-HCC dataset.Although there existed data gaps of county-level crop production, the statistics of major crops were available.Grain crops (maize, wheat, and rice) contributed the largest share (75.9%) of total HC (figure 2(a)), and the collected countylevel statistical production accounted for 68% provincial production averaged from 2001 to 2019 (figure S2(a)).The sugarcane and oilseed were the second and fourth largest contributors of the total HC (i.e.11.1% and 4.3%), and the county-level statistical production accounted for 76% of the provincial statistical production.Therefore, these three crops accounted for more than 90% of HC, and the available county-level statistical production accounted for 72% of provincial production.

Model-data comparison
Based on the TED-HCC dataset of HC generated by this study, we further evaluated the model performance of seven terrestrial ecosystem models for simulating HC.On average, the simulated HC by seven models ranged from 0.14 Gt C yr −1 to 0.87 Gt C yr −1 through 2001-2019, and there were 10.4% to 143.2% differences compared with the HC derived from statistical data (0.36 Gt C yr −1 ) (figure 5(c)).The mean simulations by CLM5.0 (0.39 Gt C yr −1 ) (figure 5(c)) were quite close to HC derived from statistical data.However, the simulations by other six models showed the large differences with HC derived from statistical data (figure 5(c)).The simulated HC trends by seven models ranged from 0.0012 Gt C yr −1 to 0.018 Gt C yr −1 2001-2019, with CLM5.0 still demonstrating the trend closest to the statistical results (figure 5(d)), while the trends simulated by the other six models continue to exhibit significant differences from the statistical data.Furthermore, although there is a substantial variation in the inter-annual changes of the six models compared to the statistical data, a closer inspection reveals that the trends of ISBA, ORCHIDEE, and IBIS models are quite similar to each other, as depicted in figures 5(a) and (b).
Our results also revealed significant variations in the spatial distribution of simulated HC among seven terrestrial ecosystem models when compared to the HC derived from the TED-HCC dataset (figures 6 and S3).The majority of models can characterize the main spatial pattern with the large intensity of HC  in Northeast, North, Southeast, and Southwest China (figure 6).However, several models (e.g.JSBACH and OCN) failed to accurately depict the spatial distribution compared with the gridded dataset generated by this study (figures 6(c) and (e)).For example, the JSBACH model only showed Southeast China as the main region of HC in China and lost other regions such as Northeast China (figure 6(c)).While most models demonstrated a significant intensity of HC in North China, the most substantial variations among models were also observed in this region (figure 6).
We further analyzed the model performance for reproducing the trend of HC compared with the observed trend.In general, seven ecosystem models showed large spatial differences in the simulated trends (figures 7 and S4).CLM5.0,ISBA, LPJ-GUESS, IBIS, and ORCHIDEE showed a similar spatial pattern on the trend of simulated HC, and a largely increased trend was observed in northern China (figures 7(a), (b), (d), (f) and (g)).
However, the spatial pattern showed large differences from the observed trend in the TED-HCC dataset, and large increases were found in Northeast China, and decreased trends existed in Southern China (figure 7(h)).Although the other two models showed the decreased trend of HC over some regions (figures 7(c) and (e)), the decreased regions largely differed with the TED-HCC dataset (figure 7(h)).In addition, the observed HC shows significantly increased trend from 2001 to 2019 with a rate of 0.0083 Gt C yr −1 (figure 5(d)).

Discussion
Harvested crop carbon play an important role in determining carbon budget of cropland and even entire terrestrial ecosystems in China.According to our dataset, there were 0.32 Gt C yr −1 exported from cropland ecosystem to human society averaged from 1981 to 2020, which was comparable with the terrestrial carbon sink in China (i.e.0.19-0.26Gt C yr −1 ) (Piao et al 2009).More important, the HC substantially increased by the rate of 0.006 Gt C yr −1 through the past 40 years, from 0.18 Gt C yr −1 in 1981 to 0.42 Gt C yr −1 in 2020 (figure 1).This trend is closely linked to the observed increase in crop yield benefit (tech-enhanced harvest index, climate-induced multi-cropping), as higher yields directly translate to greater amounts of harvested crop carbon.As HC is one of the largest carbon fluxes in cropland ecosystems, the increased HC substantially impacts carbon budget.Based on eddy covariance measurements of rotation cropland in the North China Plain, a study estimated a winter wheat ecosystem to be a large carbon sink (359 g C m −2 yr −1 ), but only a carbon sink of 90 g C m −2 yr −1 after accounting for HC (Wang et al 2015).In addition, numerous studies highlighted the harvest index (the ratio of grain to aboveground biomass) largely increased (Yang and Zhang 2010), which implied more aboveground biomass allocated to grain and reduced the magnitude of straw resulting in less accumulation of soil organic carbon in cropland ecosystems.
Although the total harvested crop carbon highly increases over the entire China during the past 40 years, there are large spatial heterogeneity of changes in HC.In general, the harvested crop carbon showed the increased trend in the north China, but decreased trend in south China.Previous study supported our conclusion and showed substantial regional shifts in crop cultivation and the subsequent impacts on crop production in China (Yuan et al 2018).They found that South China have lost large planting areas of cropland in order to pursue industry and commerce, but Northeast and Northwest China have witnessed increases in planting areas (Yuan et al 2018).The competition for resources between agricultural and nonagricultural sectors and across regions has been intensified by the rapid industrialization, urbanization, strong income growth, and population expansion.
Model-data comparison conducted in this study showed the large uncertainties of state-of-the-art terrestrial ecosystem models for simulating HC (figures 5-7, S3 and S4). Figure S5 showed the provincial crop areas derived from LUH2 dataset was close to the statistical areas at most provinces, and which implied the model capability was the more important reason for the uncertainties of simulated HC.Simulating harvested crop carbon is a considerable challenge for terrestrial ecosystem models (Lombardozzi et al 2020).The primary difficulty arises from the need to accurately represent crop management at regional and global scales (Licker et al 2010).Substantial efforts have been made to produce regional and global maps of cropland management (Xiang et al 2020).However, there are still large uncertainties due to the lack of statistical and survey datasets (Siebert et al 2010).For example, a recent study found that the widely used FAO irrigation dataset (i.e.Global Map of Irrigation Areas) in China (Xiang et al 2019) has low accuracy.In addition, several other important factors highly determined crop yield, but there was no regional and global information yet (Cabas et al 2010).Previous studies highlighted the improvement of crop variety contributed substantially to the rising of crop yield (Xiong et al 2014).A recent study showed that satellite-based light use efficiency incorporating regional information on crop variety can significantly improve the estimates of crop yield (Dong et al 2020).However, spatial information on crop variety was lacking in most regions.Therefore, most process-based terrestrial ecosystem models did not integrate several important crop management information (Pongratz et al 2018), which led to large uncertainties in simulating crop production.
TED-HCC, therefore, has an important implication for carbon budget estimates as the HC largely determined the carbon budget of cropland ecosystems (Zhang et al 2015).Therefore, a national dataset of HC generated by this study had an important role in improving the estimates of country and global-level carbon budgets.Currently, Global Carbon Budget provided several important forest and cropland disturbance information, which were difficult to simulate by terrestrial ecosystem models (Friedlingstein et al 2022).For example, the LUH2 dataset provided gridded annual harvest forest carbon (Chini et al 2021), which was quite similar to harvest carbon in the cropland ecosystems in terms of strong human decision.However, Global Carbon Budget did not provide harvest crop carbon due to data scarcity (Friedlingstein et al 2022).Our results showed the averaged harvested crop carbon in China is 0.32 Gt C yr −1 from 1981 to 2020, which is larger than the harvested forest carbon reported by the LUH2 dataset in China (0.04 Gt C yr −1 averaged from 1981 to 2020).More importantly, the former will emit into the atmosphere at a faster rate compared with harvested forest carbon.In addition, this dataset (i.e.TED-HCC) is urgently needed for the top-down method (i.e.atmospheric inverse model) to deduct HC.As an important carbon budget assessment method, atmospheric inverse models have been widely used to estimate the regional and global terrestrial carbon sink (Chen et al 2022a).It should be noticed that atmospheric inverse models estimated carbon flux between land and atmosphere, but did not simulate the process of crop harvest, and the carbon fixed in the crop grain will not be removed from cropland.Therefore, atmospheric inverse models may overestimate the carbon sink in the cropland because they did not simulate the emission of harvested crop carbon.P Ren et al

Conclusion
HC, a crucial component of the carbon budget in cropland ecosystems, poses a well-documented challenge for terrestrial ecosystem models.This study collected statistical production of 10 crop types from 1981 to 2020 and calculated the HC with crop harvest.The results showed that the total HC increased about 2.3 times during the past 40 years, from 0.18 Gt C yr −1 in the1981 to 0.42 Gt C yr −1 in 2020.The grain-type crop (i.e.rice, wheat, and maize) makes the largest contribution (68.3%) to total increases of HC in China.In addition, based on the statistical crop production data, this study generated the gridded dataset of HC (TEC-HCC, i.e.Terrestrial Ecosystem Disturbance-Harvested Carbon in Cropland Ecosystems) over entire China (did not include Hongkong, Macau, and Taiwan) jointly using crop distribution maps and satellite-based vegetation index.The spatial resolution of this dataset is 10 km × 10 km.Based on this newly produced dataset, and we evaluated the performance of the seven terrestrial ecosystem models for simulating HC in China.Several models could reproduce the spatial patterns of HC, but all models failed to reproduce the temporal variations during the past 40 years.This study highlighted a substantial increased trend of harvested crop carbon during the past 40 years, and the produced dataset has a large potential for estimating the national carbon budget and to be a benchmark of state-of-the-art terrestrial ecosystem models.

P
Figure 1.Long-term changes of harvested carbon in cropland ecosystem in China.

Figure 2 .
Figure 2. Percentage share of various crops in harvested carbon and contribution to long-term changes in China.(a) Share of various crops to the total harvested carbon.(b) Share of various crops to long-term changes of harvested carbon.(c), (d) are similar to (a), (b) but indicate provincial shares, and the abbreviations indicate the province (figure S1).

Figure 3 .
Figure 3.The relationship between the change trends of the total planting area of all crops in each province of China from 1981 to 2020, the change trends of crop yield, and the change trends of crop production, where the horizontal axis represents the area, the vertical axis represents the yield, and the size of the circle represents the change trends of harvested carbon.

Figure 4 .
Figure 4. Intensity (a) and trend (b) of harvested crop carbon from 2001 to 2019.The small bar chart I on the lower left corner of panel b shows the proportion of four situations: significantly decreased (p < 0.05), insignificantly decreased, insignificantly increased and significantly increased.The bar chart II displays the distribution of grid point trends in >10, 0-10, −10-0, and <−10 g C m −2 yr −1 .

Figure 5 .
Figure 5.Comparison between harvested carbon derived from statistical data and simulations of seven models.(a) The temporal variations of harvested carbon; (b) the detrended temporal variations of harvested carbon; (c) the harvested carbon averaged through 2001-2019, (d) the harvested carbon trends averaged through 2001-2019.

Figure 6 .
Figure 6.The mean intensity of harvested carbon derived from seven terrestrial ecosystem models from 2001 to 2019.Figures (a)-(g) show the simulations of seven models, and (i) shows the harvested carbon derived from the TED-HCC dataset.

Figure 7 .
Figure 7.Long-term trends in harvested carbon from 2001 to 2019 simulated by seven process-based ecosystem models.Figures (a)-(g) show the simulations of seven models, and (h) shows the harvested carbon trends derived from the TED-HCC dataset.

Table 1 .
Carbon content of 10 crop types.
. Although there were 17 ecosystem models in Global Carbon Budget, 7 terrestrial ecosystem models provided the output of harvested crop carbon: the Community Land Model (CLM5.0;Lawrence et al 2019), Integrated BIosphere Simulator (IBIS; Yuan et al 2014), Interaction Sol-Biosphère-Atmosphère Model (ISBA; Noilhan and Mahfouf 1996), the land component of the MPI Earth System Model (JSBACH; Reick et al 2021), Lund-Potsdam-Jena General Ecosystem Simulator model (LPJ-GUESS; Sitch et al 2003), a newly developed version of the terrestrial biosphere model (OCN; Zaehle and Friend 2010), and the Organizing Carbon and Hydrology In Dynamic Ecosystems (ORCHIDEE; Krinner et al 2005).