This study proposed a multi-target hierarchical detection (MTHD) method to simultaneously and automatically detect multiple directional land cover changes. MTHD used a hierarchical strategy to detect both abrupt and trend land cover changes successively. First, Grubbs’ test eliminated short-lived changes by considering them outliers. Then, the Brown-Forsythe test and the combination of Tomé’s method and the Chow test were applied to determine abrupt changes. Finally, Sen’s slope estimation coordinated with the Mann-Kendall test detection method was used to detect trend changes. Results demonstrated that both abrupt and trend land cover changes could be detected accurately and automatically. The overall accuracy of abrupt land cover changes was 87.0% and the kappa index was 0.74. Detected trends of land cover change indicated high consistency between NDVI (Normalized Difference Vegetation Index), change trends from LTS (Landsat Thematic Mapper and Enhanced Thematic Mapper Plus time series dataset), and MODIS (Moderate Resolution Imaging Spectroradiometer) time series datasets with the percentage of samples indicating consistency of 100%. For cropland, trends of millet yield per unit and average NDVI of cropland indicated high consistency with a linear regression determination coefficient of 0.94 (p < 0.01). Compared with other multi-target change detection methods, the changes detected by the MTHD could be related closely with specific ecosystem changes, reducing the risk of false changes in the area with frequent and strong interannual fluctuations.

Reconstructing normalized difference vegetation index (NDVI) time series datasets is essential for monitoring long-term changes in terrestrial vegetation. Here, a temporal–spatial iteration (TSI) method was developed to estimate the NDVIs of contaminated pixels, based on reliable data. The NDVIs of contaminated pixels were first computed through linear interpolation of adjacent high-quality pixels in the time series. Then, the NDVIs of remaining contaminated pixels were determined based on the NDVI of a high-quality pixel located in the same ecological zone, showing the most similar NDVI change trajectories. These two steps were repeated iteratively, using the estimated NDVIs as high-quality pixels to predict undetermined NDVIs of contaminated pixels until the NDVIs of all contaminated pixels were estimated. A case study was conducted in Inner Mongolia, China. The accuracies of estimated NDVIs using TSI were higher than the asymmetric Gaussian, Savitzky–Golay, and window-regression methods. Root mean square error (RMSE) and mean absolute percent error (MAPE) decreased by 16.7%–86.6% and 18.3%–33.0%, respectively. The TSI method performed better over a range of environmental conditions, the variation of performance by the compared methods was 1.4–5 times that of the TSI method. The TSI method will be most applicable when large numbers of contaminated pixels exist.

Determining the driving climatic factors at critical periods and potential legacy effects is crucial for grassland productivity predictions on the Qinghai–Tibet Plateau (QTP). However, studies with limited and ex situ ground samples from highly heterogeneous alpine meadows brought great uncertainties. This study determined the key climatic factors at critical plant developmental stages and the impact of previous plant growth status for interannual aboveground net primary productivity (ANPP) variations in different QTP grassland types. We hypothesize that the impact of climatic factors on grassland productivity varies in different periods and different vegetation types, while its legacy effects are not great. Pixel-based partial least squares regression was used to associate interannual ANPP with precipitation and air temperature at different developmental stages and prior-year ANPP from 2000 to 2019 using remote sensing techniques. Results indicated different findings from previous studies. Precipitation at the reproductive stage (July–August) was the most prominent controlling factor for ANPP which was also significantly affected by precipitation and temperature at the withering (September–October) and dormant stage (November–February), respectively. The influence of precipitation was more significant in alpine meadows than in alpine steppes, while the differentiated responses to climatic factors were attributed to differences in water consumption at different developmental stages induced by leaf area changes, bud sprouting, growth, and protection from frost damage. The prior-year ANPP showed a non-significant impact on ANPP of current year, except for alpine steppes, and this impact was much less than that of current-year climatic factors, which may be attributed to the reduced annual ANPP variations related to the inter-annual carbon circulation of alpine perennial herbaceous plants and diverse root/shoot ratios in different vegetation types. These findings can assist in improving the interannual ANPP predictions on the QTP under global climate change.

This study develops a practical methodology to assess the accuracy of multi‐temporal change detection using a trajectory error matrix (TEM). In this error matrix one axis represents the land‐cover change trajectory categories derived from single‐date classified images, and the other represents the land‐cover change trajectories identified from reference data. The overall accuracies of change trajectories and states of change/no‐change are used as indices for accuracy assessment. As the number of possible land‐cover change trajectories can be enormous, a practical processing flow for computing accuracy assessment indices has also been developed to avoid listing all possible change trajectories in the error matrix. A case study using this method was conducted to assess the accuracy of land‐cover change over a period with five observations in a study area in China's arid zone. This method simplifies the process of estimating overall accuracy in the change trajectory analysis, and provides a more realistic and detailed assessment of the results of multi‐temporal change detection using post‐classification comparison methods.