According to the 2019 Global Risks Perception Survey [1
], the top three “global risks” of highest concern in terms of likelihood are: extreme weather events, failure of climate-change mitigation and adaptation, and natural disasters. These three concerns, together with water crises, also appear in the top five risks in terms of impact on multiple countries and sectors [1
]. Drought, characterized as a natural departure from expected water availability [2
], is a major factor in each of these risk categories of concern. Over the past 20 years, the National Centers for Environmental Information (NCEI) identified 17 droughts in the United States that have each resulted in more than one billion-dollar loss per event. These losses often originate in the agricultural sector [3
]. This paper presents a strategy to use remote sensing to quantify the impacts of drought on forage biomass and to predict total growing season biomass from drought indices measured early in the growing season.
With 92% of the total land area used for farming and ranching, Nebraska’s agricultural production annually contributes approximately $
25 billion to the state’s economy, accounting for more than one-fifth of the state’s gross domestic product. Cattle are the state’s leading commodity in terms of value. Beef production is the largest agricultural sector and generates approximately $
7.2 billion in annual cash receipts [4
]. This makes Nebraska one of the largest beef producers in the United States and the world. Roughly 45% of Nebraska’s total area is classified as rangeland and pastures, half of which are located in the Sandhills region (approximately 12 million acres) that covers much of the north-central part of the state [5
]. Beef production in this region primarily relies upon grassland forage, which makes grassland productivity of the Sandhills important to Nebraska’s economy. Harsh environmental conditions, such as drought, can have direct consequences for ranching operations, including reduced forage yield, livestock losses, and water quality and quantity issues. To replace forage lost to drought, producers often purchase commercial feed that increases costs for ranching operations. These impacts potentially lead to other indirect issues like physical and emotional stress [6
] or financial and social instability [7
], often characterized as secondary drought impacts.
Large portions of Nebraska have experienced an extreme or exceptional drought four times over the past 15 years, with nearly 95% of Nebraska’s pastures and rangeland in “poor” or “very poor” conditions during the severe 2012 drought [8
]. Drought is an event that every land manager must consider and try to prepare for, especially in regions with lower water-holding capacity like the Sandhills, where interannual changes in precipitation greatly influence forage production [10
]. Drought planning can help manage these climate shocks. According to the concept of the three pillars of drought risk management [13
], the key areas that should be addressed are: (1) monitoring and early warning, which identifies drought status in a timely fashion; (2) vulnerability and impact assessment, determining the location and what is at risk of drought and why; and (3) mitigation and response, describing actions and measures needed to mitigate drought impacts and respond to drought emergencies. Our research contributes to the establishment of connections between these three pillars by describing links between drought indices and drought impacts, with the latter quantified as expected changes in biomass and forage.
Monitoring and early warning are important components of drought planning strategies. Over the past 20 years, the accuracy and precision of drought monitoring has improved due to technological advancements in meteorological instrumentation and the ability to archive, analyze, and disseminate the available data [14
]. However, in-situ observations are spatially and temporally limited, and are unable to meet the increasing monitoring demand, especially on large spatial scales. Compared to ground-based observations, satellites provide a solution with global, near-real-time observations over larges areas, consistent data records, and improved spatial resolution [15
]. Over the course of the last few decades, a suite of indices and tools measuring the severity and extent of a drought has been developed. Some of these are based on in-situ measurements of various parts of the hydrologic cycle, while others are a blend of input variables originating from different data sources, including satellite remote sensing.
Time-consuming, ground-based monitoring of range conditions using visual estimates or destructive sampling methods is often performed on a small scale. Applications that utilize modeling approaches to monitor and estimate annual range conditions on a large scale are being developed [16
]. So far, however, the spatial resolution or the meaning (e.g. quantity of forage) of the provided range condition information is not appropriate for use by individual land managers. Information on location-specific productivity is important, especially in hydrologically complex and land cover diverse areas such as the Sandhills, where relatively high and low productivity can be found on a small spatial extent (see more information about upland areas and wet meadows in the Study Area section). Therefore, enhancing the monitoring of range conditions on a landscape scale with sufficient spatial resolution, while providing meaningful and simple information for range management decisions, is needed to more effectively mitigate and respond to drought.
Satellite remote sensing techniques and applications have been widely used for landscape assessment of vegetation health, as well as estimation of vegetation biomass and grass forage [18
]. However, changes in vegetation health and the amount of produced biomass can be difficult to interpret because the remotely sensed signal does not capture information about the specific causes of these changes [23
]. When studying the impact of drought on biomass production, it is important to separate the influence of climate from other factors. While weather refers to short-term variations (i.e., minutes to days) in variables such as temperature, precipitation, and wind; climate describes conditions over larger regions and for extended periods of time. Climate extremes, such as drought, are characterized by the deviation of climate statistics from what is expected over a given period of time (e.g., a month, season, or year). Indices are frequently calculated to describe, monitor, and project the state of the climate.
The goal of this study is to establish a relationship between annual biomass production and climate conditions, particularly those related to drought that occurs before and during the growing season to identify the climate variables and indices that explain the majority of interannual biomass variability. Biomass production depends on various site characteristics, such as soil properties, topography, long-term climate, and historic land use and management [24
]. The interannual variation in biomass production is mainly influenced by climate conditions that occur before or during the growing season and by disturbances such as fire, grazing, or management [25
]. Therefore, when studying the impact of drought on biomass production, it is important to use a methodology that separates the influence of climate from other factors.
The Normalized Difference Vegetation Index (NDVI) has been widely used for the assessment of ecosystem performance. NDVI values averaged over the growing season (GSN) have proven to have a strong relationship with ground-based observations of biomass productivity [26
]. We separate the influence of seasonal climate on the amount of total seasonal biomass from other factors with a previously established methodology [29
] that uses historic remotely sensed NDVI, climate data, and regression tree modeling techniques. The product of the model is the Expected Ecosystem Performance (EEP) that represents the annual biomass expected to be produced at a specific site under certain climate conditions, without the influence of other factors.
The first (1) objective of this study is to develop an annual EEP model and corresponding maps for the years 2000 to 2016 at 250 meter (m) resolution for the upland grasslands of the Sandhills Major Land Resource Area (MLRA), which represents roughly half of the rangeland and pastures in Nebraska. These maps serve the purpose of isolating the effect of climate on growing season total forage production. This is achieved by using model inputs that are related only to climate and site characteristics. The output of the model also provides information about the overall importance of the different explanatory input variables. The second (2) objective is to examine the relationship between various drought indices, the timing of drought, and biomass production derived from the model. This objective uniquely combines ecosystem performance methodology with the use of various drought indices and tools that are based on diverse sources of information. We identify the indices and time frames that explain the majority of interannual seasonal biomass variability. The third (3) objective is to create a piecewise regression tree model that uses drought indices, summarizing the moisture conditions before and during the early stages of the growing season to predict annual forage production. The result of this study will improve the knowledge available to land managers for more informed decision making during the early stages of the growing season.
The current study provides a landscape-scale understanding of seasonal and interannual climate effects on the dynamics of grassland forage production in semi-arid upland locations of the Nebraska Sandhills. The EEP has been previously found to capture the variation of grassland productivity due to spatial changes in site potential and temporal changes in climate, while minimizing the influence of disturbances and land management [29
]. Other studies, however, have focused on locating the impacts of land management practices and disturbances creating EEP anomalies, subtracting EEP from AEP (e.g., [29
]). The strong use of site potential in the EEP model suggests that biomass production is dependent on a specific location [29
] and corresponds well to the west-east productivity gradient [26
]. Summer precipitation was the most important variable in the EEP model, appearing in all stratification and prediction rules. Total summer precipitation was previously found to explain majority of the time-integrated NDVI variation for warm-season grasses, the dominant grasses of the Sandhills uplands. The inclusion of summer season total precipitation in net primary productivity models from the Central Great Plains was supported [67
The EB maps were found to successfully capture the wet and dry years, as well as the productivity gradient. The maps of interannual biomass deviation provide information about specific spatial distributions of below or above normal biomass for each year caused by interannual climate variability. A differential response of vegetation in various areas to climate conditions can influence the amount of produced biomass. However, we tried to eliminate this factor by choosing only one MLRA that should have similar biophysical characteristics (soils, climate, vegetation, etc.) [31
], and by eliminating areas of wet meadows and other land cover classes. Maps of EB and the interannual biomass deviation, if provided as growing season biomass estimates during the early phases of growing seasons, might help land managers make decisions about appropriate stocking rates and other management considerations. For example, in 2007, a land manager from the drought-affected northwestern part of the Sandhills could have sought additional hay supply from the southern portion of the region, where there was above-average biomass production (Figure 4
and Figure 5
). A moderately high spatial resolution of our product could also provide information about areas with higher forage availability that would be suitable for potential relocation of animals during severe drought conditions, and be especially beneficial for Bureau of Land Management (BLM) lands or private landowners with large-area ownership.
To validate our findings from the EEP model, we compared modelled biomass data with long-term grass clipping data from two locations in the Sandhills. The relationship between these two datasets was relatively modest. The R2
values in our analysis were in the range of values observed in other studies, despite the fact that the ratio of field-observed versus remotely sensed area was much larger in those studies, compared to the area of our validation [22
]. Considering the relatively small sample size (compared to the potential variability in the area) and the difference in plot sizes of grass clipping (0.25-m2
) and the EB data (62,500-m2
), the model captured well the interannual variability in biomass production, perhaps with some underestimation of biomass during severe drought conditions. It is important to note that ground clipping observations can also introduce certain errors. Additional errors could have been introduced when converting the EEP to biomass using an empirical equation developed for a much larger area than the Sandhills [26
]. Developing a biomass equation for a more specific area could lead to improvement of biomass estimates. The regional statistics of the SSURGO range productivity dataset further validated the overall model performance. The regression line’s close alignment with the 1:1 line in Figure 8
indicated minimal bias, although we observed a slight overprediction of lower productivity values. Models tend to underpredict high values and overpredict low values. We mitigated this tendency using various classes of productivity and a long period of observed NDVI [70
]. Selecting more training points from wet and dry years that are characterized by higher and lower biomass, respectively, or using more classes of productivity stratification could lead to an improvement in the model.
We examined a suite of drought indices driven by different climate variables (SPI—precipitation, EDDI and ESI—evaporation, VegDRI—strong NDVI component, USDM—a blend of various indices and expert observations) and their relationship with biomass production. The correlation analysis revealed strong relationships between certain drought indices and EB, derived from the model during various stages of the growing season. The highest correlation was observed for the three-month ESI, capturing the period of May, June and July. This period of time has been found to be sensitive to precipitation variability in Sandhills grasslands [12
]. Longer drought index timescales had stronger explanatory power with regards to the interannual variation of EB. Strong relationships were achieved when the index covered the period of June and July, which is the peak growth period for warm-season dominant grasslands in Nebraska Sandhills [71
]. Weaker relationships were observed for periods that captured the moisture amounts during the greenup and senescence of grasslands [12
The USDM percentile index proved to be a successful method to compare the original categorical data derived from the USDM with interannual biomass anomalies. The strongest relationship was found for the summer and fall integrated index. Two years (2004 and 2013) that were classified as severe drought in the USDM, however, showed near-normal productivity values. Both of these years followed severe drought conditions in the previous years. One of the properties of the USDM is that it captures both short- and long-term drought, which is delineated in the weekly USDM online maps, but is not captured in the provided geographic information system (GIS) data. Long-term drought might not be entirely alleviated by shorter-term precipitation events during the growing season; however, this moisture supply can be crucial for forage production. Therefore, there is a need to distinguish between short- and long-term drought conditions when explaining grassland biomass variability using the USDM. We excluded the years 2004 and 2013 from our analysis, which notably improved the correlation coefficients for all seasons. Non-linear relationships are also to be noted, with stronger relationships in the lower-left quadrants of the scatterplots, indicating good relationships during the dry years, which is most likely caused by the inability of the USDM to capture wet conditions. This approach should be considered when using the USDM with respect to biomass production in semi-arid grasslands, for example in the Livestock Forage Program (LFP). The LFP program’s payment amounts for Nebraska statewide were comparable in years 2012 and 2013, despite the differences in total annual precipitation and the amount of biomass derived from our model. We recognize that ecosystem performance is not a perfect depiction of ecosystem health; other processes not captured by remote sensing can influence ecosystem health. Grassland species’ composition might be affected by severe drought. For example, in 2013, following the severe drought in 2012, the Sandhills grasslands showed more annual grasses and forbs than in other years [72
], which might have an effect on the nutritional value or palatability of forage. The relationships described in this study might be different in other regions that are characterized by different physical and biological properties [73
Our predictive model correctly captured the interannual variability in EB. The overall importance of variables in the model showed the importance of spring moisture for the exponential growth phase of warm-season grasses [74
]. We examined the training and testing mean absolute error for overfitting (testing MAE–training MAE)/training MAE), which showed a slightly higher, but still acceptable value (10.6%, while overfitting is considered negligible when <10%). We also developed an additional predictive model that used expected biomass anomalies as the dependent variable without the use of site potential as an explanatory variable. This model yielded weaker regression, but was still significant (R2
= 0.88, MAE = 73.8, p
< 0.01). We did not include the VegDRI dataset in the predictive model because of its short period of record and the high use of NDVI in the VegDRI modeling process, which our predicted variable is derived from. We will continue to improve the predictive model using larger sample size, biomass stratification, and longer time series. A measure of soil moisture in the model is currently not represented because the sandy soils do not have high water-holding capacity. Therefore, it might not be important for sites like the Sandhills. We plan to further include soil moisture in the model, especially for areas characterized by higher water-holding capacity. We also plan to investigate other equivalent NDVI products that are being developed in the event of a possible MODIS decommission. Further research is needed to test the applicability of this methodology in other semi-arid grasslands.
When compared to other forage prediction tools, this model ties information about biomass production to specific locations, which allows the exclusion of areas that are not representative of major land cover. In the case of Sandhills, the excluded areas were wet meadows with different species composition, different growth curves, and a connection to near-surface groundwater levels that can be decoupled from climate variations for one or more years.