Atmospheric Phase Screen (APS) maps generation from Copernicus Sentinel-1 SAR data.

We have worked on :

  1. Refining and completing algorithm to generate APS maps from Copernicus Sentinel-1 SAR images
  2. Applying the implemented software to generate maps over the use cases of South Africa and Uganda. The SAR images selection has been based on the location of GNSS stations and the availability of GNSS observations, which is a critical factor for the SAR maps calibration phase.

South Africa Use Case

Uganda Use Case


Soil Moisture Mapping

The soil moisture retrieval algorithm developed by STARLAB  uses a Neural Network to predict time series of surface soil moisture maps from Sentinel-1 and -2 imagery. The Neural Network is trained with synthetic data generated from model simulations. A radiative transfer model of synthetic-aperture radar backscatter is calibrated with in situ surface soil moisture data over Ghana, from the Trans-African Hydro-Meteorological Observatory (TAHMO) stations, including the new stations installed in the framework of the present TWIGA project. The model is defined by a semi-empirical polarimetric backscattering model for bare soil surfaces (Oh) coupled with a Water Cloud model (WCM). The WCM takes into account the effects of vegetation on backscatter and therefore moisture estimates are available even when vegetation is present. The calibration is based on a hierarchical Bayesian regression to take into account the expected variations across land cover types and across in situ stations themselves. Every ~6 days, a 30m spatial resolution soil moisture raster map was generated for each new Sentinel-1  SAR backscatter image.

Illustration of four displays of a time series of surface soil moisture (m3/m3) maps, from 1 April 2019 to 1 August 2019 on a region in the north of Ghana (yellow square on the left panel). White represents non-data areas, non-observed or without retrievals, e.g. over water or built-up areas.


Monitoring crop growth and health conditions with citizen science and low-cost sensors

The rainy season in Sub-Saharan Africa coincides with regular cloud cover that results in satellite data gaps during long periods of the crop growing seasons. In addition, African smallholder farmers frequently practice intercropping that is not yet distinguishable by current satellite products, even at the rather high-resolution of commonly available satellite data (e.g. ~10m, Sentinel 2A). To sustain data collection for crop modeling the VegMon app has been developed to monitor crop properties. As smartphones do not feature the same sensor arrays that are available on-board satellites, vegetation metrics were selected that can be derived from regular photos.

The VegMon app is maturing, and farmers in Kenya, Ghana, and Uganda have been trained on using the app (Plates 4-7). The app has been updated to collect additional information such as planting day and harvest day. Data collection with the VegMon app continues in Ghana, Uganda and Kenya.

Algorithms have been designed to pre-process incoming images and store the extracted vegetation metrics for data transfer to the TWIGA data hub. The algorithms developed are written in the open source programming language R and are available through GitHub here

Calibration & Integration

Calibration standards for low-cost sensors and citizen science

TWIGA utilizes a suite of monitoring methods such as ground-based sensors (e.g. TAHMO), airborne sensors, and smartphone apps. The different in-field calibration methods range from installation guidelines for TAHMO stations, camera calibration tools as well as citizen science training, harmonization meetings to online calibration and post-processing tools

Optimization scheme for monitoring locations

Identifying the optimal number and location of monitoring stations is essential for better planning and operating monitoring networks. Especially for resource-scarce national monitoring services, an indication of where additional observations are most valuable is important. Within D5.5, the optimal number and location of rainfall stations were determined using rainfall products based on satellite data and modeling. Compared with existing station networks (Figure 42), this enables users to identify regions where additional stations are required.

Platform Development

Agile development and incremental deployment

This task focuses on the continuous development and enhancement of the platform, with incremental deployments. In the first 18 months of the project, a lot of work has been done on the server-side of the platform, such as preparing the data standards, creating the API, and implementing the authentication and authorization frameworks. This was all foundational work yet it had few visible results. From M19 onwards the focus has shifted towards extending and improving the portal (front-end) in the new micro-services architecture where users can interact with the TWIGA data sources.

The .NET framework used for the micro-services has been updated from version 2.2 to version 3.1 LTS, this is the long-term support version from Microsoft. The upgrade of the .NET framework has resulted in increased performance of the micro-services.