A Simple Method to Extract Tropical Monsoon Forests Using NDVI Based on MODIS Data:A Case Study in South Asia and Peninsula Southeast Asia

Distribution of monsoon forests is important for the research of carbon and water cycles in the tropical regions. In this paper, a simple approach is proposed to map monsoon forests using the Normalized Difference Vegetation Index(NDVI) derived from the Moderate Resolution Imaging Spectroradiometer(MODIS) data. Owing to the high contrast of greenness between wet season and dry season, the monsoon forest can be easily discriminated from other forests by combining the maximum and minimum annual NDVI. The MODIS-based monsoon forest maps(MODMF) from 2000 to 2009 are derived and evaluated using the ground-truth dataset. The MODMF achieves an average producer accuracy of 80.0% and the Kappa statistic of 0.719. The variability of MODMF among different years is compared with that calculated from MODIS land cover products(MCD12Q1). The results show that the coefficient of variation of total monsoon forest area in MODMF is 7.3%, which is far lower than that in MCD12Q1 with 24.3%. Moreover, the pixels in MODMF which can be identified for 7 to 9 times between 2001 and 2009 account for 53.1%, while only 7.9% of MCD12Q1 pixels have this frequency. Additionally, the monsoon forest areas estimated in MODMF, Global Land Cover 2000(GLC2000), MCD12Q1 and University of Maryland(UMD) products are compared with the statistical dataset at national level, which reveals that MODMF has the highest R2 of 0.95 and the lowest RMSE of 14 014 km2. This algorithm is simple but reliable for mapping the monsoon forests without complex classification techniques.

Distribution of monsoon forests is important for the research of carbon and water cycles in the tropical regions. In this paper, a simple approach is proposed to map monsoon forests using the Normalized Difference Vegetation lndex （NDVI） derived from the Moderate Resolution Imaging Spectroradiometer （MODIS） data. Owing to the high contrast of greenness between wet season and dry season, the monsoon forest can be easily discriminated from other forests by combining the maximum and minimum annual NDVI. The MODIS-based monsoon forest maps （MODMF） from 2000 to 2009 are derived and evaluated using the ground-truth dataset. The MODMF achieves an average producer accuracy of 80.0% and the Kappa statistic of 0.719. The variability of MODMF among different years is compared with that calculated from MODIS land cover products （MCD 12Q 1）. The results show that the coefficient of variation of total monsoon forest area in MODMF is 7.3%, which is far lower than that in MCD12Q1 with 24.3%. Moreover, the pixels in MODMv which can be identified for 7 to 9 times between 200l and 2009 account for 53.1%, while only 7.9% ofMCD12QI pixels have this frequency. Additionally, the monsoon forest areas estimated in MODMF, Global Land Cover 2000 （GLC2000）, MCDI2Q1 and University of Maryland （UMD） products are compared with the statistical dataset at national level, which reveals that MODMv has the highest R^2 of 0.95 and the lowest RMSE of 14 014 km^2. This algorithm is simple but reliable for mapping the monsoon forests without complex classification techniques.