Data fusion for wireless sensor networks (WSN) can reduce the energy consumption of sensor nodes and prolong the network lifetime, so that it has attracted wide spread attention in a variety of applications. The existing algorithms for spatial data fusion that have been used in agricultural monitoring always aggregate the data within a certain area into one value by means of averaging. However, in addition to redundancy resulted from correlation, the sensed data also has variance due to larger monitoring area, more monitoring nodes and larger amount of data in farmland environment. Hence, data fusion in farmland monitoring should retain the differences of data while eliminating the redundancy. The idea that applying data fusion algorithm on WSN cluster head to aggregate spatially correlated data by clustering was proposed. While the parameters whose values are quite different will be clustered into different categories so that differences between the data can be reserved. An improved adaptive K-means clustering algorithm was proposed to be used in cluster head. Simulation results indicate that, the amount of data uploaded with fusion algorithm was decreased by 41.19% compared with that without fusion algorithm,and the maximum error before and after the proposed fusion algorithm is less than 36% of that before and after the averaging fusion method.When there is no clear accuracy requirement,the proposed algorithm can reduce the amount of data uploaded and maintain the relative error less than 10%, 〖JP3〗avoiding enormous error caused by improper number of clusters.When there are specific accuracy requirements, the relative error produced by the proposed algorithm can meet the error requirements strictly.