Seasonally divergent early-warning indicators of Amazon hydrological change

Abstract Variance, lag-1 autocorrelation, skewness, kurtosis and entropy are used as early-warning indicators of critical transitions, on the premise that systems losing stability fluctuate more widely and recover more slowly. We computed them for about 45 years of precipitation, evapotranspiration, soil moisture and extreme streamflow records across the Amazon Basin and its eight sub-basins, in wet (December–February) and dry (June–August) seasons, against the fraction expected by chance under stationary surrogates. Wet-season change concentrates in the south: precipitation along the western Solimões–Juruá–Purus–Madeira corridor, and root-zone soil moisture in Purus and Tapajós, where 60% and 44% of the area show four or more concordant indicators against 6% expected by chance. Dry-season change shifts eastward: the Xingu–Tapajós arc, least affected in the wet season, becomes the most affected in precipitation, while minimum streamflows give the strongest result of all, five indicators aligning at 14% of gauges, twenty-six times chance. Evapotranspiration is weakest in both seasons. Resolving the soil column by layer places the soil moisture signal at 7–100 cm depth, and the areas flagged in the two seasons avoid one another, overlapping over 2.9% of the basin against 3.8% under spatial independence. Soil moisture declines over most of the basin in the dry season; in the wet season, the areas where it now fluctuates least are not those drying fastest. The flagged areas are smaller than in an aggregated, sign-based analysis because each indicator is gated on a significant trend, the seasons are resolved separately, and every fraction is read against chance; this stricter accounting narrows the flagged area while sharpening its location, and should not be read as indicating a lesser degree of change where indicators do converge.