Rice Mapping Using Time-Series AI Algorithms – A SAR Based Approach
Rice is the single most important human food crop in the world, directly feeding more people than any other crop. Rice is a staple food that is a necessity for most people living in the Asian regions.
Precise rice mapping is of paramount importance for ensuring food security and maintaining continual economic development. In Indonesia, rice production is an essential part of its national economy. Indonesia holds the title of the third-largest producer of rice in the world.
Skymap Global has built its capability on mapping paddy across south-east Asian countries. One such example illustrated in this article is the paddy mapping of Java, Indonesia. The use of optical images was churned out as most parts of Indonesia had cloud coverage, and SAR images have the ability to penetrate cloud covers to derive valuable insights.
Our time-series satellite imagery analysis and rice growing stage
In Java, there are two seasons for planting and harvesting rice. The first season starts in October-November and ends in March-April of the following year. The second season begins in May-June and ends in August-September.
The typical timeline has a duration of 150 days, as illustrated below. During that period, several growing stages are observed. Owing to a revisit time of 12 days, it is possible to continuously monitor the growth of the plants and detect the presence of anomalies. Operating in two modes (ascending and descending), it is even possible to monitor the growing stage with a temporal sampling of 6 days. In other words, Skymap Global can use 25 images to capture the plants’ health all along the season.
Our time-series analysis and information output
Through time series analysis, we are able to curate a graph representing the evolution of the signal recorded by each pixel (10*10m) of the paddy field, as illustrated below.
Early March, rice is entering a growing stage (tillering). The cursor can be positioned anywhere inside the field. Values indicated represent the area concerned in ha.
The curve indicates a perfect growth. No disease or issue was recorded. The high temporal sampling also allows the distinction of the plants’ type throughout the year.
Our machine learning algorithms have to be trained to distinguish rice from other types of plants owing to their temporal signature. The classification result is presented here in a red color. The accuracy which we received was in the order of 94% compared to published government data, which is excellent, adding to the fact that the whole process was done automatically through the power of AI/ML.