Analysis of spectral measurements in paddy rice field: Implications for land use classification

Abstract

Rice is one of the most produced crops in the world consumed by over half of the entire population according to the FAOSTAT (2008). In addition to the agricultural value of rice, paddy-farming method used to grow rice raises a number of environmental issues, such as methane emissions and high fresh water consumption, associated with the production of rice. Monitoring of paddy fields at large scale is thus important to investigate their impacts on available water resources and greenhouse gas emissions as well as to predict and plan rice production of the year. Due to the unique spectral features of paddy rice fields during the transplanting season (i.e., strong signals in the short-wave infrared range caused by the flooded soil of the paddy fields) the Land Surface Water Index (LSWI) has been used in conjunction with the Normalized Difference Vegetation Index (NDVI) to map paddy rice fields using satellite imagery (e.g., Xiao et al. (2005); Xiao et al. (2006)). To date, satellite images of spatial scales ranging from 30 m (e.g., Landsat) to 500 m (e.g., Moderate Resolution Imaging Spectroradiometer, MODIS) or coarser scales were directly compared with ground survey data to classify the land use in the mixed paddy rice and other cropping fields. However, due to the small size of typical rice fields in Asia where majority of rice fields are located, only a fraction of each satellite pixel is covered by the rice field. Moreover, the strong short-wave infrared signal from flooded soil may drop significantly as growing rice canopy masks signals from the background soil. In order to develop more sophisticated spectral models for satellite-based land use classification, these complex features of the paddy fields need to be analysed using measurements made directly on the rice fields in various growing stages. We present spectral measurements collected during the Australia-India Land Surface Parameterisation Experiment for Remote Sensing (AILSPEX-RS1) conducted in February 2011 in the Musi catchment, India. Spectral signals at six wavelengths (i.e., 530, 570, 650, 855, 1240 and 1640 nm) were measured over 16 different land cover types during the local transplanting season of rice. Conditions of rice paddy fields varied from no-rice flooded soil to moderately grown rice canopies as tall as approximately 50 cm. LSWI of paddy rice linearly decreased with NDVI while the index in the non-paddy-rice fields increased linearly with NDVI. The linearly decreasing trend of LSWI of paddy rice continued until the NDVI values reached ∼0.8, after which LSWI of paddy rice increased with NDVI following the trend of the other crops. The analysis of the explicit NDVI-LSWI trends can be applied to map paddy rice fields until rice grows to ∼ 35 cm while the simple comparison of the indices by their magnitude can be applied to the paddy rice of approximately 20 cm.