Extended Data Fig. 4: Contributions of different factors to the drought recovery.
From: Widespread spring phenology effects on drought recovery of Northern Hemisphere ecosystems
Contribution of different factors was estimated by random forest models dependent on drought timing and vegetation types for drought recovery occurring within a single growing season (a, RSGS) or within the subsequent growing season (b, RMGS), respectively. Red and blue colours in a and b indicate positive and negative correlation coefficients between predicting variables and drought recovery, and darker colours indicate stronger correlation. Only factors with contribution >2% (averaged contribution of all factors) are displayed. The predrought period represents the six months preceding the extreme drought events. We did not consider the vegetation recovery under late-growing season drought for the RSGS scenario, to prevent the problem of overfitting caused by insufficient training dataset samples. We regrouped the extreme drought events into early-growing season (Early-GS) droughts, mid-growing season (Mid-GS) droughts, and late-growing season (Late-GS) droughts. Four kinds of vegetation types were considered, including evergreen forests (Evergreen), deciduous forests (Deciduous), shrubs, and grasses. A variety of bioclimatic and edaphic factors were introduced into this analysis, including start of growing season of drought year (SOSdrought_year), start of growing season of subsequent year (SOSsubsequent_year), total precipitation in preceding drought period (Precippre), postdrought period (Precippost) and drought response lag period (Preciplag), vapour pressure deficit in preceding drought period (VPDpre), postdrought period (VPDpost), and drought response lag period (VPDlag), NDVI during preceding drought period (NDVIpre), dormant length, mean temperature and snow water equivalent during dormancy period (Tempdorm and SWEdorm), both mean values and variability of mean annual temperature (MAT and TempCV) and mean annual precipitation (MAP and PrecipCV), multiyear mean water balance, drought sensitivity, drought response lag, as well as sand fraction.
