Abstract: The stem water content of plants can be one of the most important parameters to assess the crop response to the drought stress in arid and semi-arid areas, particularly during irrigation scheduling. But, only a few sensors can be selected to in-situ and non-invasively measure the stem water content of the thinner plant branches. In this study, a parallel-plate probe sensor was optimized and designed to in-situ and non-invasively determine the branch water content, when operating at the frequency domain (FD) of 100 MHz. A long strip (40 mm length, 5 mm width, and 0.05 mm thickness) was improved for the parallel-plate probe sensor to better fit the apple tree stem. Simulation and experimental measurements were conducted to assess the electric field distribution and the effective depth of the parallel-plate probe sensor. The capacitance, the sensitivity to the stem diameter, and the accuracy of the sensor were determined in the performance test for the branch water content of three apple trees in a greenhouse environment. A High-Frequency Structure Simulator (HFSS) was selected for the simulation analysis and the step removal wood chips test. The results showed that the effective depth of measurement was 10 mm, indicating the better performance of the parallel-plate probe sensor for the branch water content, according to the previous research (The zones of main water storage between the periderm and the xylem for the 20 mm apple tree stem should be less than that for the 30 mm apple tree stem). The sensitivity test showed that the parallel-plate probe sensor (4.23 mV/mm, and R2= 0.892) was less susceptible to the stem diameter variation. Specifically, the slope of the sensor indicated that the 1 mm variation of stem diameter resulted in the 4.23 mV measurement error by the sensor. The performance test showed that the parallel-plate probe sensor behaved the excellent stability and repeatability. There was also an excellent linear relationship between the sensor output and capacitance value (R2=0.985). The output fluctuation values of the parallel plate electrode sensor with/without the fixed load were about 0.8, and 0.5 mV, respectively. The maximum difference was 8 mV for the sensor output at the largest difference of 0.431 cm3/cm3 in the repeatability experiment. The output of the parallel-plate probe sensor in three repeated experiments was not more than 5 mV under the water content of the same branches. The temperature coefficient of the parallel-plate probe sensor was 3 mV/℃, indicating the output error of the sensor was about 3 mV for every 1 ℃ change in the external temperature. A linear equation was also obtained between the sensor output and branch water content, indicating the better performance of the sensor for the measurement requirements. The water dynamics demonstrated that the improved sensor can be expected to detect the variation in the branch water content of the potted apple tree in the greenhouse, particularly for a daily dehydration-rehydration cycle. There was no fluctuation with the full irrigation in the averaged branch water content from midnight to predawn. In the water deficit identification experiment, there was no rehydration process of AT-2 before the midnight of Sept 15th, where the branch water content increased from 0.422 to 0.495 cm3/cm3 after irrigation near the midnight of Sept 16th. Both AT-1 and AT-2 may experience a water deficit before the irrigation, according to the maximum daily trunk shrinkage (MDS). Consequently, the improved sensor can be used to in-situ and non-invasively detect the water content of the apple branch with the branch’s diameter of 5-8 mm. The water deficit of apple trees can also provide great support to guide the irrigation in the field.