At present, underwater temperature measurement of seawater is a hot research topic, because knowledge about seawater temperature is of great importance in many fields. The laser Raman spectroscopy is a feasible method for measuring the vertical profiling of seawater temperature in large water areas. However, the real-time remote sensing of underwater temperature has not been reported. In this paper, a low-cost and practical Raman Lidar seawater temperature remote sensing system is constructed, and a real-time spectra acquisition and temperature determining software system is developed. Firstly, a background subtraction algorithm which combines the spatial accumulation of the array CCD with the exposure time integral is used to effectively enhance the signal to noise ratio of Raman spectra and improve the detection sensitivity of this Raman Lidar system. Usually the Raman spectra measured on-site are in low signal to noise ratio and baseline drift conditions. In this case, the “area ratio” (i. e. the ratio of the integrated Raman spectrum at low wavelength to the integrated Raman spectrum at higher wavelength) is a good temperature indicator. In this paper, we comprehensively studied the influence of Raman spectra area ratios split positions and fitting methods on the temperature measurement accuracy. More than 500 groups normalized Raman spectra at different temperatures are experimentally measured in the process of water temperature rising continuously. The area ratio SHB/SNHB and SNHB/SHB are used as the spectra characteristics to relate with the water temperature respectively, and both linear and second-order polynomial fitting algorithm are analyzed. The results show that the split positions have a great influence on area ratio variation range, and the fitting order has a great influence on the accuracy of fitting relationship between area ratio and seawater temperature. Both of them will eventually affect the water temperature measurement error. In order to objectively and directly reflect the influences of different area ratio methods, split position and fitting order on the water temperature measurement error, we further analyze the temperature measurement error at different conditions. The results show that the temperature measurement error is less affected by the split position, while is greatly influenced by the area ratio method and the fitting order. For the same split position and the same area ratio method, the results using order polynomial fitting are better than that using linear fitting. The results also show that linear fitting thearea ratios SHB/SNHB with water temperatures is a good choice, because it can obtain good measurement accuracy, and at the same time it has the advantage that the fitting parameters are simple and easy to be adjusted. Furthermore, the influences of different area ratio method and split position on the anti-interference of the system are studied. The results show that the anti-interference of SHB/SNHB method reduces with the decrease of the split wavelength, while the anti-interference of SNHB/SHB method enhances with the decrease of the split wavelength. The research results are used to inform the parameter setting of water temperature determining method, and improve Raman Lidar system temperature measurement accuracy. Considering all these results above, we choose the large wavelength 649.3 nm as the split location to calculate the Raman spectra area ratios SHB/SNHB, and linear fitting them with the water temperatures. Finally, the continuous temperature measuring performance of this Raman Lidar seawater temperature remote sensing system is verified experimentally. The experiment results show that the temperatures measured by Raman Lidar system are in good agreement with that by synchronous temperature sensor which is dipped in the sample tap-water and connected to the computer. The maximum measurement error is about ±0.5 ℃, and the standard deviation of measurement error is about 0.21 ℃.