Threshold- and trend-based vegetation change monitoring algorithm based on the inter-annual multi-temporal normalized difference moisture index series: A case study of the Tatra Mountains

Numerous algorithms are used in remote sensing to detect changes in vegetation. Majority of them require several tunable parameters or can only detect abrupt forest disturbances. The aim of this study was to develop a new threshold- and trend-based vegetation change monitoring algorithm (TVCMA) that can detect abrupt and gradual changes in vegetation within forested and non-forested areas. To test the algorithm, the Polish and Slovak Tatra Mountains were used as the study area. Strong winds and bark beetle outbreaks (BBOs) are the primary causes of vegetation disturbances in this region. An annual time series of vegetation indices from 1984 to 2016 was used as the input. The long time span necessitated the use of scenes from the Landsat Thematic Mapper (TM), Enhanced Thematic Mapper Plus (ETM+), and Operational Land Imager (OLI). Fifty-one images were atmospherically and topographically corrected. The collected in situ data included the chlorophyll content, leaf area index, absorbed photosynthetically active radiation, and spectral signatures of non-forest vegetation, dwarf pine, and forest stands in 190 sample plots. To select the vegetation indices (VIs) most suitable for disturbance detection, ten satellite-based VIs were correlated with the acquired field data. The normalized difference moisture index (NDMI) was found to be more sensitive to vegetation disturbances and more resistant to data noise than any other tested index. The TVCMA uses two separate approaches, namely, thresholding, which indicates where and when the disturbances occurred, and a regression analysis, which presents the general trend in the time series for each pixel. The number of detected disturbances, the Spearman's correlation coefficient between the modeled trend line and satellite observations, and p-values were calculated. Different threshold values were tested to identify the value that yielded the most accurate results. By using 200 randomly selected validation points, we achieved an 83.3% producer's accuracy for disturbances (PAD), 46.3% user's accuracy for disturbances (UAD), and 97.8% overall accuracy (OA). These results confirm the potential of TVCMA for monitoring abrupt and gradual changes in vegetation. Moreover, the simplicity and data-driven character of the proposed algorithm make it suitable for multi-temporal analyses of other types of satellite data.