Tajikistan faces severe climatic and geographic challenges. With less than 10% of its territory suitable for irrigated agriculture, food security remains a persistent concern. Land degradation caused by deforestation and overgrazing further weakens ecosystems, while floods and landslides regularly threaten lives and assets – particularly in mountainous areas. Rural communities are highly vulnerable to these hazards, as well as to changing weather that directly affect their agriculture-based livelihoods, health and overall safety.
Despite this vulnerability, accurate weather forecasting, agronomic decision support and early warning systems remain limited. The country's complex topography exacerbates this challenge, requiring a dense observation network to capture local climatic conditions, yet Tajikistan's Hydrometeorological Service (Tajik Hydromet) has long been constrained by scarce resources. As a result, it must largely rely on sparse observations and global forecasting products.
To address this gap, Caritas Switzerland initiated a project in 2021, jointly funded with the Swiss Agency for Development and Cooperation, including a peer-to-peer partnership between Tajik Hydromet and MeteoSwiss. Running until 2025, the project strengthens local forecasting capacity and feeds Tajik observation data into the ECMWF assimilation, targeting improvements to both local and global forecasts.
Low-cost stations to provide local forecasts and services
The project prioritised the development of low-cost, easy-to-maintain weather stations that could be managed by Tajikistan Hydromet. Previous experience had shown that high-end imported equipment was often unsustainable due to costly maintenance. By contrast, the new stations can be serviced locally since they do not rely on vendor expertise, are built on an open-source design, and yet still rely on proven sensor technology. Their performance has been validated against MeteoSwiss' surface measurement network stations. Currently, around 320 stations across Tajikistan measure temperature, humidity and air pressure, while around 30 sites also capture precipitation, wind and solar radiation. This dense, locally managed network provides data quality sufficient for substantially improving local forecasting.
Thanks to this expanded observation base, Tajik Hydromet now benefits from downscaled local forecasts generated through a pragmatic processing chain developed by MeteoSwiss. Applying ensemble model output statistics including topographic predictors based on ECMWF's IFS ENS forecasts, the system now provides locally specific temperature forecasts for Tajikistan's varied terrain. The results consistently outperform available benchmarks, including forecasts from private providers, and are shared via a dedicated forecasting dashboard accessible to hydrometeorologists and civil defence authorities. Pilot programmes have demonstrated significant benefits such as optimised planting schedules, improved irrigation efficiency and timely warnings for extreme weather events.
Use of the data at ECMWF
Thanks to ECMWF's recent work to acquire and use additional surface observations (https://www.ecmwf.int/en/newsletter/176/news/increased-use-surface-observations), it has become possible to incorporate new SYNOP data from Tajikistan. Once the new stations were made available to 4D-Var through the acquisition system, the observation quality was monitored and assessed using statistics of their departures from short-range forecasts, know as the 'background'. When a station's data meet quality standards, it is added to a list of stations approved for assimilation during the monthly data selection procedure.
ECMWF has gradually incorporated more Tajik stations. Currently, 4D-Var assimilates surface pressure and 2-metre temperature observations from about 17% of the 'low-cost' stations, while the land surface assimilation system (LDAS) assimilates 2-metre temperature from around 50% of the stations. However, quality is not the only criterion for assimilation. Factors such as redundancy or significant elevation differences between the station and the corresponding model grid cell can also lead to exclusion.
Enhancing observational coverage in a data-sparse and heterogeneous region like Tajikistan makes these new stations exceptionally valuable for assimilation, as they significantly improve the accuracy of initial conditions for numerical weather prediction.