Accurate photometric redshifts for the CFHT Legacy Survey calibrated using the VIMOS VLT Deep Survey
We present photometric redshifts for an uniquely large and deep sample of 522286 objects with i'_{AB}<25 in the Canada-France Legacy Survey ``Deep Survey'' fields, which cover a total effective area of 3.2 deg^2. We use 3241 spectroscopic redshifts with 0<z<5 from the VIMOS VLT Deep Survey as a calibration to derive these photometric redshifts. We devise a robust calibration method which removes systematic trends in the photometric redshifts and significantly reduces the fraction of catastrophic errors. We use our unique spectroscopic sample to present a detailed assessment of the robustness of the photometric redshift sample. For a sample selected at i'_{AB}<24, we reach a redshift accuracy of σ_{Δz/(1+z)}=0.037 with η=3.7% of catastrophic error. The reliability of our photometric redshifts is lower for fainter objects: we find σ_{Δz/(1+z)}=0.029, 0.043 and η=1.7%, 5.4% for samples selected at i'_{AB}=17.5-22.5 and 22.5-24 respectively. We find that the photometric redshifts of starburst galaxies in our sample are less reliable: although these galaxies represent only 18% of the spectroscopic sample they are responsible for 54% of the catastrophic errors. We find an excellent agreement between the photometric and the VVDS spectroscopic redshift distributions at i'_{AB}<24. Finally, we compare the redshift distributions of i' selected galaxies on the four CFHTLS deep fields, showing that cosmic variance is already present on fields of 0.8 deg^2.