To solve the problems occurring in surveying the farmland environment such as limited energy and different distances of communication among the nodes, a monitoring system with adaptive transmitting power for rice fields based on received signal strength indication (RSSI) and packet loss rate (PLR) was designed and a long-time, stable and highly reliable wireless sensor network (WSN) running this algorithm with low power consumption was established. In the aspect of hardware, to extend the network coverage and improve the stability of the system, the method of frequency heterogeneity was adopted to cluster the nodes and configure different antennas for different nodes. In the aspect of software, the mini embedded free real-time operating system (FreeRTOS) was selected to improve the utilization of CPU and prolong the working life of the clusters. Additionally, to reduce the power consumption, the mode of timed sleeping and waking as well as the power-down control of sensor were selected as the working pattern and the adaptive low-power-consumption mechanism were used on the basis of data deviation perception. The result of tests showed that the system can adaptively adjust the transmitting power to the minimum value which can meet the demand of the stability of communication (PLR was less than 1.3%) within the communication distance of 150m according to the quality of current communication. In this situation, when the algorithms of adaptive transmitting power was adjusted to 8dBm, 6dBm and 3dBm, respectively, from 10dBm, the battery duration of nodes was increased by 11.9%, 21.4% and 33.3%. As the result revealed, the communication performance of the nodes was significantly better than those of three other nodes with different adaptive transmission power algorithms. Thus it can be seen that the system was highly stable and practical in the application of WSN for the monitoring of farmland environment.