DETECTION OF START OF SEASON DATES OF RICE CROP USING SAR AND OPTICAL IMAGERY, CENTRAL LUZON, PHILIPPINES

摘要

Rice plays a crucial role to food security in the Philippines. Information on rice such as where and when it is planted is essential in planning and decision making. Knowing the start of season (SoS) date is very important for proper management such as scheduling of irrigation, fertilizer application, and rice production estimation. In this study, we used Sentinel-1A/B and Landsat-8 imagery to derive the rice planted areas and SoS dates. We processed data for two cropping seasons in 2016-2017 in Central Luzon, Philippines. We used the MAPscape-RICE? software to generate the rice and SoS maps, and compared results with ground observations. We performed 3 main steps: (1) basic processing. (2) classification based on rule-based detection, and (3) accuracy assessment. In the rule-based detection of SoS date, we used SAR backscatter value and vegetation index (NDVI) derived from Landsat-8. The SoS date is detected based on the weights used to compute the reliability coefficient (RC). The following are the parameters: (1) backscatter value from SAR polarizations. (2) backscatter increase after the SoS, (3) correspondence of SoS date in SAR input polarizations, (4) consistency of backscatter with NDVI value, and (5) value of local incidence angle (LIA) in overlapping areas. The RC is obtained by summing up the contribution of each input multiplied by its own weight factor. For the accuracy of the rice area classification, we used 120 validation points per season and for the start of season, we used 70 (DS) and 80 (WS) ground observations. The rice area classification results have an overall accuracy of 93.3% (kappa = 0.87) for DS and 91.7% (kappa = 0.83) for WS. The estimated SoS dates derived from SAR correlated strongly with actual dates (R~2=0.81 for DS and R~2=0.80 for WS). The average deviation is 12 days for DS and 3 days for WS. This is acceptable since the maximum revisit period of Sentinel-1 is 12 days. With such high accuracy, this approach of detecting the area and start of season of rice can support better targeting of appropriate interventions and their timing, assessment of areas at risk and damages brought by typhoons, and vulnerability to pests and diseases.