A remote sensing and artificial neural network-based integrated agricultural drought index: Index development and applications

Abstract Reliable drought monitoring is critical for evaluating drought risk and reducing potential agricultural losses. However, many existing drought indices developed by a single indicator may not properly describe the complex features of agricultural drought. Here, we propose a new drought index—the integrated agricultural drought index (IDI), which describes the relationship between multiple variables and agricultural drought conditions. The derivation of IDI is based on the remote sensing data and the back-propagation (BP) neural network, capable of identifying the non-stationary relationship of drought conditions. Development of IDI involves the following meteo-hydrological variables: precipitation, land surface temperature (LST), normalized difference vegetation index (NDVI), soil water capacity, and elevation. The lagging effect of NDVI with respect to precipitation and LST changes can also be captured by the proposed IDI. Our results indicate that the IDI based on a machine learning method can relax the assumption used in many existing indices that the input and output data are linearly correlated. Results also demonstrate that the IDI is close to SPI-3 and SPEI-3 in a case study of the North China Plain (NCP). Moreover, we found the drought condition in the NCP area is highly correlated with 10 cm depth soil moisture at 8 agrometeorological stations and the newly developed IDI can effectively monitor the drought in terms of onset, duration, extent, and intensity of a drought episode. Additionally, the IDI provides spatial information about root zone soil moisture that can facilitate agricultural drought monitoring. The proposed framework of IDI can also be applied in other regions of the world for agriculture management.