The Extraction Of Spatial And Temperal Information Of Economic Man-made Forests Based On NDVI Time Series

Vegetation is a very important part of the global terrestrial ecosystem.It not only provides the ecosystem with the power of energy to maintain its normal operation,but also gives habitats of other species of this ecosystem.However,during the recent few decades,the activities of human being have been more and more vigorous,which at the same time brought tremendous influence on the coverage of global vegetation.As a big amount of primeval forests were being cutting down,more and more man-made forests were being planted.The excessive man-made plantations not only disturbed the original balance of the ecosystem,but also limited the biodiversity of this region.Therefore,using the techniques of remote sensing to figure out the plantation condition of man-made forests is desperately needed for protecting the functions of vegetation in the ecosystem.Researchers were used to using MODIS data,which has a good temporal resolution but bad spatial resolution,to extract economic man-made forests,using the traditional algorithms,City Block,Standardized Euclidian Distance and Bounding Envelope.However,these algorithms were not suitable to be applied in extraction of economic man-made forests in small-moderate scale.In this study,we collected Landsat 5 TM,Landsat 7 ETM and HJ-1 data to reconstruct NDVI time series and proposed a new algorithm,Inverted Triangel Area(ITA).ITA was successfully applied to extract the spatial and temporal distribution information of eucalyptus in the cities of Shanguan,Heyuan and Ganzhou.In addition,based on the analysis of the relationship between the growing period types and the earliest acquisition date of images,we indirectly estimated the planting date of eucalytus,relatively accurately.The main conclusions include: 1.The feasibility of building NDVI time series based on LANDSAT 5 TM,ETM and HJ 1A/B imagesWe additionally chose some ETM and HJ 1A/B images to make up the vacancy due to the deficiency and low quality of LANDSAT 5 TM images in some years or months,to build a complete NDVI time series over 15 years.As images from different sensors have different spectral response functions,satellite orbits,solar altitude angles when acquiring images,observation angles and so on,we did correlation analysis between LANDSAT 5 TM and HJ 1A/B.Four couples of LANDSAT 5 TM and HJ 1A/B in 2012 of the same regions were downloaded and compared,which showed that the NDVI values of any of the four couples of images had over 98% correlation among over 100 homologous pixel couples with each other.It proved NDVI obtained from LANDSAT 5 TM,ETM and HJ 1A/B images have difference,but quite small,which would not affect much on the classification results.It was proved that using LANDSAT 5 TM,ETM and HJ 1A/B images to build NDVI time series had high feasibility.2.The feasibility of using the ITA algorithm to identify eucalyptus in small-scale regionsWith the consciousness of too long interval of the NDVI time series,we fully analyzed the regular pattern of the reference NDVI time series of eucalyptus and created a new algorithm,the Inverted Triangle Area Algorithm.Compared with the other three traditional algorithms,the ITA algorithm avoids the description about the single time steps which vary within a huge range,but makes full use of the stability of the inverted triangle area to describe the entire changing trends of the NDVI time series.To demonstrate its feasibility,we used a composite UAV photograph to train the ITA and three other algorithms,and determine the optimum thresholds.The results showed that both of the producer accuracy and user accuracy of the ITA reached up to nearly 80%,while that of the other three algorithms were only about 50%.For further validation,we used GF-1 data in December of 2014 to test the ITA in a larger region,to find that the producer accuracy and user accuracy still remained in high level,about 76%.It was proved that using the ITA algorithm to identify eucalyptus in small-scale regions has high feasibility.3.The feasibility of using the time interval between the earliest image acquisition date during one eucalyptus pixel's growing period and its clear-up date to estimate eucalyptus planting dateAfter analyzing the relationship between the earliest image acquisition dates of 135 reference NDVI time series of eucalyptus and their types of growing periods(one year long and two years long),we did a bold assumption: a new parameter,?t,was introduced.We used this ?t to estimate the planting date of eucalyptus.After getting the preliminary results,we did a correction on ?t using a iteration,to develop the estimation accuracy.The final results showed that,the root mean square error between the estimated planting date and actual planting date was about 3 months,which was acceptable.4.Spatial and temporal information of the cities of Shaoguan,Ganzhou and Heyuan in Guangzhou ProvinceWhen figuring out the spatial and temporal information of the cities of Shaoguan,Ganzhou and Heyuan in Guangzhou and Jiangxi Province,we applied the ITA and BE algorithm to extract eucalyptus pixels and our own estimation method to estimate the planting date.On the basis of these,the spatial and temporal information of the three cities of Guangdong Provence from 2000 to 2014 was acquired.The result showed that during the 15 years,the eucalyptus plantation area was changing all along,reached up to the peak in 2007 and 2008,which was over 250km~2 and in other years it was only about 100km~2.