Understanding rainfall variability and extremes over the Amazon to improve their future projection

The Amazon, as the world's largest rainforest region, is important for global carbon cycle, climate change, and ecosystem. This dissertation aims to investigate rainfall variability and the related atmospheric and land processes in tropical South America using both observations and climate models. The first two studies concern the changes and variability of the dry season over the southern Amazon, a domain with vulnerable tropical rainforest. Ground-based observation from 1979 to 2011 indicates the dry season length (DSL) has increased at a rate of 6.5 ± 2.5 days per decade, mainly caused by a delay of the dry season ending (DSE). A delay of the DSE has a clear impact on fire counts in austral spring. Moreover, the DSE also has large interannual variation and thus has a strong impact on drought and flood. Three dry season conditions are crucial for determining its trend and interannual variation. (i) A poleward shift of the subtropical jet over South America (SJ[subscript SA]) can prevent cold frontal systems from moving northward into the Amazon. This delays cold air incursion and results in late DSE over. (ii) An anomalous anticyclonic center, which enhances westerly wind at 850 hPa (U850) and the South American Low Level Jets (SALLJs), leads to moisture export from the southern Amazon to La Plata basin and reduces convective systems that provide elevated diabatic heating. (iii) Smaller convective available potential energy (CAPE) and larger convective inhibition energy (CIN) limit local thermodynamically driven convection. However, global climate models can capture neither the drying trend nor the variability of the DSL and DSE. Thus, the third and fourth studies deal with understanding the reported systematic dry bias over the Amazon in climate models. Our evaluation of the Coupled Model Intercomparison Project phase 5 (CMIP5) models indicates the positive feedback between reduced cloudiness, too much surface solar radiation, high Bowen ratio, and suppression of rainfall appears to cause the dry bias. A further investigation on Geophysical Fluid Dynamics Laboratory (GFDL) models demonstrates the Amazon dry season dry bias origins from biases in complicated processes, highlights the importance of the northern Andes on moisture flux convergence over the Amazon in the dry season, and suggests using at least 1° atmosphere-only model to understand the remaining dry season dry bias. The fifth study proposed a new method to assess CMIP5 climate projections for rainfall extremes. The results indicate future heavy rainfall changes in the Amazon due to anthropogenic forcing may be underestimated by multi-model ensemble mean.