Warming climate is altering climate-control mechanisms of terrestrial carbon sequestration1,2,3,4. The direct observational evidence5 provided by a global network (FLUXNET) of continuous in situ measurements of land-atmosphere exchanges of CO2, water vapour and energy across biomes and continents, indicates that terrestrial CO2 fluxes are: (1) strongly limited by mean annual temperature (T) of less than16°C at mid- and high-latitudes; (2) strongly limited by dryness at mid- and low-latitudes; and (3) co-limited by both temperature and dryness around the mid-latitudinal belt (45°N). The sensitivity of terrestrial CO2 fluxes to T breaks down at ~16°C, a threshold value above which no further increase of CO2 uptake with increasing temperature was observed and dryness influence overrules temperature influence. Here, we examine a hypothesis that the threshold-latitudinal belt at which T is 16°C is shifting poleward as the Earth's surface warms and hence the areas of dryness-control of terrestrial CO2 fluxes (T > 16°C) is expanding. We use global land temperature data6,7 (1948–2012) to test this hypothesis and examine the potential consequences of warming-induced extension of the dryness-controlled area on climate change.

We calculated land area where T is higher than or equal to 16°C for each year during the period from 1948 to 2012 using mean monthly surface temperature data (0.5° × 0.5° resolution) from the National Centers for Environmental Prediction-National Center for Atmospheric Research (NCEP/NCAR) reanalysis dataset. We refer to land area with T ≥ 16°C as dryness-controlled areas where terrestrial CO2 fluxes are limited by dryness and not by temperature5. A pronounced increase in the dryness-controlled area occurred following a slight drop before 1976, mirroring the variation with land warming (Fig. 1a). About 90% of the variance in the extension of the dryness-controlled area was accounted for by land warming (R2 = 0.90, p < 0.0001, see Fig. 1b).

Figure 1
figure 1

Relationship between dryness-control area (%)and land temperature (°C) (1948–2012): (a) the evolution of dryness-control area (blue line) and land average temperature (red line); and (b) correlation between annual dryness-control area and annual land-average temperature (R2 = 0.90, P < 0.0001).

The dryness-control area refers to the total area of regions where mean annual temperature was higher than or equal to 16°C and terrestrial CO2 fluxes are controlled by dryness rather than temperature based on the direct observational evidence provided by a global monitoring network5. The annual dryness-control area and annual land surface temperature during the period from 1948 to 2012 were derived from mean monthly temperature data at surface (0.5° × 0.5° resolution) from the National Centers for Environmental Prediction-National Center for Atmospheric Research (NCEP/NCAR) reanalysis data set6,7. The black lines indicate the trends of dryness-control area that was similar to that of land-average temperature (omitted): a slight drop between 1948–1975 and then a striking increase during 1976–2012. The striking increase in temperature is a direct result of increased greenhouse gases in the atmosphere31. The red arrows in (a) indicate El Niño years with oceanic Niño index (ONI) greater than +1.0, while blue arrows in (a) indicate La Nina with ONI less than −1.0 ( 70% El Niño years were consistent with warmer years, while 80% La Nina years were consistent with cooler years.

We assume that net ecosystem-atmosphere exchanges of CO2 (NEE) in the areas close to the cold side (T < 16°C) of the shifted boundary are controlled by both temperature and dryness5, written as NEEB that is predicted by equation (1) (see Methods Summary). For NEE in the area on the warm side (T > 16°C) of the shifted boundary, it is written as NEED, that is determined by dryness alone through equation (2). How will increasing T affect NEE in area that shifts from control by both T and dryness (T < 16°C) to control by dryness alone (T > 16°C)? We estimated the difference between NEEB and NEED by applying a NEE model (see Methods Summary) that was derived from datasets collected by a worldwide, tower-based, observational network5 to the shifted area. The climate data used in the model were T and dyness averaged over the period 1948–2010 with 0.5° × 0.5° spatial resolution in the shifted area. Dryness was calculated from the monthly datasets of net incoming short-wave radiation and net long-wave outgoing radiation of the NCEP/NCAR reanalysis8 and the gridded monthly terrestrial precipitation datasets9. This data-driven estimate indicates that CO2 transfer from the atmosphere to the biosphere is reduced by 27% in the shifted area where T changed from less than to greater than16°C. Qualitatively, the model prediction reveals a positive feedback mechanism: climate warming extends the dryness-control area, which reduces CO2 transfer from the atmosphere to biosphere. Thus, because the atmopheric CO2 concentration will increase at a greater rate, the climate will warm at an accelerating rate due to the positive feedback. If the global area under dryness-control (Fig. 1a) continues to increase at only the same rate as occurred over the preceeding half-century, the warming-induced dryness-controlled area will double by 2050.

With climate warming much of Earth's land has been moderately drying since 1976, averged over all land areas, based on annual Palmer Drought Severity Index (PDSI) estimates (Fig. 2a) derived from the monthly self-calibrated PDSI data 0.5° × 0.5° resolution over the spatial range from 60°S to 70°N10,11. Our analysis finds that the drying trend in the shifted area was strongest and in land areas where T is above 16°C was second strongest (Fig. 2b). The land area where T > 16°C encompasses low latitudes in the northern hemisphere, most of Africa, Middle and South America, Australia, South- and Southeast Asia (Fig. 3). In these regions, tower-based FLUXNET observations12 document that at ground-level these large land areas indeed are drying up and this is confirmed by remote sensing data13. This drying is attributed to increased evaporation and evapotranspiration due to warming. If the trend of drying up over the large land area where T > 16°C continues a strong positive feedback on warming is suggested because of reduced CO2 transfer from the atmosphere to land (expansion of the brown areas in Fig. 3) via NEE that is limited substantially by water availability (see Fig. 2b in Ref. 5), thus inducing additional warming. In contrast, in the land area where mean annual temperature is below 16°C (green area in Figure 3) a trend toward greater wetness has been observed with climate warming (Fig. 2).

Figure 2
figure 2

The Palmer Drought Severity Index (PDSI).

(a) Links between PDSI and ENSO events.The red curve shows PDSI for the area with temperature above 16°C; the green curve for the area with temperature below 16°C; the grey curve for the whole area of the land; and the blue curve for the shifted area from below 16°C to above 16°C during 1948–2012. The red arrows indicate El Niño years with oceanic Niño index (ONI) greater than +1.0, while blue arrows indicate La Niña with ONI less than −1.0 ( (b) The trends of PDSI. The filled circles are five-year moving average of the PDSI data shown in (a). Mean annual land surface temperature during the period from 1948 to 2012 was derived from mean monthly temperature data at surface (0.5° × 0.5° resolution) from the National Centers for Environmental Prediction-National Center for Atmospheric Research (NCEP/NCAR) reanalysis data set6,7. Annual PDSI data were derived from the monthly self-calibrated PDSI data (0.5° × 0.5° resolution, spatial range from 60°S to 70°N,,12. Drought classification by PDSI are: [−0.49, +0.49] → normal; [−0.5, −0.99] → incipient dry spell; [−1.0, −1.99] → mild drought; [−2.0, −2.99] → moderate drought; [−3.0, −3.99] → severe drought10. The PDSI behaviours to the ENSO events were different between: (1) the area above 16°C (red curve in (a)), 90% El Nino years were dryer, while 70% La Nina years were wetter; and (2) the area below 16°C (green curve (a)), 50% El Nino years were wetter, while 40% La Nina years were dryer (see Table 1).

Figure 3
figure 3

Map of mean annual temperature (1948–2012): below 16°C in light green regions; above 16°C in light red regions and shift from below 16°C to above 16°C in purple regions.

The map was produced based on the NCEP/NCAR ranalysis data6,7 ( and created using Matlab. The information of vegetation distribution and precipitation (P) are summarized in boxes for the shifted areas (purple color) in each of seven framed regions marked on the map. The vegetation is coded according to the IGBP classification: GRA, grassland; CRO, cropland; MF, mixed forest; OSH, open shrubland; WSA, woody savanna; SAV, savanna; EBF, evergreen broad-leaf forest; and BAR, Barren or sparsely vegetated.

Two large areas between the cold (<16°C, green color in Fig. 3) and warm zones (>16°C, brown color in Fig. 3) have different performances during El Niño/Southern Oscillation (ENSO) events (Fig. 2a, Table 1). In this analysis we included El Niño years with an oceanic Niño index (ONI) greater than +1.0 during the period between 1948 and 2012 and La Niña years with ONI less than −1.0 (Table 1). Half of the El Niño years were consistent with the cold phases (dips of temperature curve) of the cold part of the land (CPL, green area in Fig. 3), while 70% of the La Niña years were consistent with the warm phases (peaks of temperature curve) of the CPL (Table 1). The CPL warm/cold phases appeared to be opposite of the warm/cold phases of the ENSO cycle. However, the global land area followed the warm/cold phases of the ENSO cycle very well. The CPL wet/dry phases appeared different from that of the global land area (Table 1). However, the wet/dry phases of the warm part of the land (WPL) were very consistent with that of the ENSO cycle, i. e. 90% El Niño years were in the dry phases, while 70% La Niña years were in the WPL wet phases (Fig. 2a). We could not find a better relationship of the WPL warm/cold phases with the ENSO cycle. However, if we assume that WPL temperature responses to the ENSO cycle lag by a year, 90% of El Nino years coincided with the WPL warm phases, while 80% of La Niña years coincided with the WPL cold phases. These fascinating coincidences, that became obvious after lagging the data by a year, can be understood at least theoretically in the following way. In the WPL wet phases, a much larger fraction of net radiation is used for evapotranspiration as latent heat and hence potential warming is reduced, while in the WPL dry phases, comparitively less net radiation is used as latent heat, so the temperature is increased. This energy budget adjustment may need about a year to reach equilibrium for about half of Earth's land. Temperature responses to the ENSO cycle of the shifted area (purple color in Fig. 3) were similar to the responses of the total global land area because the temperature of the shifted area is close to the land-average temperature. However, the pattern with 60% of El Niño years being wetter while 30% La Niña years were dryer for the shifted area is coincident with the typical precipitation patterns of the ENSO cycle reported by NOAA14.

Table 1 Temperature, PDSI and ENSO events from different areas classified in Figure 3

The shifted areas are transitional zones where not only is the climate-control mechanism of NEE switched as discussed above, but also meteorological conditions are more variable and vegetation is highly vunerable to climate changes and weather extremes. Dominant vegetations in the shifted regions (purple color in Fig. 3) are open shrublands (25%), croplands (22%), grasslands (7%) and desert (13%) (see Supplementary Table 1). Except for the shifted areas in southeastern China (box 4 in Fig. 3) and southeastern United States (box 1 in Fig. 3), most shifted areas are arid and semi-arid land with typical vegetation of open shrublands and grasslands. The annual NEE of these ecosystems is quite sensitive to climate conditions of low precipitation and high evapotranspiration rates15,16,17,18.

The locations of shifted areas in the northern hemisphere are coincident with the descending branch of the Hadley cell (HC) and are consequently associated with low precipitation and high evaporation rates19. Several lines of evidence indicate that the HC has intensified and expanded poleward over the past three decades as a consequence of climate warming20,21,22 and that the HC expansion in the northern hemisphere is stronger than in southern hemisphere23. The contemporaneous poleward shift of both the HC and the WPL (significant since late 1970s) and location of the WPL with respect to the HC (nothern desending branch of the HC), strongly suggests that the WPL migration poleward is driven by global warming. The drying trend of the shifted area (Fig. 2a–b) should be expected to result in vegetation cover shift, with decreased biodiversity and desertification. A line of evidence from remote sensing imagery indicates that drying is accelerating the degradation of vulnerable shrublands in some semiarid Mediterranean area24,25.

Division of the land into the WPL and CPL by threshold value (16°C) of annual mean temperature based on 64 years (1948–2012) climate data brings new insights into the warming of Earth's surface. The two parts of the land behave almost opposite to each other in the phases of the ENSO cycle and differ in climate control mechanisms of carbon sequestration5,26,27. The WPL has expanded poleward significantly (Fig. 1) and has become dryer (Fig. 2) in the past four decades. The trend of warming-induced drying of the WPL, by reducing NEE thereby reducing withdraw of CO2 from the atmosphere, contributes a positive feedback on global warming. Furthermore, as lands are shifted from CPL to WPL becoming more arid and subject to desertification, they also release soil carbon adding additional CO2 to the atmosphre. It is estimated that 19–29 Pg of carbon were added to the atmosphere from vegetation and soil carbon pools globally by desertification28. The frontal boundary (or the shifted area) of the WPL has been transformed by global warming into more vunerable regions where weather gradients are stronger (Fig. 2), ecosystems are more sensitive to even slight increases in water deficit (Fig. 3)25, crop yield is reduced by extreme heat waves29 and vegitated land cover and pastroral population are reduced. For instance, in Australia, where wide areas are becoming not suitable for sheep breeding due to reduced precipitation and increased soil salinity. An expansion of the global network30 monitoring NEE to target the identified shifted areas would provide data that could improve our ability both to model these regions as they undergo further transitions and to assess the likely impacts on climate as a consequence of altered NEE and increased soil aridity. The present work raises the following two questions: (1) what atmospheric circulation mechanisms support the hypothesis of a year time lag between the WPL temperature response and the ENSO water phases; and (2) is the synergistic poleward expansion of the frontal boundary of the WPL with the HC a long-term or a short-term behavior and what are the consequences of this synergy for global NEE and for the rate of change in atmospheric CO2?


Details of calculating land temperature, precipitation, net radiation and PDSI are given in the Supplementary Information. Here we summarize the method used to estimate NEE difference induced by the switch of climate control from the CPL to WPL. For the case of the shifted area in the CPL (purple in Fig. 3), its NEEB is limited by both temperature (T) and dryness (D) and is estimated using a bivariate-nonlinear regression model

where D is defined as Rn/(λP), Rn is mean annual net radiation MJ m−2 yr−1, P is mean annual precipitation mm yr−1 and λ ( = 2.5 MJ kg−1) is the enthalpy of vaporization. For another case of the shifted area in the WPL where NEED is limited by dryness along (T > 16°C), we use the regression model of D-limited group in Ref.5,

All the regression coefficients in equations (1)(2) are estimated from the published eddy-covariance data (see supplementary Table S1 in Ref. 5).