A simple method to improve field-level rice identification: toward operational monitoring with satellite remote sensing

Abstract

Discriminating crops by remote sensing remains reasonably complex and expensive for many agricultural land managers. The current study was conducted to facilitate the operational use of remote sensing for field-level rice monitoring in Australia by determining (i) whether existing methods relating to simple moisture-based rice classification could be further simplified, and (ii) whether the high accuracies resulting from that moisture-based methodology could be further increased. First, the impact of removing the most complicated processing step, atmospheric correction, on rice classification accuracies was assessed for the 2000–01 summer growing season at the Coleambally Irrigation Area, New South Wales. The primary error sources of rice classification were then identified and simple rules developed in an attempt to reduce errors associated with confusion between unharvested winter cereals and flooded rice paddies early in the summer growing season. These newly defined rules were then used on imagery acquired in the subsequent summer growing season (2001–02) in order to assess their repeatability. The assessment of atmospheric correction showed that during the critical time frame associated with high rice identification (October–November), using non-atmospherically corrected data increased overall accuracy, although the improvement was small (about 1%). Overall accuracy also increased for every case tested for both growing seasons as a result of the rule-based classification (ranging from about 1 to 14%), revealing that the methods were sufficiently repeatable. This study moves per-field rice monitoring at the Coleambally Irrigation Area closer to an operational application and shows that simple rule-based remote sensing classifications can be very effective when site practices are known.