0.5° × 0.625° gridded daily soil temperature at four depths in Ethiopia during 1993–2023

Forest soils play a key role in land-atmosphere interactions through driving physical, chemical and ecological processes of the underlying surfaces and influencing energy and carbon exchange between land and atmosphere. Daily soil temperature is the critical environmental parameter to govern essential soil processes such as nutrient cycling, microbial activity, and water retention, directly affecting plant growth, seed germination, and root growth in forest ecosystems (Verma et al, 2011). Canopy density and light interception levels vary among forest ecosystems, determining the amount of light reaching the soil and thus influencing soil temperature (Saher et al, 2025). Since accurate gridded daily soil temperature data can help to improve numerical weather forecasts and optimize key infrastructure design in forest regions, addressing pressing environmental and societal challenges (Lipon et al, 2025), generation of high-resolution gridded daily soil temperature datasets can greatly support the development of sustainable forest ecosystems (Qi et al, 2019). However, direct measurement of daily soil temperatures in large-scale forest regions is always impractical due to the expensive and time-consuming processes for any developing countries like Ethiopia.Ethiopia is distinguished by its remarkable topographic diversity, varying from as much as 116 m below sea level in the Danakil depression to 4,620 m above sea level in the Ras Dashen, resulting in a variety of forest ecosystems (Tesfaye et al, 2017). Such a dramatic altitudinal variation shapes Ethiopia's forest ecosystem, resulting in three ecological zones: Dega, Woina Dega, and Kola.Every ecological zone has the similar soil type, vegetation covers, topography and agricultural management. The highland Dega zone, ranging from 2,300 to 3,200 meters above sea level, is characterized by cooler temperatures, fertile soils, and higher rainfall; the Woina Dega zone, located between 1,500 and 2,300 meters, experiences moderate temperatures and favorable conditions for agriculture; the Kola zone encompasses lowland, semi-arid regions ranging from 500 to 1,500 meters, characterized by high temperatures, lower rainfall, and heightened susceptibility to drought (Etana et al., 2020).The combination of diverse topography of Ethiopia and recent climatic warming results in complex soil temperature variations. Although soil temperature plays a pivotal role in shaping forest ecosystem dynamics and supports healthy and sustainable development of forests, soil temperatures in Ethiopia have been only measured to a very limited extent. Among all seventeen meteorological stations in the National Meteorological Agency (NMA) of Ethiopia, soil temperature is measured only in eleven stations (see Table 1). These meteorological stations include Arba Minch, Addis Ababa, Awassa, Gondar, Hosana, Jijiga, Jimma, Konso, Neghele, Debre Markos, and Dire Dawa, In this data report, we used various ensemble boosting learning techniques to develop an optimal soil temperature forecasting model through the link between observed meteorological factors and soil temperature at multiple depths (5 cm, 10 cm, 20 cm, and 50 cm) in these 11 meteorological stations. Following that, by inputting the observed meteorological data from all 17 meteorological stations into this optimal soil temperature model, we not only filled missing values in soil temperature measurement in these 11 meteorological stations, but also produced a soil temperature dataset for the remaining 6 stations without observed soil temperature data. Since all meteorological stations in Ethiopia are sparse and irregularly distributed, these data are not enough to produce a complete soil temperature distribution map for Ethiopia. In order to solve these issues, we utilized the NASA POWER gridded meteorological data at a 0.5° × 0.625° resolution as the input of our optimal soil temperature model and we generated a high-resolution gridded soil temperature dataset in the whole Ethiopia.Ensemble boosting techniques use decision trees as their base learners and sequentially combine them to improve forecast accuracy, where each new tree is trained to correct the errors made by the previous trees. In order to generate a high-resolution gridded soil temperature dataset for the whole of Ethiopia, we adopted four advanced ensemble boosting models: XGBoost, CatBoost, Gradient Boosting, and LightGBM to establish the link between observed meteorological factors and soil temperature at multiple depths. Gradient Boosting is the first widely-used boosting technique through the connection between stagewise additive expansions and steepest-descent minimization, producing competitive, highly robust, interpretable procedures for regression issues (Friedman, 2001). XGBoost, as an improvement of Gradient Boosting, is known for its scalability and regularization techniques for handling sparse data and its implementation in parallel and distributed computing (Chen and Guestrin, 2016). CatBoost incorporates ordered boosting, a permutation-driven alternative to the classic algorithm, and an innovative algorithm for processing categorical features The performance of boosting models on soil temperature predictions can be significantly influenced by different input scenarios and time windows. For this study, the input of soil temperature prediction models should be meteorological factors which impact significantly on soil temperatures, so we selected five meteorological factors, including daily minimum air temperature (MIN), daily maximum air temperature (MAX), daily mean air temperature (MEAN), daily rainfall (RAIN), and daily humidity (HUM) as model inputs. Four kinds of input scenarios were tested in the performance of soil temperature prediction: the MEAN-MAX-MIN combination scenario is to consider only air temperature dynamics; the MEAN-MAX-MIN-HUM combination scenario is to further add humidity effects on evaporative cooling; the MEAN-MAX-MIN-RAIN combination scenario is to consider temperature-rainfall interactions; and the MEAN-MAX-MIN-HUM-RAIN scenario is to consider all meteorological factors. These input scenarios can make the boosting models simulate physical processes that influence soil thermal properties. We also considered five types of time window: one-day, two-day, three-day, four-day, and five-day.From January 1, 2010, to December 31, 2023, all meteorological factors were measured at 17 meteorological stations of the National Meteorological Agency (NMA) of Ethiopia. However, soil temperature was only measured in 11 stations. Our method for generating a high-resolution gridded soil temperature dataset in the whole Ethiopia was as follows:Step 1: Establishment of an optimal soil temperature forecast modelFor 11 meteorological stations where both climate factors and soil temperature were measured simultaneously, after removing periods with missing data, we divided the remaining dataset using an 80-20 split: 80% for training the models and 20% for testing. We employed four ensemble boosting models (XGBoost, CatBoost, Gradient Boosting, and LightGBM) with five types of time window and four input scenarios to stimulate soil temperature at different depths. By comparing different combinations of boosting models, time windows and input scenarios, we identified the optimal combination of ensemble learning model, input scenario, and time window size for forecasting daily soil temperature at different depths for each of 11 meteorological stations.Step 2: Filling missing soil temperature data in 11 meteorological stationsUsing the optimal soil forecast model established in Step 1 for the 11 meteorological stations with partial soil temperature measurements, we filled missing soil temperature data during periods where meteorological factors were measured but soil temperature was not. We input the observed meteorological factors into the optimal soil temperature forecast model to generate estimates of soil temperature at different depths for the missing periods, thereby creating complete daily soil temperature time series for all 11 stations. Step 4: Generation of high-resolution gridded soil temperature data for the whole of Ethiopia.Since all meteorological stations in Ethiopia are sparse and irregular distributed, these data are not enough to produce a complete soil temperature distribution map for Ethiopia. Therefore, we adopted NASA POWER gridded meteorological data at 0.5° × 0.625° resolution as model input; these datasets were downloaded from https://power.larc.nasa.gov/. Correlation-based matching was performed between meteorological factors at each NASA grid cell and all 11 stations with measured soil temperatures. For a given grid cell P, when meteorological data in Station Q achieved the highest correlation coefficients with those in grid cell P, this demonstrated that grid cell P and Station Q had the strongest meteorological similarity and was located in the same ecological zone.Therefore, we drove the optimal soil temperature forecast model for Station Q by inputting NASA POWER meteorological data at grid cell P and then generated daily soil temperature estimates at multiple depths at grid cell P. When we repeated the same process for all grid cells at 0.5° × 0.625° resolution, we produced a high-resolution gridded soil temperature map distribution at different depth for Ethiopia.Soil temperature data were measured twice a day at 5 cm, 10 cm, 20 cm, and 50 cm in 11 Based on these optimal soil temperature models, using the method described in Step 2, we filled the missing daily soil temperature records at 5 cm, 10 cm, 20 cm, and 50 cm depths for the 11 meteorological stations. Using the correlation-based matching and ensemble boosting approach inmeteorologicalStep 3, we generated daily soil temperature data for the remaining six meteorological stations without soil measurement.Based on the correlation-based matching of meteorological conditions between each NASA grid cells and 11 meteorological stations in Ethiopia, we input NASA POWER meteorological data (0.5° × 0.625° resolution) into optimal soil temperature prediction models and produced daily soil temperature estimates at 5 cm, 10 cm, 20 cm, and 50 cm depths. Daily soil temperatures are key drivers of soil ecological processes and they affect healthy growth of forest ecosystems. Owing to limited soil measurements, Ethiopia has urgently needed to capture daily soil temperatures at multiple depths in order to improve its environmental monitoring and enhance climate resilience in its forest ecosystems. In this data report, we utilized correlation-based matching and boosting-based learning to generate high-resolution gridded daily soil temperature datasets at 5 cm, 10 cm, 20 cm, and 50 cm depths throughout the whole of Ethiopia. These insights will be crucial for addressing challenges related to forest conservation and climate resilience, making our gridded daily soil temperature dataset valuable for research and informed decisionmaking in Ethiopia.