No abstract

Abstract. Soil moisture is highly variable in space and time, and deficits (i.e., droughts) play an important role in modulating crop yields. Limited hydroclimate and yield data, however, hamper drought impact monitoring and assessment at the farm field scale. This study demonstrates the potential of using field-scale soil moisture simulations to support high-resolution agricultural yield prediction and drought monitoring at the smallholder farm field scale. We present a multiscale modeling approach that combines HydroBlocks – a physically based hyper-resolution land surface model (LSM) – with machine learning. We used HydroBlocks to simulate root zone soil moisture and soil temperature in Zambia at 3 h 30 m resolution. These simulations, along with remotely sensed vegetation indices, meteorological data, and descriptors of the physical landscape (related to topography, land cover, and soils) were combined with district-level maize data to train a random forest (RF) model to predict maize yields at district and field scales (250 m). Our model predicted yields with an average testing coefficient of determination (R2) of 0.57 and mean absolute error (MAE) of 310 kg ha−1 using year-based cross-validation. Our predicted maize losses due to the 2015–2016 El Niño drought agreed well with losses reported by the Food and Agriculture Organization (FAO). Our results reveal that soil moisture is the strongest and most reliable predictor of maize yield, driving its spatial and temporal variability. Soil moisture was also a more effective indicator of drought impacts on crops than precipitation, soil and air temperatures, and remotely sensed normalized difference vegetation index (NDVI)-based drought indices. This study demonstrates how field-scale modeling can help bridge the spatial-scale gap between drought monitoring and agricultural impacts.

Abstract. Soil moisture is highly variable in space, and its deficits (i.e. droughts) plays an important role in modulating crop yields and its variability across landscapes. Limited hydroclimate and yield data, however, hampers drought impact monitoring and assessment at the farmer field-scale. This study demonstrates the potential of field-scale soil moisture simulations to advance high-resolution agricultural yield prediction and drought monitoring at the smallholder farm field-scale. We present a multi-scale modeling approach that combines HydroBlocks, a physically-based hyper-resolution Land Surface Model (LSM), and machine learning. We applied HydroBlocks to simulate root zone soil moisture and soil temperature in Zambia at 3-hourly 30-m resolution. These simulations along with remotely sensed vegetation indices, meteorological conditions, and data describing the physical properties of the landscape (topography, land cover, soil properties) were combined with district-level maize data to train a random forest model (RF) to predict maize yields at the district- and field-scale (250-m) levels. Our model predicted yields with a coefficient of variation (R2) of 0.61, Mean Absolute Error (MAE) of 349 kg ha−1, and mean normalized error of 22 %. We captured maize losses due to the 2015/2016 El Niño drought at similar levels to losses reported by the Food and Agriculture Organization (FAO). Our results revealed that soil moisture is the strongest and most reliable predictor of maize yield, driving its spatial and temporal variability. Consequently, soil moisture was also the most effective indicator of drought impacts in crops when compared with precipitation, soil and air temperatures, and remotely-sensed NDVI-based drought indices. By combining field-scale root zone soil moisture estimates with observed maize yield data, this research demonstrates how field-scale modeling can help bridge the spatial scale discontinuity gap between drought monitoring and agricultural impacts.

Cropland expansion is expected to increase across sub-Saharan African (SSA) countries in the next thirty years to meet growing food needs across the continent. These land transformations will have cascading social and ecological impacts that can be monitored using novel Earth observation techniques that produce datasets complementary to national cropland surveys. In this study, we present a flexible Bayesian data synthesis workflow on Google Earth Engine (GEE) that can be used to fuse optical and synthetic aperture radar data and demonstrate its ability to track agricultural change at national scales. We adapted the previously developed Bayesian Updating of Land Cover (Unsupervised) algorithm (BULC-U) by integrating a shapelet and slope thresholding algorithm to identify the locations and dates of cropland expansion and implemented a tiling scheme to allow the processing of large volumes of imagery. We apply this approach to map annual cropland change from 2000 to 2015 for Zambia (750,000 km2), a country that is experiencing rapid growth in agricultural land. We applied our cropland mapping approach to a time series of unsupervised classifications developed from Landsat 5, 7, 8, Sentinel-1, and ALOS PALSAR within 1476 tiles covering Zambia. The annual cropland changes maps reveal active cropland expansion between 2000 to 2015 in Zambia, especially in the Southern, Central, and Eastern provinces. Our accuracy assessment estimates that we have identified 27.5% to 69.6% of the total cropland expansion from 2000 to 2015 in Zambia (commission errors between 6.1% to 37.6%), depending on the slope threshold. Our results demonstrate the usefulness of Bayesian data fusion and shapelet, slope-based thresholding to synthesize optical and synthetic aperture radar for monitoring agricultural changes in situations where training data are scarce. In addition, the annual cropland maps provide one of the first spatially continuous, annually incremented accounts of cropland growth in this region. Our flexible, cloud-based workflow using GEE enables multi-sensor, national-scale agricultural change monitoring at low cost for users.

No abstract