In this report, we describe the implementation/development and evaluation of 2 models for land cover and cropping system analysis underpinned by remote sensing data. The models provide information on crop types and water use, as well as other land covers relevant to environmental monitoring in northwest Bangladesh. This is instrumental to assess changes in water use and crop areal extent that may lead to groundwater level declines as a result of a combination of factors including rainfall declines and over-use, particularly in the Barind tract of northwest Bangladesh. Both models developed here relied on freely available remote sensing reflectance data archives via Google Earth Engine (GEE). Much of the pre-processing was performed in the GEE, thus facilitating the implementation of the models in this large geographical domain. The data underpinning both models were the Enhanced Vegetation Index (EVI) and the Global Vegetation Moisture Index (GVMI).

First, a monthly actual evapotranspiration (ETa) model based on the CMRSET (CSIRO MODIS ReScaled EvapoTranspiration) model, driven by MODIS reflectance data (500 m), was implemented from 2000 to 2016. CMRSET only requires multitemporal EVI and GVMI to derive a crop factor and meteorological data for its implementation. The model provides the first long-term (> 15 years) consistent ETa time-series for the entire northwest Bangladesh region, at spatial (500 m) and temporal resolutions (monthly) that are useful to assess ETa changes. The ETa estimates were evaluated against a crop factor model driven by areal crop data aggregated at the district scale (second administrative tier). Both models were similar both in terms of seasonality and magnitude, and the largest difference between models in any month assessed was around 5%. The models’ estimates being similar while having different rationales provide a level of confidence for them to be used as inputs to other hydrological models. Second, a machine learning supervised Random Forest (RF) model, driven by Landsat reflectance data (30 m)