干旱区生态环境知识资源中心

The main objective of this work was to detect changes in a semi-arid forest using Landsat satellite multi-temporal imagery and to circumvent the lack of data on landscape temporal changes. The differential-algebraic method (subtraction) was used on 12 remote sensing variables obtained at different fixed periods. Before applying the differentiations on remote sensing data, a geometric correction and a radiometric normalization of satellite images were carried out to allow images comparison. Differentiation of individual bands (TM4 and TM5), ratios (TM4/TM5 and TM7/TM5), indexes (NDVI and NDMI), Tasseled Cap differentiation (e.g. TCB [Brightness], TCG [Greenness] and TCW [Wetness]) and finally the differentiation of the first three ACPs (ACP1, APC2 and ACP3) were carried out in order to select the best data for the detection of changes. Many radiometric thresholds were tested (11 values) to identify which ones had the ability to determine the best changes in the different studied periods. The most significant values of global precision (GP) (greater than 72% for the four studied periods) obtained in the error matrices were those of the NDVI variable using 0.9 (=standard deviation) at the mean. This allowed the calculation of NDVI changes and the positive, negative and stable proportions for each period. The use of recent very high resolution spatial images for validation allowed us to highlight the causes of changes and the impact of silvicultural works during the period of forest management.