Monitoring agriculture crops using sentinel -2 satellite: A case study of coffee in Hoa Dong, Dak Lak province

Abstract

In Vietnam, coffee is one of the most valuable traded commodities, which is planted in high altitude areas such as Tay Nguyen, Lam Dong, Cau Dat, etc. Among those, Dak Lak province, located in Central Highland, is one of the largest coffee producers in Vietnam and worldwide. However, it is listed as one of the plants which takes lots of intensive management as well as internal and external resources for growing and developing. In addition, traditional management methods are expensive and take huge efforts to monitor enormous crops effectively. Thus, farmers need an alternative way which is cost-effective and requires less effort. To overcome difficulties when dealing with traditional management, along with the rapid development of satellite technology, remote sensing approaches these recent years have been applied widely in monitoring and managing agriculture crops. They provide precise cropland maps, indicating cropland and non cropland areas, which are baselines for developing higher-level products, such as crop’s irrigation and rainfed, crop intensities, crop types, cropland status of nutrients, disease, as well as for assessment of cropland productivity (productivity per unit of land), and crop water productivity (productivity per unit of water) on a very large scale of area with slightly manual interception. However, those precise and accurate maps require data from data with high spatial resolution (less than 5m) to moderate spatial resolution (between 5m and 60m) [18] . Recently, Sentinel-2 has been used widely in precise agriculture due to its free availability to moderate spatial resolution of 10m. Therefore, in this research, I present an approach for mapping coffee extent at moderate spatial resolution (10m) by using the 10-day, 10 to 20m, Sentinel-2 data. Two of the most important tasks are to detect coffee crops and extract information from them. For the detection task, some common classification models were compared and the model with the best accuracy was chosen to be the main classifier. For the extracting information task, a map of vegetation indices used in precise agriculture was created to get the information about the crop’s status. The independent accuracy assessment showed the highest weighted overall accuracy of 99.85% with the corresponding producer’s accuracy of 99.97% and the corresponding user’s accuracy of 99.87% for the coffee class. Consequently, limitations within the implementations and analysis of the crop extent are discussed later in detail.