Forest phenology extraction across Northern Mid–High latitudes from 1982 to 2020 based on daily LTDR NDVI

• High temporal resolution greatly improves forest phenology accuracy. • Generated forest phenology dataset (1982–2020) for mid- to high-latitude. • Mitigated the discontinuities found in existing high-latitude phenology products. • Comparative analyses reveal strong sensitivity of phenology to temporal resolution. Phenological metrics derived from remote sensing data are highly sensitive to the temporal resolution of input time series. However, most existing large-scale and long-term phenology products are generated from multi-day composite datasets, which may obscure short-term vegetation dynamics. The Advanced Very High Resolution Radiometer (AVHRR) Long-Term Data Record (LTDR), characterized by its daily temporal resolution, provides an opportunity to improve the precision of phenology extraction. Nevertheless, the existing LTDR-derived phenology product (AVH09_phe) shows anomalous spatial patterns in high-latitude regions. Using AVHRR LTDR-based background field reconstruction NDVI (LTDR-BFR-NDVI), we systematically compared eight extraction strategies combining two temporal resolutions (daily and biweekly composites) and four algorithms: the first derivative (FD), second derivative (SD), and dynamic threshold methods (DT_0.2 and DT_0.5). Spatially representative field observations were used to assess their performance. The results showed that phenology derived from daily NDVI exhibits substantially stronger consistency and lower uncertainty compared to that from biweekly composites. For SOS, daily-based methods achieved higher correlations (R = 0.55–0.62) and lower errors (RMSE = 15.14–17.86 days), while biweekly-based methods showed weak correlations (R < 0.4) and larger uncertainties. For EOS, overall performance was lower, but daily-based methods still outperformed biweekly composites, with the second derivative method providing relatively balanced accuracy (R = 0.41, RMSE = 21.98 days, MAE = 13.92 days, bias = 1.14 days). The combination of daily NDVI and the second derivative method achieved the highest accuracy and was adopted to produce a new forest phenology product, LTDR-BFR-FP. Comparative analyses with other published satellite-derived phenology products revealed that LTDR-BFR-FP mitigates high-latitude artifacts in AVH09_phe. Furthermore, the comparison between daily and biweekly NDVI time series demonstrated that temporal resolution exerts a significantly influence both phenology accuracy and trends.