Improved NDVI Time Series To Access The Temporal And Spatial Dynamics Of Post-fire Forest Recovery

The Greater Xing'an Mountains,the largest natural forest area in China,are rich in forest resource and play a significant strategic role in China's forestry production.However,there are frequent forest fires in the region.Fires often quickly destroy forest resources with large-scale continuous or patched burned areas.By contrast,the post-fire recovery of vegetation is a long-term process,in which the early stages are the key to restoration of structure and function of forest ecosystems.Consequently,the use of remote sensing in monitoring the progress of post-fire forest recovery in burned areas with different fire intensities is of great significance.With the assistance of sateliite observations,we could measure forest specific responses to fires,understand the temporal and spatial characteristics of forest restoration,and learn about ecosystem processes of regenetated forests and their changes in ecosystem functions for sustainable managements.Yet,satellite imagery is influenced by TM sensor itself,atmospheric conditions and seasonality of forests,so that the available inter-annual image series is often not continuous and ready to use.Based on the 1987 catastrophic fire on May 6-June 2 in the Greater Xing'an Mountains,this study takes the normalized difference vegetation index(NDVI)as an important characterization parameter to examine the post-fire forest greenness dynamics,which reveals forest recovery along the inter-annual NDVI trajectory.Firstly,three NDVI reconstrument methods are tested and their performances are assessed: 1)the TM Pair method that builds a regression model between the tested image and the peak-season(July-August);2)the GIMMS method that extracts the NDVI differences from the 15-day GIMMS NDVI data in the days between the test image and the peak-season image,then implements this NDVI difference to the tested image;and 3)the Meteorology method that performs spatial interpolation based on station data to adjust the NDVI change of the tested image.A multi-year NDVI time series in growing season is constructed via the three methods,and the best results are applied for forest recovery analysis in next step.Secondly,combining the fire intensity and forest types,the spatial distribution of forest recovery trend and its recovery trajectories in different burned areas in 1987-1999 was analyzed,and the NDVI recovery processes of forest types under different fire disturbances were monitored.The primary findings are as follows:(1)By comparing the residual distributions under different fire intensities based on thesame seasonal adjustment,it is found that the residuals of the projected NDVI in Mayand October were larger than those in June and September,indicating that images inMay and October are not optimal for multi-year NDVI time series reconstruction.Inthese periods,the uncertain weather conditions(cold wave,snow events)in this coldregion may lead to worse seasonal adjustment than those in June and September.(2)By comparing the residuals and accuracy of the three NDVI seasonal adjustments inthe same month,it is found that the residuals between the reference NDVI and theprojected NDVI were smallest using the TM Pair method,followed by theMeteorology method,then the GIMMS method.(3)The linear relationship between TM NDVI and GIMMS NDVI show high variationof the GIMMS data in October(19881007 and 19951011),confirming that theGIMMS method was not suitable for the correction of TM NDVI in dates close to theend of growing season.(4)Based on the growing-season NDVI time series from 1987 to 2009,the greennessrecovery showed the remarkable stages: NDVI recovered most significantly in heavyand moderate fire zone in 2 to 5 years after fire;NDVI continued to significantlyincrease after 6 to 13 years after fire;13 years later,the statistical difference betweenNDVI of burned and unburned areas became not significant.(5)The 13 years after the fire was a key period for the recovery of forest greenness.Therewas a clear positive correlation between forest recovery rate and fire intensity;theseverer the intensity of fire,the faster the NDVI growth rate and more significant thegrowth trend.(6)Post-fire recovery rate varied with forest types: the white birch(Betula platyphylla)became dominant in many burned areas,which grew faster and had higher NDVIvalues than the recovery of coniferous forest dominated by larch(Larix gmelinii).TheNDVI values were similar between the two forest types in unburned forest with highcanopy covers.