North of 30°N, the reported 1982–2011 weakening of the NDVI–temperature link is within year-shuffle noise in PKU GIMMS NDVI to 2022, and the April–October partial correlation did not fall

Piao et al. (2014) reported that north of 30°N, the year-to-year correlation between growing-season NDVI and temperature became weaker from 1982 to 2011 (moving 15-year windows, P < 0.01). We repeated the analysis with PKU GIMMS NDVI v1.2, which corrects satellite drift. We used the AVHRR-only version (1982-2015) and the AVHRR-MODIS version (1982-2022), and CRU TS v4.10 climate data, at 30,036 cells of 0.5 degrees. We added a permutation test that shuffles the years with the same order for all variables and cells. This test keeps the overlap of the windows and the spatial correlation. The simple correlation of the area-weighted series decreased from 0.77 (1982-1996) to 0.64 (1997-2011). Piao et al. reported 0.72 to 0.32. The partial correlation, which controls for precipitation and cloud cover, increased from 0.53 to 0.64. We ran 24 variants (product, season, temperature data, latitude band, continent, cell mask). Of 94 moving-window trend tests, 47 had naive p < 0.05, but none had permutation p < 0.05 (smallest 0.09). A phase-randomized null that keeps autocorrelation gave the same result. At cell level, the naive test found a significant decrease in 33% of the area, and shuffled years gave 32% (5-95% range 25-39%). Cell decreases from 1982-1996 to 1997-2011 did not continue: those cells gained +0.17 in R by 2008-2022, as noise predicts. The decrease of the simple correlation agrees with the decrease of interannual temperature variance (SD 0.32 to 0.21 °C) at a constant NDVI slope per °C. The AVHRR-MODIS product has a -0.010 NDVI step at the 2003 change to MODIS. This step lowers the correlation in windows that include 2003, unless it is removed. Main limitations: we did not use GIMMS NDVI3g or CRU TS 3.2. The CRU cloud field correlates with temperature. The tests have low power, so the result does not show that the relation is constant.

NDVI · vegetation greening · temperature sensitivity · moving window correlation · permutation test · PKU GIMMS NDVI · CRU TS · northern ecosystems · replication · sensor discontinuity · partial correlation · field significance