Remote sensing-based drought severity modeling and mapping using multiscale intelligence methods

Drought as a natural disaster is one of the human's ecological, hydrological, agricultural, and economic concerns. In this study, multiscale intelligence methods were proposed for drought severity detection and mapping in the northwest part of Iran for the years of 2007 to 2020. In the modeling process two scenarios were considered and in-situ and remote sensing datasets were adopted with two machine learning models namely M5 Pruning tree (M5P) and Random Forest (RF). In the first scenario, the in-situ datasets including the precipitation, relative humidity, evaporation, and temperature were used as inputs of the intelligence models to assess drought severity in terms of the Standardized Precipitation Index. In the second scenario, the SM2RAIN-ASCAT precipitation product and Normalized Difference Vegetation Index (NDVI) and Land Surface Temperature (LST) products of the MODIS were considered as inputs. During the drought severity modeling process, the input time series were first broken down into several subseries via the Variational Mode Decomposition; then, the most effective subseries were selected and imposed into the M5P and RF as inputs. Also, the potential of the relatively new TemperatureVegetation Water Stress Index (T-VWSI), which has developed based on the NDVI and LST, was assessed in drought severity monitoring. The results proved the appropriate efficiency of the proposed multiscale methods in effectively detecting drought severity. Also, it was observed that the T-VWSI could be successfully used for detecting drought occurrences in areas without meteorological datasets.