Sub-Saharian African agriculture is perceived as a high-risk activity as is mainly rainfall-dependent and climate variability is increasing.
Within the framework of the EU Horizon2020 funded TWIGA project, together Starlab Space and TAHMO, we intend to contribute in reducing uncertainties related to climatic information by providing improved estimates of surface soil moisture. Accurate risk related information is key to foster investment into the agricultural sector, facilitate the access to insurance products and ultimately promote food security.
Soil moisture is one of the key soil characteristics as it is closely related with crop vegetation development. Current available soil moisture products are generated and calibrated for global or regional scale applications and at spatial resolutions (~1km2) usually too coarse for monitoring common small farms.
We present an approach to generate a soil moisture maps at 30 m spatial resolution, every 3-6 days, by using time series of Sentinel-1 synthetic-aperture radar backscatter and Sentinel-2 optical reflectance data, and in-situ data. Soil moisture is estimated with a neural network trained with synthetical data generated from simulations from a radiative transfer model. Neural network are useful to ensure computationally affordable estimates over large areas while keeping the accuracy high. The radio transfer model is calibrated with in-situ soil moisture measurement in order to capture the local relationships between soil moisture and radar and optical data, and soil and vegetation characteristics.
We performed an independent validation with leave-one-out cross sampling approach. The observed good accuracy, particularly in homogeneous areas, encourages us to undertake further work in order to be able to map soil moisture on any region of the Globe, by integrating the available global in-situ station networks.
Bellow you can see an example of our soil moisture maps over a 80x80 km2 square in the north of Ghana for four months.
