Spatial Application of Southern U.S. Pine Water Yield for Prioritizing Forest Management Activities

Abstract

Forest management depends on forest condition data and the ability to quantify the impacts of management activities to make informed decisions. Spatially quantifying water yield (WY) from forests across large landscapes enables managers to consider potential WY changes when designing forest management plans. Current forest water yield datasets are either spatially coarse or too restricted to specific sites with in situ monitoring to support some project-level forest management decisions. In this study, we spatially apply a stand-level southern pine WY model over a forested landscape in the Florida panhandle. We informed the WY model with pine leaf area index inputs created from lidar remote sensing and field data, a spatial and temporal aridity index from PRISM and MODIS data, and a custom depth to groundwater dataset. Baseline WY conditions for the study area were created using the Esri and Python tools we developed to automate the WY workflow. Several timber thinning scenarios were then used to quantify water yield increases from forest management activities. The results of this methodology are detailed (10 m spatial resolution) forest WY raster datasets that are currently being integrated with other spatial datasets to inform forest management decisions.

1. Introduction

Prioritizing and planning forest management activities depends on reliable, reproducible, high-quality ecological data, combined with novel computational analyses to support informed decision making [1]. Forest management plans use decision support systems to assess current forest conditions and predict how these conditions will change under alternative or competing management scenarios [2]. Decision support systems almost always use an information system, such as a database or geographic information system (GIS), where spatial ecological data are used in computational analysis [3]. Decisions about where, when, and how to conduct forest management activities such as timber thinning and prescribed fire must consider multiple uses and services that forests provide such as timber, wildlife habitat, recreation, non-timber commercial forest products, clean water supply, and carbon sequestration. These uses can be related spatially across a landscape, which explains the ubiquitous use of GIS within decision support systems [3]. The existing constraint on these decision support systems is often the availability of quality data that can quantify all these elements at the appropriate spatial and temporal scales [3]. Without these data, forest management plans cannot accurately predict the downstream effects of their management decisions or compare between different management scenarios.

An existing constraint on some forest management decision support systems is the lack of high spatial resolution water yield data. Water yield (WY) is rainfall minus evapotranspiration (ET), and the biotic attribute most directly linked with evaporation and transpiration in a vegetated landscape is leaf area [4]. Leaf area is typically quantified using a leaf area index (LAI), which has been found to strongly predict the ET of a vegetated landscape [4,5,6]. Therefore, access to high spatial resolution LAI datasets can inform models of water yield within a forest management decision support framework.

Modeling water yield is important for forest managers as forested lands are a large source of water, and forests require larger quantities of water than other land surface ecosystems [7]. Forests have a high ET rate due to the elevated rates of leaf interception and transpiration from the large biomass of trees [8]. Forests also supply a more reliable and higher-quality water source than any other land use [9]. They promote high soil water infiltration capacity [10], have high soil organic matter, and reduced soil compaction compared to other land use types [11]. In the conterminous United States, forested lands provide surface water yield at a much higher rate than the percent land surface area they occupy [7,12]. Several studies have found that the majority of surface water yield in the conterminous U.S. is provided from forested lands, which cover only a third of the land surface area [7,9,12]. In the southeastern U.S., forested lands provide 36% of the total annual drinking water supply and 34% of the overall water supply [13,14]. Therefore, land managers concerned with water quality and water quantity in these regions should strive to balance ET reductions with increased water yield, while maintaining forested lands that protect water quality.

Globally, reduction in forest cover, and, therefore, LAI, have been shown to increase water yield [15]. Forest management activities such as timber thinning, and to a lesser extent, prescribed fire, allow land managers to manipulate the water yield balance [8,16,17]. However, the specific amount and type of LAI reduction needed to produce increased water yield is site-dependent [18]. In the southeastern U.S., it has been shown that naturally regenerated pine stands managed for lower basal area have lower values of ET and can have higher water yields when compared to dense, even-aged pine plantations. McLaughlin et al., 2013, found pine stands in the southeastern U.S. that are managed for lower basal area and recurrent understory fire, and, therefore, lower LAI, result in 64% more water yield over 25 years than high-density pine plantations [8]. Other studies have found lower values of ET in natural slash pine stands which ranged from 17–85% less than slash pine plantations in Florida [19,20]. As well, the models of Susaeta and Alavalapati, 2021 [21], suggest that natural longleaf pine stands produce 4% more water yield than privately managed longleaf pine stands.

Water yield from forested lands is coming under threat from climate change, leading to more ET and alterations in precipitation regimes [7,14]. Climate change is also contributing to increases in fire frequency and severity, altered tree species distribution, and flooding in forested lands [22,23,24]. In the southern U.S. specifically, drier conditions could threaten wetland habitats, increase tree mortality, and increase wildfire size and frequency [22,24,25,26]. Management decisions on forested lands can help mitigate climate change impacts by targeted reduction in forest LAI to increase water yield [21]. Increase in water yield from forests can provide more water for the forest ecosystem itself, accomplishing multiple management objectives [27]. Reduction in forest LAI can also be conducted for the benefit of ecosystem and groundcover restoration, wildlife habitat improvement, wildfire prevention, and timber production. To realize multiple benefits of forest LAI reduction across a landscape at a high spatial resolution, an accurate quantification of water yield from the forested landscape is necessary.

Many studies quantify water yield at the basin or watershed scale using flow data and hydrological models [28,29,30,31]. Within-stand monitoring of ET and precipitation can also be conducted to provide a stand-level water yield dataset, although the cost is prohibitive [5]. While these methods are informative at the basin or watershed level, in the case of watershed models they are spatially coarse. While stand models have higher spatial resolution, they also have a small spatial extent. Neither of these methods are ideal for use in a forest management decision support system that requires both a large spatial extent and high spatial resolution. Previous studies have recognized this gap in the spatial quantification of forest water yield, and the need to help direct management decisions regarding specific multiple-use priorities, such as timber thinning to support climate change mitigation [7], or for mapping ecosystem services across a forested landscape [32]. Integrating a water yield dataset at the spatial resolution and extent where forest management decisions are made is an essential advancement for land managers [5].

Recent research into southern pine water yield modeling [5] and advances in analysis of remote sensing data for forest structure [33] have provided the basis for a high-resolution spatial analysis of water yield at a landscape scale. Acharya et al., 2022, provide a water yield model as a function of forest structure, hydrogeological setting, and climate [5]. The development of this stand-level pine water yield model was in direct response to Florida’s public land managers needing a simple, yet robust method to measure pine forest water yield at a stand level. The spatial application of this model across a larger range of southern pine in Florida is the natural progression for integration into a decision support system. A high-resolution landscape-level spatial water yield output can be developed from this stand model using the latest forest structure and climate datasets developed from remotely sensed data. A spatial pine water yield estimate allows for integration into a decision support tool that can compare different thinning scenarios, multiple climate scenarios, and adjustments for use in different locations. This spatial decision support tool will enable managers to quantify the impacts of management decisions on their area’s hydrology and allow them to integrate this with other spatial datasets used to quantify the multiple uses of forests. Advancements in remote sensing technology and spatial analysis techniques that incorporate free, open-source software have lowered the entry barriers for creating and maintaining spatial decision support tools and related datasets [34]. Since many entry-level forestry jobs now require geospatial knowledge and skills [35], a GIS-based decision support tool will likely be familiar to many forest land managers.

The primary goal of this study is to create a practical, reproducible workflow for creating high-resolution, landscape-level spatial pine water yield data from the stand-level pine water yield model presented in [5]. This workflow focuses only on pine tree species, as they are the dominant tree species in the study area, as well as being the primary focus of forest management operations. Additionally, the stand-level pine water yield model we use was developed to quantify southern pine water yield and is not meant to be a comprehensive model of water yield [5]. The goal of this workflow is to spatially quantify the water yield lost to pine evapotranspiration (ET) at a high spatial resolution. This type of data helps fill the existing gap in water yield data and can be incorporated into a decision support system to help inform forest management decisions. To that end, this study will demonstrate the use of spatial pine water yield outputs within a decision support framework in several ways. First, after creating a baseline scenario, we derive pine water yield outputs from three timber thinning scenarios to directly compare pine water yield increases at various spatial scales. That said, this workflow will make no determination on where this additional water yield will be allocated within the hydrologic system. Second, we demonstrate the usefulness of this workflow for quantifying the impact of climate variability on pine forest water yield. Finally, we develop accessible geospatial tools that allow forestry practitioners to easily adopt and customize their own decision support framework. This study provides packaged water yield workflows as an Esri ArcGIS Pro geoprocessing toolbox for forestry practitioners comfortable with that GIS software, and as a free, open-source Python script that does not require a GIS software license.

2. Materials and Methods

2.1. Study Area

The Apalachicola River study region of Northwest Florida comprises 11 counties in the panhandle and totals just over 1.9 million hectares (Figure 1). The humid subtropical climate includes hot summers, mild winters, and a long growing season with an average precipitation of approximately 140 cm per year. This region is bisected by the lower Apalachicola–Chattahoochee–Flint (ACF) River basin, and includes drainage areas from the Choctawhatchee, Ochlocknee, and Aucilla–Wacissa River basins. The Apalachicola River is the largest river in Florida in terms of flow volume, and the region includes several first- and second-order magnitude springs as well as other karst geologic features that provide hydrologic connectivity to the Floridan Aquifer.

The Apalachicola Region is also designated as one of the top five biodiversity hot-spots in North America [36]. The area provides habitat for endemic, threatened, and endangered species [36]. Currently listed species include the frosted flatwoods salamander (Ambystoma cingulatum), red-cockaded woodpecker (Leuconotopicus borealis), striped newt (Notophthalmus perstriatus), and flora populations such as Harper’s beauty (Harperocallis flava), Florida skullcap (Scutellaria floridana), white-birds-in-a-nest (Macbridea alba), and Godfrey’s butterwort (Pinguicula ionantha), which are in decline from the loss of habitat and available freshwater in their preferred environment [37]. Forests are the dominant land cover type in the study area (73%) according to Florida Fish and Wildlife Conservation Commission’s Cooperative Land Cover dataset, version 3.4 [38], with 46% being pine-dominated forests, including the large publicly held lands of the Apalachicola National Forest (230,337 ha) and Tate’s Hell State Forest (85,902 ha). This forested landscape is a major source of freshwater discharge to the Apalachicola River, Bay, and Estuary originating in Florida [37]. This area was also impacted by Hurricane Michael, a category 5 hurricane, making landfall near Mexico Beach, Florida, on 10 October 2018, leading to an estimated 19.18% loss of forested basal area [39].

Historically, the Apalachicola Region was part of the extensive longleaf pine (Pinus palustris) ecosystem in the southeastern U.S., characterized by a range of pine-dominated natural community types from wet flatwoods and savannahs to xeric sandhills and upland pine forests, having open canopies with low basal areas, extremely diverse groundcover, and frequent fire return intervals that limit understory encroachment from shrubs and hardwood species. Past management practices throughout the Apalachicola Region, however, resulted in the conversion of much of this extensive longleaf pine ecosystem to dense pine plantations used for industrial timber production. Alterations to the study area’s hydrology and fire regime to maximize timber production included large wetland areas being ditched, drained, and bedded to accommodate the growth of slash pine (P. elliottii), while the exclusion of fire further degraded habitat conditions. In recent years, land management agencies have prioritized the restoration of longleaf pine across its historic range by applying consistent silvicultural practices aimed at improving ecosystem condition.

2.2. Stand-Level Water Yield Model

Acharya et al., 2022, introduced a southern pine water yield model that employed temporally and spatially dense in situ data to predict water yield as a function of forest structure, hydrogeological setting, and climate [5]. In situ data were collected on five southern U.S. pine forests in Florida. Six sites were selected at each location for a total of 30 unique plots, with various levels of ownership, different dominant pine species, stocking densities, and management practices to capture a wide range of Florida’s pine forests. Sites were monitored for four years, with collected data ranging from soil characteristics such as soil moisture, texture, root biomass, and water table depth, to standard forest metrics, such as basal area, LAI, species, and density, and meteorological data including temperature and potential evapotranspiration (PET).

The forest structural metric that was the strongest predictor of water yield was LAI, along with a binary classification of groundwater depth, and aridity index (a ratio of PET to precipitation). The stand-level water yield model was specifically designed using these three readily measurable site properties with the intention of further application to new locations. The final fixed-effect general linear model, with an R² value of 0.78, for estimating water yield in centimeters per year is shown in Equation (1).

WY = 197.6 + (−9.7 ∗ LAI) + (16.8 ∗ DTW) + (−137.2 ∗ ARID),

(1)

where WY is water yield (cm/yr), LAI is leaf area index, DTW is depth to water table, and ARID is aridity index. WY in this model is defined as the difference between precipitation and ET where ET includes interception and soil ET [5].

2.3. Geospatial Data

Spatial application of the stand model to produce water yield outputs that can be used in a multi-scalar approach to forest management relies on the best available remotely sensed data; however, any effort to extrapolate from a stand-level study to a large landscape using remotely sensed data will inherently produce estimation errors. In this study, four raster datasets, shown in

Table 1. Summary of geospatial datasets used in the spatial water yield (WY) workflow and their corresponding stand model variable. Spatial resolutions, pixel units, and data sources are also included.

Stand WY Model Variable	Geospatial Dataset	Spatial Resolution	Pixel Units	Source
LAI	Pine Basal Area	5 m	ft2/acre	[33]
DTW	Depth to Water Table	30 m	feet, msl	[40]
Aridity Index	Potential Evapotranspiration	500 m	mm/yr	[41]
	Mean Annual Precipitation	4 km	mm/yr	[42]

, were compiled to develop water yield estimates for the Apalachicola Region: (1) pine species basal area, (2) depth to water table, (3) annual potential evapotranspiration, and (4) annual precipitation. All geospatial inputs required pre-processing steps that were conducted in Esri’s ArcGIS Pro version 2.9.1 to a gridded TIFF format, with a uniform study area extent, spatial resolution of 10 m, and a WGS 1984 UTM Zone 16N coordinate system. Additional pre-processing specific to each dataset is discussed below.

2.3.1. Pine Basal Area

A pine-specific basal area raster was developed using 2018 United States Geological Survey 3D Elevation Program (3DEP) lidar and 246 forested field plots, also collected during the spring and summer of 2018. Multiple height and relative density variables derived from the lidar point cloud and field plots inform a general linear model of estimated basal area. The specific modeling methodology used is detailed in [33]. The model was applied over the entire study area. The resulting raster output units are pine basal area in square feet per acre, which were converted to square meters per hectare, with a spatial resolution of 5 m, shown in Figure 2. The general linear model of pine basal area has a RMSE value of 24.05 and an adjusted R² value of 0.68. The pine basal area model’s relationship is not asymptotic and does not saturate like some reflectance-based remotely sensed datasets.

2.3.2. Depth to Water Table

A continuous depth to water table (DTW) raster for the entire state was obtained from Florida Geological Survey (FGS) [43]. This dataset was created for the FGS Floridan Aquifer Vulnerability Assessment project [40]. DTW, shown in Figure 3, is a categorical variable in the stand water yield model; therefore, we reclassified this as a binary raster where 0 equals a shallow water table (<5 m) and 1 is representative of a deep water table (≥5 m). DTW was tested as a continuous variable in other stand model iterations, but water table depths across field sites showed a bi-modal distribution and, thus, performed better as a binary variable [5].

2.3.3. Aridity Index

A regional aridity index, the ratio of PET to precipitation, was created using the PRISM Climate Group mean annual precipitation and Moderate Resolution Imaging Spectroradiometer (MODIS) PET gridded datasets [41,42]. These datasets were selected because they have the highest spatial and temporal resolution for publicly available datasets that cover the full extent of the study area. Data were compiled from 2013 to 2020 for both inputs, as these years show recent, short-term variation in hydroclimate for the study area. In addition to standard pre-processing, raw PET rasters were converted using the appropriate scale factor (0.1) and gaps were filled using a 5 × 5 focal statistic. Both precipitation and PET underwent multiple rounds of downscaling and smoothing before aridity index (PET/precipitation) was calculated for each year and mosaicked to a multiband dataset. One year, 2016, was removed from the stack due to projection and scale factor issues with the MODIS PET dataset.

The average value for each pixel in the multiband dataset was saved to an individual grid, shown in Figure 4a as Mean ARID, with a range of 0.84 to 1.45. This mean aridity index was used as the default aridity index input when calculating WY in the timber thinning scenarios. Additionally, to simulate the shift in climate to more drought-like conditions that are projected for the southeastern U.S., the highest observed value for each pixel across the seven years included in the development of this dataset was saved to a grid, shown in Figure 4b as Max ARID with a range of 0.97 to 1.79. This maximum aridity index was used as an example climate scenario to estimate potential water yield under projected drought conditions using aridity values already experienced in the study area.

A constant value for aridity was tested to create a simplified workflow and reduce data acquisition and processing. However, due to the size of the study area, variation in aridity index extended beyond one standard deviation noted in [5] and, therefore, would adversely affect final water yield estimates. We elected to use the spatially explicit aridity indexes as input to the spatial water yield estimation workflow.

2.4. Spatial Water Yield Estimation Workflow

The stand water yield model uses LAI as the forest attribute most directly correlated with forest ET and water yield in southern pine forests [5]. Therefore, the first step in our water yield estimation workflow was to convert the existing pine basal area raster to LAI. The pine basal area raster was converted from square feet per acre to square meters per hectare, then we used Equation 2 which shows the model provided in Cohen et al., 2018 [17], to estimate LAI from pine basal area (R² = 0.65).

LAI = 0.073 BAm + 0.2761,

(2)

where LAI is leaf area index and BAm is basal area in square meters per hectare. Recorded BA from [5] field plots ranged from 0 to 52 m²ha⁻¹ (0 to 225 ft²ac⁻¹), and LAI did not exceed 5 m²m⁻², which was consistent with the modeled values of the Pine BA dataset used in this study. In contrast to previous studies [8], saturation of LAI with increased basal area was not observed in the development of this relationship [17].

After calculating LAI, a baseline water yield estimate was produced by applying the general linear model from [5] shown in Equation (1) spatially across the study area. The output of this workflow (Figure 5) is an estimate of how much water (in centimeters per year) is predicted to return to the hydrologic system, across the study area at 10 m spatial resolution. We refer to this baseline output as the ‘current’ water yield, which reflects existing water yield conditions before any management activities are conducted.

2.5. Timber Thinning and Climate Scenarios

To estimate the effect of timber thinning on water yield, we reduced pine basal area in three different thinning scenarios to a maximum value of 7, 11, and 18 m²ha⁻¹ (30, 50, 80 ft²ac⁻¹, respectively). These basal area values were selected because they represent the low, middle, and high ends of excellent-condition southern open pine ecosystems [44]. A spatial LAI raster was calculated for each timber thinning scenario and applied to the stand water yield model. Separate water yield estimates were created from the two aridity indexes to emulate current climate conditions in the study area, as well as climate changes to drier conditions.

Finally, the difference between water yield estimates from each timber thinning scenario and the ‘current’ water yield estimate, referred to as potential water yield gain (PWYG), were exported. These PWYG rasters are useful to land managers because they help identify where timber thinning activities would be most impactful and quantify how much water yield could be associated with particular forest management strategies.

2.6. Workflow Automation Tools

We automated the workflow steps by creating a toolset in Esri’s ArcGIS Pro ModelBuilder (version 2.9.1). These tools utilize the input geospatial data to apply the stand water yield model and output estimated water yields for baseline conditions and the various timber thinning scenarios. The included tools are (1) Basal Area to LAI, (2) Current Water Yield Calculation, and (3) Scenario Water Yield Calculation. The tools were designed to be informed with a user’s own input datasets, i.e., study area-specific LAI, DTW, and aridity index. Additional information on how to download and use the toolbox, as well as data input requirements, can be found in Appendix A.

We also developed an open source set of water yield tools contained within a Python script (Appendix B). This script relies on the NumPy and GDAL Python libraries to spatially apply the stand water yield model from user-supplied raster inputs [5].

3. Results

Processing time using the ArcGIS Pro tools end-to-end in one PWYG scenario was approximately 10 min for the entire study area, and approximately one hour for the seven scenarios included in this analysis. Processing time using the Python workflow, which has fewer preprocessing steps than the ArcGIS Pro workflow, was 3 min to complete and save all three thinning scenario rasters. There were very slight differences in the outputs from the two methods due to a known round up error from the NumPy library used in the Python workflow. Though this difference was not significant, we chose to use the outputs from the ArcGIS Pro tool for the analysis.

The water yield outputs generated in this analysis consist of seven rasters representing the ‘current’ WY using the mean aridity raster, the WY expected from the three thinning scenarios (also using the mean aridity index), and then the WY expected from the three thinning scenarios under maximum aridity conditions. Each output was 1.34 GB covering the 1.9 million hectare study area at 10 m spatial resolution with pixel values in centimeters per year. These seven outputs are detailed, with summary pixel value statistics, in

Table 2. Summary of output spatial datasets produced for timber thinning and climate scenarios, with variation in inputs for ARID. The mean (and range) summary statistics are from individual pixel values in units of cm per year.

Output	Description	Pixel WY
wy_current	Baseline; ‘current’ BA conditions; Mean ARID	52.8 (−22.2–96.0)
wy_18_mean	BA reduced to 18 m2ha−1; Mean ARID	53.3 (−11.5–96.0)
wy_11_mean	BA reduced to 11 m2ha−1; Mean ARID	54.2 (−9.7–96.0)
wy_7_mean	BA reduced to 7 m2ha−1; Mean ARID	55.2 (−9.2–96.0)
wy_ 18_max	BA reduced to 18 m2ha−1; Maximum ARID	16.9 (−63.6–76.6)
wy_ 11_max	BA reduced to 11 m2ha−1; Maximum ARID	17.9 (−58.6–76.6)
wy_7_max	BA reduced to 7 m2ha−1; Maximum ARID	18.9 (−55.8–76.6)

. The range of pixel WY values decreases across timber reduction scenarios with the removal of upper LAI values. Inversely, mean WY values increase with the reduction in BA and LAI. The decrease in mean values for scenarios with the maximum ARID input underscores the effect departures from average aridity can have on regional water yields.

The ‘current’ water yield result informed with mean aridity (Figure 4a), binary depth to water (Figure 3), and 2018 pine basal area (Figure 2) is shown in Figure 6a. This ‘current’ water yield raster was used as the baseline to compare against the three timber thinning and climate scenarios resulting in six PWYG rasters. The PWYG raster showing the difference between wy_7_mean, the highest thinning scenario to 7ba, and wy_current is shown in Figure 6b.

PWYG datasets were summarized over two meaningful sub-regional boundaries commonly used by local land managers for further prioritization: USGS Watershed Boundary Dataset (WBD) hydrologic units at the 12-digit subwatershed level (HUC12) and Apalachicola National Forest (ANF) forest management compartment boundaries, shown in Figure 7 and

Table 3. Summary of HUC12s and ANF compartments with the highest PWYG from three timber thinning scenarios to maximum BA values of 18, 11, and 7 m²ha⁻¹, total area and area of dense pine stands (BA ≥ 18 m²ha⁻¹) in hectares, and annual WY in millions of cubic meters per year (Mm³ yr⁻¹). Proportion of area that exceeds each timber thinning threshold included (%) next to annual WY.

	Hectares		WY (Mm3 yr−1)
	Total	Dense Pine	Current	18	11	7
HUC12
Cat Creek	10,740	4040	49.16	50.86 (38%)	53.45 (58%)	55.39 (70%)
Central Lake Talquin	9041	2108	36.78	37.61 (23%)	39.22 (47%)	40.57 (60%)
Cerser Swamp	8514	2487	54.76	55.94 (29%)	57.57 (48%)	58.85 (59%)
Eagle Nest Bayou-Wetappo Creek Frontal	17,983	3156	120.24	121.54 (18%)	123.81 (33%)	125.76 (44%)
Fisher Creek	9219	2806	35.32	36.31 (30%)	38.70 (72%)	40.81 (88%)
Florida River	12,866	2513	76.72	78.02 (20%)	79.64 (31%)	80.94 (41%)
Harrison Swamp-Intercoastal Waterway Frontal	19,641	3267	121.92	123.23 (17%)	125.49 (30%)	127.37 (38%)
Kennedy Creek	13,368	2877	76.89	78.07 (22%)	80.28 (46%)	82.33 63%)
Lake Munson	11,061	2337	41.93	42.70 (21%)	44.70 (51%)	46.48 (63%)
Little Owl Creek	12,183	2468	69.65	70.75 (20%)	72.66 (45%)	74.62 (69%)
ANF Compartment
27	889	548	4.52	4.87 (62%)	5.21 (85%)	5.43 (94%)
68	1839	539	10.20	10.49 (29%)	10.88 (55%)	11.23 (79%)
73	1536	591	8.42	8.77 (39%)	9.19 (67%)	9.51 (78%)
106	1083	466	6.01	6.24 (43%)	6.58 (77%)	6.84 (88%)
205	1676	792	5.75	6.16 (47%)	6.75 (89%)	7.20 (97%)
206	2308	709	10.08	10.41 (31%)	11.06 (73%)	11.57 (84%)
217	1637	436	6.54	6.70 (27%)	7.20 (80%)	7.60 (93%)
248	1307	416	4.84	5.03 (32%)	5.41 (75%)	5.71 (88%)
328	1699	377	7.49	7.68 (22%)	8.06 (63%)	8.42 (86%)

. Subwatershed boundaries were used as means of addressing watershed health across property ownership boundaries during restoration planning, a standard strategy employed by land managing agencies. ANF compartments were used to prioritize results based on boundaries regularly used by the USDA Forest Service to manage forest resources. Summarized water yields are reported as water volume in cubic meters per year, a 1:1 conversion of pixel values (cm/yr) due to the 10 m spatial resolution of the outputs.

Ten subwatersheds appeared near the top for PWYG in all three scenarios. These subwatersheds are shown in Figure 7a. Cat Creek, a subwatershed that is wholly contained within Tate’s Hell State Forest, had the highest PWYG in every timber thinning scenario from the analysis. With approximately 38% of its area exhibiting a dense pine BA of 18 m²ha⁻¹ or higher, Cat Creek presents the greatest opportunity to increase WY via BA/LAI reduction in the study area. The results of the analysis in this subwatershed show pine tree thinning to a basal area of 18 m²ha⁻¹ could return an estimated 1.7 million cubic meters of water to the hydrologic system annually in an average aridity scenario, whereas heavier thinning to 7 m²ha⁻¹ could return 6.2 million cubic meters of water annually in the same average aridity.

On the ANF, nine compartments were identified among the top PWYG across all three timber thinning scenarios and are shown in Figure 7b. Similar to Cat Creek, compartment 27 has the largest percentage of its area (62%) with a dense pine BA (≥18 m²ha⁻¹) and a PWYG of over 910,000 m³ in a BA reduction to 7 m²ha⁻¹; however, compartment 206 offers the greatest total PWYG of all compartments under the same thinning scenario of nearly 1.5 million cubic meters annually.

Table 3 shows a detailed breakdown of the ten HUC12 watersheds and nine timber-management compartments that the analysis identified as having the highest PWYG from timber thinning operations. Pine tree LAI modeled from pine basal area is a primary component of the water yield model, so the area within a HUC12 watershed or forest compartment that contains pine basal area above the high thinning scenario of 18 m²ha⁻¹ is also detailed in Table 3.

PWYG for each of the three timber thinning scenarios, using the highest aridity value experienced as a drought simulation, was calculated to show the effect of the aridity extreme on water yield and how science-based timber thinning decision-making could mitigate these impacts. The results were analyzed at the ANF compartment level using the heaviest thinning scenario—7 m²ha⁻¹ for discussion below. Patterns in PWYG across the three thinning levels were similar at each reduction level when summarized by compartment. Western compartments adjacent to the Apalachicola River experiencing the lowest amount of water loss when aridity values are high, less than 2 million m³yr⁻¹ compared to eastern compartments that could experience water losses greater than 13 million m³ annually.

4. Discussion

4.1. Applications of WY in Forest Management

The predictive planning workflow and codified tools presented in this study provide a solution to estimate potential water yield impacts resulting from changes in vegetative cover prior to the implementation of proposed forest management practices (e.g., timber thinning, planting, and growth). This spatial workflow employs established, peer-reviewed, and field-verified models [5,33] that, in turn, permit the user to predict potential water yield changes under various scenarios across the landscape at multiple spatial scales. Specifically, it produces region-wide outputs that can be applied effectively to practical, project-level extents within user-defined boundaries, or in the case of this study, to forest management compartments, hydrologic units, and specific forest stands where silvicultural prescriptions and land management decisions are routinely made. This spatial scale and extent are a unique aspect of these results not found in previous water yield models of this area [45]. The extent and resolution of these results allows a single pine water yield assessment to inform multiple land management decisions across multiple scales, extents, and boundaries, which will influence short and long-term management strategies and potentially help identify partnering opportunities on adjacent public and/or private lands. As well, this type of assessment process accommodates other management strategies that could reduce total LAI and PET (e.g., prescribed fire, mechanical, or invasive species treatments) allowing for greater water yield while also improving habitat conditions for threatened and endangered species within a priority management compartment.

Essentially, the results of this spatial water yield analysis are meant to be integrated into a decision support framework with other administrative, ecological, hydrologic, and infrastructure datasets including, but not limited to, roads, trails, bridges, culverts, low water crossings, wildlife management areas, wilderness locations, recreation sites, and high-resolution lidar-derived hydrographic features. This spatial workflow can also be applied as a practical complement to other water yield models used within a decision support system, in part because the spatial resolution is comparatively finer than other hydrologic models [46,47] which may not provide high-resolution outputs due to their inherently coarse spatial scales. Region-wide water modeling can also be somewhat complex, cumbersome, and time-consuming, requiring numerous input variables that might be dependent on streamgages and water measuring devices that may or may not exist [48]. This is particularly the case in the lower Apalachicola Region, which can essentially be considered a “data desert” due to the significant lack of streamgages or other water flow measuring devices [49]. Further, the stand model’s focus on southern pine forests, that are common regional management targets, also fills an important knowledge gap that water yield models using overall land surface LAI or ET cannot fill [50]. Finally, the workflows created for the water yield process are flexible, free, and open source, thus allowing it to be easily transferred, reproduced, and updated as new water yield models are developed.

A simple example of how this assessment can work within a spatial decision support framework to inform management decisions alongside other spatial datasets is shown in Figure 8. Using a GIS software system, Figure 8 displays a detailed view of the water yield results within forest compartment number 68 on the Apalachicola NF. Compartment 68’s boundary exists entirely within the Little Owl Creek subwatershed, a HUC12 that repeatedly appears near the top of every PWYG scenario and is also near the top of each PWYG scenario for compartments. Water yield increases from timber thinning scenarios within compartment 68 range from roughly 290,000 m³ to nearly 710,000 m³, depending on the level of thinning conducted in the 29% of area above the dense pine threshold, see Table 3. The inclusion of geospatial data from other Forest Service projects included in Figure 8, such as road infrastructure condition surveys from 2017, enhanced hydrologic flow lines derived from 2018 USGS 3DEP lidar, and known locations of federally listed threatened and endangered species, can assist land managers in developing a comprehensive restoration plan that maximizes resources to help address interconnected and collocated problems within the compartment.

Spatial water yield estimates such as this can also be used to predict the impacts of timber thinning, or even regional disturbances such as hurricanes and their impacts on infrastructure potentially due to long-term increases in water yield or changes to hydrology. As the impacts of climate change increase, these types of assessments will be invaluable for updating infrastructure requirements and identifying specific at-risk features. To use Figure 8 as an example, a land manager could utilize this information to determine whether thinning operations in the western and southern portion of the compartment may require road infrastructure improvements for the four poor condition stream crossings, red points, that are along the southern border of the compartment to prevent further deterioration or catastrophic failure at those sites. A combination of factors such as the heavy usage of the roadway for logging operations and increasingly wet conditions in that area enhance the likelihood of continued damage to the already poor condition stream-road crossings.

In determining restoration priorities at the HUC12 level, three potentially high-yielding watersheds present in Table 3, Cerser Swamp, Eagle Nest Bayou, and Harrison Swamp, are directly adjacent to Mexico Beach, FL, where Hurricane Michael, a category 5 hurricane, made landfall on 10 October 2018. This storm led to the significant loss of basal area in the study area [39]. Analysis in these watersheds did not account for this loss of pine basal area from Michael, due to the remotely sensed data being collected before the storm. The water yield estimate presented here should be updated using pine basal area rasters developed from remotely sensed data obtained post-Hurricane Michael to reflect the existing water yield and allow for a comparison of pre- and post-hurricane water yield. Post-hurricane lidar was flown in Summer 2020 and the final product was recently delivered to USGS and made available to the public. The tools and workflows presented in this study and those developed earlier [33] will allow us to easily create updated water yield analyses as new data become available. It is important to note that these thinning scenarios will merely provide temporary decreases in LAI. Persistent increases in water yield will only be accomplished through sustained maintenance activities, such as ongoing prescribed fire and intermittent thinning.

The workflow presented in this study incorporates important spatial variability in factors such as depth to water and aridity, both of which can be held as constant at the stand level. It is critical to incorporate this spatial variability when extrapolating to the landscape level, as can be seen in Figure 4 and Figure 6. These figures highlight the effect higher aridity values can have when estimating water yield, as shown by the lower yields in the eastern portion of the study area by Figure 6a, but also how they can help prioritize forest management activities in areas where timber removal would have the most impact, as with Figure 6b.

This spatial variability becomes even more important if land managers are interested in adapting management plans with longer-term restoration goals. Our results demonstrate that climate, and potential climate change, can play an overwhelmingly important role in the water yield balance. Within the climate scenario, using the highest-observed values in the study area for aridity index to simulate drought-like conditions, potential water yield gain, or loss, can be estimated to further prioritize restoration efforts where the most change is likely to occur. Figure 9 shows that an aggressive timber thinning plan on a public land tract as large as the Apalachicola National Forest would not be able to prevent reductions in water yield due to increases in aridity. However, by preparing for future conditions and reducing forest LAI in suitable locations, land managers could lessen the impact of this water yield decline by millions of cubic meters of water each year. Monitoring changes in aridity and its effect on water yield, both spatially and temporally, can be captured with annual collection of PET and precipitation datasets and the generation of updated water yield outputs.

4.2. Model Limitations and Future Research

While pine tree LAI was the strongest predictor of water yield in the stand model [5], the workflow presented here used a high spatial resolution pine basal area dataset and converted it to LAI using the conversion function in [17]. This was undertaken as basal area is a commonly used forest metric collected by forester managers and has been modeled at high spatial resolutions using remotely sensed products and field plots [33,51,52,53]. Pine basal area is a dataset that forest managers are more likely to have available at high spatial resolution as compared to LAI. Therefore, the inclusion of pine basal area in this workflow will likely facilitate the adoption and use of this workflow. However, a potential improvement to this water yield estimate would be to estimate LAI directly from lidar point clouds, instead of converting from basal area. Using LAI directly from lidar point clouds would require a water yield model that includes all tree species, or a method of extracting only the pine tree species LAI from the lidar point cloud. A more general in situ model might add uncertainty to the water yield estimates as hardwood-dominated stands could have a significantly different ET rate than pine-dominated stands. Making this change would require further research on water yield from different species, species groups, and stand compositions.

To further validate the results of this study’s pine water yield estimates, a complete water yield model that includes all landscape LAI, as well as surface and groundwater dynamics, would be required. By modeling only water yield from pine tree species, in areas where pine trees are the dominant source of LAI, model output estimates will closely match the water yield outputs from a model using total vegetation LAI. However, in areas with large hardwood encroachment or high understory LAI, the pine water yield estimate will account for much less of the total water yield.

Important data points for a comprehensive water yield model would include long-term streamgages that measure discharge and water levels, which this study area has considerable lack of, most noticeably on smaller streams and tributaries [49]. Improving the monitoring network by increasing streamgage sites would help to validate the presented spatial methodology. Furthermore, streamgage data, when analyzed with groundwater monitoring well information, could help determine where the increased water yield is going within a holistic landscape water yield model. Currently, no assumptions are made as to the destination of estimated increases in water volume. This study only quantifies WY lost to pine ET and does not establish groundwater recharge into the surficial or Floridan aquifers, or discharge into any streams, rivers, or bays within the study area.

5. Conclusions

Owing to climate and land use changes, freshwater is an at-risk resource which will worsen as water demand increases, particularly where climate projections show hotter and drier conditions [5,8]. An integrative and predictive water yield method that incorporates recognized, peer-reviewed, and field-verified models to produce multi-scale high-resolution water yield data offers land managers a valuable operational management tool to help prioritize terrestrial and hydrologic restoration projects. Selecting and prioritizing restoration activities in suitable locations on forested landscapes can efficiently and effectively satisfy multiple management objectives while directly and indirectly impacting water availability.

The pine water yield model used in this study was developed for southern pine ecosystems in Florida and is likely applicable to other pine-dominated ecosystems across the Southeastern Coastal Plain. Acharya et al., 2022, found that water yield differences in dominant pine species were not significant at the stand level, meaning that the model and the tools developed for this study could be used to predict water yield in many areas where southern pine trees are dominant [5]. The largest barrier for using this workflow in other southern forests would be the creation of a pine basal area raster such as the one developed for this study. However, our prior work has shown that these types of raster datasets can be created with operational accuracy from U.S. Forest Service, Forest Inventory and Analysis data which are available across the U.S., often in combination with local lidar data [51], airborne imagery [52], or satellite imagery [53] at various spatial resolutions.

The tools we document in the appendices offer a transferable workflow that can be easily reproduced as new remote sensing data become available. This method can also be modified with new coefficient values if water yield models, like Equation (1), need to be adjusted for dissimilar species groups or different study areas. The ability to spatially quantify water yield and associate the results with other land management activities for integration into a decision support system, independent of ownership boundaries, provides public land managers and private forest landowners the opportunity to make spatially informed decisions to help improve and protect the natural resources, and particularly the freshwater sources of the Apalachicola Region of Florida.