Bias-corrected and spatially disaggregated seasonal forecasts: a long-term reference forecast product for the water sector in semi-arid regions

Abstract

Abstract. Seasonal forecasts have the potential to substantially improve water management particularly in water-scarce regions. However, global seasonal forecasts are usually not directly applicable as they are provided at coarse spatial resolutions of at best 36 km and suffer from model biases and drifts. In this study, we therefore apply a bias-correction and spatial-disaggregation (BCSD) approach to seasonal precipitation, temperature and radiation forecasts of the latest long-range seasonal forecasting system SEAS5 of the European Centre for Medium-Range Weather Forecasts (ECMWF). As reference we use data from the ERA5-Land offline land surface rerun of the latest ECMWF reanalysis ERA5. Thereby, we correct for model biases and drifts and improve the spatial resolution from 36 km to 0.1∘. This is performed for example over four predominately semi-arid study domains across the world, which include the river basins of the Karun (Iran), the São Francisco River (Brazil), the Tekeze–Atbara river and Blue Nile (Sudan, Ethiopia and Eritrea), and the Catamayo–Chira river (Ecuador and Peru). Compared against ERA5-Land, the bias-corrected and spatially disaggregated forecasts have a higher spatial resolution and show reduced biases and better agreement of spatial patterns than the raw forecasts as well as remarkably reduced lead-dependent drift effects. But our analysis also shows that computing monthly averages from daily bias-corrected forecasts particularly during periods with strong temporal climate gradients or heteroscedasticity can lead to remaining biases especially in the lowest- and highest-lead forecasts. Our SEAS5 BCSD forecasts cover the whole (re-)forecast period from 1981 to 2019 and include bias-corrected and spatially disaggregated daily and monthly ensemble forecasts for precipitation, average, minimum, and maximum temperature as well as for shortwave radiation from the issue date to the next 215 d and 6 months, respectively. This sums up to more than 100 000 forecasted days for each of the 25 (until the year 2016) and 51 (from the year 2017) ensemble members and each of the five analyzed variables. The full repository is made freely available to the public via the World Data Centre for Climate at https://doi.org/10.26050/WDCC/SaWaM_D01_SEAS5_BCSD (Domain D01, Karun Basin (Iran), Lorenz et al., 2020b), https://doi.org/10.26050/WDCC/SaWaM_D02_SEAS5_BCSD (Domain D02: São Francisco Basin (Brazil), Lorenz et al., 2020c), https://doi.org/10.26050/WDCC/SaWaM_D03_SEAS5_BCSD (Domain D03: basins of the Tekeze–Atbara and Blue Nile (Ethiopia, Eritrea, Sudan), Lorenz et al., 2020d), and https://doi.org/10.26050/WDCC/SaWaM_D04_SEAS5_BCSD (Domain D04: Catamayo–Chira Basin (Ecuador, Peru), Lorenz et al., 2020a). It is currently the first publicly available daily high-resolution seasonal forecast product that covers multiple regions and variables for such a long period. It hence provides a unique test bed for evaluating the performance of seasonal forecasts over semi-arid regions and as driving data for hydrological, ecosystem or climate impact models. Therefore, our forecasts provide a crucial contribution for the disaster preparedness and, finally, climate proofing of the regional water management in climatically sensitive regions.

Future changes in precipitation over the upper Yangtze River basin based on bias correction spatial downscaling of models from CMIP6

“…The effectiveness of BSCD method is based on the assumption that large-scale weather exhibits strong influence on small-scale weather (Maraun et al 2010). In our study, we assumed that the largescale (1.6°) does not interpret any small-scale (0.08°) variability, which is consistent with (Wood et al 2004, Wang and Chen 2014, Lorenz et al 2021. For example, Maurer and Hidalgo (2008) used BCSD to downscale GCMs from 1.4°to 0.125°.…”

Combined Modes of the Northern Stratosphere, Tropical Oceans, and East Asian Spring Rainfall: A Novel Method to Improve Seasonal Forecasts of Precipitation

“…The maximum rainfall anomaly correlation between the observations and corrected forecasts using COP‐EPP is around 0.4–0.5 at a lead time of 1–3 months (Khajehei et al., 2018; Ma et al., 2016). Other methods such as machine learning and spatial‐disaggregation approach can also improve the NMME rainfall forecasts (Lorenz et al., 2021; Xu et al., 2019). The machine learning approach can improve the maximum rainfall ACC to around 0.6 at some very local regions; in contrast, the rainfall ACC with error correction using the MCA/SVD method in this study exceed 0.6 (see Figure 4) in a broader region in East Asia.…”

Interrelations of vegetation growth and water scarcity in Iran revealed by satellite time series

“…Furthermore, air temperature, humidity and pressure are corrected to account for the altitude differences between ERA5 and ERA5-Land grids 107 . Since its release, ERA5-Land has been extensively compared to similar datasets or in situ data [108][109][110] while other studies used ERA5-Land as hydrometeorological reference data for bias-correcting seasonal forecasts 111 , as driving data for modeling photovoltaic power 112 , or for deriving agricultural drought indicators 113,114 . Within a similar context, Zandler et al (2020) 115 compared the performance of ERA5-Land for assessing NDVI anomalies across peripheral conservation areas of Central Asia and concluded that such reanalysis-based datasets outperform gauge-or satellite-based products and their combinations as they are highly variable and may not be applicable in the analyzed regions.…”