Limited driving of elevated CO₂ on vegetation greening over global drylands

We use the average of three satellite-observed LAI records (reprocessed Moderate-resolution imaging spectroradiometer (MODIS), Global Land Surface Satellite (GLASS) and GLOBMAP LAI data sets) to analyze the changes in vegetation over drylands during the period of 2000–2019. The reprocessed MODIS version 6 LAI products (Myneni et al 2015, Yuan et al 2020) based on two-step integrated method (Yuan et al 2011) perform much better than the original product (Yuan et al 2011, 2020). The GLASS LAI product (Version 6) (Ma and Liang 2022) was derived from MODIS surface reflectance data using the bidirectional long short-term memory model. The missing data of the GLASS LAI for January and February 2000 was substituted by the monthly climatology of LAI during 2000–2019. The GLOBMAP LAI (Version 3) (Liu et al 2012) was made up by combination of AVHRR LAI (1981–2000) and MODIS LAI (2000–2020). The MODIS LAI series was generated from MODIS C6 land surface reflectance data. The reprocessed MODIS, GLASS, and GLOBMAP LAI data were interpolated to 1° spatial resolution based on the linear interpolation method to match the spatial resolution of the Global Land Data Assimilation System (GLDAS) 2.1 forcing data (Rodell et al 2004).

In this study, we use GLDAS2.1 atmospheric forcing fields to drive the Noah-MP. We use the global 1 km hybrid State Soil Geographic Database and the United States Geological Survey (USGS) 24–category vegetation data to determine the soil and vegetation parameters, respectively. Both the 1 km data sets are aggregated to 1° with the dominant soil and vegetation types to match the spatial resolution of the GLDAS2.1 forcing data. The soil and vegetation parameters are then determined for each soil type and vegetation type through the look-up tables of Noah-MP.

To compare the performance of Noah-MP and dynamic global vegetation models (DGVMs) to simulate the interannual variations and trends in LAI, we used outputs from S3 experiment of 13 DGVMs from TRENDYv10 datasets (Friedlingstein et al 2022), including CABLE-POP, CLM5.0, IBIS, ISAM, ISBA-CTRIP, LPJ-GUESS, LPX-Bern, OCN, ORCHIDEE, ORCHIDEEv3, SDGVM, VISIT and YIBs. For the comparison with satellite-observed and Noah-MP modeled LAI, LAI simulation of CLM5.0, ISAM, LPJ-GUESS, LPX-Bern, ORCHIDEE, ORCHIDEEv3, SDGVM and VISIT was regridded to 1° using linear interpolation method.

All the model experiments were conducted with the Noah-MP LSM (Niu et al 2020). The model describes the carbon, energy and water flux exchanges between the land surface and the atmosphere controlled by terrestrial ecohydrological processes and the corresponding storage terms. Noah-MP include a short-term dynamic model to predict LAI, which is converted from the predicted leaf carbon mass controlled by carbon allocation of the assimilated carbon, through photosynthesis of sunlit and shaded leaves, maintenance and growth respirations, leaf turnover and death due to water and temperature stresses (Niu et al 2011). Noah-MP explicitly represents plant water storage supplied by dynamic root water uptake through hydrotropic root growth to meet the transpiration demand (Niu et al 2020). It allows more carbon translocation to roots when the plants are under water stress and represents root hydrotropism through more carbon translocation to roots but less turnover in wetter soil layers. The model improved the resilience of plants to droughts by using the water in shallow soil and groundwater through capillary rise. In this study, we drive the model at 1° spatial resolution and 3 hour time step from 2000 to 2019 over global drylands with a spin-up of 5 loops. The model physical options are listed in table 1.

To determine the relative contributions of the driving factors to the LAI trends, we conducted three experiments using Noah-MP to quantify the relative contribution of climate change, CO₂, and LULCC to the LAI trends. The first experiment was forced by the 3-hourly GLDAS2.1 climatic forcing data during 2000–2019 and varying atmospheric CO₂ (hereafter 'ALL'). The second experiment was forced by the same varying atmospheric CO₂ but 3-hourly climatology of climatic forcing during 2000–2019 (simulation 'CO₂'). The third experiment was forced by the GLDAS2.1 climatic forcing data during 2000–2019 but with a constant CO₂ concentration at the level of year 2000 (hereafter 'CLI'). The difference between ALL and CO₂ (ALL–CO₂) and that between ALL and CLI (ALL–CLI) were used to evaluate the contribution of climate change and CO₂ to LAI trends, respectively. Since the Noah-MP does not incorporate modules that account for the effects of LULCC on LAI, the difference between satellite-based LAI trends and modeled LAI trends from experiment 'ALL' was then explained as the contributions of LULCC.

Our study focuses on vegetation dynamics during the 'growing season', when the surface air temperature (based on the GLDAS2.1 monthly climatology between 2000 and 2019) is greater than 0 °C for both the satellite-observed and modeled LAI. Potential evapotranspiration was firstly calculated with the Penman–Monteith equation (Allen et al 1998) and the GLDAS2.1 forcing data from 2000 to 2019. Aridity index (AI) was then computed as the ratio of mean annual precipitation to mean annual potential evapotranspiration. Drylands are defined as where AI < 0.65 (Middleton and Thomas 1997, Huang et al 2015, Berg and McColl 2021), which include dry subhumid, semi-arid, arid and hyperarid regions (figure 1(a)). Besides, bare ground and water bodies represented by the USGS vegetation data, and areas without vegetation cover represented by the satellite-observed and modeled LAI are not included in the study. The vegetation types of the study area are shown in figure 1(b).

Zoom In Zoom Out Reset image size

We used a non-parametric Mann–Kendall method to determine the linear trends in the precipitation and the growing season LAI for both the satellite-based and modeled LAI at the 95% significance level. We used the Pearson correlation coefficient to evaluate the correlation between variables. When calculating the area, we took into account the changing grid box size with latitude. To calculate the mean LAI over dryland and the 12 selected regions, we used a spatially weighted mean.

Our results show that global drylands cover ∼40% of the global land area, and the extent is similar to the findings of Feng and Fu (2013), Burrell et al (2020), and Berg and McColl (2021). It was reported that global drylands have expanded towards the north of the 60° N parallel and cover ∼45% of the Earth's terrestrial area (Huang et al 2015, Prăvălie 2016). Due to the uncertainty of the GLDAS2.1 data, this characteristic is not captured by our method.

The LAI linear trends derived from the average of the three satellite LAI data sets show positive values ('greening') over a large portion of the drylands (71.72%) during the period of 2000–2019 (figure 2(a)). Globally, 29.47% of the drylands show a significant greening trend (p <0.05) including the southwestern North America, southern South America, Sahel, Middle East, southern and eastern Central Asia, South Asia, Tibet Plateau, Mongolia, and Northern China. The satellite-based LAI shows a significant decreasing trend (p < 0.05) over only 2.94% of the drylands. About 67.59% of the drylands have no significant (p <0.05) change from 2000 to 2019. It is worth noting that the satellite-based LAI shows a decreasing trend ('browning') spreading over the eastern South America, southern Africa, northwestern Central Asia and northwestern Australia; though the trend is not significant at 95%.

Zoom In Zoom Out Reset image size

The Noah-MP (experiment ALL) with a representation of hydrotropic root has an excellent performance in simulating the spatial patterns of LAI trend (figure 2(b)). Overall, the LAI trends modeled by Noah-MP agree with those of satellite observations over 71.26% of the drylands (52.60% for greening and 18.66% for browning). The disagreement between Noah-MP and satellite observations accounts for 28.74% of the drylands (figure 2(c)), which spreads over the southwestern United States, a large part of the southern Africa, northwestern Central Asia, southern India, a few areas of south central and east central Australia. The LAI trends modeled by Noah-MP and the multi-model ensemble mean (MMEM) LAI trends of DGVMs in the TRENDY project agree with those of satellite observations over 70.14% and 73.6% of the drylands (only the regions covered by all the 13 DGVMs of TRENDY project, three satellite-observed LAI products, and Noah-MP simulations were selected), respectively (figure 3). Compared with TRENDY results, the Noah-MP has comparable performance in simulating the spatial distributions of LAI trends over drylands.

Zoom In Zoom Out Reset image size

We compare annual anomalies of the growing season LAI modeled by Noah-MP with those of the average of the three satellite observations. For the greening trend (52.60% of drylands) (figure 2(c)), averaged over global drylands, the modeled LAI trend (0.0048 m² m⁻² yr⁻¹) agrees well with the satellite-based LAI trend (0.0048 m² m⁻² yr⁻¹) (figure 4(a)). The model also agrees very well with the satellite observations in the LAI interannual variations with a correlation coefficient of 0.93 (p <0.01). For the browning trend (18.66% of drylands) (figure 2(c)), the satellite-based LAI decreases at −0.0033 m² m⁻² yr⁻¹, while the modeled LAI shows a greater decreasing trend (–0.0074 m² m⁻² yr⁻¹) (figure 4(b)). The model also matches satellite observations in the interannual variations for these areas (r = 0.86, p < 0.01). Over the 12 selected regions (figures 4(c)–(n)), the model also captures the interannual variations of LAI, and the correlation coefficients between the two are greater than 0.74 over all the selected regions except the southern South America. We observe that there are differences between the interannual variations of modeled and satellite-observed LAI, mainly because the Noah-MP does not include the modules that account for the effects of LULCC on LAI. We then compare the interannual variability of regional LAI modeled by Noah-MP with the DGVMs in TRENDY project, and it shows that the Noah-MP has better performance (figure 5).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The LAI changes can be attributed to climate change, elevated CO₂ and LULCC in the study. Nitrogen deposition is also a driver for vegetation change, while this factor is not represented in Noah-MP, like most of other LSMs or ecosystem models. In fact, nitrogen deposition played little role in dryland areas (Bobbink et al 2010, Zhu et al 2016). Ours results show that climate change, elevated CO₂ and LULCC contribute 55.07%, 44.86% and 0.07% to the LAI changes over areas with a greening trend, respectively, and climate change dominates the LAI changes over areas with a browning trend, followed by LULCC (figure 6(a)). Rising atmospheric CO₂ concentration promotes the leaf scale photosynthesis. In addition, the elevated CO₂ leads to enhanced plant water use efficiency (Donohue et al 2013, Driscoll et al 2020), because the amount of water loss through transpiration during photosynthesis is reduced due to partial closure of the plant stomata. Hence, the rising atmospheric CO₂ concentration generally has a positive effect on LAI changes over drylands (figure 7(b)) and therefore has been regarded as the main driver of the greening over drylands (Burrell et al 2020) and global continents (Zhu et al 2016). Different from a previous study (Burrell et al 2020), our results show that elevated CO₂ dominates the trend of LAI over only 13.89% of drylands (figure 6(b)). Water resources should be a major limiting factor for vegetation growth over drylands (Zhang et al 2022). Changes in precipitation explain the year-to-year variations of LAI (p <0.05) over all selected regions except the Tibet Plateau (figures 4(c)–(n)). This indicates that there is a potential linkage between climate change and vegetation change. Our results of the primary drivers are further confirmed by previous regional studies. In the following paragraphs, we will compare the results of model simulation with regional studies according to different dominant factors.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

3.2.1. The role of climate change

3.2.2. The role of LULCC

The modeled LAI trends disagrees with the satellite-based LAI trends over ∼28.74% of the global drylands. Here we select four typical regions (southern California, southern Africa, northwestern Central Asia, and southern India) to investigate the causes to the disagreement.

The satellite-based LAI decreases from 2000 to 2019 in southern California (figure 2(a)), whereas the modeled LAI shows an opposite trend (figure 2(b)). The southern California was under three severe droughts during 2002–2003, 2007–2009, and 2012–2016 (MacDonald 2010, Lund et al 2018). The droughts along with fires and pathogenic outbreaks increased forest and woodland mortality (Dennison et al 2014). For example, over 102 million trees died from 2010 to 2016 on California's drought stricken forests (www.fs.usda.gov/news/releases/new-aerial-survey-identifies-more-100-million-dead-trees-california). Besides, agriculture practices in the Central Valley highly depends on irrigation (Tindula et al 2013) and plays an important role in vegetation changes. However, the version Noah-MP used in this study does not represent these processes by driven only by climate forcing and CO₂ concentration.

The satellite-based LAI increases during the period 2000–2019 in portions of the southern Africa (figure 2(a)), while the Noah-MP-modeled LAI decreases (figure 2(b)) because of great decrease in precipitation (figure 1(c)). These areas, including Botswana, Zimbabwe, northeastern Namibia and the eastern south Africa, underwent significant net gains in WPE (Saha et al 2015, Venter et al 2018) and the WPE enhanced vegetation greening and water use efficiency in these areas (Deng et al 2021). The WPE has not been represented in Noah-MP, causing the model to underestimate vegetation greenness.

Desertification of grassland is observed across the northwestern Central Asia (figure 2(a)), while the modeled LAI shows an increasing trend (figure 2(b)). It was reported that the observed browning trend mainly attributes to the drying climate and persistent drought in the northwestern Central Asia (Zhang et al 2018). The Noah-MP may overestimate grassland resiliency to droughts due to uncertainties in the rooting depth (too deep), making the model unable to simulate the LAI trend in this region.

The satellite-based LAI increases significantly from 2000 to 2019 in the southern India (figure 2(a)), where the dominant vegetation type is irrigated cropland (figure 1(b)). The modeled LAI decreases in these regions (figure 2(b)) due mainly to great reduction in precipitation (figure 1(c)). The anthropogenic effects, such as heavy fertilizer use for agriculture production (Lu and Tian 2017) and groundwater extraction for irrigation (Rodell et al 2009), have greatly contributed the greening of the southern India (Chen et al 2019). However, the model does not consider the impact of these human activities on vegetation changes, so it underestimates the increasing LAI trend.