Irrigation accounts for ~70% of global freshwater withdrawals and ~90% of consumptive water use, driving myriad Earth system impacts. In this Review, we summarize how irrigation currently impacts key components of the Earth system. Estimates suggest that more than 3.6 million km² of currently irrigated land, with hot spots in the intensively cultivated US High Plains, California Central Valley, Indo-Gangetic Basin and northern China. Process-based models estimate that ~2,700 ± 540 km³ irrigation water is withdrawn globally each year, broadly consistent with country-reported values despite these estimates embedding substantial uncertainties. Expansive irrigation has modified surface energy balance and biogeochemical cycling. A shift from sensible to latent heat fluxes, and resulting land–atmosphere feedbacks, generally reduce regional growing season surface temperatures by ~1–3 °C. Irrigation can ameliorate temperature extremes in some regions, but conversely exacerbates moist heat stress. Modelled precipitation responses are more varied, with some intensive cropping regions exhibiting suppressed local precipitation but enhanced precipitation downstream owing to atmospheric circulation interactions. Additionally, irrigation could enhance cropland carbon uptake; however, it can also contribute to elevated methane fluxes in rice systems and mobilize nitrogen loading to groundwater. Cross-disciplinary, integrative research efforts can help advance understanding of these irrigation–Earth system interactions, and identify and reduce uncertainties, biases and limitations.

Groundwater depletion will reduce cropping intensity in India, and canal irrigation cannot fully substitute for this loss.

Remote sensing, or Earth Observation (EO), is increasingly used to understand Earth system dynamics and create continuous and categorical maps of biophysical properties and land cover, especially based on recent advances in machine learning (ML). ML models typically require large, spatially explicit training datasets to make accurate predictions. Training data (TD) are typically generated by digitizing polygons on high spatial-resolution imagery, by collecting in situ data, or by using pre-existing datasets. TD are often assumed to accurately represent the truth, but in practice almost always have error, stemming from (1) sample design, and (2) sample collection errors. The latter is particularly relevant for image-interpreted TD, an increasingly commonly used method due to its practicality and the increasing training sample size requirements of modern ML algorithms. TD errors can cause substantial errors in the maps created using ML algorithms, which may impact map use and interpretation. Despite these potential errors and their real-world consequences for map-based decisions, TD error is often not accounted for or reported in EO research. Here we review the current practices for collecting and handling TD. We identify the sources of TD error, and illustrate their impacts using several case studies representing different EO applications (infrastructure mapping, global surface flux estimates, and agricultural monitoring), and provide guidelines for minimizing and accounting for TD errors. To harmonize terminology, we distinguish TD from three other classes of data that should be used to create and assess ML models: training reference data, used to assess the quality of TD during data generation; validation data, used to iteratively improve models; and map reference data, used only for final accuracy assessment. We focus primarily on TD, but our advice is generally applicable to all four classes, and we ground our review in established best practices for map accuracy assessment literature. EO researchers should start by determining the tolerable levels of map error and appropriate error metrics. Next, TD error should be minimized during sample design by choosing a representative spatio-temporal collection strategy, by using spatially and temporally relevant imagery and ancillary data sources during TD creation, and by selecting a set of legend definitions supported by the data. Furthermore, TD error can be minimized during the collection of individual samples by using consensus-based collection strategies, by directly comparing interpreted training observations against expert-generated training reference data to derive TD error metrics, and by providing image interpreters with thorough application-specific training. We strongly advise that TD error is incorporated in model outputs, either directly in bias and variance estimates or, at a minimum, by documenting the sources and implications of error. TD should be fully documented and made available via an open TD repository, allowing others to replicate and assess its use. To guide researchers in this process, we propose three tiers of TD error accounting standards. Finally, we advise researchers to clearly communicate the magnitude and impacts of TD error on map outputs, with specific consideration given to the likely map audience.

Remote sensing offers a low-cost method for developing spatially continuous crop production statistics across large areas and through time. Nevertheless, it has been difficult to characterize the production of individual smallholder farms, given that the land-holding size in most areas of South Asia (<2 ha) is smaller than the spatial resolution of most freely available satellite imagery, like Landsat and MODIS. In addition, existing methods to map yield require field-level data to develop and parameterize predictive algorithms that translate satellite vegetation indices to yield, yet these data are costly or difficult to obtain in many smallholder systems. To overcome these challenges, this study explores two issues. First, we employ new high spatial (2 m) and temporal (bi-weekly) resolution micro-satellite SkySat data to map sowing dates and yields of smallholder wheat fields in Bihar, India in the 2014–2015 and 2015–2016 growing seasons. Second, we compare how well we predict sowing date and yield when using ground data, like crop cuts and self-reports, versus using crop models, which require no on-the-ground data, to develop and parameterize prediction models. Overall, sow dates were predicted well (R² = 0.41 in 2014–2015 and R² = 0.62 in 2015–2016), particularly when using models that were parameterized using self-report sow dates collected close to the time of planting and when using imagery that spanned the entire growing season. We were also able to map yields fairly well (R² = 0.27 in 2014–2015 and R² = 0.33 in 2015–2016), with crop cut parameterized models resulting in the highest accuracies. While less accurate, we were able to capture the large range in sow dates and yields across farms when using models parameterized with crop model data and these estimates were able to detect known relationships between management factors (e.g., sow date, fertilizer, and irrigation) and yield. While these results are specific to our study site in India, it is likely that the methods employed and the lessons learned are applicable to smallholder systems more generally across the globe. This is of particular interest given that similar high spatio-temporal resolution micro-satellite data will become increasingly available in the coming years.

Food security will be increasingly challenged by climate change, natural resource degradation, and population growth. Wheat yields, in particular, have already stagnated in many regions and will be further affected by warming temperatures. Despite these challenges, wheat yields can be increased by improving management practices in regions with existing yield gaps. To identify the magnitude and causes of current yield gaps in India, one of the largest wheat producers globally, we produced 30 meter resolution yield maps from 2001 to 2015 across the Indo-Gangetic Plains (IGP), the nation’s main wheat belt. Yield maps were derived using a new method that translates satellite vegetation indices to yield estimates using crop model simulations, bypassing the need for ground calibration data. This is one of the first attempts to apply this method to a smallholder agriculture system, where ground calibration data are rarely available. We find that yields can be increased by 11% on average and up to 32% in the eastern IGP by improving management to current best practices within a given district. Additionally, if current best practices from the highest-yielding state of Punjab are implemented in the eastern IGP, yields could increase by almost 110%. Considering the factors that most influence yields, later sow dates and warmer temperatures are most associated with low yields across the IGP. This suggests that strategies to reduce the negative effects of heat stress, like earlier sowing and planting heat-tolerant wheat varieties, are critical to increasing wheat yields in this globally-important agricultural region.