Sentinel-1A level-1 high resolution GRD data have 20X22m resolution (rgXaz)with 5X1 number of looks. While Landsat-8 data have 30m resolution. My question is how can I compare the resolution differences between them? In other words, which one of them has higher resolution? When I digitize a lake size from the two sensors, lake size from Landsat-8 is always greater than Sentine-1A. Note that its a same lake; images were taken on the same acquisition date or with not much of time difference. What could be the reasons behind for differences in the lake sizes? Is it because of resolution difference? If yes, How can I interpret the results?
This is a bit of an apples and oranges comparison. The Sentinel-1A sensor is an active radar system carrying a C-band synthetic aperture radar array. Whereas, Landsat 8 is a passive spectral system with 16-bit radiometric resolution across 0.43 - 2.29 micrometers (excluding the 100m2 IR bands). The characteristics of the sensors will dictate the feature resolution of the objects being detected.
Because the Landsat 8 sensor is carrying a blue edge band (0.43 - 0.45 um), it is likely able to discriminate water/wetland features more effectively than the backscatter signal from Sentinel-1A. This would account for the differences in feature size, regardless of the on the ground cell resolution differences. If you want to use radar SAR data for water/wetland classification it would be better to use L-band data from ALOS PALSAR. This would provide much better discrimination than Sentinel-1A C-band data.
I would have to add here that you really need to consider a classification algorithm. It is impossible to visually Elucidate information contained in both of these data. There could very well be information contained in the data that is providing feature discrimination that is not readily apparent in a visual assessment. The difference in the results could be associated with user bias and not information content. This is particularity relevant in a multivariate context where there may be very relevant information, contained across multiple bands that are not being used in and RGB composite, comprising the image backdrop being used in heads up digitizing. You are effectively throwing away data. In the radar data there is likely a pixel gradient that is not apparent in your visual interpretation. The image can also be seriously compromised based on the type of stretch used in display.
Based on the information you provided, Sentinel-1A has a higher spatial resolution. Each pixel from the Sentinel is 20m x 22m, while each pixel from Landsat 8 is 30m x 30m. For a 100m x 100m area, there would be approximately 25 pixels in the Sentinal image and approximately 9 pixels in the Landsat image.
If you're classifying the image, your analysis may be yielding a larger area from the Landsat image because of the mixed pixel problem. Pixels on the edge of the lake cover some lake and some land. But when you classify the image, you force it to pick one or the other. The smaller spatial resolution from the Sentinel image would minimize this problem. Because of the larger Landsat pixel size, there is likely a larger area mistakenly classified as lake when it is really land.
If you're manually digitizing the image, I don't have a quick answer.
EDIT: Sentinel is an active sensor, which is going to change how you interpret the data. If you're just looking for the lake area, I would use a passive sensor and not an active sensor. The lake water will have very low reflectance across all Landsat bands.