Soil moisture seasonality alters vegetation response to drought in the Mongolian Plateau

2.1.1. Vegetation indices and land cover map

2.1.2. Climate and drought indices data

2.1.3. Soil moisture and terrestrial water storage data

2.1.4. Livestock data

Pearson correlation and regression analysis were performed to understand relationships between ${\text{NI}}{{\text{R}}_{\text{V}}}$ and environmental factors, as well as the impacts of drought on animal husbandry. An unpaired t-test was used to determine whether mean annual precipitation differed significantly (p < 0.05) between the two time periods (2000–2009 vs.1950–1999). All statistical analyses were performed using MATLAB R2019a software.

The decadal severe drought was identified by precipitation data and two drought indices (the 12 month SPEI12, and the scPDSI) (figure 1). The average annual precipitation in 2000–2009 was 14% lower than that in 1950–1999, 42.5 mm, and the difference is statistically significant (p = 0.0002). Both SPEI12 and scPDSI indicate drought (SPEI12 < −0.5, scPDSI < −1) in most years of 2000–2009 for this region. Moreover, satellite-based TWS, showed a declining trend during the decadal drought (figure 1(c)). This drought was alleviated by increased precipitation and decreased temperature in the 2010s (figure 1). Then, based on annual precipitation records from CRU, short-term droughts revisited the MP in 2014 (262 mm yr⁻¹), 2015 (260 mm yr⁻¹) and 2017 (236 mm yr⁻¹), which are close to the mean annual precipitation during the 2000–2009 drought (259 mm yr⁻¹). Both SPEI12 and scPDSI confirm these short-term droughts in the 2010s (figure 1(c)). The long-term and short-term drought pattern in the two sub-regions (MN and IM) are the same as for MP (figure S2).

Zoom In Zoom Out Reset image size

During the severe 10 year drought, ${P^{{\text{GS}}}}$ was 15% lower than after drought (2000–2009 vs. 2010–2018). Also during the severe 10 year drought (2000–2009), ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{GS}}}$ was 8.9% lower than during the post-drought period 2010–2018. Indeed, we found an overall significant positive relationship between ${P^{{\text{GS}}}}$ and ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{GS}}}$ (R² = 0.53, p < 0.001), which was significant in 78.0% of the grid cells (78.5% and 77.1% in MN and IM, respectively) (figure S5(a)). The lowest ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{GS}}}$ of the study period occurred in 2007, consistent with a persistent decrease in scPDSI and SPEI until 2007. Then, ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{GS}}}$ started to recover from the severe drought with increasing ${P^{{\text{GS}}}}$ beginning in 2008, but the ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{GS}}}$ anomaly remained negative until 2010. The increase in ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{GS}}}$ by 0.06 between the periods 2010–2018 and 2000–2009 was mainly attributed to increased ${\text{NI}}{{\text{R}}_{\text{V}}}$ during the summer (${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{JJA}}}$; 68.2%) and spring (${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{AM}}}$; 19.3%) (table 1). Summer still plays the leading role when focusing on the sub-region scale, and the contribution proportions are 76% and 61% for MN and IM, respectively (table S2). The spatial pattern of ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{GS}}}$ recovery shows that summer is the dominant season driving recovery in 83.5% grid cells (85.0% and 81.3% in MN and IM, respectively) (figure S4(c)). As the correlation results show, summer precipitation was the dominant factor determining ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{JJA}}}$, and ${P^{{\text{JJ}}}}$ explained more of the residual variation (64%) than ${P^{{\text{JJA}}}}$ (45%) and ${\text{S}}{{\text{M}}^{{\text{May}}}}$ (21%) (table 2). The partial correlation coefficient between ${\text{S}}{{\text{M}}^{{\text{May}}}}$ and ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{JJA}}}$ was significant when controlling for the effects of summer precipitation and temperature for MP and IM (p < 0.05).

Table 1. Seasonal contribution to the change in the mean integrated growing season NIR_V between two periods (2000–2009, 2010–2018). Growing season: ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{GS}}}$; spring: ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{AM}}}$; summer: ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{JJA}}}$; autumn: ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{SO}}}$.

	Integrated NIRV (2000–2009)	Integrated NIRV (2010–2018)	Difference in integrated NIRV between 2010–2018 and 2000–2009	Contribution (%)
${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{GS}}}$	0.62	0.67	0.06	100.00
${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{AM}}}$	0.12	0.13	0.01	19.32
${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{JJA}}}$	0.37	0.41	0.04	68.15
${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{SO}}}$	0.13	0.13	0.01	12.53

Despite the overall strong relationship between precipitation and vegetation productivity, some years with similar ${P^{{\text{GS}}}}$ (220–240 mm) had very different levels of vegetation productivity (~50% of total 2000–2018 range). For example, in 2003, despite high ${P^{{\text{GS}}}}$ (42.7 mm above the 2000–2018 average, figure 2(b)), the anomaly of vegetation productivity was negative ('anomaly' in the study refers to the deviation from the mean). A comparison between 2007 and 2017, which had similarly low ${P^{GS}}$ (209 mm and 219 mm, respectively), provides an example of the mismatch of NIR_v and P. Spring precipitation (${P^{{\text{AM}}}}$) in 2017 was 32% lower than that in 2007, but ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{AM}}}$ in 2017 was 12% higher than in 2007. This higher ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{AM}}}$ despite lower spring precipitation in 2017 is related with the higher SM and TWS in March of 2017 (figures 3(c) and (d), table S3). The positive SM anomaly in 2017 lasted from March to June, alleviating the ${P^{{\text{AM}}}}$ deficit in 2017 (figure 3(d)). On the other hand, the negative ${\text{S}}{{\text{M}}^{{\text{Mar}}}}$ anomaly in 2007, propagated from the depleted soil water in the previous consecutive drought years (2000–2006), contributed to low ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{AM}}}$ in 2007. The TWS anomaly in 2007 from reconstructed GRACE data also showed a deficit of water storage in March, resulting from the previous drought years. Furthermore, as the correlation results show (table 2; Table S4), the soil moisture in March (${\text{S}}{{\text{M}}^{{\text{Mar}}}}$) is highly correlated with spring ${\text{NI}}{{\text{R}}_{\text{V}}}$ (${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{AM}}}$) for all three regions (MP, MN and IM; p < 0.05). The partial correlation between ${\text{S}}{{\text{M}}^{{\text{Mar}}}}$ and ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{GS}}}$ is significant for MP and IM (p < 0.05) after controlling for the effect of GS precipitation and temperature. Furthermore, ${\text{S}}{{\text{M}}^{{\text{Mar}}}}$ explained 35% and 29% of the residual variation in ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{GS}}}$, predicted from ${P^{{\text{GS}}}},$ for MP and IM, respectively (figure 2(f)). This relationship was significant over 20% and 28% of regional grids for MP and IM, respectively (figure S5(b)). For MN, the partial correlation between ${\text{S}}{{\text{M}}^{{\text{Mar}}}}$ and ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{GS}}}$ are only marginally significant (R = 0.47; p = 0.06) driven by a weaker relationship between spring ${\text{NI}}{{\text{R}}_{\text{V}}}$ and GS ${\text{NI}}{{\text{R}}_{\text{V}}}$ (R = 0.43, p = 0.06).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Soil moisture in the previous Octoberx (${\text{S}}{{\text{M}}^{{\text{Oct}}}})$ is highly correlated with ${\text{S}}{{\text{M}}^{{\text{Mar}}}}$ for all three regions (p < 0.001). The ${\text{S}}{{\text{M}}^{{\text{Oct}}}}$ was 12%, 11%, and 13% higher in 2010–2018 than in 2000–2009 for MP, MN, and IM, respectively. For IM, there is a direct and significant correlation between maximum SWE and ${\text{S}}{{\text{M}}^{{\text{Mar}}}}$ (p < 0.05), and the correlation is stronger after controlling for the effect of ${\text{S}}{{\text{M}}^{{\text{Oct}}}}$ (p < 0.01). For MN, though there is not a significant correlation between SWE and ${\text{S}}{{\text{M}}^{{\text{Mar}}}}$, SWE is significantly correlated with the difference between ${\text{S}}{{\text{M}}^{{\text{Apr}}}}$ and ${\text{S}}{{\text{M}}^{{\text{Mar}}}}$(p = 0.02), and the relationship is stronger when controlling for the ${\text{S}}{{\text{M}}^{{\text{Oct}}}}$ (p = 0.01).

In MN, livestock numbers dropped in 2000–2002 (figure 4(a)), following severe drought onset in 2000 (figure 1(c)). The drop ended a persistent increase since the 1970s (except for camel). From 2000 to 2002, the total livestock number declined by 37% relative to 1999. These declines were 51% for cattle, 37% for horse, 30% for sheep, 29% for camel, and 17% for goat. Since 2003, the population of the three most commonly reared livestock types, sheep, goat, and cattle, has increased slowly, while horse and camel population numbers remain relatively unchanged. By 2009, the number of sheep and goats exceeded that before the severe drought, but camel, horse and cattle numbers were below pre-drought level. As vegetation productivity recovered from drought during 2010–2018, the total livestock number increased at a rate twice that in 2002–2009 (figure 4(a)). The change in livestock number was closely linked to vegetation productivity in MN. The detrended livestock number was significantly correlated with the detrended 3 year moving average ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{GS}}}$ in most aimags in central and eastern MN (figure 4(b)). The livestock populations in these regions make up 55% of the total national livestock units. Furthermore, the relationship between livestock number and vegetation productivity was strongest in the areas with the most livestock (figure 4(c)).

Zoom In Zoom Out Reset image size

The temporal pattern of ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{GS}}}$ in IM was similar to the pattern in MN (R = 0.77, p < 0.001, figure S7). However, the severe decadal drought in the early 2000s had relatively less impact on livestock numbers in IM (figure 5). Total livestock number decreased in the early years of the drought (2000 and 2001), but increased by 45% from 1999 to 2010. As a result, the detrended total livestock number was negatively correlated with the detrended 3 year moving average ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{GS}}}$ (R = −0.45, p = 0.08).

Zoom In Zoom Out Reset image size

Evaluated with scPDSI (SPEI), the severe decadal drought reached its worst point in 2007. It still remained below −2 (−1) in 2008–2009, suggesting the continuation of the drought. This unprecedented long-term drought caused a significant decrease in pasture productivity and damaged the livestock sector (Nandintsetseg and Shinoda 2013), and also caused this region to shift from a carbon sink to a carbon source (Lu et al 2019). Starting in 2010, the drought indices rebounded, and the ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{GS}}}$ anomaly changed to positive while ${\text{NDV}}{{\text{I}}^{{\text{GS}}}}$ returned to its 1982–2016 average level (figure S6). Although vegetation productivity was still suppressed by drought in 2008 and 2009, soil water started to recover, which contributed to vegetation recovery since 2010 (figure 2(d)). The 2 or 3 years required for recovery of grassland vegetation productivity after this severe drought is generally shorter than what would be expected of forests, due to the different hydraulic traits, water use strategies and growth rate of herbaceous- vs. woody-plant dominated ecosystems (Anderegg et al 2019, Schwalm et al 2017, Kolus et al 2019).

Several previous studies have demonstrated the importance of post-drought precipitation for recovery of productivity (Antofie et al 2015, Schwalm et al 2017, He et al 2018), which is also evident in our study. In the MP, most of the precipitation occurs in summer (64%, 63%, and 66% for MP, MN and IM, respectively). Interestingly, summer was also the season with the most precipitation increase between 2000–2009 and 2010–2018 (61%, 65%, and 58% of the annual precipitation increase for MP, MN and IM, respectively). As a result, the strong correlation between ${P^{{\text{JJ}}}}$ and ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{JJA}}}$ played a determined role for the increase in summer vegetation productivity between the two periods. However, the effect of spring precipitation on the vegetation productivity should not be overshadowed. For example, in the two transitional years, 2009 and 2010, though rainfall only increased during the spring (27 mm) and autumn (11 mm) and actually decreased in the summer (−11 mm) of 2010 compared with 2009, ${\text{NI}}{{\text{R}}_{\text{V}}}$ of the three seasons was higher in 2010 than in 2009 (figure 2). Since the ${\text{S}}{{\text{M}}^{{\text{Mar}}}}$ is similar in 2 years (183, 182 mm), it is the increase in spring precipitation that enhanced grassland productivity in spring and summer. The spring precipitation exerts its influence on vegetation productivity in three aspects: (a) promoting spring productivity, (b) leaving water for plants to use in summer, which can be seen from the significant partial correlation between ${P^{{\text{AM}}}}$ and ${\text{S}}{{\text{M}}^{{\text{May}}}}$ (after removing the effect of ${\text{S}}{{\text{M}}^{{\text{Mar}}}}$) as well as the significant partial correlation between ${\text{S}}{{\text{M}}^{{\text{May}}}}$ and ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{JJA}}}$, (c) enabling plants to expand the leaves to better photosynthesize and develop the root system to fully utilize the available water during the late months (Bates et al 2006, Peng et al 2013). Thus, in this region spring rainfall may be more critical for post-drought vegetation productivity recovery than the same amount of rainfall in summer or autumn.

Our results suggest that pre-GS SM plays a key role in spring grassland vegetation productivity. For example, by comparing 2007 and 2017, we showed that a deficit of water storage in March resulting from the previous drought years could amplify the drought effect of deficit of ${P^{{\text{GS}}}}$ on vegetation productivity in 2007; while a positive pre-GS SM anomaly helped alleviate the drought impact in 2017 and led to higher ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{AM}}}$. Higher pre-GS SM advances leaf-out phenology of grasslands (Shen et al 2011, Ganjurjav et al 2016) and provides effective moisture for spring growth (Shen et al 2011, Wang et al 2018). Both higher residual SM from previous GS and higher winter snowfall resulted in higher ${\text{S}}{{\text{M}}^{{\text{Mar}}}}$ in 2011–2013, compensating for the spring precipitation deficits (figures 2(b) and (d)). In contrast, long-term drought in the 2000s exacerbated the deficit of ${\text{S}}{{\text{M}}^{{\text{Mar}}}}$ by amplifying the deficit of residual SM from the previous GS (figure 2(c)). Snowfall, an important moisture source, could help vegetation resist drought in arid and semi‐arid regions (Peng et al 2010, Li et al 2020). In IM, because of warmer early spring temperatures, the snow has melted and permeated into the soil by March, reflected by the significant correlation between SWE and ${\text{S}}{{\text{M}}^{{\text{Mar}}}}$, promoting vegetation productivity in April (significant correlation between SWE and ${\text{NI}}{{\text{R}}_{\text{V}}}^{{\text{Apr}}}$). In contrast, in MN, the melting process starts later and can lag into April, reflected by the significant correlation between SWE and the difference in SM between March and April. In addition to the benefits of snow as a water source for vegetation, deep snow is also an effective insulator, inhibits wind soil erosion, and increases nutrient availability by enhancing soil temperature and facilitating microbial activity and mineralization (Schimel et al 2004, Schmidt and Lipson 2004, Tomaszewska et al 2020).

Using stable isotopes of hydrogen and oxygen at one grassland site in IM, Chi et al (2019) demonstrated that contribution of snow to plant water uptake continues throughout the whole GS. However, in our study, the significant correlation between ${\text{S}}{{\text{M}}^{{\text{Mar}}}}$ (${\text{S}}{{\text{M}}^{{\text{Apr}}}}$ for MN) and the monthly ${\text{NI}}{{\text{R}}_{\text{V}}}$ ends in June. We thus infer that the legacy of snow water lasts no later than June in our study. This shorter legacy period could be because the regional mean SWE is in the range of 5–40 mm in this study, while the maximum SWE was ~159 mm in the snow-addition plot in Chi et al (2019). The difference in snow depth between the two studies indicates that the legacy time of snow water uptake depends on the amount of SWE. By conducting isotope observation at one grassland site in Tibetan Plateau, Hu et al (2013) claimed that once the monsoon season starts, plants tend to use topsoil moisture mainly from rainfall rather than residual moisture from snow (Hu et al 2013). How long the legacy effect of snow water for vegetation growth can last in the GS is not well known yet, and it is relevant to local climate and water partition between species (Yang et al 2011, Trnka et al 2015, Potopová et al 2016). Hence, further observation, manipulative experiments and modeling are needed.

Drought-caused changes in grassland vegetation productivity can impact the stock number of cattle, sheep, and goats (Nardone et al 2010, O'Mara 2012). The impact of drought on livestock is particularly clear for MN (figure 4). During the severe decadal drought in the 2000s, livestock number in MN decreased (figure 4), and milk and meat production declined by 42% and 40% from 1999 (figure 6). The livelihoods of herders were threatened and some were forced to give up herding completely due to economic losses during the severe decadal drought (Rao et al 2015). In contrast, the number of livestock and livestock products increased during the 2000s in IM (figures 5 and 6). The increase may be attributed to increased forage import and crop residues compensating the shortage of grass feed during the drought (National Bureau of Statistics of China 2019). Although the short-term droughts in the 2010s had smaller impacts than the long-term drought in the 2000s on the livestock sector in MN, more effective mitigation measures and policies will be required for the livestock sector in MN to survive the predicted future drying trend in MP (Chen et al 2015, 2018, Hessl et al 2018). The different responses to droughts between MN and IM shown in this study could offer important insights for future drought mitigation (Chen et al 2015). In addition, climate change could lead to more harsh winters (Cohen et al 2014), which increase livestock mortality (Fernandez-Gimenez et al 2015). The ultimate impacts of the trade-off between the benefit of snow water for vegetation growth and the damage of harsh winters on livestock remains uncertain and needs further investigation.

Zoom In Zoom Out Reset image size