Blue III-nitride light-emitting diodes (LEDs) are widely used nowadays in solid-state lighting, as white light can be produced by combining yellow phosphors and blue LEDs. This technology has the advantage to have a higher luminous efficiency than incandescent light bulbs and fluorescent tubes. This high efficiency is attributed to InGaN quantum wells (QWs), which exhibit an internal quantum efficiency (IQE) higher than 90 \% at 300 K. However, even if blue LEDs are industrially produced, the high IQE of InGaN QWs is still barely understood. In InGaN/GaN-based LEDs, an InGaN layer grown before the InGaN/GaN QW is often found. This InGaN layer named an underlayer (UL) is used in all commercial devices and is known to increase drastically the efficiency of LEDs. However, the physical mechanism behind the efficiency improvement is debated in the literature. In the first part of this work, a study of the different existing hypotheses proposed is conducted. The conclusion is that the role of the InGaN UL is to capture point defects before they reach the InGaN QWs. Indeed, points defects react with InGaN and are transformed into non-radiative centers when they are buried in the InGaN alloy. We call such point defects reacting with indium "surface defects" (SDs). To explore the SD capture by the InGaN UL, we studied different kinds of UL: InGaN UL, InAlN UL, and low-temperature GaN UL. We found that indium atoms play a key role in the capture of SDs. Indeed, the number of SDs buried in the UL depends on its thickness and indium composition. As a consequence, GaN layers are not able to bury SDs as efficiently. A model based on segregation of SDs was proposed based on our observations. As InGaN UL are present in all LEDs, we concluded that SDs are problematic for all LEDs independently of the type of reactor. The growth temperature of GaN was identified as the parameter which generates SDs. Indeed, for temperatures higher than 820 \degres C in MOVPE, SDs are created. Different impurities were searched by secondary ion mass spectrometry measurements, but no correlation was found between the sample efficiency and the impurity concentration. Our hypothesis is that the stability of the GaN surface is affected by the temperature and therefore leads to the formation of SDs, which could be intrinsic defects such as nitrogen vacancies. However, we do not have any experimental proof about the actual nature of SDs yet. In the last part of this work, near UV LEDs with InAlN UL were designed and characterized. Single QW (SQW) LEDs emitting around 405 nm with an InAlN/GaN superlattice UL exhibit an IQE around $\sim$70 \%, which is higher than for a SQW LED without UL ($\sim$10 \%). The $I$-$V$ characteristic is similar for both samples thanks to a highly doped region placed at the interface between the InAlN/GaN superlattice UL and the GaN spacer.