Anomalous changes in groundwater have been long reported prior to some sizable earthquakes. Here we examined the water temperature and water level data at the Lijiang well in Yunnan, China from November 1, 2007 to September 30, 2017, and obtained the correlation coefficients between them varying with time by using different time windows. We then use Molchan’s error diagram to obtain the area skill scores for different alarm sizes and distances, which are used for evaluating the prediction efficiency of local sizable earthquakes. We found that (i) when the time window is 90 days and the correlation coefficient is greater than 0.10, a sizable earthquake (M ≥ 5.0) is more likely to happen within 200 km during the following 45 days; (ii) when the time window is 90 days and the correlation coefficient is greater than 0.65, a sizable earthquake (M ≥ 5.0) is more likely to happen within 150 km during the following 45 days. Because there are two separate aquifers and the water temperatures become warmer with depths larger than 200 m, we speculate that normally the well water is recharged mainly by colder rainfall through the shallow aquifer, and increasing water level causes decreasing water temperature and vice versa. However, before a local sizeable earthquake (M ≥ 5.0) within 200 km, there may be deep hotter water flowing into and out of the well through the deep aquifer and breaking the negative correlation between them, and positive correlation coefficients are even observed. This study may provide some useful clues for short to medium-term forecast of sizable earthquakes in the Lijiang region and can provide methodological reference for similar studies in other regions.