Figure 1.
Map of the study area and location in China.
Table 1.
Characteristics of vegetation and soil.
Table 2.
Choice of precipitation events.
Figure 2.
A Conceptual diagram of the threshold-delay model based on Eqs. 1, 2, and 3 [11].
(A) is the relationship between δt (the magnitude of the increase in the response) and yt–1 (the previous state of the response variable), where ymax is the maximum potential value of the response variable and δ* is the maximum potential response increase; (B) is the relationship between δ* and rainfall size at lag τ (days). RL is the lower threshold below which rain events do not stimulate a response. RU is the upper threshold above which rain events than do not yield additional benefits; and (C) provides a hypothetical response curve.
Figure 3.
Frequency distributions of rainfall and rainfall events.
(A) From April to October in 2007; (B) From April to October in 2011; (C) From April to October between 2004 and 2012.
Table 3.
Statistical analysis of the changes in NDVI in response to rainfall pulses.
Figure 4.
Intra-seasonal precipitation patterns in the threshold-delay data for NDVI in response to rainfall pulses.
Error bars represent standard error for all the pixels in the study area. (A) The mean NDVI of the desert study area response to rainfall pulses for Terra and Aqua satellites. (B) The mean NDVI of the dune study area response to rainfall pulses for Terra and Aqua satellites. (C) The Maximum NDVI of this two-habitat study area's response to rainfall pulses for Terra and Aqua satellites.
Table 4.
The Pearson's correlation coefficients among the response increase of NDVI, rainfall event size, maximum and mean precipitation intensity, duration of rainfall and dry interval.
Figure 5.
Relationship between NDVI increments and rainfall event sizes in different habitats.
(A) Dune ecosystem: n = 34, adjusted r2 = 0.59, p<0.0001; (B) Desert ecosystem: n = 34, adjusted r2 = 0.60, p<0.0001. The relationships show significant linear trends for dune and desert ecosystems. The observations used for these analyses come from both the Terra and Aqua satellites. Due to the lag, here, the NDVI increment is the the maximum response to a rainfall event. The error bars represent standard errors for all the pixels in the study area.
Figure 6.
Percentage increase in NDVI, from the day before rain to the day of peak NDVI, in response to different rainfall size classes.
Letters above columns represent significantly different treatments (Tukey's HSD test, P<0.05). 5–10 mm: n = 20; others: n = 12. The observations used for these analyses come from both the Terra and Aqua satellites.
Table 5.
The parameters of the threshold-delay model for changes in NDVI in response to rainfall pulses.
Figure 7.
Patterns of mean temperature in response to rainfall pulses.