TL;DR
This study demonstrates that satellite-based soil moisture data, particularly from active microwave sensors like ASCAT, can effectively detect and map agricultural irrigation patterns in China, which are often poorly represented in land surface models. By comparing satellite data with a model that doesn't account for irrigation, the researchers show that microwave remote sensing offers a promising tool for understanding and managing water resources in irrigated regions.
What Problem Does This Paper Address?
Accurate soil moisture information is crucial for understanding Earth's water and energy cycles, especially in the face of climate change and increasing human interference. While land surface models (LSMs) are commonly used to simulate soil moisture, they often fail to adequately represent human-induced modifications to the hydrological cycle, such as agricultural irrigation. This omission can lead to biases in model simulations, particularly in heavily irrigated areas. Detecting and quantifying these unmodeled irrigation processes is therefore a significant challenge for improving hydrological modeling and agricultural management. This paper aims to address this gap by evaluating the potential of various satellite-based microwave soil moisture products to identify and map irrigation patterns across China.
Methods
The researchers employed a multi-faceted approach to evaluate the utility of satellite soil moisture data for detecting irrigation.
- Validation with In-Situ Data: They first validated three satellite soil moisture products (AMSR-E/AMSR2, ASCAT, and ESA CCI) against ground-truth measurements from a nationwide agro-meteorological network. Beyond standard correlation metrics (Pearson and Spearman), they introduced entropy-based mutual information (MI). MI was found to be more suitable for evaluating soil moisture anomalies caused by irrigation, especially under wetter conditions, as it can capture more of the uncertainty in the "ground truth" data.
- Detecting Irrigation Signals: To identify irrigation patterns, they used the two-sample Kolmogorov-Smirnov (K-S) test. This statistical test compares the probability distribution functions (CDFs) of soil moisture time series. They compared the CDFs of satellite-derived soil moisture (which implicitly includes irrigation signals) with those from the ERA-Interim reanalysis model, which does not incorporate irrigation. A significant difference in CDFs, as indicated by the K-S statistic, suggests the presence of irrigation.
- Spatial and Temporal Analysis: The K-S test was applied to identify spatial irrigation hotspots and to examine the temporal response to irrigation. They also investigated the sensitivity of the detection capabilities to different satellite overpass times (ascending vs. descending orbits).
Datasets
- Satellite Soil Moisture Products:
- AMSR-E and AMSR2: Passive microwave observations from the Advanced Microwave Scanning Radiometer (AMSR-E) and its successor (AMSR2), using the Land Parameter Retrieval Model (LPRM) algorithm.
- ASCAT: Active microwave observations from the Advanced Scatterometer (ASCAT), using an empirical change-detection approach.
- ESA CCI: A blended multi-sensor soil moisture product from the European Space Agency, combining passive and active microwave data.
- Benchmark Model:
- ERA-Interim: A global land surface reanalysis dataset that does not account for irrigation. This served as a baseline to isolate the irrigation signal.
- In-Situ Observations:
- Nationwide Agro-meteorological Network: Ground-truth soil moisture measurements collected across China, with a high density in the Huang-Huai-Hai (HHH) Plain.
- Ancillary Data:
- Areas Equipped for Irrigation (AEI): Gridded data from FAOSTAT (circa 2005) indicating areas with irrigation infrastructure.
- Actual Irrigation Areas: The Irrigated Area Map Asia and Africa (IWMI-IAMAA) was used as a benchmark for identifying actual irrigated areas.
Key Insights and Ideas
- Mutual Information for Irrigation Anomalies: Conventional correlation coefficients (Pearson, Spearman) can be misleading when evaluating soil moisture under irrigation. Entropy-based mutual information (MI) provides a more robust metric for capturing soil moisture anomalies induced by irrigation, especially when soil moisture significantly exceeds the norm.
- K-S Test as a Powerful Irrigation Detector: The K-S test is highly effective at discerning differences in soil moisture distributions between irrigated and non-irrigated areas. This method is less sensitive to the absolute values of soil moisture and more focused on the overall distribution, making it ideal for detecting the systematic impact of irrigation.
- ASCAT's Superiority in Irrigation Detection: Among the tested satellite products, ASCAT (active microwave) generally showed the highest performance in detecting irrigation intensity, particularly in the Huang-Huai-Hai Plain. This suggests that active microwave sensors might be more sensitive to the hydrological impacts of irrigation.
- Diurnal Cycle Matters for Passive Microwave: Passive microwave sensors like AMSR showed greater discrepancies between ascending and descending orbits. This is attributed to their retrieval algorithms being more influenced by surface temperature, which is affected by irrigation and exhibits a strong diurnal cycle.
- Microwave Remote Sensing as a Complement to LSMs: Satellite soil moisture data can effectively capture the "unmodeled" irrigation process, providing a valuable complement to land surface models that currently lack detailed irrigation representations.
Key Results
- Validation: Approximately 60% of the uncertainty in soil moisture anomalies from "ground truth" could be resolved by satellite-based retrievals, with ASCAT showing the best performance. ERA-Interim's performance was lower in this regard.
- Irrigation Detection: The K-S test successfully identified irrigation hotspots across China. The Huang-Huai-Hai (HHH) Plain was identified as a major irrigation hotspot by all three microwave products.
- ASCAT highlighted intensive irrigation in the mid-south HHH Plain, likely due to its sensitivity to flooded rice fields.
- AMSR detected irrigation in the north HHH Plain and areas in Xinjiang.
- ESA CCI also detected irrigation in Hebei, showing good agreement with AMSR.
- Accuracy: Confusion matrices indicated excellent irrigation detection accuracy for all three data sets, with producer's accuracy generally above 70%. ASCAT achieved the highest producer's accuracy (88%) in detecting actual irrigation areas.
- Temporal Dynamics: The study demonstrated that the SM2RAIN algorithm, using ASCAT data, could estimate irrigation amounts, showing good agreement with recorded values in the Weishan irrigation district.
- Orbit Sensitivity: While AMSR showed some sensitivity to ascending/descending orbit differences, ASCAT exhibited less noticeable discrepancies, suggesting its detection capabilities are more robust across different overpass times.
Limitations and Future Work
- Uncertainties in Retrieval Algorithms: The paper acknowledges that uncertainties exist in the retrieval algorithms for all satellite products, which can affect detection accuracy.
- Influence of Background Factors: The K-S test relies on the assumption that background factors are similar between irrigated and non-irrigated areas, which may not always hold true. Other factors like vegetation density and radio frequency interference (RFI) can also influence soil moisture signals.
- Quantifying Irrigation Amount: While the study demonstrates the potential for detecting irrigation patterns and intensity, further research is needed to accurately quantify the amount of water used for irrigation.
- Algorithm Development: More efforts are required to reduce uncertainties and develop advanced algorithms for quantifying irrigation water use.
- Model Integration: The findings have significant implications for improving numerical model development and agricultural management, suggesting the need to better incorporate irrigation processes into LSMs.
Summary updated