Rice extraction in the Central Valley using multi-temporal Landsat images

Abstract

Paddy rice is one of the most important crops to offer daily food for human beings and has a strong influence on food security and market stability. Therefore, it is of significance to obtain first-hand information related to rice such as rice planting extent. In order to timely and accurately obtain the spatial distribution and temporal variations of rice plants, remote sensing techniques have been introduced and applied to achieve this goal, among which the phenological method is unsupervised and can be used to automatically extract rice. However, the phenological methods are strongly dependent on the weather conditions, and the resultant extraction results might be damaged or incomplete. Additionally, the machine learning-based methods are less influenced by the weather contaminations but are restricted by the availability of training samples. In consideration of the challenges of each approach, this study combines these two approaches to extract rice by using the initial phenology-based rice samples as the training samples of machine learning models, thus obtaining more complete results and high accuracy. This strategy is tested in the Central Valley of California, USA, one of the main rice-producing areas in the United States, and is compared with the rice maps derived from Cropland Data Layer (CDL) from the U.S. Department of Agriculture (USDA). The results show that through the incorporation of the phenology- and machine learning-based methods, the derived rice maps present high consistency with the CDL-rice map, and high accuracy is achieved in terms of OA and Kappa values reaching up to 0.96 and 0.86 respectively, while the purely phenology-based rice map shows great misclassifications thus leading to a sharp decrease in the Kappa value (0.49). This comparison proves the effectiveness of this strategy. The satisfactory results in this study show the possibility of extending this strategy to other regions for rice mapping.