Previous studies into the relation between human consumption and indirect water resources use have unveiled the remote connections in virtual water (VW) trade networks, which show how communities externalize their water footprint (WF) to places far beyond their own region, but little has been done to understand variability in time. This study quantifies the effect of inter-annual variability of consumption, production, trade and climate on WF and VW trade, using China over the period 1978-2008 as a case study. Evapotranspiration, crop yields and green and blue WFs of crops are estimated at a 5 × 5 arc-minute resolution for 22 crops, for each year in the study period, thus accounting for climate variability. The results show that crop yield improvements during the study period helped to reduce the national average WF of crop consumption per capita by 23%, with a decreasing contribution to the total from cereals and increasing contribution from oil crops. The total consumptive WFs of national crop consumption and crop production, however, grew by 6% and 7%, respectively. By 2008, 28% of total water consumption in crop fields in China served the production of crops for export to other regions and, on average, 35% of the crop-related WF of a Chinese consumer was outside its own province. Historically, the net VW within China was from the water-rich South to the water-scarce North, but intensifying North-to-South crop trade reversed the net VW flow since 2000, which amounted 6% of North's WF of crop production in 2008. South China thus gradually became dependent on food supply from the water-scarce North. Besides, during the whole study period, China's domestic inter-regional VW flows went dominantly from areas with a relatively large to areas with a relatively small blue WF per unit of crop, which in 2008 resulted in a trade-related blue water loss of 7% of the national total blue WF of crop production. The case of China shows that domestic trade, as governed by economics and governmental policies rather than by regional differences in water endowments, determines inter-regional water dependencies and may worsen rather than relieve the water scarcity in a country.

Water Footprint Assessment is a fast-growing field of research, but as yet little attention has been paid to the uncertainties involved. This study investigates the sensitivity of and uncertainty in crop water footprint (in m3 t−1) estimates related to uncertainties in important input variables. The study focuses on the green (from rainfall) and blue (from irrigation) water footprint of producing maize, soybean, rice, and wheat at the scale of the Yellow River basin in the period 1996–2005. A grid-based daily water balance model at a 5 by 5 arcmin resolution was applied to compute green and blue water footprints of the four crops in the Yellow River basin in the period considered. The one-at-a-time method was carried out to analyse the sensitivity of the crop water footprint to fractional changes of seven individual input variables and parameters: precipitation (PR), reference evapotranspiration (ET0), crop coefficient (Kc), crop calendar (planting date with constant growing degree days), soil water content at field capacity (Smax), yield response factor (Ky) and maximum yield (Ym). Uncertainties in crop water footprint estimates related to uncertainties in four key input variables: PR, ET0, Kc, and crop calendar were quantified through Monte Carlo simulations. The results show that the sensitivities and uncertainties differ across crop types. In general, the water footprint of crops is most sensitive to ET0 and Kc, followed by the crop calendar. Blue water footprints were more sensitive to input variability than green water footprints. The smaller the annual blue water footprint is, the higher its sensitivity to changes in PR, ET0, and Kc. The uncertainties in the total water footprint of a crop due to combined uncertainties in climatic inputs (PR and ET0) were about ±20% (at 95% confidence interval). The effect of uncertainties in ET0was dominant compared to that of PR. The uncertainties in the total water footprint of a crop as a result of combined key input uncertainties were on average ±30% (at 95% confidence level).