Study Area

Yuma has over 450 small, family-owned agricultural companies [4]. For simplicity, only one plot from a farmer in this region was selected. Mellon Farms grows a range of vegetables and fruits, focusing on winter greens. This farm is located near the border of the USA and Mexico, on the western side of Arizona. The map below (to the left) gives the study plot, and the point of interest used for gathering observations is -114.73391700474117,32.68231003417071 (Long, Lat).

Generally speaking, this region of the USA does not get much rain. When we look at the precipitation for the study point, 0.027 meters of total precipitation were recorded by Era5 for 2022 and 0.161 meters of precipitation for 2023. Though not a large amount of rain overall, the accumulation seen in 2023 is almost six times what was seen in 2022. The graph below (top right) gives the NDVI values for 2022 and 2023. Note there is a regular cadence in the NDVI with peaks in December/January and April/May. January of 2022 had a significantly higher peak than the peak in January of 2023. High NDVI aligns with the growing schedule posted on the Mellon Farms website; lettuce, broccoli, cauliflower, and celery are grown between November and May [5]. Google Earth Engine (GEE) was used to pull the data using the COPERNICUS/S2_SR image collection. The second graph (lower right) gives the average daily temperature readings for 2022 and 2023 at the point of interest. Both years follow a similar pattern, with roughly the same overall highs and lows. 2023 looks like it has a slightly steeper pitch to the peak than 2022.

Method

For the purpose of this study, gross primary productivity (GPP) is our primary metric for comparing the effects of the El Nino on crop growth. GPP is the rate at which solar energy is absorbed and then used to create sugar molecules during photosynthesis. GPP is a common metric for plant productivity. Net plant productivity (NPP) is GPP, excluding the energy required for plant respiration. The rule of thumb for respiration is 50% of the GPP. Using remote sensing inputs, GPP can be estimated at spatial and temporal scales for cropland applications. For this study, the SCOPE model, in conjunction with retrieval, will be used to calculate GPP.

Data

There are three data sources used in this study used for three aspects. Google Earth Engine (GEE) was the primary tool to gather and transform data. ERA5-Land Daily Aggregated was used for the meteorological inputs, specifically the ‘ECMWF/ERA5_LAND/HOURLY’ image collection. The physiological properties of the vegetation were derived from Copernicus’ Sentinel-2 multi-spectral images, specifically the image collection ‘COPERNICUS/S2_SR’. Finally, the validation data for plant productivity results came from the World Association for Public Opinion Research (WaPOR). For this research net primary production (Global- Dekadal – 300m) from the WaPOR v3 Time series will be used.

Retrieval

Retrieval is the process of getting the physical properties of an object through its reflectance values. The chart below (left) shows some possible retrieval methods. The radiative transfer model (RTM) is one of many retrieval methods.  RTM links remote sensor readings to the physical properties plants are known to have. When RTM is performed from the starting point of observed radiance, as done here, it is known as an inverted radiative transfer model. The chart below (right) depicts the inverted RTM process. For this research, the RTM used was retrieval_rtmo-itc2020 (RTMO), a MATLAB-based script that takes in Sentinel-2 data.  RTM is a fitting process, as multiple combinations of plant properties could produce similar results. Both LUT-based inversion and numerical inversion were attempted for this study. RTMO has a setting for both LUT and numeric inversion. 

Accuracy and Outliers

Given that RTM is a process of fitting a curve through optimization or values in a table, results are subject to some error. Root mean squared error (RMSE) was chosen as the metric to help address the level of accuracy seen in the model. The threshold for accuracy in this study is higher than .03. In addition to filtering observations based on RMSE, Sentinel 2 imagery was also used to validate points visually.  Normalized Difference Vegetation Index (NDVI) is a metric that measures the density of vegetation using sensor data. A high LAI should happen in conjunction with a high NDVI. In places where there was a mismatch between LAI and NDVI, Copernicus Sentinel 2 images were used to see if clouds or other factors led to an error in the readings. It was removed if the satellite image did not match the expected LAI value, whether too high or too low.

Results

Developing GPP

The LUT method output was unusable as 112 out of 117 records in 2022 had a root mean squared error (RMSE) higher than .03. 2023 was equally unusable, with 113 out of 114 records having an RMSE higher than .03. That does not leave enough records to develop a proper time series. Even if the RMSE threshold were raised to .05, that would still eliminate 104 and 106 records for 2022 and 2023, respectively. Thus, the results derived from only the NO method were used. 

Validating with WAPOR

The graph below and to the left shows the NPP value from WaPOR. The graph to the right shows the scope results compared to the WaPOR values. Observations made using the derived SCOPE GPP are supported by the WaPOR validation data set. The magnitude of change and seasonal shift were seen in the WaPOR data for the same time period, validating that there was a difference in plant productivity between 2022 and the following El Nino year of 2023. When comparing WaPOR to SCOPE, the RMSE was .83. One could argue that this is too large, given its proportion to the data. However, the goal of this study was not to predict WaPOR data but to validate the overall pattern.  Future research could be dedicated to more accurate results or adding additional points of interest to bolster the results further.  

Conclusion

The goal of this study was to compare the plant productivity of a Mellon Farms field before and during an El Nino year. In both the values found by SCOPE and WaPOR, there was a large change between the years. There was a change in magnitude as well as the timing of plant productivity. The max GPP was cut in half between 2022 and 2023. In 2022, the max GPP occurred during the spring plant growth, while in 2023, it occurred in January. There was less plant productivity overall, but the spring growth cycle was especially affected during the El Nino year. Spring growth came later and was less productive than the previous year. El Nino is known for higher temperatures and more rain.

El Nino makes the season warmer and wetter than it typically would. Mellon Farms irrigates its crops using the Colorado River. Even when there is not enough rain, the crops in this area have access to enough water. This means that crops in this area do not benefit from the extra rain that comes with an El Nino year. Though, the farmer's water bill might have. The drop in plant productivity could have been due to the increased heat associated with the El Nino year. Lack of water and heat are the main stressors on a plant's productivity. Drier conditions would mean that the stomata remain closed more often, and therefore, the plant as a whole would not be as productive, hence the drastic drop in the magnitude of productivity.

Like any remote sensing endeavor, clouds and other data processing issues may have influenced the results. The presence of clouds and how SCOPE handles its parameters could have led to the negative GPP and jagged values. Better data processing and gathering could lead to smoother results, but the stark difference between years can not be denied. Seeing a large change in plant productivity with both SCOPE and WaPOR validates that there could have been a change during the El Nio year for the crops in this field.

References

1. National Geographic Society. (n.d.). El Niño. National Geographic Education. Retrieved [January 18, 2024], from https://education.nationalgeographic.org/resource/el-nino/

2. NOAA Ocean Service. (n.d.). What are El Niño and La Niña? Retrieved [January 18, 2024], from https://oceanservice.noaa.gov/facts/ninonina.html

3. Jones, B. (2023, April 18). You — yes, you — are going to pay for the century-old mistake that’s draining the Colorado RiverVox. Retrieved [January 18, 2024], from https://www.vox.com/the-highlight/23648116/colorado-river-lake-mead-agriculture-leafy-greens

4. DeWit, M. (2019, January 22). Farming in the Border Town of Yuma, Arizona. SBA Office of Advocacy. Retrieved [January 18, 2024], from https://advocacy.sba.gov/2019/01/22/farming-in-the-border-town-of-yuma-arizona/

5. Mellon Farms. (n.d.). Ranch locations. Retrieved [January 18, 2024], from http://mellonfarms.com/custom-farming/ranch-locations/