As one of the most important staple foods globally, rice sustains nearly half of the world's population. Accurate and timely paddy rice mapping is, thus, essential for rice-related policy-making to ensure food security in the context of anthropogenic, environmental and climate changes. However, paddy rice mapping remains a challenging task since it usually has similar spectral characteristics to other land covers. In this research, for the first time, an entirely new approach, called RiceTColour, was proposed for mapping rice fields within the Commission Internationale de l'Eclairage (CIE) colour space based on their unique spectra during the rice transplanting period as observed in remotely sensed imagery. We demonstrate that transplanted rice fields, representing a mixture of soil, water and rice seedlings, consistently exhibit relatively low spectral values in both SWIR and NIR bands across various geographical locations, leading to their unique dark green colours in the false-colour image composed of SWIR, NIR and Red bands. Based upon this, we transformed these three spectral bands into the CIE colour space where paddy rice was found to be readily and completely separated from the other land covers. Straightforward, but specific classification criteria were established within the CIE colour space to differentiate paddy rice from the other land covers. The proposed RiceTColour, thus, represents a new approach for paddy rice identification, that is mapping paddy rice using the CIE colour space based upon the previous underexplored remotely sensed spectra of paddy fields during the transplanting season. The effectiveness of the proposed method was investigated over five rice-planting regions distributed across different geographical regions, characterised by different climates, rice cropping intensities, irrigation schemes and cultural practices. Specifically, the mapping criteria established in a training site (S1) were directly generalised to the other four sites (S2 to S5) for paddy rice mapping. Experimental results demonstrated that the RiceTColour method consistently achieved the most accurate and balanced classifications across all five sites compared with four benchmark comparators: a SAR-based method, an index-based method and two supervised classifier-based methods. In particular, the RiceTColour method performed relatively stable, producing an overall accuracy exceeding 95% in the training site (S1) as well as the four generalised sites (S2 to S5), which is an encouraging result. Such efficient yet stable rice mapping results across various rice-planting regions suggest a very strong generalisation capability of the proposed RiceTColour method. In consideration of the relatively large planting area of paddy rice fields globally, the proposed parameter-free, efficient, and generalisable RiceTColour method, thus, holds great potential for widespread application in various rice-planting areas worldwide.
Read full abstract
Sep 1, 2024
Pavel Exner + 1
Open Access
