A novel evapotranspiration based irrigation quantification method using the hydrological similar pixels algorithm

Globally, the agricultural sector is the largest consumer of fresh water, despite the increased efficiency in irrigation. Remote sensing is a valuable tool to monitor agricultural water use. In this study, we demonstrate a novel algorithm that computes high-resolution (10 m) remote sensing-based evapotranspiration (ET) data linked exclusively to irrigation, i.e. the incremental evapotranspiration (ETincr). The methodology compares the ET of irrigated agricultural pixels to the weighted average ET of a subset of natural Hydrological Similar Pixels (HSP). The hydrological similarity is based upon a set of features derived from DEM, soil texture, reference evapotranspiration, and precipitation datasets. The difference in ET between the subset of hydrological similar natural pixels and the corresponding irrigated agricultural pixel is explanatory for the amount of ET related to irrigation (ETincr). These results are then converted to the water use (m3) per agricultural field. The method is validated for three study areas in South Africa, Spain, and Australia. Comparing the monthly and seasonal water use estimates to water meter observations in the Hex Valley (South Africa), yielded an R2 of 0.751 and 0.780, respectively. For the Ebro (Spain) and Namoi (Australia) study areas, the accuracy of the monthly estimates decreased. In Australia, this was a result of the water meters being linked to local reservoirs, instead of the direct use of the irrigation systems. In total, 8 out of the 27 validation fields with monthly data showed a Kling-Gupta Efficiency (KGE) larger than 0.5, which highlights that the temporal variability can be captured well by the model. Generally, seasonal estimates showed to be most accurate, which makes the product suitable for comparison with seasonal water allocations and could help to monitor overconsumption in water-scarce environments.