Semi-arid North-central Namibia has high potential for rice cultivation because large seasonal wetlands (oshana) form during the rainy season. Evaluating the distribution of surface water would reveal the area potentially suitable for rice cultivation. In this study, we detected the distribution of surface water with high spatial and temporal resolution by using two types of complementary satellite data: MODIS (MODerate-resolution Imaging Spectroradiometer) and AMSR-E (Advanced Microwave Scanning Radiometer–Earth Observing System), using AMSR2 after AMSR-E became unavailable. We combined the modified normalized-difference water index (MNDWI) from the MODIS data with the normalized-difference polarization index (NDPI) from the AMSR-E and AMSR2 data to determine the area of surface water. We developed a simple gap-filling method (“database unmixing”) with the two indices, thereby providing daily 500-m-resolution MNDWI maps of north-central Namibia regardless of whether the sky was clear. Moreover, through receiver-operator characteristics (ROC) analysis, we determined the threshold MNDWI (−0.316) for wetlands. Using ROC analysis, MNDWI had moderate performance (the area under the ROC curve was 0.747), and the recognition error for seasonal wetlands and dry land was 21.2%. The threshold MNDWI let us calculate probability of water presence (PWP) maps for the rainy season and the whole year. The PWP maps revealed the total area potentially suitable for rice cultivation: 1255 km² (1.6% of the study area).

Evaluating land surface models (LSMs) using available observations is important for understanding the potential and limitations of current Earth system models in simulating water- and carbon-related variables. To reveal the error sources of a LSM, five essential climate variables have been evaluated in this paper (i.e., surface soil moisture, evapotranspiration, leaf area index, surface albedo, and precipitation) via simulations with the IPSL (Institute Pierre Simon Laplace) LSM ORCHIDEE (Organizing Carbon and Hydrology in Dynamic Ecosystems) model, particularly focusing on the difference between (i) forced simulations with atmospheric forcing data (WATCH Forcing Data ERA-Interim – WFDEI) and (ii) coupled simulations with the IPSL atmospheric general circulation model. Results from statistical evaluation, using satellite- and ground-based reference data, show that ORCHIDEE is well equipped to represent spatiotemporal patterns of all variables in general. However, further analysis against various landscape and meteorological factors (e.g., plant functional type, slope, precipitation, and irrigation) suggests potential uncertainty relating to freezing and/or snowmelt, temperate plant phenology, irrigation, and contrasted responses between forced and coupled mode simulations. The biases in the simulated variables are amplified in the coupled mode via surface–atmosphere interactions, indicating a strong link between irrigation–precipitation and a relatively complex link between precipitation–evapotranspiration that reflects the hydrometeorological regime of the region (energy limited or water limited) and snow albedo feedback in mountainous and boreal regions. The different results between forced and coupled modes imply the importance of model evaluation under both modes to isolate potential sources of uncertainty in the model.

The fate of a boreal forest may depend on the trend in its normalized difference vegetation index (NDVI), such as whether the NDVI has been increasing significantly over the past few decades. In this study, we analyzed the responses of two Siberian larch forests at Spasskaya Pad and Elgeeii in eastern Siberia to various waterlogging-induced disturbances, using satellite-based NDVI and meteorological data for the 2000–2019 period. The forest at Spasskaya Pad experienced waterlogging (i.e. flooding events caused by abnormal precipitation) during 2005–2008 that damaged canopy-forming larch trees and increased the abundance of water-resistant understory vegetation. By contrast, the forest at Elgeeii did not experience any remarkable disturbance, such as tree dieback or changes in the vegetation community. Significant increasing NDVI trends were found in May and June–August at Elgeeii (p < 0.05), whereas no significant trends were found at Spasskaya Pad (p > 0.05). NDVI anomalies in May and June–August at Elgeeii were significantly associated with precipitation or temperature depending on the season (p < 0.05), whereas no significant relationships were found at Spasskaya Pad (p > 0.05). Thus, the 20 year NDVI trend and NDVI–temperature–precipitation relationship differed between the two larch forests, although no significant trends in temperature or precipitation were observed. These findings indicate that nonsignificant NDVI trends for Siberian larch forests may reflect waterlogging-induced dieback of larch trees, with a concomitant increase in water-resistant understory vegetation.

Surface water monitoring with fine spatiotemporal resolution in the subarctic is important for understanding the impact of climate change upon hydrological cycles in the region. This study provides dynamic water mapping with daily frequency and a moderate (500 m) resolution over a heterogeneous thermokarst landscape in eastern Siberia. A combination of random forest and conditional generative adversarial networks (pix2pix) machine learning (ML) methods were applied to data fusion between the Moderate Resolution Imaging Spectroradiometer (MODIS) and the Advanced Microwave Scanning Radiometer 2, with the addition of ancillary hydrometeorological information. The results show that our algorithm successfully filled in observational gaps in the MODIS data caused by cloud interference, thereby improving MODIS data availability from 30.3% to almost 100%. The water fraction estimated by our algorithm was consistent with that derived from the reference MODIS data (relative mean bias: −2.43%; relative root mean squared error: 14.7%), and effectively rendered the seasonality and heterogeneous distribution of the Lena River and the thermokarst lakes. Practical knowledge of the application of ML to surface water monitoring also resulted from the preliminary experiments involving the random forest method, including timing of the water-index thresholding and selection of the input features for ML training.