A low-cost rice mapping remote sensing based algorithm

Egypt faces a great challenge, limited water resources and increasing water demand. The agriculture sector consumes about 83% of the available water resources. The main water-consuming crop planted in summer is rice. Thus for any better water resources management, rice mapping is required. Remote sensing can be utilized for rice mapping. This will potentially save money and effort. The most differentiating feature of rice is being flooded in the transplanting period. Xiao (2005) developed a rice mapping algorithm by studying the dynamics of three vegetation indices, the Land surface water index (LSWI), the normalized difference vegetation index (NDVI), and the Enhanced vegetation Index (EVI). The key assumption is that a moisture sensitive index, as LSWI, will capture the flooding of rice and will temporal lily exceeds or approaches NDVI, or EVI, thus signaling rice transplanting. Xiao utilized MODIS (500 m spatial resolution, twice a day temporal resolution) free satellite imagery. However, its coarse resolution combined with Egypt heterogeneous and fragmented land ownership raised the need for the algorithm modification. In the current research a low-cost rice-mapping algorithm was developed. The accuracy of rice mapping from MODIS satellite imagery was enhanced by making use of LANDSAT imagery. This was achieved by developing a novel decision tree classifier that classifies land cover into its four main classes namely: vegetation, desert, bare land or urban, and water utilizing LANDSAT imagery. The non-vegetation area is then used to refine the rice area calculated from MODIS. Another challenge of rice mapping from MODIS is that in rice fields the reflectance is a combination of water, vegetation, soil, and ditches thus not always the LSWI will exceed the EVI or the NDVI as proposed in the literature, but instead it will approach it in the transplanting period. In order to reflect this, a ∆-parameter was introduced. The adopted criteria for rice mapping was LSWI + ∆\u3e NDVI or LSWI + ∆\u3e EVI. The ∆-parameter was obtained as best fit for each rice-growing region. The ∆-parameter is different for EVI and NDVI. The ∆EVI for Kafrelsheikh and Dumyat was found to be 0.04. Daqehleya, Gharbeya and Sharqeya ∆-parameter was calculated as 0.05. While Behera governorate ∆-parameter was estimated to be 0.07. While ∆--NDVI parameter for KafrElsheikh was 0.174, for Dumyat was 0.178, for Sharqeya was 0.18, for Gharbeya was 0.197, for Behera was 0.23, and for Daqhleya the ∆- NDVI parameter was 0.155. The developed rice-mapping algorithm was applied to the Delta region in Egypt to predict the rice cultivated areas in the year 2009. The resultant rice areas map was validated using randomly selected points, and local knowledge of rice planting practices, against very high-resolution (60 cm) imagery. The overall accuracy of the main land cover mapping was 90%. The rice areas map and probable transplanting dates conforms to local knowledge of rice planting practices. The results of this study indicate that the developed rice-mapping algorithm can be applied as an economic way for rice area mapping on a timely and frequent basis. However mapping rice fields prior to flooding would have been revealed more information for water management. More research should be directed to the early mapping of rice transplanting in the future